Using Machine Learning to Develop a Predictive Model for Future Fire Seasons

White, Andrew T ; Hain, Christopher R ; Schultz, Christopher J ; Case, Jonathan L ; White, Kristopher D

关键词: DEEP NEURAL NETWORKS;    DRYING;    MACHINE LEARNING;    PREDICTIVE MODELS;    REGIONS;    SEASONS;    SOIL MOISTURE;    WILDFIRES;   

RP-ID  :  MSFC-E-DAA-TN72933

学科分类：人工智能

美国|英语

来源: NASA Technical Reports Server

【 摘 要 】

The deep learning model shows promise for predicting areas of high wildfire potential. Full evaluation of the model performance is ongoing. Currently, the developed deep learning model is better overall at predicting the number of fires over the acres burned. Acres burned is dependent on location, suppression plan, and current conditions. Antecedent conditions are only one piece of the equation. In-season changes are not accounted for. An ignition source is required, which further complicates the model training and prediction.