Neural Network Downscaling to Obtain Local Precipitation Scenarios in the Italian Alps: A Case Study

Predicting local precipitation patterns over the European Alps remains an open challenge due to many limitations. The complex orography of mountainous areas modulates climate signals, and in order to analyse extremes accurately, it is essential to account for convection, requiring high-resolution climate models’ outputs. In this work, we analyse local seasonal precipitation in Trento (Laste) and Passo Tonale using high-resolution climate data and neural network downscaling. Then, we adopt an ensemble and generalized leave-one-out cross-validation procedure, which is particularly useful for the analysis of small datasets. The application of the procedure allows us to correct the model’s bias, particularly evident in Passo Tonale. This way, we will be more confident in achieving more reliable results for future projections. The analysis proceeds, considering the mean and the extreme seasonal anomalies between the projections and the reconstructions. Therefore, while a decrease in the mean summer precipitation is found in both stations, a neutral to positive variation is expected for the extremes. Such results differ from model’s, which found a clear decrease in both stations in the summer’s mean precipitation and extremes. Moreover, we find two statistically significant results for the extremes: a decrease in winter in Trento and an increase in spring in Passo Tonale.