Prescribed Fire Optimization. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

Insufficient application of wildfire prevention techniques has resulted in a dangerous buildup of wildfire fuel. To reduce this build up, fire districts will conduct prescribed burns on areas in their zone of authority. One obstacle to executing prescribed burns is the inability to accurately quantify the risk of a candidate prescribed burn. Currently the preburn risk appraisal is solely dependent on the experience of fire scientists. This project brings a data-driven approach to quantifying the risk of a candidate prescribed burn. This approach utilizes logistic regression to produce calibrated risk scores. The logistic regression is trained on data generated by a fire simulation software called QUIC-Fire. Atmospheric conditions such as wind speed, wind direction, and surface moisture are the features which the model uses to produce a risk score. A proof-of-concept user interface was developed to show how risk scores could be integrated into the go/no-go decision-making process of a prescribed burn. The end goal is for BurnPro3D to integrate the risk scores into their user interface.