Student Research Reports

View Research Report

Wildfires have the capacity to destroy forests, homes, and the health of ecosystems. As a result of rising global temperatures, wildfires have experienced increased frequency and severity, leading to devastating outcomes throughout sections of North America. All of these factors have led to an increase in demand for tools that can accurately and efficiently identify potential risk areas, particularly after the devastating California wildfires of the 2010s and 2020s. This project aims to analyze pictures taken by NASA's GLOBE Observer app to identify land cover types in order to classify their potential contribution to wildfire risk in any given region. Additionally, past wildfires in California from 2000-2018 were also considered as another factor for future wildfire risk. A python model was then created using the FEMA Wildfire Risk Database and the GLOBE Observer database. The two features of land cover type and historical wildfires are paired with the FEMA wildfire risk database to determine the wildfire risk. The random forest algorithm helped ensure that each decision tree in the algorithm contributes to the final prediction, with the most frequent risk level chosen as the output. This approach ensures robustness and accuracy by combining the insights from multiple trees. The model uses these features to make predictions about wildfire risk, assigning a high, moderate, or low risk level based on the patterns it learned during training. In the end, the results showed that forests were least susceptible to wildfire spread while Urban areas presented a significant threat to wildfire risk. Scientists can use this algorithm, which provides real-time, ad-hoc data to analyze the risk of wildfires at times when satellites may not accurately capture land cover imagery. Keywords: wildfire risk, remote sensing, NASA, GLOBE, land cover types