Regional scale wildfire occurrence forecasting through AI techniquesAlemayohu, Samuel Brhane (Avtor) Škerjanec, Mateja (Mentor) Senatore, Alfonso (Komentor) De Rango, Alessio (Komentor) master thesiswildfire forecastingCalabrialinear regressionmachine learningdeep learningAI in environmental engineeringTransformerWildfires are a growing threat in the Mediterranean region. This thesis focuses on daily wildfire prediction in Calabria, southern Italy, by comparing five modeling techniques: Linear Regression, Gaussian Process Regression (GPR), Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and Transformer. The analysis is based on a spatiotemporal dataset prepared at a 100-meter grid resolution, by combining clustered wildfire aggregations and daily climate variables for eight zones and the entire region. All models were calibrated, validated, and evaluated using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R²), residual diagnostics, and Exact-Match percentage. At the zone level, the most stable model was XGBoost, which topped the rankings with up to 86% Exact-Match rates and up to 0.595 R² values. Linear Regression produced the highest zone-level performance for Zone 1, showing strong linear relationships between climate and fire in that area. The deep learning models (LSTM and Transformer) offered no improvement over the simpler methods, and GPR generally produced lower accuracy. At the regional level, GPR produced the highest overall regional fit with an R² of 0.836 and an RMSE of 2.41 fires/day, and XGBoost produced the highest daily-level accuracy with a MAE of 1.42 fires/day and 41.3% Exact-Match rate. Results show that tree-based ensemble classifiers with lagged features offer the most reliable choice for wildfire forecasting in Calabria, making highly accurate predictions, remaining easy to interpret, and being computationally efficient. Careful feature engineering reduces the need for complex models and ensures that simpler approaches remain competitive.[S. B. Alemayohu]20252025-10-04 08:45:04Magistrsko delo/naloga174575UDK: 614.841.42:004.896(450.78)(043.2)VisID: 175579COBISS_ID: 251910915sl