As climate change and urbanization increase wildfire risks for around 115 million people, traditional models like Rothermel’s are inadequate for predicting fire propagation. To tackle this, I created a three-dimensional cellular automaton (CA) wildfire model that simulates fire spread using: (1) landscape generation from satellite imagery and elevation data; (2) a CA with a graph neural network for heat transfer and ember transport; and (3) a Bayesian optimization model to locate high-risk zones for fire suppression. Applied to the 2020 Bobcat Fire in Los Angeles, the model replicated mid-stage and final burn areas with 93.2% and 89.9% accuracy, respectively, reducing the burn area by 26.6%. This approach offers a valuable tool for predicting wildfire behavior and optimizing response strategies.