Fire Risk Simulation Model (FRSM) Outputs ================================================================ ========================================================================================================== Overview ========================================================================================================== The Fire Risk Simulation Model (FRSM), developed at the University of California Merced, is a suite of statistical probability models that estimate the likelihood of occurrence, size, and severity of large wildfires (> 1000 acres or 400 ha) across California. Fire risk is simulated as a function of climate conditions and landscape, including vegetation and urban development. In standalone FRSM runs, initial historical vegetation and land cover conditions exported from LUCAS remain fixed over time. There are no feedbacks between burned areas and future fire probability, and no sub-scenarios for vegetation management, land-use change, or urban growth. This configuration is designed to isolate the influence of climate and weather conditions on wildfire probability, size, and severity. ========================================================================================================== Model Characteristics ========================================================================================================== - Model Type: Fire Risk Simulation Model (FRSM); large fire occurrence probability, size, and burn severity fractions. - Spatial Resolution: 3 km - Temporal Resolution: Semi-monthly for most variables; outputs are aggregated to annual time steps for summary products (e.g. for fire counts) - Temporal Span: 1982–2099 (WRF-downscaled climate simulations); 2000–2100 (LOCA2-downscaled climate simulations), 1951-2020 (WRF-downscaled historic ERA5 climate observations) - Iterations: 2,000 per climate simulation - Vegetation Dynamics: Static baseline (no vegetation feedback or land cover change) for uncoupled FRSM runs; dynamic annual vegetation and urban growth (with fire-vegetation-land use feedbacks enabled) for coupled LUCAS-FRSM runs ========================================================================================================== Model Inputs ========================================================================================================== Dynamically Downscaled Climate Data (WRF - UCLA) ------------------------------------------------ FRSM was run using Weather Research and Forecasting (WRF) climate simulations produced by UCLA and bias-adjusted by the Scripps Institution of Oceanography. Model Scenario Ensemble Member CESM2 SSP3-7.0 r11i1p1f1 CNRM-ESM2-1 SSP3-7.0 r1i1p1f2 EC-Earth3-Veg SSP3-7.0 r1i1p1f1 FGOALS-g3 SSP3-7.0 r1i1p1f1 WRF-ERA5 Historical reanalysis (1951–2020) — Statistically Downscaled Climate Data (LOCA2 - Scripps) ------------------------------------------------------- The model was also run using LOCA2-Hybrid downscaled CMIP6 projections for 2000-2100, which include multiple SSP and ensemble configurations. Model Scenarios Ensemble Members ACCESS-CM2 SSP2-4.5, SSP5-8.5 r1i1p1f1 CNRM-ESM2-1 SSP3-7.0, SSP5-8.5 r1i1p1f2 EC-Earth3-Veg SSP2-4.5, SSP3-7.0, SSP5-8.5 r4–r5i1p1f1 HadGEM3-GC31-LL SSP5-8.5 r1, r3i1p1f3 INM-CM5-0 SSP2-4.5 r1i1p1f1 MPI-ESM1-2-HR SSP2-4.5 r1i1p1f1 MPI-ESM2-0 SSP3-7.0 r4i1p1f1 Vegetation and Urban Footprint ------------------------------------------------ Standalone FRSM simulations use a static vegetation and land-use layer derived from the 2020 LUCAS historical vegetation dataset. Vegetation and urban extent are held constant over time. No sub-scenarios for vegetation management or urban growth are applied. Synthetic Relative Humidity Datasets ------------------------------------ Relative humidity is a key driver variable for the FRSM. Three daily datasets, each matching the model’s spatial grid and projection, are used: Dataset Description Source ERA5-WRF Historical ERA5-WRF (1950–2021), daily COG rasters UC Merced ERA5-WRF Data UCLA WRF (Bias Corrected) Daily COG rasters organized by GCM/SSP/RunID UC Merced WRF Bias-Corrected Data LOCA2 Daily COG rasters for each LOCA2 climate scenario UC Merced LOCA2 Data Climate Scenario Tiers ---------------------- Climate scenarios were categorized into five tiers. For standalone FRSM runs, only Tier 1, Tier 2a, Tier 2b, and Tier 3 scenarios were used. Tier 4 scenarios were not selected for modeling in this project. Key characteristics of the selected tiers include: - Tier 1: Climate simulations that span the range of plausible mid-century outcomes (through 2064), capturing mid-century climate extremes such as drought and precipitation anomalies. These include the most skillful and high-quality models as determined by the Scripps team (n = 5). - Tier 2a: Includes additional mid-century extremes and high-quality/skilled models identified by Scripps (n = 8). - Tier 2b: Scenarios that include dynamically downscaled data (e.g., hourly WRF) and have all necessary variables available in both dynamical and statistical downscaling (n = 1). - Tier 3: Simulations representing significant variability in end-of-century climate outcomes. ========================================================================================================== Model Outputs ========================================================================================================== Fire Event List (fireList) -------------------------- Each model run produces a comprehensive list of all simulated fires across 2,000 iterations. Files are provided in .csv and .rds (R data serialization) formats and include the following variables: - FireID: Unique identifier for each simulated fire - x, y: Centroid coordinates of the fire (Teale Albers projection) - year, month, day: Date of the simulated fire (“day” refers to semi-monthly time step) - Iteration: Model iteration in which the fire occurred - Size: Total size of the fire (hectares). - sizeLow, sizeMod, sizeHigh: Area burned at low, moderate, and high severity (hectares) - cleanSize: Fire size capped at 3,000,000 hectares - cleanLow, cleanMod, cleanHigh: Severity areas based on capped size - fracLow, fracMod, fracHigh: Fraction of the total fire area by severity (0–1) - US_L3CODE, US_L3NAME: EPA Level III Ecoregion code and name - iterStart / iterEnd - Range of model iterations included - yearStart / yearEnd - Simulation years covered Note: These files are large (1–5 GB depending on format and scenario). Note: The exact Teale Albers projection using the PROJ4 string is "+proj=aea +lat_1=34 +lat_2=40.5 +lat_0=0 +lon_0=-120 +x_0=0 +y_0=-4000000 +ellps=GRS80 +towgs84=0,0,0,0,0,0,0 +units=m +no_defs” Annual Burn Rasters (BurnRasters/) ---------------------------------- Each burn raster summarizes the expected area burned in each 3 km grid cell across all iterations for a given year and scenario. Files are stored as Cloud-Optimized GeoTIFFs (COG) using the Teale Albers projection. File name format: burned_3km_TA_.tif - Annual raster showing total area burned across all iterations. Each raster is generated from the corresponding fire list outputs. Fires are grouped by year and include all 2,000 model iterations for a given scenario (climate, vegetation management, urban footprint). Each simulated fire is placed on the landscape in its originating grid cell and then spread outward in a roughly circular pattern. Each cell can burn up to 85% of its area (approximately 765 hectares) before fire spread continues outward from the center cell. All fires simulated within a given year are then summed to produce a single raster representing the total area burned in each cell across all iterations. Dividing the raster values by the number of iterations (2,000 for the climate scenarios) yields the expected burned area per cell (hectares) for that year. Large Fire Presence Maps (PresenceRasters/) ------------------------------------------- Each presence raster summarizes the number of large fires centered in each 3 km grid cell across all iterations for a given year. Files are stored as Cloud-Optimized GeoTIFFs (COG) using the Teale Albers projection. File name format: presence_3km_TA_.tif (Annual raster showing the total number of large fires centered on each pixel across all iterations.) Each raster is a count of the number of times a large fire ignition was centered on that grid cell in a given year, aggregated over all model iterations. Dividing this count by the total number of iterations (2,000) yields the probability of a large fire being centered on that pixel in that year. ========================================================================================================== Data Organization ========================================================================================================== Uncoupled FRSM Runs ------------------- Standalone FRSM datasets use a static vegetation and land-use layer derived from the 2020 LUCAS historical baseline. Repository directories for LOCA2 and WRF outputs are structured as: //WesterlingFireModel/ Each directory contains the following subfolders: - fireList/ - Tabular listings of all simulated fire events - BurnRasters/ - Annual rasters summarizing expected area burned - PresenceRasters/ - Annual rasters indicating large fire occurrence File Formats ------------ All raster outputs are provided as Cloud-Optimized GeoTIFFs (COGs) with embedded coordinate reference system (CRS) information, ensuring compatibility with standard GIS software and cloud-based analysis workflows. ========================================================================================================== Citations and Acknowledgements ========================================================================================================== This modeling and data processing effort is part of the California Energy Commission-funded research on wildfire probability modeling at UC Merced, conducted by the Pyregence consortium (pyregence.org).