ANN-SOM approach for satellite data pre-processing in rainfall-runoff modeling

The use of artificial neural network (ANN) models in water resource applications as rainfall-runoff modeling has grown considerably over the last decade. In order to obtain more accurate models, the qualification of applied data must be improved. Satellite data as a source of proper data in field of rainfall measurement over a watershed is utilized in this paper. Doubtlessly, spatial pre-processing methods can promote the quality of precipitation data. In the current research the self organizing map (SOM) is used for spatial pre-processing purpose. A two-level SOM neural network is applied to identify spatially homogeneous clusters of the satellite data in order to choose the most operative and effective data for the Feed-Forward Neural Network (FFNN) model which is trained by the Levenberg-Marquardt algorithm and considering only one hidden layer. The results indicate that the imposition of spatial pre-processed data to the FFNN model lead to promising evidence in the improvement of rainfall-runoff model.