Data driven approach to estimating fire danger from satellite images and weather information

Wildfires cause extensive damage to nature and human developments. Substantial funds are spent preparing for and fighting them. This work develops a data driven approach to modeling the probabilistic risk of a currently burning fire becoming large and dangerous. We based our model upon observations of fire, weather and surrounding extracted from remote satellites. Data driven models reached good recognition accuracy in predicting fire danger in the coming day or two. We intend using the predictions in planning algorithms, e.g. flight plans for unmanned fire surveillance aircraft, to fight the fires in a more efficient and timely manner.