Downscaling fire weather extremes from historical and projected climate models

Abstract

An important aspect of predicting future wildland fire risk is estimating fire weather—weather conducive to the ignition and propagation of fire—under realistic climate change scenarios. Because the majority of area burned occurs on a few days of extreme fire weather, this task should be able to resolve fire weather extremes. In this paper, we present a statistical downscaling procedure based on distribution based scaling (DBS) to bias correct the Fire Weather Index (FWI), part of the Canadian Forest Fire Danger Rating System, as calculated from modeled climate data. Our study area is western Canada (British Columbia and Alberta) and we consider both an historical control period (1990–2000) and three future time periods (2020–2030, 2050–2060, and 2080–2090). The historical data used to calibrate the DBS procedure comprises weather station data and weather from the North American Regional Reanalysis (NARR), whereas the future climate projections are the output of three regional climate models, corresponding to different model parameterizations and downscaled from the NCAR Community Earth System Model under the RCP 8.5 scenario. By fitting a truncated Weibull distribution to observed and modeled FWI values, our method is able to reproduce historical extremes in fire weather indices as determined by the distribution of annual potential spread days, which are defined as days with FWI values greater than 19. Moreover, by calibrating the DBS procedure with gridded reanalysis data as well as station observations, we are able to project future spread day distributions over the entire study area. The results of this study show the DBS procedure leads to a greater number of projected annual spread days at most locations compared with estimates using the uncorrected model output, and that all three RCM models show positive increases in potential annual spread days for the 2050–2060 and 2080–2090 time periods.