Development of a Data-Driven Model for the Estimate of Drought Risk for Agriculture

This thesis draws inspiration from the European Drought Risk Atlas to quantify the impacts of agricultural drought on maize production in Italy. The goal is to develop a preliminary predictive model that can be refined and expanded in future research. Historical climatic data, including temperature and precipitation, were sourced from the VHR-REA_IT dataset, while projections under the RCP 8.5 scenario (2030–2070) were obtained from the VHR-PRO_IT dataset. The provincial-level maize cultivation data, sourced from ISTAT, focused on Northern Italy, where sufficient agricultural records were available. A regression analysis was conducted to assess yield anomalies, with five regression models implemented, optimized, and evaluated using various metrics. Although overall model performance leaves room for improvement, the Extra Trees Regressor was identified as the most reliable model based on its performance on the baseline data and its consistency with existing literature. This model was then used to make future predictions under the RCP 8.5 scenario. The average yearly yield anomaly was calculated for 2040 and 2060, with results indicating an average annual production loss between 0% and 6%, aligning with findings from the European Drought Risk Atlas. While the model is still a prototype and requires further refinement, particularly with the addition of more suitable features, it provides a solid foundation for future development of drought risk estimation tools in agricultural settings.