Use of artificial neural network for predicting stage-discharge relationship and water quality parameters for selected Hawaii streams

Abstract

The goal of the study is to examine the efficacy of the artificial neural network (ANN) to develop stage-discharge relationship and to simulate water quality parameters. With the data from the United States Geological Survey (USGS) for two Hawaii stream gaging stations: the Manoa Stream at Kanewai field and the Waiakeakua Stream at Honolulu, ANN will be used to predict discharge for those two stations, but only Manoa Stream at Kanewai Field will be analyzed for the water quality parameters such as dissolved oxygen, dissolved organic carbon, solids residue and suspended sediment. For both stations, the performance of ANN is superior to the rating curves currently used by USGS. The network with one or two hidden layers does not make significant difference for modeling those two rating curves, but it was found that the selection of the test data set is very important. For simulating water quality parameters, the network fails to learn the relation between input and target due to the insufficient input parameter and short length of record. For the station of Waiakeakua Stream at Honolulu, the most important input parameters are hydraulic radius and conveyance of the cross section where the discharge was measured. But for Manoa Stream at Kanewai Field, it is one of the antecedent gage heights (H(t-2)) that contributes significantly to the network performance.