Best performing ANN-based models.

The table gives the in- and output parameters, the network type (Feed-Forward Artificial Neural Network: FFW; Radial Basis Function Artificial Neural Network: RBF), the number of hidden units for FFW and number of radial basis functions for RBF, and the root-mean-squared error of the networks (RMSE) summed up for the training and test data set at convergence of the training procedure for the best performing ANN-based models as evaluated using the evaluation data set. Additionally, the coefficient of determination (r2), the y-axis intercept, and the slope (k) of the linear least-squares regression analysis between observed and predicted values computed for the combined training and test data set as well as for the spatial data set are shown (see also Figs. 1, 2).