Coverage

The Almanac is published in two forms that differ in spatial extent and temporal resolution. California includes all years 1985-2025; CONUS includes 1990, 2000, 2010, 2020 and 2024. Both are released.

Released · v2026.1

Wildland Almanac - California

All 21 properties, water years 1985–2025, statewide at 30 m. The full stack, openly available now. How to access →

Released · v2026.1

Wildland Almanac - CONUS

Conterminous U.S. coverage, applying the same code nationally - 17 properties for water years 1990, 2000, 2010, 2020, and 2024, now on Source Cooperative. Disturbance is reported cumulatively between snapshots (rather than annually as in California) and the mask is water-only. Get the CONUS data →

California and CONUS are not yet on the same layer roster. The California release was revised in August 2026: five water and dieoff layers were renamed to the FixedPrecip / ObservedPrecip convention, the two FlamMap fire layers were replaced by three ELMFIRE ones, and canopy height became a standalone layer. CONUS still carries the earlier names (WaterFlux_AETmax, WaterFlux_AETrealized, WaterFlux_PminusETmax_SPI0, Fire_FL, Fire_ROS) and does not include the dieoff or canopy-height layers. The layer descriptions on this page describe California. CONUS will be rebuilt onto the California vocabulary in a future release; until then, read the CONUS README for its own roster.

Vegetation cover & structure 5 layers

Veg_TreeFrac · Veg_ShrubFrac · Veg_HerbFrac · Veg_BareFrac units fractional cover ×10,000 · period annual, summer-weighted

Fractional cover of tree, shrub, herbaceous, and bare. The four layers sum to 10,000 at each valid pixel - a Veg_TreeFrac of 9,000 means trees account for 90% of cover. All cover is viewed from above; lower strata hidden beneath the canopy are not counted.

Method: XGBoost on synthetic Landsat imagery (COLD / CCDC), trained on RAP and USFS data.

Veg_CanopyHt units centimeters (areal mean) · period annual, summer-weighted

Areal-mean canopy height: a straight arithmetic mean across the 900 one-metre sub-pixels of an airborne-lidar canopy height model, gaps and bare ground included. A pixel that is half closed canopy at 20 m and half bare ground has an areal-mean height near 10 m. This is the height to use for vegetation structure, biomass allometry and change analysis, and it is the height the Carbon_AGB calibration and the dieoff layers are built on. See the note below — it is not the same quantity as the CH band of Fire_LCP.

Method: XGBoost on synthetic Landsat, trained on the ALS canopy-height data of Allred et al. (2025).

Two canopy heights, and how to choose. Canopy height is published twice, under two different definitions. Veg_CanopyHt is the areal mean - the average over the entire 30 m pixel including gaps and bare ground, in centimeters. The CH band of Fire_LCP is a stand height - an estimate of the height of the taller trees in the pixel, in decimeters, which is what a fire model needs: fire models carry canopy openness separately through canopy cover, so a gap-diluted mean would count openness twice and under-shelter the surface fuels. The two differ by roughly 2-6× on the same ground, with the largest divergence at low canopy cover. Use Veg_CanopyHt for structure, biomass and change; use the Fire_LCP CH band for fire modeling. They are not interchangeable and should not be compared directly.

Hydrology 6 layers

ObservedPrecip and FixedPrecip, and how to choose. The water and dieoff properties come in two forms. The ObservedPrecip layers use the precipitation that actually fell, and answer what conditions were in a particular year - these are the ones to compare against independent observations such as stream gauges. The FixedPrecip layers hold precipitation at a reference so that variation in space and time comes from the vegetation alone - these are the ones for tracking change caused by management, fire or regrowth. The reference differs by layer: WaterFlux_AET_FixedPrecip holds precipitation non-limiting, WaterFlux_Runoff_FixedPrecip holds it at the long-term mean (SPI-48 = 0), and Vulner_TreeDieoff_FixedPrecip holds it at severe drought (SPI-48 = −2).

WaterFlux_AET_FixedPrecip units mm/yr · period annual sum, water year; precipitation held non-limiting

Evapotranspiration if water were never limiting, given the observed vegetation structure. Depends on structure, not year-to-year precipitation, so its series is more stable than the observed-precipitation layer - useful for isolating the hydrologic effect of vegetation change (management, fire, regrowth). Formerly published as WaterFlux_AETmax.

Method: regressions on AmeriFlux observations, monthly Landsat, and PRISM meteorology.

WaterFlux_AET_ObservedPrecip units mm/yr · period annual sum, water year

Best estimate of evapotranspiration that actually occurred, accounting for both vegetation structure and the precipitation that fell. Always ≤ the fixed-precipitation layer; equal in wet years, lower in dry years. Can be compared to independent observations of gauged runoff. Formerly published as WaterFlux_AETrealized.

Method: as above, with a precipitation-limited monthly water balance.

WaterFlux_Soilmoisture units mm water-equivalent, full rooting zone · period end of water year (Sep)

End-of-water-year soil moisture, from the same monthly water balance that produces WaterFlux_AET_ObservedPrecip and WaterFlux_Runoff. Potentially useful for diagnosing plant-water stress, the controls on water yield, and live fuel moisture.

WaterFlux_SoilmoistureFrac units fraction of max rooting-zone storage ×10,000 (0–10,000) · period end of water year (Sep)

End-of-water-year soil moisture as a fraction of the maximum rooting-zone storage: 10,000 is a full soil column at the seasonal low point, lower values are progressively drier. Normalizes for differences in storage capacity between pixels, which the absolute WaterFlux_Soilmoisture layer does not.

WaterFlux_Runoff units mm/yr · period annual sum, water year

Best estimate of annual discharge (surface runoff plus subsurface percolation) calculated as precipitation minus WaterFlux_AET_ObservedPrecip within the monthly water balance.

WaterFlux_Runoff_FixedPrecip units mm/yr · period annual sum, precip fixed at long-term mean (SPI-48 = 0)

Annual discharge predicted for an average-precipitation year (48-month SPI = 0), computed from the fixed-precipitation AET and the long-term-mean precipitation with no net soil-moisture change. Precipitation is held constant, so the layer isolates how vegetation density alone affects water yield. Formerly published as WaterFlux_PminusETmax_SPI0.

Fire hazard & behavior 4 layers

The fire layers are the fuels equivalent of FixedPrecip. Flame length and rate of spread use the same ensemble of 240 weather scenarios in every year; burn probability uses that same ensemble and the same ignition locations, seeding and threading, so nothing random varies between years. The fire models are not tuned over time or space. These layers therefore change from year to year for one reason only: the fuels changed. That is what makes them usable for tracking the effect of management, fire and regrowth — and it is also why they are not forecasts for any particular fire season.

Fire_LCP units 8 single-band Int32 COGs (FARSITE/FlamMap landscape) · period bands 4–8 annual; bands 1–3 static

The components of a FARSITE/FlamMap landscape for fire-behavior models such as FlamMap, delivered as eight single-band COGs rather than one multi-band file (five vary per year, three are static). The bands, in LCP order: (1) elevation m, (2) slope deg, (3) aspect deg, (4) fuel model (Scott & Burgan FBFM40, categorical), (5) canopy cover %, (6) canopy height dm, (7) canopy base height dm, (8) canopy bulk density kg/m³×100. Bands 1-3 are reprojected directly from LANDFIRE; band 4 is from LightGBM; bands 5-8 from XGBoost/LightGBM predictions drawing on LANDFIRE and on internally processed observations such as the canopy cover described above. Clip and stack the bands to assemble a landscape for an area of interest - see the Use page for the recipe. To convert to SI: divide CH and CBH by 10 for metres, and CBD by 100 for kg/m³; elevation, slope, aspect and canopy cover are already in their stated units. Band 6 is a stand height, not an areal mean.

Fire_ELMFire_FL units meters ×100 · period annual fuels, frequency-weighted across 240 weather scenarios

Characteristic flame length, computed with ELMFIRE in pixel-local mode over each year's Fire_LCP. Rather than one fixed weather condition, the layer weights 240 weather scenarios by how often each occurs, so it reflects the fire behavior a pixel can be expected to support across the weather it actually experiences.

Fire_ELMFire_ROS units (m/min) ×100 · period annual fuels, frequency-weighted across 240 weather scenarios

Characteristic rate of spread, from the same ELMFIRE runs as Fire_ELMFire_FL; the same interpretation applies.

Fire_ELMFire_BurnProbabilityRelative units relative BP × 1,000,000 (int16) · period annual fuels; simulated fire growth · native 120 m, delivered on the 30 m grid

The fraction of simulated fires that reached each pixel, from ELMFIRE in level-set (fire-growth) mode. Roughly 500,000 ignitions are drawn uniformly over each simulation domain (30 km-buffered USGS ARD tiles) and grown under a weather scenario drawn and weighted from the local climate distribution, with initial and extended suppression. This captures exposure from fire arriving from elsewhere, which the pixel-local flame-length and spread-rate layers cannot. The int16 ceiling of 32,767 corresponds to a relative burn probability of 0.032767. See the caveat below before using it.

Relative burn probability is not an annual probability. The magnitude of Fire_ELMFire_BurnProbabilityRelative is set by the number of simulated ignitions - a modeling choice, not a property of the landscape. It is comparable across space and across years within this dataset, and it is not a probability of burning in a given year: do not read 0.0005 as a 0.05% annual chance. The layer is calibrated so that simulated burned-area shares by fuel class match the FPA-FOD observed record; the absolute rate is not calibrated.

The FlamMap layers have been withdrawn. Earlier releases carried Fire_FL and Fire_ROS, computed with FlamMap under a single fixed weather condition (20 mph uphill wind, constant dry fuel moistures). They have been replaced by the three ELMFIRE layers above, which weight across the observed weather distribution rather than assuming one scenario. The old layers are no longer published for California; they remain in the CONUS release pending its rebuild.

Carbon 2 layers

Carbon_AGB units metric tons / hectare (total mass, not C) · period annual, summer-weighted

Aboveground live biomass, mainly tree; total mass, not carbon. A hybrid product combining FIA plot observations with internally processed remote sensing such as the canopy height, co-registered into the Almanac stack. Useful for nature-based-solutions accounting and biomass-dynamics tracking. Caveat: comparison against plot inventories shows Carbon_AGB may underestimate both absolute biomass and year-to-year increments in high-AGB stands — see the AGB benchmarks before using it for regional accumulation accounting.

Carbon_GPP units g C / m² / yr · period annual sum, water year

Annual gross primary production (mass of carbon). Derived from NIRv, flux-tower regressions, and PRISM datasets. The layer to reach for when you need productivity patterns or the climatic and vegetation controls on carbon uptake.

Forest dieoff risk 2 layers

Vulner_TreeDieoff_ObservedPrecip units unitless index, 0–32,767 (int16) · period rolling four-water-year deficit

“Is dieoff likely here, now?” Canopy height multiplied by the rectified four-water-year cumulative water deficit, using the precipitation that actually fell. Varies with both vegetation and recent drought, which makes it the layer for predicting which pixels are most at risk in a given year, and the one to compare against independent observations such as aerial dieoff mapping. It is correspondingly poor at isolating the effect of vegetation density, or at showing the benefit of management. Read as a relative index; the units are not interpretable in absolute physical terms.

Vulner_TreeDieoff_FixedPrecip units unitless index, 0–32,767 (int16) · period annual, precip fixed at severe drought (SPI-48 = −2)

“How exposed is this stand if a severe drought arrives?” — the planning layer. The same canopy-height × deficit calculation, but with precipitation held at the SPI-48 = −2 climatology rather than what fell. Because the drought is held constant, all variation comes from the vegetation, which makes this the layer for isolating the effect of density and for showing the benefit of management. It is correspondingly poor at predicting which pixels will die off in a particular year. Note this is a specified severe drought, not the worst possible one. Replaces the earlier Vulner_TreeDieoff_SPI-2.

Dieoff spin-up years. Vulner_TreeDieoff_ObservedPrecip uses a rolling four-water-year window that cannot be filled at the start of the record. Water years 1985-1988 sum the available years and scale to a four-year equivalent (1985 is a single year scaled ×4). They are retained rather than nulled, but they are not comparable with 1989 onward and should be treated as questionable.

Disturbance severity 2 layers

Observed event-based loss at pixels identified as disturbed by COLD. Pixels with no disturbance in a given year are encoded as 0; −9999 marks pixels outside the study area, masked, or where the pre/post reference is unavailable. Available for water years 1986–2024 - the first and last years of the series cannot be computed.

Disturbance_TreeFrac units Δ fraction ×10,000 (positive = loss) · period event delta

Annual loss of tree fractional cover at disturbed pixels. A value of 3,000 means an absolute loss of 30% tree cover (e.g. 40% → 10%). Disturbance timing for some pixels may be attributed to a slightly different year than the actual occurrence - more often, but not always, the following year.

Disturbance_AGB units Δ tons/ha ×10 (positive = loss) · period event delta

Annual loss of aboveground biomass at disturbed pixels, as the change in Carbon_AGB across pre- and post-disturbance years. Caveats as for Disturbance_TreeFrac. Watch the scaling: Carbon_AGB is stored in tons/ha with no scaling, but Disturbance_AGB is stored at ten times that — divide by 10 before comparing it against a difference you computed yourself from Carbon_AGB.

Before you analyze

Check the benchmarks first. The quality benchmarks PDF sets out, layer by layer, what each was compared against and how well it agreed - the independent reference data, the comparison method, and the result. It is the right place to judge whether a given layer is good enough for what you intend to do with it. Some layers are better constrained than others, and the differences matter.

v2026.1 stabilized on 29 August 2026 - re-sync if you pulled it earlier. This is the first complete and stable release. Reaching it took several in-place revisions through June and August 2026, the last of which corrected the fire-behavior layers (Fire_ELMFire_FL, Fire_ELMFire_ROS, Fire_ELMFire_BurnProbabilityRelative) and changed their values materially. A copy pulled before 29 August 2026 is not the same data as v2026.1 today; copies pulled on or after that date are current. From here, any change that alters a quantitative result will be published as v2026.2 alongside it.

Use one release end-to-end. Every annual version reprocesses the full 1985-onward series under one code base, so values for a given year may differ between versions. Do not splice years from different versions in a time-series analysis.

Masking. Layers are masked to the California wildland extent. Excluded and set to no-data: open water, ocean and inland, from the LANDFIRE Existing Vegetation Type layer; everything outside California; urban, agricultural, water, barren and unclassified land (CALFIRE FVEG WHR types 0 and 46–49); and a set of non-target EPA Level-IV ecoregions (the Central Valley and specified deserts). This release drops a vegetation-richness threshold used in earlier CECS products, so slightly more pixels are retained. The mask is not necessarily complete. For landscape-scale statistics (e.g. total biomass over a watershed), treat masked pixels as missing, not zero.

Grid. EPSG:5070 (NAD83 / CONUS Albers) at 30 m, on the same continental grid as the CONUS release. California is 45,000 rows × 30,000 columns, extent −2,415,585, 1,214,805 to −1,515,585, 2,564,805. All layers are co-registered to sub-pixel precision, so the 21 properties can be treated as a single stack. Files are LZW-compressed BigTIFF COGs with overviews — mode resampling for the categorical fuel-model band, nearest for aspect, average for everything else.

No-data and recent years. No-data is −9999 for most layers, but 0 for all seven fire layers — the eight Fire_LCP bands by FARSITE convention, and Fire_ELMFire_FL, _ROS and _BurnProbabilityRelative, in which masked ground is stored as zero rather than as a fill value. On those three, genuinely non-burnable ground inside the mask is also 0, so a zero cannot be distinguished from no-data on the raster alone. Some recent water years are less constrained - WY2023 reflects extreme California snowpack, and WY2025 is subject to refinement in future releases. For year-specific work, compare against nearby years.

License & citation

Released under Creative Commons Attribution 4.0 (CC BY). Free to use, share, and adapt with attribution. Offered as is, with no promise of technical support; please file feedback and reports of errors on the GitHub issue tracker.

University of California disclaimer. The University of California ("UC") makes these materials available pursuant to the following disclaimers: the materials are offered "as is"; user assumes any and all risks, of any kind or amount, of using these materials; user shall use the materials only in accordance with law; user releases, waives, discharges and promises not to sue UC, its directors, officers, employees or agents, from liability from any and all claims, including the negligence of UC, resulting in personal injury (including death), accidents or illnesses, property loss, as well as any and all loss of business and/or profit in connection with user's use of the materials; and user shall indemnify and hold UC harmless from any and all claims, actions, suits, procedures, costs, expenses, damages, and liabilities, including attorney's fees, arising out of user's use of the materials and shall reimburse UC for any such incurred expenses, fees or costs. Use of these materials implies user consent to these terms.

Goulden, M.L. (2026). The Wildland Almanac - California (Version v2026.1). Source Cooperative. DOI: [pending – EZID]. Released under CC BY.

The data