Investigating the role of saliency analysis with a neural network rainfall-runoff model

Software tools are available which translate neural network solutions into standard computer languages and source code. This conversion process enables trained networks to be implemented as embedded functions within existing hydrological models or assembled into stand-alone computer programs. In addition to this primary use, embedded functions can also provide new opportunities for dynamic testing and for the internal investigation of the model's function. Saliency analysis, the disaggregation of a neural network solution in terms of its forecasting inputs, is one approach which is explored here. Saliency analysis is used to investigate the performance of a neural network one-step-ahead hydrological forecasting model using different combinations of input data for testing and validation. (C) 2001 Elsevier Science Ltd. All rights reserved.