Prediction of seasonal wildland fire severity in South Dakota using artificial neural networks

Seasonal wildfire severity is in large part determined by the local climatic conditions. Large wildland fires across the Western US occur more frequently during periods of drought and/or excessively warm temperatures; however, the effects of long-term precipitation and temperature anomalies on fire season severity across the varied fuel models of South Dakota are less understood. In this study, 20 years of temperature, precipitation, and wildfire acreage burned data from both forest and grassland ecosystems within South Dakota are exploited to elucidate the relationships between climate and fire season severity. After a preliminary data mining step, a multilayer feedforward artificial neural network was trained (using backpropagation) to predict fire severity. Preliminary results are encouraging and demonstrate that by utilizing antecedent temperature and precipitation conditions, skillful predictions of fire season severity may be possible with lead times ranging from months to seasons in advance.