Projected increases in western US forest fire despite growing fuel constraints

The study addresses how much fuel limitations arising from prior fires (fire-fuel feedbacks) can constrain anticipated climate-driven increases in western US forest-fire area over the near term (2021–2050). Contextually, burned area in western US forests has increased roughly ten-fold over the past half-century, driven by factors including fire suppression legacies, management practices, and especially increased fuel aridity associated with anthropogenic warming. Climate change is intensifying vapor pressure deficit, reducing snowpack, and lengthening fire seasons, all of which increase fuel aridity (F) and fire-weather extremes. A key uncertainty for projecting future fire is whether decreasing available fuels following fires will impose nonlinear constraints on climate–fire relationships. The purpose of this work is to quantify how dynamic fire-fuel feedbacks of varying strength and longevity may moderate climate-driven increases in forest-fire area at the scale of the entire western US forested region, and to assess implications for variability and extreme fire years relative to the record 2020 season.

Prior research demonstrates strong links between interannual fuel aridity and burned area in western US forests and globally, with warming increasing aridity and fire-weather extremes. Empirical models driven solely by climate often project large future increases in burned area but implicitly assume constant fuel availability. Evidence indicates that increased burning can initiate feedbacks that reduce future burn potential by removing biomass, disrupting fuel continuity, and in some cases causing post-fire conversion to non-forest vegetation that carries different fire regimes. Studies at local to regional scales have shown that prior fires can limit subsequent fire spread or severity and that drought conditions can shorten the period of fuel limitation. However, the combined effects of climate change and fire-fuel feedbacks have not been comprehensively evaluated at the macroscale of the entire western US forest domain. This work builds on these studies by explicitly incorporating alternative forms and strengths of fuel-feedback processes into climate–fire models and testing their implications for near-term projections.

Data: The study used monthly PRISM climate data (max/min temperature, dew-point, precipitation) at 1/24° resolution (1950–2020) and ERA5 shortwave radiation and 10-m wind at 0.25° interpolated to the PRISM grid. For future climate, outputs from 30 CMIP6 models (one ensemble member each) were used: historical (1950–2014) and SSP2-4.5 (2015–2050), interpolated to 1.0°. Burned area came from MTBS (1984–2018, fires ≥404 ha; excluding unburned-to-low severity and prescribed burns) and MODIS Collection 6 (2001–2020) adjusted to match MTBS during overlap; forest/woodland defined via LANDFIRE environmental site potential. Forest-fire area (FFA) prior to the satellite era was taken from published estimates for 1916–1983 to initialize dynamic variables.

Fuel aridity indices: Three aridity proxies were computed: VPD (mean Mar–Sep), Penman–Monteith reference evapotranspiration ETo (sum Mar–Sep, modified for elevated CO2 impacts on surface resistance following Yang et al. 2019), and climatic water deficit CWD (Jan–Dec). Gridcell indices were standardized to 1991–2020 (z-scores). The regional fuel aridity F is the average of gridcell aridity over forest/woodland, re-standardized to 1991–2020 and bias-corrected via equidistant quantile mapping. CWD was selected for projections due to strong historical relationships with FFA and more conservative future increases versus VPD/ETo.

Model design: A macroscale approach modeled annual western US FFA using statistical relationships with F. A static model assumed constant fuel availability: log(FFA(t)) = α1 + β1·F(t) + εt, where ε are residuals resampled via Monte Carlo to capture variability not explained by F (e.g., ignitions, extreme weather).

Dynamic models incorporated fuel constraints via L(t), the fraction of contemporary forested land ineligible to carry forest fire, reducing the effective response of log(FFA) to F: log(FFA(t)) = α2 + β2·F(t) + εt, with the response moderated by L. L = L1 + Lt combines:

• Semi-permanent post-fire regeneration failure L1: fraction p of burned forest converting to non-forest based on 3-year post-fire mean aridity F3y: p=0 for F3y<1; p=(F3y−1)/2 for 1≤F3y≤2; p=1 for F3y>2; L1(t) is the cumulative sum of p·FFA/T since 1984 (T = total contemporary forest area). Parameter μ was set such that L1 ≈ 0.01 by 2050 under baseline assumptions, indicating minor contribution to mid-century feedbacks.
• Temporary feedback Lt from recent fire history with longevity r: two formulations over a nominal ~30-year window represented using 7-year history proxies. Constant feedback: Lt(t)=γ·Σ(FFA(i))/T over the previous 7 years; Fading feedback: higher weights on the last 5 years and sinusoidally decreasing weights for years 6–7. Feedback strength γ took values 0.5 (weak), 1.0 (moderate), or 1.5 (strong). Stronger γ implies larger areas rendered temporarily ineligible than the area recently burned, representing mosaic effects and reduced connectivity.

Drought modulation of longevity: Longevity r shortens with higher contemporaneous aridity: r=30 years for F<0, linearly decreasing to r=20 years at F≥2, reflecting observations that drought and extreme fire weather can overcome recent fuel limitations and allow reburning at shorter intervals.

Model execution: Dynamic models were initialized using LANDFIRE mean fire return interval (MFRI) to approximate pre-1916 baseline FFA and run through 2050. Static and dynamic models were driven by each CMIP6 model’s F for 1950–2050, using observed FFA to update L through 2020, then model-derived FFA thereafter. Both models limited maximum annual burning: in the static model, a cap at the estimated 100-year return period for the observational era (7.5% of T); in dynamic models, 7.5% of eligible area (1−L)·T. Monte Carlo resampling (n=1000) of ε generated replicated FFA time series; reported statistics use the median across replicates for each climate model.

Validation and diagnostics: Cross-validation split (1984–2000 train; 2001–2020 test) assessed bias, coefficient of efficiency, and correlation of log(FFA). Weak to moderate feedback dynamic models retained skill comparable to the static model; very strong feedbacks (γ≥2) were excluded due to substantial underprediction during validation. Interannual variability metrics included the interquartile range (IQR), generalized extreme value analysis for recurrence of 2020-sized seasons, and the percentage of years with FFA below the 1991–2020 median.

• Historical relationship: A strong nonlinear relationship exists between fuel aridity F and log(FFA) for 1984–2020 (r^2=0.80). A 1.0σ increase in mean F from 1984–1999 to 2000–2020 coincided with a ~4-fold increase in the upper quartile of annual forest-fire area.
• Projected aridity: CMIP6 multi-model mean projects F increasing by 0.66σ for 2021–2050 relative to 1991–2020 (IQR +0.45 to +0.80σ), primarily due to warming-driven evaporative demand.
• Static model (constant fuels): Mean annual FFA doubles (+107%) in 2021–2050 vs 1991–2020; cumulative 2021–2050 FFA ≈ 35% of contemporary forest area. 26/30 climate models project ≥50% increases. Interannual variability (IQR) doubles, and the probability of years exceeding the record 2020 season roughly triples. Nonetheless, approximately one-third of 2021–2050 years are projected to have FFA below the 1991–2020 median.
• Dynamic models (with fuel-feedbacks): Despite feedbacks, substantial increases persist. Weak and moderate feedbacks yield mean annual FFA increases of ~82–90% and ~63–75%, respectively; the strongest tested feedback yields a +46% increase relative to 1991–2020. Feedbacks increase L over time, constraining eligible area but only modestly attenuating climate-driven growth. Variability and the likelihood of 2020-exceeding years still increase (though less than in the static model). Alternative parameterizations and aridity indices show similar qualitative outcomes; e.g., weak feedback formulations reduce projected FFA by ~14–19% relative to the static model. Using VPD as F produces larger increases than CWD.
• Fuel eligibility trends: Modeled L declined through much of the 20th century (fire deficit), heightening sensitivity of FFA to aridity, but has begun increasing with recent large fire years, thereby incrementally constraining future eligible area.

The results indicate that while fire-fuel feedbacks exist and will expand the fraction of forested lands temporarily or semi-permanently ineligible to carry forest fire, these feedbacks only moderately reduce climate-driven increases in forest-fire area through mid-century. The dominant thermodynamic influence of anthropogenic warming on fuel aridity, combined with the nonlinear response of FFA to F, drives substantial projected increases in mean burned area, variability, and the frequency of extreme years comparable to or exceeding 2020. The aggregate, macroscale approach is justified by the strong regionwide coherence in aridity and fire response, though finer-scale coupled fire–vegetation models could elucidate spatial heterogeneity and additional feedbacks. Other disturbance agents (insects, drought mortality, windthrow, invasive grasses, harvest) can interact with fire in complex ways not explicitly modeled here, and fuel-feedback longevity shortens under drought, enabling reburning at shorter intervals. Management implications include recognizing that some increased fire—often at low to moderate severity—can contribute to ecosystem restoration toward pre-colonization fire regimes, but escalating megafire risk necessitates proactive mitigation and adaptation to reduce negative socio-ecological impacts.

This study shows that western US forest-fire area is very likely to continue increasing over the next three decades due to rising fuel aridity from anthropogenic climate change, even when accounting for a wide range of plausible fire-fuel feedbacks. Under constant fuels, mean annual FFA roughly doubles; with dynamic fuel constraints, increases remain substantial (+46% to +90% depending on feedback strength). Interannual variability and the likelihood of extreme years similar to or exceeding 2020 also increase. Fire-fuel feedbacks alone are unlikely to reverse the trend by mid-century. These results underscore the need for proactive adaptation and mitigation strategies, including expanding prescribed fire and mechanical fuel treatments to increase L in targeted ways, improving preparedness for extreme fire-weather seasons, and reducing greenhouse gas emissions to limit aridity increases. Future research should better constrain the form, magnitude, and spatial variability of L, integrate additional disturbance interactions, employ higher-resolution coupled fire–vegetation models across subregions, and assess operational windows and scaling for prescribed fire under changing climate.

• The fuel-limitation term L is poorly constrained and likely varies spatially and temporally; its parameterizations (strength γ, longevity r, forms of Lt, regeneration failure p) are simplified and may not capture local complexities.
• Other disturbance agents (insects, drought mortality, windthrow, invasive grasses, timber harvest) and their interactions with fire are not explicitly modeled, though they can influence fuels and fire regimes.
• The models assume time-invariant human and management influences embedded in regression parameters; potential future changes in suppression strategies, ignitions, and policies are not dynamically represented.
• Daily to sub-seasonal meteorology and specific extreme fire-weather patterns are not explicitly included beyond what is captured by the seasonal aridity index F.
• A cap on maximum annual burned area (7.5% of T or eligible area) imposes a constraint that is not fully physically quantified.
• Pre-satellite burned area estimates and LANDFIRE MFRI inputs used for initialization carry uncertainties; MODIS-to-MTBS adjustments add additional uncertainty for 2019–2020.
• Results are aggregated over the western US forests; subregional variability in climate–fire relationships and feedback strengths may yield different local outcomes.
• The CO2 effect on ETo (surface resistance) is uncertain at regional scale.
• CMIP6 model uncertainties in projecting aridity remain, and using different aridity indices (e.g., VPD vs CWD) changes the magnitude of projected increases.