Suppression Cost Forecasts in Advance of Wildfire Seasons

Abstract

Abstract Approaches for forecasting wildfire suppression costs in advance of a wildfire season are demonstrated for two lead times: fall and spring of the current fiscal year (Oct. 1–Sept. 30). Model functional forms are derived from aggregate expressions of a least cost plus net value change model. Empirical estimates of these models are used to generate advance-of-season forecasts. Cost forecasts involve estimation of suppression cost equations by geographical region based on a time series of historical data (1977–2006) of costs, a time trend, and climate variables, forecasts of the next season's suppression costs, by region and in total across all regions, and generation of suppression cost forecast probability distributions by region and in aggregate. The forecasts are also evaluated historically for their goodness of fit using cross-validation techniques. The two lead time forecast models are compared with the 10-year moving average of suppression costs, currently used as a budget request formula by the US Forest Service. Results show that the spring forecast of suppression costs is statistically no better than the fall forecast for predicting the coming season's costs. However, both the spring and fall forecasts significantly outperform the 10-year moving average, reducing forecast errors by approximately 60%.