Estimation of Downed Woody Time-Lag Fuel Loadings with Multimodal Remote Sensing Data and Ensemble Machine Learning Regression Model

Riyaaz Uddien Shaik, Mohamad Alipour, Eric Rowell, Bharathan Balaji, Adam C. Watts, Ertugrul Taciroglu

Abstract: Accurate fuel condition assessment is crucial for predicting fire behavior, enhancing operational decision support, and improving overall fire management. Our approach utilizes diverse data sources, such as Landsat-8 optical imagery, Sentinel-1 (C-band) SAR imagery, PALSAR (L-band) SAR imagery, and terrain features, to estimate time-lag fuel loadings (1 hour, 10 hours, and 100 hours). Optical data mainly captures the characteristics of leaf and forest canopy, while SAR data is more sensitive to forest vertical structures due to its strong penetrability. An ensemble model was trained on the Forest Inventory and Analysis (FIA) plots and spectral indices. Followed by, feature importance analysis and the inclusion of polynomial features were undertaken. The ensemble strategy, involving neural networks, decision trees, gradient boosting, and ensemble methods, achieved R 2 values of 0.72, 0.70, and 0.60 for 1-hour, 10-hour, and 100-hour fuel loads. Extensive experimentation in the 2021 Dixie Fire incident validates the effectiveness of our approach, emphasizing the value of leveraging multimodal data and ensemble machine learning models for real-time fuel load estimation.