A comparative study for prediction of direct runoff for a river basin using geomorphological approach and artificial neural networks

Abstract

Traditional techniques for estimation of flood using historical rainfall–runoff data are restricted in application for small basins due to poor stream gauging network. To overcome such difficulties, various techniques including those involving the morphologic details of the ungauged basin have been evolved. The geomorphologic instantaneous unit hydrograph method belongs to the latter approach. In this study, a gamma geomorphologic instantaneous unit hydrograph (GGIUH) model (based on geomorphologic characteristics of the basin and the Nash instantaneous unit hydrograph model) was calibrated and validated for prediction of direct runoff (flood) from the catchment of the Dulung-Nala (a tributary of the Subarnarekha River System) at Phekoghat station in the state of West Bengal in the eastern part of India. Sensitivity analysis revealed that a change in the model parameters viz., n, RA and RB by 1–20% resulted in the peak discharge to vary from 1.1 to 27.2%, 3.4 to 21.2% and 3.4 to 21.6%, respectively, and the runoff volume to vary from 0.3 to 12.5%, 2.1 to 2.6% and 2.2 to 2.7%, respectively. The Nash–Sutcliffe model efficiency criterion, percentage error in volume, the percentage error in peak, and net difference of observed and simulated time to peak which were used for performance evaluation, have been found to range from 74.2 to 95.1%, 2.9 to 20.9%, 0.1 to 20.8% and −1 to 3 h, respectively, indicating a good performance of the GGIUH model for prediction of runoff hydrograph. Again, an artificial neural network (ANN) model was prepared to predict ordinates of discharge hydrograph using calibrative approach. Both the ANN and GGIUH models were found to have predicted the hydrograph characteristics in a satisfactory manner. Further, direct surface runoff hydrographs computed using the GGIUH model at two map scales (viz. 1:50,000 and 1:250,000) were found to yield comparable results for the two map scales. For a final clarification, the probability density function of the actual and predicted data from the two models was prepared to compare the pattern identification ability of both the models. The GGIUH model was found to identify the distribution pattern better than the ANN model, although both the models were found to be ably replicating the data patterns of the observed dataset.