Artificial Neural Network-Based Forest Fire Spread Prediction Models in the Mediterranean Climate Zone: Simulation Accuracy of Wind-Topography Interaction

Forest fires, particularly in the Mediterranean climate zone, constitute one of the most severe natural disasters threatening ecological balance and societal well-being. Accurate prediction of fire spread dynamics is critically important for the effectiveness of early warning systems and the optimization of response strategies. This study comprehensively investigates the extent to which artificial neural network (ANN)-based forest fire spread prediction models can accurately simulate the wind-topography interaction specific to the Mediterranean climate zone. Within the scope of this research, multilayer perceptron (MLP), convolutional neural networks (CNN), recurrent neural networks (RNN-LSTM), and hybrid deep learning architectures were comparatively evaluated. Model performances were analyzed using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), and structural similarity index (SSIM) metrics. The findings reveal that the CNN-LSTM hybrid model demonstrated statistically significantly superior performance in simulating wind-topography interaction compared to other architectures (R² = 0.91; RMSE = 12.4 m). However, a notable decline in model accuracy was observed on slopes exceeding 35% gradient and under conditions of sudden wind direction changes. This research contributes uniquely to the literature by elucidating the strengths and limitations of ANN-based approaches for fire spread modeling specific to Mediterranean geography.