Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Leveraging Cellular Automata for Real-Time Wildfire Spread Modeling in California

Connor Weinhouse, Jameson Augustin

TL;DR

This study develops a lightweight cellular automata model for real-time wildfire spread in California by integrating NDVI-derived vegetation, wind, and terrain slope, and validates it against the Pacific Palisades 2025 burn scar using a dNBR-based burn mask. The approach achieves a recall of 0.860, precision of 0.605, and an F1 score of 0.711 after 50 parameter-optimization trials, with simulations averaging 1.22 seconds. The results demonstrate that a simple CA framework can deliver actionable, open-access data-driven fire-spread predictions suitable for emergency response, while highlighting trade-offs such as overprediction and simplified meteorology. The work argues for operational viability as a priority in wildfire forecasting and envisions integration with existing systems to support rapid evacuations and resource allocation. Future work should incorporate dynamic weather and firefighting inputs to further improve accuracy without sacrificing real-time performance.

Abstract

Wildfires are becoming increasingly frequent and devastating, and therefore the technology to combat them must adapt accordingly. Modern predictive models have failed to balance predictive accuracy and operational viability, resulting in consistently delayed or misinformed fire suppression and public safety efforts. The present study addresses this gap by developing and validating a predictive model based on cellular automata (CA) that incorporates key environmental variables, including vegetation density (NDVI), wind speed and direction, and topographic slope derived from open-access datasets. The presented CA framework offers a lightweight alternative to data-heavy approaches that fail in emergency contexts. Evaluation of the model using a confusion matrix against burn scars from the 2025 Pacific Palisades Fire yielded a recall of 0.860, a precision of 0.605, and an overall F1 score of 0.711 after 50 parameter optimization trials, with each simulation taking an average of 1.22 seconds. CA-based models can bridge the gap between accuracy and applicability, successfully guiding public safety and fire suppression efforts.

Leveraging Cellular Automata for Real-Time Wildfire Spread Modeling in California

TL;DR

This study develops a lightweight cellular automata model for real-time wildfire spread in California by integrating NDVI-derived vegetation, wind, and terrain slope, and validates it against the Pacific Palisades 2025 burn scar using a dNBR-based burn mask. The approach achieves a recall of 0.860, precision of 0.605, and an F1 score of 0.711 after 50 parameter-optimization trials, with simulations averaging 1.22 seconds. The results demonstrate that a simple CA framework can deliver actionable, open-access data-driven fire-spread predictions suitable for emergency response, while highlighting trade-offs such as overprediction and simplified meteorology. The work argues for operational viability as a priority in wildfire forecasting and envisions integration with existing systems to support rapid evacuations and resource allocation. Future work should incorporate dynamic weather and firefighting inputs to further improve accuracy without sacrificing real-time performance.

Abstract

Wildfires are becoming increasingly frequent and devastating, and therefore the technology to combat them must adapt accordingly. Modern predictive models have failed to balance predictive accuracy and operational viability, resulting in consistently delayed or misinformed fire suppression and public safety efforts. The present study addresses this gap by developing and validating a predictive model based on cellular automata (CA) that incorporates key environmental variables, including vegetation density (NDVI), wind speed and direction, and topographic slope derived from open-access datasets. The presented CA framework offers a lightweight alternative to data-heavy approaches that fail in emergency contexts. Evaluation of the model using a confusion matrix against burn scars from the 2025 Pacific Palisades Fire yielded a recall of 0.860, a precision of 0.605, and an overall F1 score of 0.711 after 50 parameter optimization trials, with each simulation taking an average of 1.22 seconds. CA-based models can bridge the gap between accuracy and applicability, successfully guiding public safety and fire suppression efforts.

Paper Structure

This paper contains 33 sections, 6 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Data
  4. Data Sources
  5. Study Area
  6. Satellite Imagery
  7. Sentinel-2
  8. VIIRS
  9. Meteorological Data
  10. Topographic Data
  11. Methods
  12. Burnt Area Mapping
  13. Normalized Burn Ratio (NBR)
  14. Differenced Normalized Burn Ratio (dNBR)
  15. Preprocessing
  16. ...and 18 more sections

Figures (4)

  • Figure 1: The Palisades Fire Footprint
  • Figure 2: RGB and NBR imagery before and after the Pacific Palisades fire
  • Figure 3: dNBR and burn mask results depicting the range and severity of the fire
  • Figure 4: Evaluation of the Palisades fire simulation