Application of artificial neural network in stage-discharge relationship

Abstract

ABSTRACT: The prediction of discharge and its variability in a river is an essential componentof surface-water planning. For that purpose, a functional relationship between stage and dischargeis established with the help of field measurement and the relationship is expressed as a ratingcurve. Stage and discharge are time-dependent and very often they exhibit random fluctuations,their relationship is not always unique. One of the solutions is using more observations fromseveral previous time steps and to build more complex data-driven model; in the present paper theapplication of an artificial neural network (ANN) is considered. Data from the dischargemeasuring station at Swarupgunj, India is used for this purpose. Comparison to a conventionalstatistical stage-discharge model show the superiority of an approach using ANN.1 INTRODUCTIONThe prediction of discharge and its variabilityin a river is an essential component of surface-water planning. For this purpose, a functionalrelationship between stage and discharge isoften established with the help of fieldmeasurements and the relationship isexpressed as a rating curve. Unfortunately, thefunctional relationship between stage anddischarge very often is not unique. Often thislimits the practical usefulness of rating curves.In a situation when a considerable amountof data about the studied process is available(as it is in the considered case), the use ofsimplified traditional techniques, like arating curve, may hardly be justified. In orderto achieve maximum result it is important touse all available data. One of the options is touse regression- and auto-correlation-basedstatistical methods like ARIMA models (Boxand Jenkins 1976). However, with the recentadvancements in artificial intelligence, datamining and soft computing, there is a choice ofbetter techniques that may help to solve theproblem. One of them is