Import: LANDFIRE CONUS disturbance history — annual disturbance 1999–present + Dist24 (pairs with MTBS)

[cboettig]

Dataset

LANDFIRE CONUS disturbance history — annual disturbance rasters, split out of #203. Designed to join with MTBS (Monitoring Trends in Burn Severity), the highest-priority fire-history add in #143.

⛔ SCOPE

Extent

CONUS only, full upstream extent. (AK / HI / PR-VI have their own annual-disturbance stacks — AKAnnualDisturbance_1999_present, HIAnnualDisturbance_2011_present, PRVIAnnualDisturbance_2015_present — deferred to the same AK/HI follow-up as #203.)

Products — verified URLs (HEAD-probed 2026-08-05)

Dataset name	Full product name	URL	Zip
annual-disturbance	LANDFIRE Annual Disturbance, 1999–present (CONUS)	https://landfire.gov/data-downloads/AnnualDist/USAnnualDisturbance_1999_present.zip	1.9 GB
dist-2024	LANDFIRE 2024 Annual Disturbance	https://landfire.gov/data-downloads/CONUS_Disturbance/LF2024_Dist24_CONUS.zip	97 MB

Also available, decide in/out before building:

• LF2024_PDist24_CONUS.zip — Preliminary Disturbance 2024
• LF2025_LDist25_CONUS.zip (42 MB) / LF2025_PDist25_CONUS.zip — Limited / Preliminary Disturbance 2025
• CONUSLimitedDisturbance_2023_Present.zip

Recommendation: publish the 1999–present annual stack as the primary product (it is the analytical layer — disturbance type, severity, and year per pixel) plus the current-year Dist24; treat Preliminary and Limited products as provisional and exclude them unless there's a specific need, since they'd otherwise silently double-count with the final release.

Structure — resolve BEFORE generating manifests

USAnnualDisturbance_1999_present is a multi-year product. Determine from the zip whether it ships as one multi-band raster, one raster per year, or an encoded type+severity+year composite, then pick the layout:

• Per-year rasters → follow the unified multi-year pattern already used for NLCD Annual (catalog/land-cover/k8s/ Ingest: NLCD Annual — land cover + land cover change (2015-2026) #453): one COG + one hex per year, consistent dataset naming (annual-disturbance/<year>).
• Single encoded raster → decode to explicit columns (disturbance type, severity, year) before hexing; a composite code is not queryable.

Record the answer in this issue before building. 27 years × (COG + 2 hex jobs) is a large commitment — confirm the layout first.

H3 + reducer

• Native resolution 10, --parent-resolutions 9,8,7,6,5,0 — matches Import: LANDFIRE CONUS vegetation & departure suite — EVT, BpS, FRG, SClass, VCC, VDep (LF2024/2020/2016) #203, Import: LANDFIRE 2024 CONUS vegetation structure & fuels — EVC, EVH, FBFM13/40, FVC/FVH/FVT, canopy CBH/CBD/CC/CH #515, and NLCD for direct h10 = h10 joins.
• Categorical (disturbance type / severity classes) → mode + fractions. fractions matters here: it answers "what share of this cell burned / was harvested", which mode cannot.
• Never sum/mean on class codes.

Join with MTBS

The point of this layer is cross-validating and extending burn history. Ensure the H3 grid and year encoding line up with whatever MTBS ingest lands (#143) — same native resolution, same year column name/type. If MTBS lands first, match it.

Naming

Full product names in all STAC titles/descriptions and classification:classes[].name (e.g. "LANDFIRE Annual Disturbance 1999–present (CONUS)"); abbreviations only in keys. Document every disturbance-type and severity code from the shipped CSV.

Bucket / license

public-landfire, license: public-domain, registered in the bucket-level STAC collection created in #203.

Acceptance criteria

• Layout question above answered in this issue before any cluster job runs.
• Annual stack (+ Dist24) on public-landfire, full CONUS, provisional products explicitly in or out.
• Per layer/year: COG + res-10 hex with mode and fractions.
• Complete classification:classes (type + severity) with color_hint; year column documented.
• Coverage gate passes (scripts/check-hex-coverage.sh); scripts/verify-stac.py --bucket public-landfire --dataset <each> exits 0.
• Joinability with MTBS (Catalog audit: Ecosystem processes (fire regimes, restoration potential) #143) verified at h10.

Split from #203. Part of #140; serves the fire-history gap in #143.