Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MFiSP: A Multimodal Fire Spread Prediction Framework

Alec Sathiyamoorthy, Wenhao Zhou, Xiangmin Zhou, Xiaodong Li, Iqbal Gondal

TL;DR

This study addresses the need for accurate, timely wildfire spread forecasts during large-scale events like the Black Summer fires, where traditional FBAn-based methods and static inputs limit predictive power. It introduces MFiSP, a multimodal framework that fuses geotagged social media signals and remote-sensing observations, and employs Monte Carlo simulations with fuel-map ensembles to forecast perimeters while adaptively aligning with the observed Rate of Spread ($ROS$) between assimilation cycles. Key contributions include formalizing fuel-load manipulation, a probabilistic data-assimilation pipeline robust to noisy inputs, and a rate-of-spread manipulation step that selects the best perturbed fuel map via the $Jaccard$ similarity to observed perimeters, improving forecast accuracy in synthetic tests. The results show that multimodal data fusion consistently outperforms FBAn-based baselines, particularly in rapidly evolving fire regimes, indicating tangible benefits for resource allocation and evacuation planning; future work will apply MFiSP to real wildfire datasets.

Abstract

The 2019-2020 Black Summer bushfires in Australia devastated 19 million hectares, destroyed 3,000 homes, and lasted seven months, demonstrating the escalating scale and urgency of wildfire threats requiring better forecasting for effective response. Traditional fire modeling relies on manual interpretation by Fire Behaviour Analysts (FBAns) and static environmental data, often leading to inaccuracies and operational limitations. Emerging data sources, such as NASA's FIRMS satellite imagery and Volunteered Geographic Information, offer potential improvements by enabling dynamic fire spread prediction. This study proposes a Multimodal Fire Spread Prediction Framework (MFiSP) that integrates social media data and remote sensing observations to enhance forecast accuracy. By adapting fuel map manipulation strategies between assimilation cycles, the framework dynamically adjusts fire behavior predictions to align with the observed rate of spread. We evaluate the efficacy of MFiSP using synthetically generated fire event polygons across multiple scenarios, analyzing individual and combined impacts on forecast perimeters. Results suggest that our MFiSP integrating multimodal data can improve fire spread prediction beyond conventional methods reliant on FBAn expertise and static inputs.

MFiSP: A Multimodal Fire Spread Prediction Framework

TL;DR

This study addresses the need for accurate, timely wildfire spread forecasts during large-scale events like the Black Summer fires, where traditional FBAn-based methods and static inputs limit predictive power. It introduces MFiSP, a multimodal framework that fuses geotagged social media signals and remote-sensing observations, and employs Monte Carlo simulations with fuel-map ensembles to forecast perimeters while adaptively aligning with the observed Rate of Spread () between assimilation cycles. Key contributions include formalizing fuel-load manipulation, a probabilistic data-assimilation pipeline robust to noisy inputs, and a rate-of-spread manipulation step that selects the best perturbed fuel map via the similarity to observed perimeters, improving forecast accuracy in synthetic tests. The results show that multimodal data fusion consistently outperforms FBAn-based baselines, particularly in rapidly evolving fire regimes, indicating tangible benefits for resource allocation and evacuation planning; future work will apply MFiSP to real wildfire datasets.

Abstract

The 2019-2020 Black Summer bushfires in Australia devastated 19 million hectares, destroyed 3,000 homes, and lasted seven months, demonstrating the escalating scale and urgency of wildfire threats requiring better forecasting for effective response. Traditional fire modeling relies on manual interpretation by Fire Behaviour Analysts (FBAns) and static environmental data, often leading to inaccuracies and operational limitations. Emerging data sources, such as NASA's FIRMS satellite imagery and Volunteered Geographic Information, offer potential improvements by enabling dynamic fire spread prediction. This study proposes a Multimodal Fire Spread Prediction Framework (MFiSP) that integrates social media data and remote sensing observations to enhance forecast accuracy. By adapting fuel map manipulation strategies between assimilation cycles, the framework dynamically adjusts fire behavior predictions to align with the observed rate of spread. We evaluate the efficacy of MFiSP using synthetically generated fire event polygons across multiple scenarios, analyzing individual and combined impacts on forecast perimeters. Results suggest that our MFiSP integrating multimodal data can improve fire spread prediction beyond conventional methods reliant on FBAn expertise and static inputs.

Paper Structure

This paper contains 14 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. MFiSP: Multi-source Fire Spread Prediction
  4. Preliminary
  5. Data Assimilation
  6. Fire Spread Prediction
  7. Rate of Spread Manipulation
  8. Experimental Evaluation
  9. Data Collection
  10. Fire Perimeters
  11. Social Media Dataset
  12. Evaluation Methodology
  13. Experimental Results
  14. Conclusion

Figures (2)

  • Figure 1: High-level overview of the Multimodal Fire Spread Prediction framework.
  • Figure 2: Predicted perimeters compared to their actual fire perimeters during hours 4 to 6 of the explosive case. 4th hour of comparisons (left), 5th hour of comparisons (middle), 6th hour of comparisons (right).