Research Symposium

BIO

Jonathan Martin, from Panama City, Florida, has research interests in the field of artificial intelligence, particularly exploring how it can be used to create positive impact. Jonathan's career goal is to become an established software engineer and use those skills to build technology that impacts people’s lives at scale.

Deep Learning–Based Surrogate Modeling of JULES-INFERNO for Accelerated Global Wildfire Prediction

Abstract

Deep learning methods are increasingly being used to accelerate computationally expensive environmental simulations. This study explores the development of deep learning surrogate models to approximate the wildfire prediction capabilities of the JULES-INFERNO model. While JULES-INFERNO provides detailed simulations of wildfire occurrence and burned area dynamics, its high computational cost limits accessibility and makes real time forecasting difficult. To address this challenge, this research develops data driven models that learn the relationship between climate variables and predicted burned area. Model performance will be evaluated using Absolute Error per Pixel (AEP) and the Structural Similarity Index Measure (SSIM) to assess both numerical accuracy and spatial consistency. Building on prior work, this study expands the input space by incorporating additional climate variables including precipitation and humidity to improve model sensitivity and generalization across different climate conditions. By reducing simulation time from hours to seconds while maintaining strong predictive accuracy, this research contributes toward the development of faster and more accessible global wildfire forecasting tools.

Keywords: Artificial Intelligence , Machine Learning, Wildfires