Neural networks application to Neretva basin hydro-meteorological data

Abstract

Neural networks application to the analysis and prediction of the hydro-meteorological data is presented. The neural networks are trained and tested with water-level and water-flow data measured at three stations in the Neretva river basin. Estimation of the water-level based on water-flow and vice versa is presented. These data are highly (byt nonlineary) correlated. The proposed approach can be used to reconstruct missed measurements caused, for example, by measurement equipment failure. In this way an accurate and complete set of measurements can be obtained. Estimation of downstream measurements based on upstream data is also analysed. It is shown that highly accurate estimations can be obtained when there is no tributaries between measurement stations.