The present invention relates to a method and apparatus, based on the use of a neural network (NN), for (a) predicting important groundwater/surface water output/state variables, (b) optimizing groundwater/surface water control variables, and/or (c) sensitivity analysis, to identify physical relationships between input and output/state variables used to model the groundwater/surface water system or to analyze the performance parameters of the neural network.