Use of neural network to predict flash flood and attendant water qualities of a mountainous stream on Oahu, Hawaii

Abstract

Frequent flash floods of Hawaii streams pose continuous threats to the coastal environment because the streams respond rapidly to high runoff and huge transport quantities of sediments, to which are sorbed nutrients, heavy metals, and persistent hydrophobic organic compounds. High-frequency stream flow and water quality estimation are essential to correctly assess water quality variations and pollutant loads during flash floods, because stream flow and turbidity in Hawaii can change by a factor of 60 and 30, respectively, in 15 min. This study shows the application of artificial neural networks (ANNs) to assess flash floods and their attendant water quality parameters using measured data of a Hawaii stream. The paper illustrates that ANNs predict stream flow with a correlation coefficient (R) greater than 0.99 and turbidity and specific conductance with R-values greater than 0.80. Although the R-values for the estimation of dissolved oxygen, pH, and water temperature were somewhat low, most of the estimated stream water quality values (turbidity, specific conductance, dissolved oxygen, pH, and water temperature) were within the limits of ±30% deviations of the 1:1 line. The R-value for the estimation of stream water qualities could have been significantly improved if high resolution (at 15 min or lower measurement frequency), noise-free, and continuous data were available for a longer period of time. The paper demonstrates that the upstream water quality parameters depend on weather forces and land use of the watershed and the downstream water quality parameters additionally influenced by oceanic tides. Stream stage is found to be an important input parameter for stream flow prediction using ANN; however, the predictive performance of ANN for the estimation of stream flow is improved if weather data, rainfall, and evapotranspiration are included in the input data set.