Spatial linear modeling and forecasting of forest fires across the United States

摘要

Data from the United States Fire Service is used to develop a model for forecasting fire severity. It presents a strong case for extending the use of remote sensing techniques in the analysis of ground conditions and fires. Only one of the predictor variables is derived from AVHRR data, and the present analysis still treats fuel models as stationary predictors. Spectral mixture analysis (SMA) of multispectral data from future sensors, such as Landsat 7 and MODIS, can be used in conjunction with ground measurements to generate much denser spatial and temporal predictors of fire occurrences. At this higher resolution, the rapid extraction of representative (fires vs. no fires) pixel populations over an extended period preceding the prediction date, becomes critical to the success of the linear predictor.