Climate influences wildfire activity at multiple temporal scales, but if and how these relationships vary through time is poorly understood. Time-varying fire-climate relationships have direct relevance for understanding the controls of fire regimes and projecting future wildfire activity. We evaluated the accuracy and predictive skill of time-varying statistical models predicting annual fire extent from climate variables, for a 12-million ha region in the U.S. Northern Rockies, from 1902-2008. Fire extent was based on a updated fire-perimeter dataset that accounted for fires > 40 ha on federally managed lands in Idaho and Montana west of the Continental Divide. Downscaled climate data, including water balance metrics and fire danger metrics from the Canadian and U.S. fire danger rating systems, were used as predictors of the natural logarithm of annual fire extent, in simple and multiple linear regression models.
Results/Conclusions
Fire-danger and water-balance metrics outperformed multivariate models based on temperature and precipitation, yielding higher combinations of calibration accuracy and predictive skill. The strength and nature of fire-climate relationships varied markedly over time. The early 20th century was characterized by strong fire-climate relationships (69-75% maximum explained variance [MEV]) and significant predictive skill. Weaker but significant relationships characterized the mid 20th century, ca. 1930-1970 (49-64% MEV), with slightly lower predictive skill. Increasingly strong fire-climate relationships towards present (75-88% MEV), ca. 1977-2008, appear unique since at least the early 20th century in their strength and nature. Models trained during this period overpredict area burned when applied to earlier periods, indicating an amplified response of annual area burned to climatic variability over the past three decades. Biomass accumulation from a 50-yr period of little burning during the mid 20th century is likely responsible for the amplified response, suggesting that when fuel limitations change, the functional link between climate and fire can be significantly modified. Future projected and realized wildfire activity, therefore, could vary widely depending on the selected calibration periods and future vegetation change.