Forecasting of Monthly Streamflows Based on Artificial Neural Networks

Artificial neural networks (ANN) have experienced a major breakthrough in civil engineering topics throughout the past 15 years, especially in the hydroinformatics field. Fewer attempts have been made to unveil any feasible physical meaning behind the ANN and their probable application for solving day to day engineering problems. This work explores the possibility of linking the weights of simple multilayer perceptrons with some physical characteristics of watersheds, by means of statistical regressions. The procedure is applied to the forecast of monthly streamflows in the central region of Colombia. Nineteen watersheds were delimited within the zone of study, using geographic information system software. Obtained results allow to foresee that watersheds characteristics such as area, length, and slope of the main stream could be connected with the ANN weights. Better results are expected when daily records and other variables such as rain, evaporation, etc. be included.