Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FUELVISION: A Multimodal Data Fusion and Multimodel Ensemble Algorithm for Wildfire Fuels Mapping

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

TL;DR

This work addresses the need for real-time, landscape-scale wildfire fuel mapping by developing FuelVision, a multimodal data fusion and multi-model ensemble framework that integrates Landsat-8 imagery, SAR data (Sentinel-1 and PALSAR), and terrain features with FIA plot labels. It introduces label propagation and synthetic data augmentation (CTGAN) to overcome limited ground-truth data, and employs a layered AutoML stacking ensemble to predict the Scott and Burgan 40 fuel models at 30 m resolution. The approach yields an overall test accuracy of 0.771 on California data and demonstrates robust results in Dixie and Caldor fire case studies, with uncertainty quantified via prediction probabilities. The combination of multimodal fusion, pseudo-labeling, and synthetic data augmentation provides a scalable path toward on-demand, near-real-time fuel maps that can improve fire risk assessment and response, supported by validation against NAIP and timber-harvest datasets.

Abstract

Accurate assessment of fuel conditions is a prerequisite for fire ignition and behavior prediction, and risk management. The method proposed herein leverages diverse data sources including Landsat-8 optical imagery, Sentinel-1 (C-band) Synthetic Aperture Radar (SAR) imagery, PALSAR (L-band) SAR imagery, and terrain features to capture comprehensive information about fuel types and distributions. An ensemble model was trained to predict landscape-scale fuels such as the 'Scott and Burgan 40' using the as-received Forest Inventory and Analysis (FIA) field survey plot data obtained from the USDA Forest Service. However, this basic approach yielded relatively poor results due to the inadequate amount of training data. Pseudo-labeled and fully synthetic datasets were developed using generative AI approaches to address the limitations of ground truth data availability. These synthetic datasets were used for augmenting the FIA data from California to enhance the robustness and coverage of model training. The use of an ensemble of methods including deep learning neural networks, decision trees, and gradient boosting offered a fuel mapping accuracy of nearly 80\%. Through extensive experimentation and evaluation, the effectiveness of the proposed approach was validated for regions of the 2021 Dixie and Caldor fires. Comparative analyses against high-resolution data from the National Agriculture Imagery Program (NAIP) and timber harvest maps affirmed the robustness and reliability of the proposed approach, which is capable of near-real-time fuel mapping.

FUELVISION: A Multimodal Data Fusion and Multimodel Ensemble Algorithm for Wildfire Fuels Mapping

TL;DR

This work addresses the need for real-time, landscape-scale wildfire fuel mapping by developing FuelVision, a multimodal data fusion and multi-model ensemble framework that integrates Landsat-8 imagery, SAR data (Sentinel-1 and PALSAR), and terrain features with FIA plot labels. It introduces label propagation and synthetic data augmentation (CTGAN) to overcome limited ground-truth data, and employs a layered AutoML stacking ensemble to predict the Scott and Burgan 40 fuel models at 30 m resolution. The approach yields an overall test accuracy of 0.771 on California data and demonstrates robust results in Dixie and Caldor fire case studies, with uncertainty quantified via prediction probabilities. The combination of multimodal fusion, pseudo-labeling, and synthetic data augmentation provides a scalable path toward on-demand, near-real-time fuel maps that can improve fire risk assessment and response, supported by validation against NAIP and timber-harvest datasets.

Abstract

Accurate assessment of fuel conditions is a prerequisite for fire ignition and behavior prediction, and risk management. The method proposed herein leverages diverse data sources including Landsat-8 optical imagery, Sentinel-1 (C-band) Synthetic Aperture Radar (SAR) imagery, PALSAR (L-band) SAR imagery, and terrain features to capture comprehensive information about fuel types and distributions. An ensemble model was trained to predict landscape-scale fuels such as the 'Scott and Burgan 40' using the as-received Forest Inventory and Analysis (FIA) field survey plot data obtained from the USDA Forest Service. However, this basic approach yielded relatively poor results due to the inadequate amount of training data. Pseudo-labeled and fully synthetic datasets were developed using generative AI approaches to address the limitations of ground truth data availability. These synthetic datasets were used for augmenting the FIA data from California to enhance the robustness and coverage of model training. The use of an ensemble of methods including deep learning neural networks, decision trees, and gradient boosting offered a fuel mapping accuracy of nearly 80\%. Through extensive experimentation and evaluation, the effectiveness of the proposed approach was validated for regions of the 2021 Dixie and Caldor fires. Comparative analyses against high-resolution data from the National Agriculture Imagery Program (NAIP) and timber harvest maps affirmed the robustness and reliability of the proposed approach, which is capable of near-real-time fuel mapping.
Paper Structure (21 sections, 5 equations, 14 figures, 7 tables)

This paper contains 21 sections, 5 equations, 14 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Research Significance
  4. Study Area and Data Used
  5. Region of Analysis
  6. FIA and its plots analysis
  7. Remote Sensing Data
  8. Methodology
  9. Label Propagation
  10. Synthetic Data Generation
  11. Importance of inclusive features
  12. Ensemble Model Training
  13. Non-Burnables Mapping: Post-Processing Step
  14. Results & Discussions
  15. Evaluation of Data Augmentation Techniques
  16. ...and 6 more sections

Figures (14)

  • Figure 1: FIA Plots with assigned fuel models in the study area of California (locations here are approximate to maintain FIA spatial confidentiality).
  • Figure 2: Number of FIA plots with fuel data assignment per year
  • Figure 3: Number of Fuel Models assignment in FIA plots for California
  • Figure 4: FuelVision Framework
  • Figure 5: Pseudo-Labelling using Jeffries-Matusita-Spectral Angle Mapping Scores
  • ...and 9 more figures