Record fires in Alaska during the summer of 2015 resulted in 5.1 million acres burned with a staggering cost of about $188M. Cost savings are possible if skillful outlooks were available to allow for better long-term planning. Since 1940, four of the ten largest fire seasons (2004,2005,2009 and 2015) have occurred in the last 15 years. The availability of the North American Multi-Model Ensemble (NMME) allows the development of seasonal forecast products that can be used to inform logistical planning.
This study employs the NOAA seasonal forecasts using the CFSv2 model to construct the FWI (Canadian Forest Fire Danger Rating System Fire Weather Index) a season in advance to evaluate the probability of an active wildland fire season in Alaska. The FWI is used by fire managers in Alaska to evaluate fire danger over regions defined as Predictive Service Areas (PSAs). The Build Up Index (BUI) is one of the FWI that characterizes available fuels and depends on the seasonal progression of temperature and precipitation. BUI calculated directly from the reanalysis data or the CFSv2 seasonal forecasts displayed large biases which necessitated correcting the forecasts. Temperatures are generally too low and precipitation is too high over most of Alaska in the CFSv2. Corrections of the forecasts were conducted at the PSA level using available meteorological station data of temperature and precipitation. Quantile mapping, a method which adjusts the model cumulative distribution function (CDF) to that of the observations, was used to correct for biases in the seasonal forecasts. The FWI was calculated from the corrected CFSv2 forecasts over the period 1982-2018, compared to station-based FWI for the 1994-2018 period. The probability of an active wildland fire season based on the March 1 (and May 1) initialized forecasts for June-August is compared to acres burned over the 1982-2018 period.
The skill of the seasonal outlooks will be evaluated and measures of uncertainty will be quantified to enhance the usability of the seasonal products by fire managers. Sensitivity tests are explored and highlight the importance of correcting both temperature and precipitation in the calculation of the BUI. The post-processed CFSv2 ensemble members will be further analyzed to identify key large-scale patterns that can be used to better identify the probability of wildland fire in Alaska. Finally, the presentation of the final output will be vetted by the fire management community to ensure its effectiveness.