Prediction of monthly flows in rivers of high Andean basins with an artificial neural network approach. Case: Crisnejas river, Peru

Abstract

Predicting the hydrological behavior in hydrographic basins composed of high Andean ecosystems that have a variety of climates, with complex geology, highly varied topography, and soils with a high content of organic matter that generate a very heterogeneous vegetation cover, is very difficult, and if it is added the scarcity of hydrometric information in hydrographic networks causes great uncertainty when planning the use of water resources. The predominant trend for prediction is through hydrological models that relate precipitation and runoff, which require historical information that is not available in most cases. The application of the artificial neural networks technique allows a methodology adaptable to the information available in each basin to analyze the relationship between precipitation and runoff. Because of its robustness, results can be obtained with great precision. This research aimed to estimate and predict the average monthly flows for the Crisnejas river basin, located in the northern region of the Peruvian Andes, for which there were historical records of 12 meteorological stations and a hydrometric station, using flow data, precipitation, temperature and normalized difference vegetation index (NDVI), with a multilayer perceptron-type artificial neural network, which achieved a goodness of fit of 81 % in the coefficient of determination. Then with the generated record, another network of the recurrent type was trained to predict monthly mean flows for eight years with a goodness of fit of 71 %.