Continued warming could transform Greater Yellowstone fire regimes by mid-21st century

Abstract

Climate change is likely to alter wildfire regimes, but the magnitude and timing of potential climate-driven changes in regional fire regimes are not well understood. We considered how the occurrence, size, and spatial location of large fires might respond to climate projections in the Greater Yellowstone ecosystem (GYE) (Wyoming), a large wildland ecosystem dominated by conifer forests and characterized by infrequent, high-severity fire. We developed a suite of statistical models that related monthly climate data (1972-1999) to the occurrence and size of fires >200 ha in the northern Rocky Mountains; these models were cross-validated and then used with downscaled (~12 km × 12 km) climate projections from three global climate models to predict fire occurrence and area burned in the GYE through 2099. All models predicted substantial increases in fire by midcentury, with fire rotation (the time to burn an area equal to the landscape area) reduced to <30 y from the historical 100-300 y for most of the GYE. Years without large fires were common historically but are expected to become rare as annual area burned and the frequency of regionally synchronous fires increase. Our findings suggest a shift to novel fire-climate-vegetation relationships in Greater Yellowstone by midcentury because fire frequency and extent would be inconsistent with persistence of the current suite of conifer species. The predicted new fire regime would transform the flora, fauna, and ecosystem processes in this landscape and may indicate similar changes for other subalpine forests.

Fig. 1.

Annual area burned by large fires in the Northern Rockies: observed (line) versus predicted from 1,000 simulations per month, aggregated across 2,309 grid cells (boxes show interquartile range; whiskers, 1.5× interquartile range; and points, extremes). Results are a composite of simulations for fire presence/absence using probabilities from cross-validated logistic regressions, simulations for fire size from cross-validated Poisson lognormal distributions conditional on fire presence and the linear estimator from the cross-validated linear regression, and fire size estimated using a generalized Pareto distribution conditional on fire occurrence and moisture deficit (see also Figs. S6–S9). The generalized Pareto is validated elsewhere (SI Text and Figs. S7 and S8).

Fig. 2.

The range (light shading), interquartile range (dark shading), and median (dotted line) of predicted area burned for 1,000 gridded monthly simulations, aggregated over the GYE by year, and the observed annual area burned (solid line) for three downscaled GCM SRES A2 climate scenarios: NCAR CCSM 3.0 (A), CNRM CM 3.0 (B), and GFDL CM 2.1 (C).

Fig. 3.

Projected fire rotations calculated for 1,000 monthly gridded simulations for four 30-y periods across three downscaled GCM SRES A2 climate scenarios.