Wildfire Smoke Detection Analysis. In Data Science & Engineering Master of Advanced Study (DSE MAS) Capstone Projects

Wildfires have caused significant damages throughout the state of California. In 2018 alone, it is estimated that approximately $150 billion of damages were caused by wildfires . The devastating impact of these fires in recent years has intensified the need for hardened infrastructure systems and more advanced fire prevention and mitigation efforts. This project explores improving a pre-existing, image classification model for smoke detection, called SmokeyNet , originally developed by the San Diego Supercomputer Center at the University of California, San Diego. SmokeyNet uses images that are captured from a network of cameras across Southern California known as the High Performance Wireless Research and Education Network (HPWREN) to predict and identify fire smoke. Improvements were explored by 1) incorporating supplementary geostationary satellite fire detection model data (Wildfire Automated Biomass Burning Algorithm, WFABBA) to increase model accuracy, and 2) programmatically geolocating the approximate origin of a fire from an image to aid in the emergency response process. Weather data was analyzed as an additional supplementary dataset to the SmokeyNet model, but retraining and retesting was not fully complete. From the ensemble of the multiple fire models, it resulted in a slight boost to overall accuracy. When programmatically geolocating the fire, by using triangulation, location approximations were able to be calculated within 3 miles of actuals; however, a small sample size of 10 was used and further data would need to be gathered and evaluated.