A 1984–2015 fuel-aridity model predicted the 2016–2024 forest area inside western US wildfire perimeters within 1%; recently burned forest reburned less for about 7 years

Question: does the exponential relation between fire-season fuel aridity and western US forest burned area (Abatzoglou & Williams 2016, fitted to 1984–2015) predict the years after 2015? Approach: we rebuilt the series from public data. Burned area is the forest area (USFS 250 m forest type groups) inside MTBS wildfire perimeters in 11 western states. Aridity is the mean of four standardized TerraClimate metrics (Mar–Sep VPD and reference ET, annual water deficit, Jun–Aug PDSI) over forest cells. We fitted log10(area) on aridity for 1984–2015 only and predicted 2016–2024. We also measured how often forest burned again, by years since the last perimeter, against its area share in 50 km blocks. Results: the fit had R2 = 0.73 (4.7 times more area per standard deviation of aridity). For 2016–2024 the model predicted 9.01 Mha and we observed 8.99 Mha (95% predictive interval 4.9–16.6 Mha). All 9 test years, including the 2020 record and the 2021 aridity maximum outside the fit range, were inside 95% prediction intervals. Out-of-sample squared error was 0.035 (log10) for aridity, 0.137 for a linear time trend and 0.358 for the fit-period mean, and aridity alone was better than trend and mean for all fit periods that end in 2000–2015. Post-2015 slope and level terms were not significant (p = 0.71). Regional models showed more burning than predicted in California (observed/predicted 1.96, mean log error p = 0.011) and less in the Southwest and Northern Rockies, within intervals. Forest burned 1–2 years earlier reburned at 0.19 (95% CI 0.14–0.26) of the expected rate, 3–5 years at 0.41, and 6–10 years at 0.98. Our upper estimate of the 2016–2024 reduction of burnable forest by recent fires was 4.9% for the west and 11% for California, small compared with the aridity effect. Limitations: perimeters include unburned patches; we used TerraClimate and four metrics, not gridMET and eight; nine test years cannot exclude moderate slope changes (95% CI −0.48 to +0.32 log10 per standard deviation); the test does not test attribution to human forcing.

wildfire · burned area · fuel aridity · vapor pressure deficit · western United States · MTBS · TerraClimate · out-of-sample test · forest · reburn · fire-fuel feedback · climate attribution