Dynamic Forest Fire Spread Prediction Via CA–ML Fusion: Integrating Cellular Automata and Stacked Ensemble Learning

Dr.P.JayaPrakash1 , Gedi Sireesha2 , Balimi Sharanya3 , Buddebaigari Mahammad Yasin4 ,Mangalampat Muniprasad5

1Associate Professor Dept of Information Technology, SV College of Engineering, Tirupati, India.

2B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

3B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

4B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

5B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India.

Email:1pokalajayaprakash@gmail.com, 2sireeshasiri039@gmail.com, 3baalimisharanya123@gmail.com, 4yasinshaikh0431@gmail.com,

5pmuni815@gmail.com

Corresponding Author*: Dr.P.JayaPrakash

ABSTRACT:A Forest fire is an uncontrolled and rapidly spreading fire occurring in a forest or vegetated area, fueled by combustible materials and influenced by weather and terrain conditions. Forest fire prediction systems traditionally rely on models like FARSITE and Prometheus, which use physical and empirical equations based on fuel type, weather, and terrain. Methods such as statistical modelling, empirical formulas, and cellular automata simulate fire spread but face limitations due to complex fire dynamics, data scarcity, and poor integration of multiple factors. These models often yield high prediction errors in burned area estimation, lack dynamic adaptability, and offer limited visualization and machine learning integration. The proposed system combines Cellular Automata with the Wang Zhengfei model to predict forest fire spread in both space and time, capturing direction and speed. A stacked ensemble of XGBoost, LightGBM, and Gradient Boosting predicts the burned area, optimized through multicollinearity checks and grid search. This CA–ML fusion enables dynamic fire spread visualization and accurate impact estimation. Validation on China’s 3.29 Forest Fire and Montesinho datasets showed higher accuracy and lower errors than FARSITE and Prometheus.KEYWORDS: FARSITE and Prometheus, Cellular Automata, and Montesinho datasets, XGBoost,LightGBM, and Gradient Boosting, multicollinearity, machine learning.