Future trends of reference evapotranspiration in Sicily based on CORDEX data and Machine Learning algorithms

In years of increasing impact of climate change effects, a reliable characterization of the spatiotemporal evolutionary dynamics of evapotranspiration can enable a significant improvement in water resource management, especially as regards irrigation activities. Sicily, an insular region of Southern Italy, has exceptionally valuable agricultural production and high irrigation needs. In this study, the ETo reference evapotranspiration in Sicily was first evaluated on the basis of historical and future climate parameters, referring for future values to two climate scenarios characterized by different Representative Concentration Pathways: RCP 4.5 and RCP 8.5. Then, the Hierarchical algorithm was used to divide Sicily into three homogeneous regions, each characterized by specific ETo features. In addition, some Machine Learning (ML) algorithms were used to develop forecasting models based on only historical data. Support Vector Regression (SVR) was used to predict the future values of Tmin and Tmax, while an ensemble model based on Multilayer Perceptron (MLP) and M5P Regression Tree was developed for the ETo forecasting. Predictions made with the ensemble MLP-M5P model were compared with the ETo computed for the RCP 4.5 and RCP 8.5 future climate scenarios. During the forecast period, from 2001 to 2091, evapotranspiration increases were observed for all three clusters. For cluster C1, along the coast, percentage increases of 7.52%, 14.64% and 10.78%, were computed for RCP 4.5, RCP 8.5, and MLP-M5P, respectively, while, for cluster C3, in the inland, percentage increases were higher and equal to 8.12%, 16.71%, and 14.98%, respectively. The ensemble MLP-M5P model led to intermediate trends between RCP 4.5 and RCP 8.5, showing a high correlation with the latter (R2 between 0.93 and 0.98). The developed approach, based on both clustering and forecasting algorithms, provided a comprehensive analysis of the reference evapotranspiration, with the detection of the different homogeneous regions and, at the same time, the evaluation of the evapotranspiration trends, both in coastal and inland areas.

Automated summary

The study evaluated the reference evapotranspiration in Sicily using historical and future climate parameters, and divided the region into three homogeneous areas using a hierarchical algorithm. Machine learning algorithms were then used to forecast future evapotranspiration. The results showed that evapotranspiration increased for all three regions during the forecast period, with higher increases observed in the inland areas. This approach provides a comprehensive analysis of evapotranspiration trends in different regions.