Cloud Native serializations of publicly available fire data from USGS and Calfire. Calfire data includes controlled burn treatments. Data through 2022
This bucket contains four kinds of wildfire layer. Fire perimeters record where fire has burned. Burn severity records how severely it burned, per 30 metre pixel, for every large fire since 1984. Incident records record how fires were fought — suppression strategy, cost, personnel and losses. Wildfire Hazard Potential is a modelled surface describing where fire is likely to burn and how intensely — not an observation, and not a map of risk to people or property. Modelled hazard and observed outcome can diverge; every collection here carries the resolution 8 H3 cell, so comparing them is a join.
Contents:
calfire-2025/firep, prior calfire-2024/firep)calfire-2025/rxburn, prior calfire-2024/rxburn)usgs-fires-2021/combined)whp-2023-*)mtbs-perimeters-1984-2024)mtbs-severity-1984-2024-*)fired-events-2001-2021)ics-209-plus-1999-2020/wf-incidents). Coverage stops at 2020 — always LEFT join to FPA-FOD.The perimeter datasets are polygon geometries processed into GeoParquet, PMTiles and H3 hex Parquet (resolution 10). ICS-209-PLUS is point geometry (incident point of origin) in the same three formats at resolution 10, matching FPA-FOD so both meet at h8, with six non-spatial companion tables joined on INCIDENT_ID. Wildfire Hazard Potential is raster, published as a WGS84 COG plus H3 hex Parquet (resolution 9 classified, 8 continuous). MTBS burn severity is raster, published as one WGS84 COG per year plus two H3 hex Parquet products at resolution 10 — dominant class and per-class fractional coverage — partitioned by year.
The 2025 CAL FIRE release (FRAP fire251gdb) is the current dataset for California fires and prescribed burns; the 2024 release is retained as a prior version.
calfire-2025/firep) — currentSource: CAL FIRE FRAP GIS (fire251gdb, layer firep25_1)
Features: 23,334 fire perimeter polygons
Coverage: California, 1878–2025
License: CC-BY-4.0
Source layer: firep (PMTiles source-layer name)
Hex aggregation note: in
calfire-2025/firep/hex/h0=*/data_0.parquet, every source attribute (includingGIS_ACRES) is repeated on each H3 cell the fire covers. NeverSUM(GIS_ACRES)on hex data. To total burned acreage, deduplicate by feature first:SELECT SUM(GIS_ACRES) FROM (SELECT DISTINCT _cng_fid, GIS_ACRES FROM read_parquet('…/firep/hex/h0=*/data_0.parquet')).
calfire-2025/rxburn) — currentSource: CAL FIRE FRAP GIS (fire251gdb, layer rxburn25_1)
Features: 11,975 prescribed burn polygons
Coverage: California, 1900–2025
License: CC-BY-4.0
Source layer: rxburn (PMTiles source-layer name)
calfire-2024/firep) — prior releaseSource: CAL FIRE FRAP GIS
Features: 22,810 fire perimeter polygons
Coverage: California, 1878–2024
Source layer: firep (PMTiles source-layer name)
calfire-2024/rxburn)Source: CAL FIRE FRAP GIS
Features: 10,675 prescribed burn polygons
Coverage: California, 1980s–2024
Source layer: rxburn (PMTiles source-layer name)
usgs-fires-2021/combined)Source: USGS ScienceBase
Features: 135,061 fire perimeter polygons
Coverage: Continental US, Alaska, Hawaii, US territories, 1878–2021
Source layer: combined (PMTiles source-layer name)
This dataset combines records from dozens of federal, state, and local data sources into a single de-duplicated national inventory. Where multiple sources cover the same fire, the highest-quality polygon is selected and source attributes are preserved in list fields.
whp-2023-*) — modelled hazard, not perimetersFour collections, because the source classifies each geographic domain against its own
distribution of the continuous index. The class breaks are different numbers in CONUS and
Alaska — a CONUS "Very High" is index > 1,985, an Alaska "Very High" is > 8,912, 4.5× apart —
so a pooled GROUP BY class across domains is meaningless arithmetic.
Classes: 1=Very Low, 2=Low, 3=Moderate, 4=High, 5=Very High, 6=Non-burnable, 7=Water. Classes 6 and 7 are the absence of burnable fuel, not hazard levels.
Hazard is not risk. WHP describes how likely and intense fire is given fuels, topography and weather. It says nothing about what is exposed to loss. A remote wilderness ridge can be Very High with nothing at stake; a subdivision can be Moderate and still top a risk ranking.
For any CONUS-vs-Alaska comparison use the continuous collections — the raw index means the same thing everywhere.
⚠️ Area statistics belong on the hex asset, not the COG. The COG is geographic and therefore not equal-area — counting its pixels is not an area measurement. The source Albers grid is the area truth, and the hex reproduces its class shares to within 0.6 percentage points.
mtbs-*) — the observational recordThree collections from the Monitoring Trends in Burn Severity program, a joint USGS and USDA Forest Service effort that has mapped every large fire in the United States since 1984.
Severity classes: 1=Unburned to Low, 2=Low, 3=Moderate, 4=High, 5=Increased Greenness, 6=Non-Processing Area Mask. Classes 5 and 6 are not severity levels — 5 is increased post-fire greenness, 6 is area excluded from mapping — so leave them out of any "share at high severity" denominator.
Not a fire census. MTBS maps only fires over 1,000 acres in the West and 500 acres in the East. Small fires are absent by design and they are most ignitions.
Reburns are signal, not duplication. The same ground burns repeatedly over four decades, so "total acres burned" has to say whether it means unique ground or fire-years.
Use hex-fractions, not hex, for shares. The mode asset keeps only each cell's dominant
class; a cell that is 40% high severity and 60% moderate reads as entirely moderate.
⚠️ Missing years. Six source mosaics could not be retrieved from the publisher, so severity is absent for CONUS 2004, 2017 and Alaska 1987, 1995, 2001, 2013. A query filtering on one of those years returns no rows, which means the severity mapping is unavailable — not that nothing burned. Perimeters cover all 41 years.
fired-events-2001-2021)Source: CU Scholar deposit — Earth Lab / CU Boulder, produced with firedpy from MODIS MCD64A1
Features: 278,569 event polygons, 666,046 km² burned
Coverage: CONUS + Alaska, ignition dates 2000-11-02 to last-burn 2021-05-01
License: CC-BY-4.0
Source layer: fired-events-2001-2021 (PMTiles source-layer name)
H3: native resolution 9, parents 8 and 0
firedpy groups MCD64A1 burned pixels that fall within the same space-time window (5 pixels, 11 days)
into one fire event. Unlike the perimeter layers here, every event carries how fast it spread —
fsr_km2_dy, fsr_px_dy — and its maximum, minimum and mean single-day growth (m[xnu]_grw_km2).
h8 joins it to mtbs-* and whp-2023-*.
This is a different census from MTBS, not a better one: FIRED has no size threshold (median event 0.43 km²) but inherits MCD64A1's blind spots — sub-pixel fires, dense canopy, cloud, and fires between monthly composites. Event counts between the two are not comparable.
Hex aggregation note: in
fired-events-2001-2021/hex/h0=*/data_0.parquetevery per-event total (tot_ar_km2,tot_pix,tot_perim,event_dur,fsr_*,m[xnu]_grw_*) is repeated on each of the 6,266,383 (event, cell) rows. NeverSUMorAVGthese on hex data. Deduplicate first:SELECT SUM(tot_ar_km2) FROM (SELECT DISTINCT id, tot_ar_km2 FROM read_parquet('…/fired-events-2001-2021/hex/h0=*/data_0.parquet')).
⚠️ ig_utm_x / ig_utm_y are not UTM and not lon/lat. They are the source MODIS sinusoidal
metres; only the geometry was reprojected to EPSG:4326. Use ST_Centroid(geom) for a lon/lat
ignition point.
⚠️ eco_mode and eco_name are computed independently and do not pair 1:1 (105 codes, 105
names, 294 observed combinations). Filter on one or the other, not both. Likewise lc_code and
lc_mode share the IGBP value domain but disagree on 4.3% of events; lc_name/lc_desc label
lc_mode.
CAL FIRE datasets:
California Department of Forestry and Fire Protection (CAL FIRE), Forest and Range Assessment Program (FRAP). Fire Perimeter and Prescribed Burn data, 2025 release (fire251gdb; 2024 release also retained). https://frap.fire.ca.gov/mapping/gis-data/
USGS dataset: Welty, J.L., and Jeffries, M.I., 2021, Combined wildfire datasets for the United States and certain territories, 1878-2019: U.S. Geological Survey data release, https://doi.org/10.5066/P9ZXGFY3
Wildfire Hazard Potential: Dillon, G.K. 2023. Wildfire Hazard Potential for the United States (270-m), version 2023, 4th Edition. Fort Collins, CO: Forest Service Research Data Archive. https://doi.org/10.2737/RDS-2015-0047-4
MTBS datasets: Eidenshink, J., Schwind, B., Brewer, K., Zhu, Z.-L., Quayle, B., and Howard, S. 2007. A project for monitoring trends in burn severity. Fire Ecology 3(1): 3-21. Picotte, J.J., et al. 2020. Changes to the Monitoring Trends in Burn Severity program mapping production procedures and data products. Fire Ecology 16:16. https://doi.org/10.1186/s42408-020-00076-y. Data accessed 2026-08-21: Burned Areas Boundaries, version 12.0, April 2025; Thematic Burn Severity Mosaics, version 9.0 (August 2024) for 1984-2022 and version 12.0 (April 2025) for 2023-2024. https://www.mtbs.gov/
FIRED: Balch, J.K.; St. Denis, L.A.; Mahood, A.L.; Mietkiewicz, N.P.; Williams, T.M.; McGlinchy, J.; Cook, M.C. FIRED (Fire Events Delineation): An Open, Flexible Algorithm and Database of US Fire Events Derived from the MODIS Burned Area Product (2001-2019). Remote Sens. 2020, 12, 3498. https://doi.org/10.3390/rs12213498
| string |
| Complex fire name (if applicable) |
| COMPLEX_ID | string | Complex fire ID |
| OBJECTIVE | int16 | Management objective code |
| GIS_ACRES | float32 | GIS-calculated area in acres |
| COMMENTS | string | Additional comments |
| FIRE_NUM | string | Historical fire number |
| Shape_Length | float64 | Perimeter length in meters |
| Shape_Area | float64 | Area in square meters |
| int16 |
| Fuel consumption class code |
| PRE_CON_CLASS | int16 | Pre-treatment condition class code |
| POST_CON_CLASS | int16 | Post-treatment condition class code |
| Shape_Length | float64 | Perimeter length in meters |
| Shape_Area | float64 | Area in square meters |
| All fire codes from contributing sources |
| Listed_Fire_IDs | string | All fire IDs from contributing sources |
| Listed_Fire_IRWIN_IDs | string | All IRWIN IDs from contributing sources |
| Listed_Fire_Dates | string | All fire dates from contributing sources |
| Listed_Fire_Causes | string | All fire causes from contributing sources |
| Listed_Fire_Cause_Class | string | Cause classification |
| Listed_Rx_Reported_Acres | string | Reported Rx acres |
| Listed_Map_Digitize_Methods | string | Digitization methods |
| Listed_Notes | string | Source dataset notes |
| Processing_Notes | string | USGS processing and QA notes |
| Wildfire_Notice | string | Data quality notice for wildfire attribution |
| Prescribed_Burn_Notice | string | Data quality notice for prescribed burn attribution |
| Wildfire_and_Rx_Flag | string | Wildfire/Rx overlap flag |
| Overlap_Within_1_or_2_Flag | string | Temporal overlap flag |
| Circleness_Scale | float64 | Shape circularity (1.0 = circle); low values may be artifacts |
| Circle_Flag | int16 | Suspiciously circular geometry flag |
| Exclude_From_Summary_Rasters | string | Whether to exclude from summary rasters |
| Shape_Length | float64 | Perimeter length in meters |