{ "stac_version": "1.0.0", "stac_extensions": [ "https://stac-extensions.github.io/table/v1.2.0/schema.json", "https://stac-extensions.github.io/raster/v1.1.0/schema.json", "https://stac-extensions.github.io/classification/v2.0.0/schema.json", "https://stac-extensions.github.io/scientific/v1.0.0/schema.json" ], "type": "Collection", "id": "inhabit-v4-2024", "title": "INHABIT v4.0 (2024) — invasive-plant habitat suitability, 12 fire-relevant species (CONUS)", "description": "Habitat-suitability surfaces for 12 fire-cycle-relevant invasive plants from the USGS Invasive Species Habitat Tool (INHABIT) v4.0, June 2024 — the ten exotic annual grasses and forbs that build continuous fine fuels, plus the two riparian woody invaders (tamarisk, Russian olive). Twelve of the release's 259 species; extending to the full release is separate work, not a silent expansion.\n\n⚠️ **This is modelled POTENTIAL habitat, not observed occurrence.** A high-suitability pixel is a statement about where a species could establish or become abundant given the environment, not a record that it is present. Every result from this collection reads \"suitable habitat for X\", never \"X is present\". The claim it supports is about invasion pressure and vulnerability, not realized invasion.\n\n**Products.** Three continuous weighted-ensemble surfaces per species — suitability for occurrence, for abundance (≥5% cover) and for high abundance (≥25% cover) — each a relative 0–100 index (`mean` reducer; aggregate with AVG, never SUM), plus a categorical `integrated-binary` map that thresholds and combines all three (`mode` reducer), at three percentile thresholds. Nine products per species, 108 layers.\n\n**Every product ships as a COG for every species.** Where a layer is also hexed, the hexed product set is the canonical one — the three `-masked` continuous surfaces plus `integrated-binary-fifth` — because those are the defaults every tabulation is meant to read. The unmasked continuous surfaces and the `first`/`tenth` thresholds are COG-only **by design, not by omission**: they exist so the masking and threshold choices stay inspectable at full resolution, and hexing them would only serve a sensitivity analysis no current work reports. Adding any missing hex later is additive and needs no rebuild.\n\n**All 12 species are hexed to H3** (48 hex assets), so an `h8` join resolves for every species in this collection — but only for 4 of the 9 products: `occurrence-masked`, `abundance-masked`, `high-abundance-masked`, `integrated-binary-fifth`. Every other product is published as a COG only, and an `h8` join against one returns nothing. That is scope, not breakage: those products serve sensitivity analysis, the COGs are complete and final, and adding their hex later is additive and needs no rebuild.\n\n**`-masked` is the default for inventoried-roadless-area work.** The `-masked` variants suppress suitability outside the MESS training envelope (multivariate environmental similarity surface, Elith et al. 2010 — at least one predictor outside its training range). Roadless country is disproportionately high-elevation, undersampled terrain, exactly where unrestricted extrapolation is least trustworthy. The restriction is substantial and varies about 3× across these species: it retains 95% of cheatgrass occurrence pixels but only 34% of buffelgrass abundance pixels, so for the most-affected species the masked-vs-plain choice moves an area total more than the threshold choice does. Per-species retention is tabulated in the README and on each asset.\n\n**`fifth` (0.05) is the canonical threshold**, named up front so a later analysis cannot pick whichever threshold supports its conclusion. `first` (0.01, inclusive) and `tenth` (0.10, targeted) are the sensitivity band.\n\n⚠️ **Class `-1` in the integrated-binary maps is a class, not fill.** It is the MESS extrapolation flag; NoData is `-128`. It reaches 48% of a raster (red brome), so a sign test that treats negatives as fill silently deletes half the map. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold.\n\n⚠️ **Road-bias audit: it fires, and it splits the species.** gHM (Global Human Modification — built substantially from roads, and in this catalog already as `global-human-modification`) is a predictor in all 12 models. Restricted to the occurrence models, which is what a distance-to-road gradient leans on, its share of predictor importance runs from 2.0% for cheatgrass (rank 6/23) to 25.0% for jointed goatgrass (rank 1/24). So a road-distance gradient is defensible for cheatgrass, medusahead (3.1%) and red brome (4.0%), and substantially circular — the model restating its own predictor — for jointed goatgrass, buffelgrass (22.4%), Russian olive (17.6%), tamarisk (17.1%) and Russian thistle (12.9%). This does not undermine the suitability surfaces: human modification is a legitimate predictor of invasion. It undermines only the inference \"suitability rises near roads, therefore roads drive invasion\" for the high-gHM species. Per-species figures are on every asset; method and per-model-group breakdown in `catalog/invasives/BUILD.md`.\n\n⚠️ **CONUS only — the largest roadless block is not covered.** INHABIT v4 excludes Alaska, where the Tongass and Chugach are the single largest roadless acreage in the National Forest System. Every national statement drawn from this collection is a CONUS statement and must be reported against the rule-affected base acreage rather than as a national total. INHABIT Global V1 (doi:10.5066/P13AJ46S) could close the gap later but is a different grid and different models, not pixel-comparable with v4 CONUS.\n\n**Grid.** Source is Albers Conical Equal Area (NAD83/GRS 1980) at 98.4693338923368 m — the FGDC `absres`, which the papers round to \"90 m\"; all 108 source rasters share one 44319×30671 grid. Warped to EPSG:4326 at 0.001096100359° (57745×25711), nearest-neighbour for the class rasters and bilinear for the continuous, with `-srcnodata` set so NoData cannot bleed into data. H3 native resolution 10 (≈1.55 source pixels per cell) with parents 9, 8 and 0; `h8` is the catalog's universal join key.\n\n**v3.0 (doi:10.5066/P9V54H5K) is not comparable and is deliberately not held.** v4 changed the ensemble from model agreement to continuous relative suitability, so v3 and v4 pixel values do not mean the same thing.\n\nSpecies: Bromus tectorum (cheatgrass, ScienceBase 66a14c62d34ec831f2c2b1cc); Bromus rubens (red brome, ScienceBase 66a14559d34ef32deb7ee1a2); Bromus japonicus (Japanese brome, ScienceBase 66a14239d34ef32deb7ede9c); Bromus arvensis (field brome, ScienceBase 669fd2e7d34eef99d5abc5eb); Taeniatherum caput-medusae (medusahead, ScienceBase 66bb7a6ad34e033882814282); Ventenata dubia (ventenata, ScienceBase 66c3adebd34e03388287af28); Aegilops cylindrica (jointed goatgrass, ScienceBase 669ee0dcd34eef99d5abbb3b); Agropyron cristatum (crested wheatgrass, ScienceBase 669ee8e0d34eef99d5abbb50); Salsola tragus (Russian thistle, ScienceBase 66c3c389d34e03388287bba9); Tamarix chinensis/ramosissima (tamarisk, ScienceBase 66bb7eccd34e033882814498); Elaeagnus angustifolia (Russian olive, ScienceBase 66a68b88d34ea6469c870807); Cenchrus ciliaris (buffelgrass, ScienceBase 66a29430d34ec831f2c2d2d7).\n\n[Published here: 108 COGs, all grid-verified, and 48 H3 hex layers. The remaining layers are published as COGs only; any later hex is additive, needs no rebuild, and is tracked outside this collection.]", "license": "public-domain", "keywords": [ "invasive species", "habitat suitability", "species distribution model", "cheatgrass", "annual grass", "fine fuels", "wildfire", "roadless", "tamarisk", "INHABIT", "USGS", "CONUS", "H3" ], "extent": { "spatial": { "bbox": [ [ -128.386308874497, 23.093342014672, -65.084314604497, 51.268044444672 ] ] }, "temporal": { "interval": [ [ "1980-01-01T00:00:00Z", "2024-06-30T00:00:00Z" ] ] } }, "sci:doi": "10.5066/P14HNEJF", "sci:citation": "Jarnevich, C.S., Engelstad, P., Williams, D.A., Shadwell, K.S., Reimer, C.J., Henderson, G.C., Prevey, J.S., and Pearse, I.S., 2024, INHABIT species potential distribution across the contiguous United States (ver. 4.0, June 2024): U.S. Geological Survey data release, https://doi.org/10.5066/P14HNEJF. Method: Jarnevich et al. 2024, NeoBiota 96:261-278, https://doi.org/10.3897/neobiota.96.134842.", "sci:publications": [ { "doi": "10.3897/neobiota.96.134842", "citation": "Jarnevich, C.S., et al. 2024. Predicted occurrence and abundance habitat suitability of invasive plants in the contiguous United States: updates for the INHABIT web tool. NeoBiota 96:261-278." }, { "doi": "10.1371/journal.pone.0263056", "citation": "Engelstad, P., et al. 2022. INHABIT: A web-based decision support tool for invasive plant species habitat visualization and assessment across the contiguous United States. PLoS ONE 17(2):e0263056." } ], "providers": [ { "name": "U.S. Geological Survey, Fort Collins Science Center", "roles": [ "producer", "licensor" ], "url": "https://doi.org/10.5066/P14HNEJF" }, { "name": "Boettiger Lab (cng-datasets COG + H3 processing)", "roles": [ "processor" ], "url": "https://s3-west.nrp-nautilus.io/public-invasives" } ], "links": [ { "rel": "self", "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/stac-collection.json", "type": "application/json" }, { "rel": "root", "href": "https://s3-west.nrp-nautilus.io/public-data/stac/catalog.json", "type": "application/json" }, { "rel": "parent", "href": "https://s3-west.nrp-nautilus.io/public-invasives/stac-collection.json", "type": "application/json" }, { "rel": "license", "href": "https://doi.org/10.5066/P14HNEJF", "type": "text/html" }, { "rel": "cite-as", "href": "https://doi.org/10.5066/P14HNEJF", "type": "text/html" }, { "rel": "describedby", "href": "https://s3-west.nrp-nautilus.io/public-invasives/README.md", "type": "text/markdown" } ], "assets": { "bromus-tectorum-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus tectorum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 95.1% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–100; 55.10% of the grid is data (610,364,110 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "cheatgrass (Bromus tectorum) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Bromus tectorum (cheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 95.1% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-tectorum-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus tectorum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 95.1% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–97; 55.06% of the grid is data (610,829,800 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "cheatgrass (Bromus tectorum) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Bromus tectorum (cheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 95.1% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-tectorum-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus tectorum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 95.1% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–95; 55.06% of the grid is data (610,823,170 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "cheatgrass (Bromus tectorum) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Bromus tectorum (cheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 95.1% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-tectorum-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Bromus tectorum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Bromus tectorum (cheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=38,329,816, 0=447,915,693, 1=90,394,262, 2=17,513,279, 3=193,202,096 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-tectorum-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "cheatgrass (Bromus tectorum) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Bromus tectorum (cheatgrass) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=38,329,816, 0=447,915,693, 1=90,394,262, 2=17,513,279, 3=193,202,096. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-tectorum-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus tectorum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 95.1% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,011,531 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus tectorum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 95.1% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–97; 57.92% of the grid is data (572,022,149 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus tectorum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 95.1% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus tectorum (cheatgrass), 0–100. Source-measured range 0–95; 57.92% of the grid is data (572,015,519 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-tectorum-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Bromus tectorum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Bromus tectorum (cheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=38,162,079, 0=344,705,305, 1=88,630,388, 2=38,874,620, 3=276,982,754 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-tectorum-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_tectorum/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "cheatgrass (Bromus tectorum) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Bromus tectorum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 2.0% of predictor importance (rank 6/23) in this species' OCCURRENCE models and 1.2% (rank 12/25) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: target. This species has NO KDE-background occurrence model; its only occurrence model uses the bias-mitigated target-background design.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Bromus tectorum (cheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=38,368,578, 0=542,781,474, 1=40,955,455, 2=12,856,456, 3=152,393,183 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-rubens-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus rubens, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 51.5% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus rubens (red brome), 0–100. Source-measured range 0–100; 29.82% of the grid is data (953,958,369 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "red brome (Bromus rubens) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Bromus rubens (red brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 51.5% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-rubens-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus rubens, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 79.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus rubens (red brome), 0–100. Source-measured range 0–100; 46.07% of the grid is data (733,049,807 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "red brome (Bromus rubens) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Bromus rubens (red brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 79.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-rubens-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus rubens, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 79.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus rubens (red brome), 0–100. Source-measured range 0–98; 46.12% of the grid is data (732,428,158 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "red brome (Bromus rubens) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Bromus rubens (red brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 79.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-rubens-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Bromus rubens, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Bromus rubens (red brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=381,452,798, 0=313,995,157, 1=5,890,663, 2=4,930,957, 3=81,085,571 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-rubens-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "red brome (Bromus rubens) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Bromus rubens (red brome) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=381,452,798, 0=313,995,157, 1=5,890,663, 2=4,930,957, 3=81,085,571. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-rubens-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus rubens, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 51.5% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus rubens (red brome), 0–100. Source-measured range 0–100; 57.91% of the grid is data (572,119,232 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus rubens, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 79.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus rubens (red brome), 0–100. Source-measured range 0–100; 57.87% of the grid is data (572,633,467 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus rubens, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 79.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus rubens (red brome), 0–100. Source-measured range 0–98; 57.92% of the grid is data (572,011,818 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-rubens-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Bromus rubens, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Bromus rubens (red brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=379,204,213, 0=212,161,826, 1=32,780,417, 2=21,171,702, 3=142,036,988 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-rubens-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_rubens/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "red brome (Bromus rubens) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Bromus rubens, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 4.0% of predictor importance (rank 8/22) in this species' OCCURRENCE models and 2.2% (rank 8/24) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Bromus rubens (red brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=381,821,716, 0=344,355,843, 1=4,615,836, 2=1,656,264, 3=54,905,487 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-japonicus-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus japonicus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 97.2% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–100; 56.28% of the grid is data (594,326,675 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Japanese brome (Bromus japonicus) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Bromus japonicus (Japanese brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 97.2% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-japonicus-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus japonicus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 98.2% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–98; 56.88% of the grid is data (586,171,002 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Japanese brome (Bromus japonicus) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Bromus japonicus (Japanese brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 98.2% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-japonicus-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus japonicus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 98.2% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–100; 56.87% of the grid is data (586,229,565 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Japanese brome (Bromus japonicus) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Bromus japonicus (Japanese brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 98.2% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-japonicus-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Bromus japonicus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Bromus japonicus (Japanese brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=20,464,051, 0=347,793,096, 1=112,292,409, 2=37,289,263, 3=269,516,327 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-japonicus-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Japanese brome (Bromus japonicus) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Bromus japonicus (Japanese brome) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=20,464,051, 0=347,793,096, 1=112,292,409, 2=37,289,263, 3=269,516,327. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-japonicus-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus japonicus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 97.2% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,036,711 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus japonicus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 98.2% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–98; 57.92% of the grid is data (571,985,778 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus japonicus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 98.2% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus japonicus (Japanese brome), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,044,341 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-japonicus-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Bromus japonicus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Bromus japonicus (Japanese brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=17,229,753, 0=143,772,823, 1=143,606,461, 2=32,280,362, 3=450,465,747 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-japonicus-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_japonicus/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Japanese brome (Bromus japonicus) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Bromus japonicus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.3% of predictor importance (rank 6/26) in this species' OCCURRENCE models and 3.5% (rank 8/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Bromus japonicus (Japanese brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=21,505,760, 0=465,296,628, 1=75,072,376, 2=35,477,322, 3=190,003,060 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-arvensis-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus arvensis, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 94.3% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus arvensis (field brome), 0–100. Source-measured range 0–97; 54.60% of the grid is data (617,175,743 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "field brome (Bromus arvensis) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Bromus arvensis (field brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 94.3% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-arvensis-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus arvensis, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 97.3% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus arvensis (field brome), 0–100. Source-measured range 0–100; 56.38% of the grid is data (592,992,127 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "field brome (Bromus arvensis) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Bromus arvensis (field brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 97.3% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-arvensis-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus arvensis, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 97.3% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus arvensis (field brome), 0–100. Source-measured range 0–100; 56.19% of the grid is data (595,471,464 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "field brome (Bromus arvensis) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Bromus arvensis (field brome), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 97.3% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-arvensis-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Bromus arvensis, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Bromus arvensis (field brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=33,469,349, 0=252,321,824, 1=201,569,792, 2=36,012,431, 3=263,981,750 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-arvensis-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "field brome (Bromus arvensis) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Bromus arvensis (field brome) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=33,469,349, 0=252,321,824, 1=201,569,792, 2=36,012,431, 3=263,981,750. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "bromus-arvensis-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Bromus arvensis, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 94.3% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Bromus arvensis (field brome), 0–100. Source-measured range 0–97; 57.88% of the grid is data (572,504,658 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Bromus arvensis, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 97.3% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Bromus arvensis (field brome), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,033,452 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Bromus arvensis, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 97.3% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Bromus arvensis (field brome), 0–100. Source-measured range 0–100; 57.73% of the grid is data (574,614,274 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "bromus-arvensis-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Bromus arvensis, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Bromus arvensis (field brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=25,872,331, 0=85,916,747, 1=257,904,639, 2=90,010,151, 3=327,651,278 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "bromus-arvensis-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/bromus_arvensis/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "field brome (Bromus arvensis) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Bromus arvensis, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 8.0% of predictor importance (rank 4/27) in this species' OCCURRENCE models and 5.0% (rank 7/29) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Bromus arvensis (field brome). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=38,627,780, 0=417,898,847, 1=108,454,968, 2=20,077,847, 3=202,295,704 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "taeniatherum-caput-medusae-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Taeniatherum caput-medusae, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 64.9% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 37.57% of the grid is data (848,579,085 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "medusahead (Taeniatherum caput-medusae) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Taeniatherum caput-medusae (medusahead), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 64.9% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "taeniatherum-caput-medusae-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Taeniatherum caput-medusae, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 69.2% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 40.09% of the grid is data (814,395,278 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "medusahead (Taeniatherum caput-medusae) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Taeniatherum caput-medusae (medusahead), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 69.2% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "taeniatherum-caput-medusae-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Taeniatherum caput-medusae, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 69.2% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 40.08% of the grid is data (814,470,961 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "medusahead (Taeniatherum caput-medusae) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Taeniatherum caput-medusae (medusahead), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 69.2% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "taeniatherum-caput-medusae-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Taeniatherum caput-medusae, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Taeniatherum caput-medusae (medusahead). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=275,370,039, 0=414,748,207, 1=32,490,899, 2=4,715,140, 3=60,030,861 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "taeniatherum-caput-medusae-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "medusahead (Taeniatherum caput-medusae) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Taeniatherum caput-medusae (medusahead) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=275,370,039, 0=414,748,207, 1=32,490,899, 2=4,715,140, 3=60,030,861. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "taeniatherum-caput-medusae-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Taeniatherum caput-medusae, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 64.9% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 57.90% of the grid is data (572,262,488 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Taeniatherum caput-medusae, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 69.2% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 57.91% of the grid is data (572,074,157 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Taeniatherum caput-medusae, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 69.2% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Taeniatherum caput-medusae (medusahead), 0–100. Source-measured range 0–96; 57.91% of the grid is data (572,149,840 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "taeniatherum-caput-medusae-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Taeniatherum caput-medusae, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Taeniatherum caput-medusae (medusahead). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=273,231,187, 0=253,335,365, 1=111,889,078, 2=26,082,421, 3=122,817,095 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "taeniatherum-caput-medusae-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/taeniatherum_caput_medusae/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "medusahead (Taeniatherum caput-medusae) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Taeniatherum caput-medusae, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 3.1% of predictor importance (rank 10/26) in this species' OCCURRENCE models and 3.4% (rank 7/26) pooled over all model groups, so a distance-to-road gradient on this species is DEFENSIBLE. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Taeniatherum caput-medusae (medusahead). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=276,106,026, 0=458,422,871, 1=16,155,144, 2=4,252,486, 3=32,418,619 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "ventenata-dubia-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Ventenata dubia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 78.3% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Ventenata dubia (ventenata), 0–100. Source-measured range 0–98; 45.34% of the grid is data (743,026,619 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "ventenata (Ventenata dubia) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Ventenata dubia (ventenata), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 78.3% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "ventenata-dubia-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Ventenata dubia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 78.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Ventenata dubia (ventenata), 0–100. Source-measured range 0–100; 45.54% of the grid is data (740,307,254 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "ventenata (Ventenata dubia) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Ventenata dubia (ventenata), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 78.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "ventenata-dubia-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Ventenata dubia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 78.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Ventenata dubia (ventenata), 0–100. Source-measured range 0–100; 45.34% of the grid is data (742,932,272 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "ventenata (Ventenata dubia) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Ventenata dubia (ventenata), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 78.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "ventenata-dubia-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Ventenata dubia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Ventenata dubia (ventenata). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=170,230,431, 0=517,134,089, 1=12,756,265, 2=33,168,539, 3=54,065,822 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "ventenata-dubia-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "ventenata (Ventenata dubia) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Ventenata dubia (ventenata) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=170,230,431, 0=517,134,089, 1=12,756,265, 2=33,168,539, 3=54,065,822. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "ventenata-dubia-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Ventenata dubia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 78.3% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Ventenata dubia (ventenata), 0–100. Source-measured range 0–98; 57.91% of the grid is data (572,084,571 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Ventenata dubia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 78.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Ventenata dubia (ventenata), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,011,817 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Ventenata dubia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 78.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Ventenata dubia (ventenata), 0–100. Source-measured range 0–100; 57.73% of the grid is data (574,636,835 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "ventenata-dubia-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Ventenata dubia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Ventenata dubia (ventenata). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=164,454,089, 0=311,417,696, 1=28,771,755, 2=60,495,190, 3=222,216,416 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "ventenata-dubia-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/ventenata_dubia/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "ventenata (Ventenata dubia) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Ventenata dubia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 6.0% of predictor importance (rank 7/26) in this species' OCCURRENCE models and 5.7% (rank 6/26) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Ventenata dubia (ventenata). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=170,676,706, 0=552,374,781, 1=8,571,330, 2=23,908,563, 3=31,823,766 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "aegilops-cylindrica-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Aegilops cylindrica, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 92.4% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–96; 53.51% of the grid is data (631,891,343 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "jointed goatgrass (Aegilops cylindrica) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Aegilops cylindrica (jointed goatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 92.4% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "aegilops-cylindrica-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Aegilops cylindrica, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 100.0% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–98; 57.92% of the grid is data (572,022,148 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "jointed goatgrass (Aegilops cylindrica) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Aegilops cylindrica (jointed goatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 100.0% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "aegilops-cylindrica-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Aegilops cylindrica, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 96.4% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–99; 55.81% of the grid is data (600,621,171 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "jointed goatgrass (Aegilops cylindrica) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Aegilops cylindrica (jointed goatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 96.4% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "aegilops-cylindrica-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Aegilops cylindrica, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Aegilops cylindrica (jointed goatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=57,930,517, 0=391,847,400, 1=197,680,156, 2=88,100,038, 3=51,797,035 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "aegilops-cylindrica-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "jointed goatgrass (Aegilops cylindrica) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Aegilops cylindrica (jointed goatgrass) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=57,930,517, 0=391,847,400, 1=197,680,156, 2=88,100,038, 3=51,797,035. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "aegilops-cylindrica-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Aegilops cylindrica, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 92.4% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–96; 57.89% of the grid is data (572,375,402 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Aegilops cylindrica, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 100.0% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–98; 57.92% of the grid is data (572,022,148 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Aegilops cylindrica, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 96.4% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Aegilops cylindrica (jointed goatgrass), 0–100. Source-measured range 0–99; 57.92% of the grid is data (571,983,808 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "aegilops-cylindrica-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Aegilops cylindrica, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Aegilops cylindrica (jointed goatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=52,758,462, 0=198,766,120, 1=141,252,797, 2=147,593,475, 3=246,984,292 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "aegilops-cylindrica-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/aegilops_cylindrica/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "jointed goatgrass (Aegilops cylindrica) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Aegilops cylindrica, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 25.0% of predictor importance (rank 1/24) in this species' OCCURRENCE models and 22.0% (rank 1/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Aegilops cylindrica (jointed goatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=59,665,571, 0=525,424,175, 1=163,308,672, 2=19,806,699, 3=19,150,029 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "agropyron-cristatum-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Agropyron cristatum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 87.1% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–94; 50.36% of the grid is data (674,755,138 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "crested wheatgrass (Agropyron cristatum) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Agropyron cristatum (crested wheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 87.1% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "agropyron-cristatum-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Agropyron cristatum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 87.7% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–100; 50.74% of the grid is data (669,647,177 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "crested wheatgrass (Agropyron cristatum) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Agropyron cristatum (crested wheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 87.7% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "agropyron-cristatum-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Agropyron cristatum, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 87.4% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–100; 50.57% of the grid is data (671,917,758 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "crested wheatgrass (Agropyron cristatum) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Agropyron cristatum (crested wheatgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 87.4% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "agropyron-cristatum-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Agropyron cristatum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Agropyron cristatum (crested wheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=96,230,905, 0=423,953,925, 1=47,791,502, 2=12,663,660, 3=206,715,154 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "agropyron-cristatum-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "crested wheatgrass (Agropyron cristatum) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Agropyron cristatum (crested wheatgrass) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=96,230,905, 0=423,953,925, 1=47,791,502, 2=12,663,660, 3=206,715,154. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "agropyron-cristatum-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Agropyron cristatum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 87.1% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–94; 57.85% of the grid is data (572,926,803 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Agropyron cristatum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 87.7% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–100; 57.83% of the grid is data (573,266,541 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Agropyron cristatum, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 87.4% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Agropyron cristatum (crested wheatgrass), 0–100. Source-measured range 0–100; 57.87% of the grid is data (572,635,534 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "agropyron-cristatum-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Agropyron cristatum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Agropyron cristatum (crested wheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=91,517,791, 0=317,913,435, 1=53,749,864, 2=12,182,329, 3=311,991,727 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "agropyron-cristatum-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/agropyron_cristatum/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "crested wheatgrass (Agropyron cristatum) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Agropyron cristatum, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 7.2% of predictor importance (rank 3/26) in this species' OCCURRENCE models and 5.0% (rank 7/28) pooled over all model groups, so treat a distance-to-road gradient on this species with CAUTION. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Agropyron cristatum (crested wheatgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=97,592,240, 0=464,986,343, 1=46,690,837, 2=7,808,189, 3=170,277,537 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "salsola-tragus-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Salsola tragus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 93.9% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 54.33% of the grid is data (620,802,029 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian thistle (Salsola tragus) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Salsola tragus (Russian thistle), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 93.9% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "salsola-tragus-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Salsola tragus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 90.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 52.41% of the grid is data (646,950,564 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian thistle (Salsola tragus) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Salsola tragus (Russian thistle), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 90.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "salsola-tragus-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Salsola tragus, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 92.0% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 53.28% of the grid is data (635,134,483 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian thistle (Salsola tragus) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Salsola tragus (Russian thistle), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 92.0% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "salsola-tragus-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Salsola tragus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Salsola tragus (Russian thistle). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=48,071,746, 0=475,417,747, 1=64,089,227, 2=14,042,797, 3=185,733,629 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "salsola-tragus-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian thistle (Salsola tragus) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Salsola tragus (Russian thistle) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=48,071,746, 0=475,417,747, 1=64,089,227, 2=14,042,797, 3=185,733,629. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "salsola-tragus-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Salsola tragus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 93.9% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 57.85% of the grid is data (572,906,060 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Salsola tragus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 90.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 57.87% of the grid is data (572,633,152 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Salsola tragus, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 92.0% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Salsola tragus (Russian thistle), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,011,801 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "salsola-tragus-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Salsola tragus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Salsola tragus (Russian thistle). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=47,928,959, 0=261,892,067, 1=186,182,389, 2=10,142,461, 3=281,209,270 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "salsola-tragus-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/salsola_tragus/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian thistle (Salsola tragus) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Salsola tragus, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 12.9% of predictor importance (rank 2/22) in this species' OCCURRENCE models and 6.1% (rank 3/25) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Salsola tragus (Russian thistle). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=48,237,521, 0=556,087,727, 1=38,588,589, 2=29,372,044, 3=115,069,265 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "tamarix-chinensis-ramosissima-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Tamarix chinensis/ramosissima, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 92.8% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–95; 53.72% of the grid is data (629,062,123 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "tamarisk (Tamarix chinensis/ramosissima) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Tamarix chinensis/ramosissima (tamarisk), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 92.8% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "tamarix-chinensis-ramosissima-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Tamarix chinensis/ramosissima, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 93.9% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–94; 54.36% of the grid is data (620,441,186 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "tamarisk (Tamarix chinensis/ramosissima) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Tamarix chinensis/ramosissima (tamarisk), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 93.9% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "tamarix-chinensis-ramosissima-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Tamarix chinensis/ramosissima, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 93.9% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–95; 54.36% of the grid is data (620,450,650 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "tamarisk (Tamarix chinensis/ramosissima) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Tamarix chinensis/ramosissima (tamarisk), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 93.9% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "tamarix-chinensis-ramosissima-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Tamarix chinensis/ramosissima, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Tamarix chinensis/ramosissima (tamarisk). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=56,885,483, 0=463,812,346, 1=41,950,795, 2=19,341,806, 3=205,364,716 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "tamarix-chinensis-ramosissima-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "tamarisk (Tamarix chinensis/ramosissima) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Tamarix chinensis/ramosissima (tamarisk) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=56,885,483, 0=463,812,346, 1=41,950,795, 2=19,341,806, 3=205,364,716. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "tamarix-chinensis-ramosissima-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Tamarix chinensis/ramosissima, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 92.8% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–95; 57.89% of the grid is data (572,365,418 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Tamarix chinensis/ramosissima, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 93.9% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–94; 57.90% of the grid is data (572,320,988 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Tamarix chinensis/ramosissima, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 93.9% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Tamarix chinensis/ramosissima (tamarisk), 0–100. Source-measured range 0–95; 57.90% of the grid is data (572,320,988 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "tamarix-chinensis-ramosissima-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Tamarix chinensis/ramosissima, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Tamarix chinensis/ramosissima (tamarisk). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=56,513,928, 0=284,935,006, 1=128,560,337, 2=15,354,208, 3=301,991,667 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "tamarix-chinensis-ramosissima-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/tamarix_chinensis_ramosissima/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "tamarisk (Tamarix chinensis/ramosissima) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Tamarix chinensis/ramosissima, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.1% of predictor importance (rank 3/25) in this species' OCCURRENCE models and 16.4% (rank 2/26) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Tamarix chinensis/ramosissima (tamarisk). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=56,954,439, 0=527,659,640, 1=23,843,267, 2=24,841,334, 3=154,056,466 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "elaeagnus-angustifolia-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Elaeagnus angustifolia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 96.1% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–100; 56.56% of the grid is data (590,458,997 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian olive (Elaeagnus angustifolia) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Elaeagnus angustifolia (Russian olive), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 96.1% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "elaeagnus-angustifolia-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Elaeagnus angustifolia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 95.7% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–96; 55.43% of the grid is data (605,823,945 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian olive (Elaeagnus angustifolia) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Elaeagnus angustifolia (Russian olive), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 95.7% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "elaeagnus-angustifolia-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Elaeagnus angustifolia, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 95.8% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–100; 55.46% of the grid is data (605,480,675 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian olive (Elaeagnus angustifolia) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Elaeagnus angustifolia (Russian olive), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 95.8% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "elaeagnus-angustifolia-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Elaeagnus angustifolia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Elaeagnus angustifolia (Russian olive). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=30,624,791, 0=387,639,025, 1=65,382,996, 2=37,748,932, 3=265,959,402 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "elaeagnus-angustifolia-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "Russian olive (Elaeagnus angustifolia) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Elaeagnus angustifolia (Russian olive) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=30,624,791, 0=387,639,025, 1=65,382,996, 2=37,748,932, 3=265,959,402. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "elaeagnus-angustifolia-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Elaeagnus angustifolia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 96.1% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–100; 58.87% of the grid is data (559,136,520 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Elaeagnus angustifolia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 95.7% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–96; 57.89% of the grid is data (572,365,418 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Elaeagnus angustifolia, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 95.8% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Elaeagnus angustifolia (Russian olive), 0–100. Source-measured range 0–100; 57.92% of the grid is data (572,022,148 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "elaeagnus-angustifolia-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Elaeagnus angustifolia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Elaeagnus angustifolia (Russian olive). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=28,623,611, 0=191,150,286, 1=141,929,360, 2=48,756,183, 3=376,895,706 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "elaeagnus-angustifolia-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/elaeagnus_angustifolia/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "Russian olive (Elaeagnus angustifolia) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Elaeagnus angustifolia, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 17.6% of predictor importance (rank 3/27) in this species' OCCURRENCE models and 11.3% (rank 3/28) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Elaeagnus angustifolia (Russian olive). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=30,887,891, 0=505,164,272, 1=25,383,600, 2=60,291,440, 3=165,627,943 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "cenchrus-ciliaris-occurrence-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/occurrence-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — occurrence suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Cenchrus ciliaris, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 56.5% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–98; 32.74% of the grid is data (914,325,055 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-occurrence-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/occurrence-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "buffelgrass (Cenchrus ciliaris) — occurrence suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for occurrence of Cenchrus ciliaris (buffelgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 56.5% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "cenchrus-ciliaris-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — abundance (>=5% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Cenchrus ciliaris, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 33.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–98; 19.48% of the grid is data (1,094,460,644 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "buffelgrass (Cenchrus ciliaris) — abundance (>=5% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for abundance (>=5% cover) of Cenchrus ciliaris (buffelgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 33.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "cenchrus-ciliaris-high-abundance-masked-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/high-abundance-masked.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — high abundance (>=25% cover) suitability, MESS-restricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Cenchrus ciliaris, **restricted to the MESS training envelope**: pixels where any predictor falls outside its training range are set to NoData rather than extrapolated. This is the collection's default variant for inventoried-roadless-area work, because roadless country is disproportionately the high-elevation, undersampled terrain where unrestricted extrapolation is least trustworthy. **The restriction is not cosmetic here: it retains 33.6% of the plain product's data pixels for this species × model group** — quantified per species in the collection README, and for the most-affected species the masked/plain choice moves an area total more than the threshold choice does. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–97; 19.45% of the grid is data (1,094,929,958 px are NoData 255 — outside CONUS, or suppressed outside the MESS training envelope). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-high-abundance-masked-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/high-abundance-masked/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "buffelgrass (Cenchrus ciliaris) — high abundance (>=25% cover) suitability, MESS-restricted (H3 res 10)", "description": "H3 resolution-10 hex of MESS-restricted relative suitability for high abundance (>=25% cover) of Cenchrus ciliaris (buffelgrass), **reducer `mean`** — the area-weighted mean of the source pixels covering each cell. Suitability is a normalized 0–100 index, so `mean` is the correct reducer and **`SUM(suitability)` is meaningless**: aggregate across cells with AVG (or MAX for a hotspot), never SUM. Cells with no valid source pixel are absent rather than zero, and the missing cells are *unknown*, not *unsuitable*. This layer is sparse where the MESS restriction applies — for this species × model group the restricted product retains 33.6% of the plain product's data pixels. **Modelled potential habitat, not observed occurrence.** Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "cenchrus-ciliaris-integrated-binary-fifth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/integrated-binary-fifth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — integrated suitability class, fifth percentile (0.05) (COG)", "description": "Categorical integrated-binary map for Cenchrus ciliaris, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.05 percentile threshold — **the canonical threshold for this collection**. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the fifth-percentile (0.05) threshold for Cenchrus ciliaris (buffelgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=342,322,727, 0=407,783,717, 1=5,649,165, 2=4,905,953, 3=26,693,584 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "cenchrus-ciliaris-integrated-binary-fifth-hex": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/integrated-binary-fifth/hex/h0=*/data_0.parquet", "type": "application/x-parquet", "title": "buffelgrass (Cenchrus ciliaris) — integrated suitability class, fifth percentile (0.05) (H3 res 10)", "description": "H3 resolution-10 hex of the integrated-binary fifth-percentile (0.05) class map for Cenchrus ciliaris (buffelgrass) — **the canonical threshold for this collection** — **reducer `mode`**, so each cell takes the dominant class of the source pixels it covers. ⚠️ **`mode` keeps only the dominant class and discards the mix, so this asset cannot answer \"how much area is high-abundance-suitable\" without undercounting to plurality cells.** At ~1.55 source pixels per res-10 cell the loss is small, but it is not zero and it is one-directional: minority classes inside a mixed cell vanish. Per-class fractional coverage is not produced by this pipeline; an area-accounting use case needs a `fractions` build. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class (MESS extrapolation), not NoData** — filter it out of suitability accounting explicitly; a `>= 0` or `>= 1` predicate does that. Source class pixel counts: -1=342,322,727, 0=407,783,717, 1=5,649,165, 2=4,905,953, 3=26,693,584. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data" ], "h3:native_resolution": 10, "h3:parent_resolutions": [ 9, 8, 0 ], "table:columns": [ { "name": "suitability_class", "type": "double" }, { "name": "h10", "type": "uint64" }, { "name": "h9", "type": "uint64" }, { "name": "h8", "type": "uint64" }, { "name": "h0", "type": "uint64" } ] }, "cenchrus-ciliaris-occurrence-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/occurrence.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — occurrence suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for occurrence of Cenchrus ciliaris, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 56.5% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for occurrence of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–98; 57.89% of the grid is data (572,375,534 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — abundance (>=5% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for abundance (>=5% cover) of Cenchrus ciliaris, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 33.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for abundance (>=5% cover) of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–98; 57.92% of the grid is data (572,022,148 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-high-abundance-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/high-abundance.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — high abundance (>=25% cover) suitability, unrestricted (COG)", "description": "Continuous weighted-ensemble relative suitability for high abundance (>=25% cover) of Cenchrus ciliaris, **unrestricted** — suitability is predicted everywhere the model can be evaluated, including where predictors fall outside the range they were trained on. **Prefer the `-masked` companion for inventoried-roadless-area work**, which suppresses exactly that extrapolation; the masked variant retains 33.6% of this product's data pixels for this species × model group, and the difference is concentrated in the undersampled high-elevation terrain roadless areas occupy. Published so the masked/unmasked spread is auditable rather than assumed. Warped to EPSG:4326 with bilinear resampling (`-srcnodata 255`, so NoData is excluded from interpolation and cannot bleed into data); overviews AVERAGE. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability", "data_type": "uint8", "nodata": 255, "unit": "relative suitability index (0–100)", "description": "Relative habitat suitability for high abundance (>=25% cover) of Cenchrus ciliaris (buffelgrass), 0–100. Source-measured range 0–97; 57.89% of the grid is data (572,375,534 px are NoData 255 — outside CONUS). **Relative index, not a probability and not an amount.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." } ] }, "cenchrus-ciliaris-integrated-binary-first-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/integrated-binary-first.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — integrated suitability class, first percentile (0.01) (COG)", "description": "Categorical integrated-binary map for Cenchrus ciliaris, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.01 percentile threshold — the most inclusive (comprehensive) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the first-percentile (0.01) threshold for Cenchrus ciliaris (buffelgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=342,124,562, 0=358,486,945, 1=27,785,188, 2=5,951,447, 3=53,007,004 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] }, "cenchrus-ciliaris-integrated-binary-tenth-cog": { "href": "https://s3-west.nrp-nautilus.io/public-invasives/inhabit-v4-2024/cenchrus_ciliaris/integrated-binary-tenth.tif", "type": "image/tiff; application=geotiff; profile=cloud-optimized", "title": "buffelgrass (Cenchrus ciliaris) — integrated suitability class, tenth percentile (0.10) (COG)", "description": "Categorical integrated-binary map for Cenchrus ciliaris, combining the thresholded occurrence, abundance and high-abundance ensembles at the 0.10 percentile threshold — the most restrictive (targeted) of the three. The three thresholds convey a gradient from inclusive to targeted; `fifth` is canonical and `first`/`tenth` are its sensitivity band, so report the spread rather than choosing the threshold that suits a conclusion. Warped to EPSG:4326 with **nearest-neighbour** resampling (bilinear on class codes invents codes with no upstream referent); overviews NEAREST. Colours are ColorBrewer YlOrRd for ranks 1–3 with a desaturated purple for the extrapolation flag, chosen here — the source ships no palette. Road-bias audit: gHM (Global Human Modification, itself built substantially from roads) carries 22.4% of predictor importance (rank 1/23) in this species' OCCURRENCE models and 13.9% (rank 3/24) pooled over all model groups, so a distance-to-road gradient on this species is SUBSTANTIALLY CIRCULAR and should not be reported as evidence that roads drive invasion. It does not undermine the suitability surface itself -- human modification is a legitimate predictor of invasion. Occurrence background samples: KDE,target.", "roles": [ "data", "visual" ], "raster:bands": [ { "name": "suitability_class", "data_type": "int8", "nodata": -128, "description": "INHABIT integrated-binary class at the tenth-percentile (0.10) threshold for Cenchrus ciliaris (buffelgrass). Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. Warped nearest-neighbour from the source Albers grid, so the class set is exactly the source's: [-1, 0, 1, 2, 3]. Source pixel counts per class: -1=342,350,597, 0=424,367,304, 1=3,697,548, 2=2,694,763, 3=14,244,934 (57.92% of the grid is data; the rest is NoData -128 outside CONUS). Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "classification:classes": [ { "value": -1, "name": "Novel environmental conditions", "description": "MESS extrapolation flag: at least one predictor at this pixel is outside the range of values used to train the model, so no suitability rank is asserted here. NOT NoData (NoData is -128) and NOT a suitability level -- exclude it from suitability accounting rather than treating it as unsuitable.", "color_hint": "7B68A6" }, { "value": 0, "name": "Unsuitable", "description": "Below the fifth-percentile threshold for all three model groups.", "color_hint": "F2F2F2" }, { "value": 1, "name": "Occurrence suitable", "description": "At or above the fifth-percentile occurrence threshold, but below the abundance threshold.", "color_hint": "FED976" }, { "value": 2, "name": "Abundance suitable", "description": "At or above the fifth-percentile abundance (>=5% cover) threshold, but below the high abundance threshold.", "color_hint": "FD8D3C" }, { "value": 3, "name": "High abundance suitable", "description": "At or above the fifth-percentile high abundance (>=25% cover) threshold.", "color_hint": "BD0026" } ] } ] } }, "table:columns": [ { "name": "suitability", "type": "double", "description": "Area-weighted MEAN relative habitat suitability, 0-100, of the source pixels covering the cell (reducer=mean). The model group -- occurrence, abundance (>=5% cover) or high abundance (>=25% cover) -- is named in this asset's own `description`; one column name is used across all three so that the shared definition is not duplicated. **Relative index, not an amount: aggregate with AVG/MIN/MAX, never SUM.** Values are each model's prediction rescaled to 0-100 and ensembled by continuous Boyce index, so they are comparable BETWEEN model groups and species only as ranks within a map, not as absolute probabilities. Source NoData (255) is excluded before averaging. **Modelled potential habitat, not observed occurrence.** Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF." }, { "name": "h10", "type": "uint64", "description": "H3 cell ID at resolution 10 (native resolution; one row per cell)." }, { "name": "h9", "type": "uint64", "description": "H3 cell ID at resolution 9 (parent rollup)." }, { "name": "h8", "type": "uint64", "description": "H3 cell ID at resolution 8 (parent rollup; the catalog's universal join key -- roadless areas, wildfire hazard and land cover all carry h8)." }, { "name": "h0", "type": "uint64", "description": "H3 cell ID at resolution 0; hive partition key." }, { "name": "suitability_class", "type": "double", "description": "Dominant (mode) INHABIT integrated-binary class of the source pixels covering the cell. **Integer codes stored as double** (an artifact of the `mode` reducer), so compare with `= 3` or `>= 1` and cast if an exact integer join is needed. Values: -1=Novel environmental conditions, 0=Unsuitable, 1=Occurrence suitable, 2=Abundance suitable, 3=High abundance suitable. Ranks are EXCLUSIVE in the raster but NESTED in the source's own tabulations: occurrence-suitable ground is `suitability_class >= 1` (classes 1+2+3), abundance-suitable is `>= 2`, high-abundance-suitable is `= 3`. Counting class 1 alone undercounts occurrence-suitable ground by every cell that also clears a higher threshold. **-1 is a class, not fill** -- it flags novel environmental conditions (MESS extrapolation, Elith et al. 2010); source NoData is -128 and is excluded. **Categorical: never SUM or AVG this column** -- count cells per class. The `mode` reducer keeps only each cell's dominant class and discards the mix, so a per-class area total from this column undercounts to plurality cells; see this asset's `description`. Source: USGS INHABIT v4.0 (June 2024), doi:10.5066/P14HNEJF.", "values": [ -1, 0, 1, 2, 3 ] } ] }