WEATHER FORECAST DOWNSCALING FOR APPLICATIONS IN SMART AGRICULTURE AND PRECISION FARMING USING ARTIFICIAL NEURAL NETWORKS

This study proposes an Artificial Neural Network (ANN) algorithm for downscaling weather forecasts of some variables useful for agriculture in Southern Italy. Using the Weather Research and Forecasting (WRF) model at 1.2 km spatial resolution, the algorithm performs downscaling at 240 m resolution using an operation similar to bilinear interpolation, but with enhanced performance. To train the ANNs, a database was built using the WRF model in Large Eddy Simulation (LES) mode with 240 m grid spacing. Particular attention was paid to defining the architecture of the ANNs and selecting the inputs. The comparison of the algorithm's performance against spline interpolation shows a reduction of the mean squared error (MSE) ranging from a minimum of 6% for solar irradiance to a maximum of 87% for surface pressure.