BLAZE TRACK: Developing Dynamic Wildfire Risk Maps Leveraging MODIS Satellite Imagery and Advanced AI Algorithms

Wildfires pose pervasive threats to communities, causing loss of lives, disrupting critical infrastructure services, and economic loss. Therefore, identifying accurate wildfire risk hotspots at granular spatiotemporal resolution and predicting the future risk for wildfires is crucial for preventive measures. To that end, we propose to develop a geoAI-based framework, BLAZE TRACK, that leverages satellite imagery data from the Moderate Resolution Imaging Spectroradiometer (MODIS). The satellite data provides information on the historical burnt areas from 2015-2022 at a weekly temporal scale where every pixel (spatial dimension 500m*500m) is labeled as burned or unburned. This data is integrated with data on vegetation, topography, elevation, and climate data. Leveraging advanced machine learning classifiers like Random Forest, AI algorithms like self-attention long short-term memory (LSTM), a dynamic risk mapping is developed for the state of California, most susceptible to wildfires due to its unique weather, topography, and vegetation cover. Our results show that the dynamic wildfire risk prediction map is ~95% accurate. Our analysis also identifies the key predictors (e.g., climate, topographical, and vegetation factors) associated with a higher wildfire risk. The high, low, and moderate wildfire risk zones and their seasonality can be detected by leveraging our proposed BLAZE TRACK, which better informs emergency managers of proactive, risk-informed preparedness measures. The analysis can be further expanded to understand how the wildfire risk varies for low-income communities by integrating socioeconomic information of the regions into the modeling framework to formulate humanitarian policy directives.