Figure 6
From: Harnessing deep learning to forecast local microclimate using global climate data
Assessment of importance of input variables. The figures provide insights into the analysis of input variable importance for the four feed forward neural networks. Each figure follows a consistent structure, with the input physical variables displayed on the Y-axis and two subplots (Swr is the shortwave radiation, Lwr is the longwave radiation, Ws is the wind speed, Tr is the reference temperature, Hr is the relative humidity, Prc is the precipitation, P is the pressure, Cc is the cloud coverm, Hs is the specific humidity). The left subplot shows the variation in \(R^2\) values, while the right subplot presents the variation in MAE values. The bars represent the average variation across 10 simulations, and the error bars indicate the relative standard deviation of the mean. (a) The figure presents the results for the feed forward neural network trained on the ERA5 database for temperature prediction. (b) The figure showcases the results for the feed forward neural network trained on the ARPA station recordings for temperature prediction. (c) The figure displays the results for the feed forward neural network trained on the ERA5 database for relative humidity prediction. Finally, (d) the figure presents the results for the feed forward neural network trained on the ARPA station recordings for relative humidity prediction.
