A recently developed statistical forecast for seasonal wildfire area burned shows significant skill for the Western United States, particularly in the Great Basin and Sierra Nevada. Canonical Correlation Analysis of pre-filtered time series of seasonal area burned and leading values of the Palmer Drought Severity Index is used to construct linear forecast models of area burned a season or more in advance. Experimental forecasts for the 2003 fire season were issued in January, February and early April. The earlier forecasts showed the most skill in the Great Basin. All three forecasts anticipated an above normal fire season in the Colorado Rockies and in mountain ranges in Arizona and New Mexico, and a below normal fire season in desert basins throughout the West.
We are extending this work to include forecasts of seasonal suppression costs for the USDA Forest Service by Forest Service administrative regions. This work compares linear regression models estimating suppression costs using area burned forecasts as predictors to Canonical Correlation Analysis models that estimate suppression costs directly from leading climate indices.