Disturbance - The Wildland Almanac

What was compared

Disturbance benchmarking figure: multi-panel comparison of Wildland Almanac against reference data. — Benchmarking panels for disturbance, from WildlandAlmanac_CA_QualityBenchmarks.pdf (v2026.1). Panel letters below correspond to the panels above.

a Biomass remaining after disturbance, WA against FIA field plots, by cause. Good agreement after fire; most insect, disease and drought loss is missed.

b WA biomass loss against independent MTBS fire severity. Loss climbs steeply, from 7 Mg/ha in the lightest class to 92 in the heaviest.

c Where WA and Hansen both see forest loss, pooled to 1 km. WA finds 66% of Hansen's loss, and 96% of shared detections agree on the year within one.

d The same against MTBS fire. WA finds 92% of high-severity fire and 81% of moderate, and dates 92% of it within one year.

Evidence

Strengths and weaknesses on each axis, as measured. Weaknesses are stated at the same level of detail as strengths.

Axis	Strengths	Weaknesses
External skill	FIA field: WA tracks how much biomass each agent removes, with fire capture 0.66 [a]. Loss rises monotonically with independent MTBS severity, 7 to 92 Mg/ha [b]. Severe fire drives post-fire AGB toward zero (10% below 25 Mg/ha).	Non-fire magnitude is soft: mortality capture 0.15, harvest 0.28 - diffuse and partial disturbance is under-registered [a]. The detection floor misses small and low-severity events.
Internal coherence	Closure holds: large AGB drops coincide with the disturbance flag, and only 2% of pixel-years are big drops with no flag. WA correctly identifies fire as the dominant agent.	That 2% is diffuse mortality the abrupt-change detector misses. The layer targets abrupt loss, not gradual decline - by design, but it bounds what it can be used for.
Cross-dataset consistency	At 1 km, WA maps and dates fire with no bias (median on the 1:1): high-severity recall 92%, Jaccard 0.58, 92% within ±1 yr against MTBS [d]. Hansen agrees on stand replacement (Jaccard 0.53, 96% within ±1 yr) [c].	MTBS and Hansen share Landsat lineage with WA, so this is consistency rather than proof - though they are independently produced, which makes agreement on where and when meaningful. Per-cell fraction density scatters (R² 0.55) where WA biomass and MTBS dNBR draw intensity differently.

Synthesis.
High confidence Fire location, timing and severity

Lower confidence Non-fire-agent magnitude and small-event detection

Specifics

WA = Disturbance, 1986-2024. References: Hansen/UMD Global Forest Change v1.12 (annual loss 2001-2024); MTBS thematic severity 1984-2024; FIA repeat-visit plots (FIADB California, 2024 release).