Quantifying Fire at High Resolution across India
Led by Dr. Meghna Agarwala (Department of Environmental Studies), this project is working to improve the accuracy of satellite-based fire detection, which currently identifies less than 30% of fire events. Global fire databases, such as MODIS fire products, report even lower accuracy—under 20%—for detecting small-scale fires, leading to significant omissions.
Fires play a crucial role in local ecological processes, such as vegetation dynamics and soil erosion, as well as global climate systems, including hydrological cycles and atmospheric change. Better fire mapping is helping forest departments manage fires more effectively, but current detection methods continue to struggle with misclassification.
The research team is building on previous work with Landsat data, which increased fire detection accuracy to approximately 78% but still faced misclassification challenges. To address this, the project is applying time-series analysis and machine learning (ML) techniques to improve classification by analysing fire trends over time rather than relying on absolute pixel values. This approach is proving effective in agricultural monitoring and forest fire detection in China, demonstrating its potential for broader application.
By refining these techniques, the team is developing a robust, nationwide fire detection model for India, strengthening fire monitoring, management, and climate resilience efforts.
