Evaluation of GFF and RBF Neural Network Models and Soil Moisture Accounting Algorithm for HEC-HMS Model in Continuous Semi-Distributed Rainfall-Runoff Simulation in Jarahi Basin

Abstract

Runoff estimation is effective way in utilization and allocation of water resources for various agricultural, drinking, hydraulic and environmental sectors. In this paper, continuous simulation of rainfall-runoff in the basin with two different models including HEC-HMS conceptual model and data-processing model (artificial neural networks) are considered to evaluate the ability and accuracy of these two models in estimating runoff. The continuous flow simulation was used to calculate soil moisture losses (SMA) in sub-basins. For calibration of the model, daily precipitation, flow, evapotranspiration data from 2001 to 2007 were used and for model accuracy period of 2008 to 2011 were used. The results showed that the HEC-HMS model, along with the SMA model, has a good ability to continuously simulations in dry and continuous periods in the basin. In order to select the input variables that affect the flow rate in artificial neural networks, a generalized feeder grid (GFF) and a radial base function grid (RBF), partial interpolation algorithm (PMI) was used. The results of using the PMI algorithm showed that the input variable influences the flow velocity at the Gargar hydrometric station, the current day flow rate at this station. The results showed that the GFF network has more efficiency and accuracy than the conceptual model of HEC-HMS and RBF network in continuous run-run run simulation in the basin. The Nash coefficient for HEC-HMS and GFF and RBF networks is 0.6, 0.6677 and 0.6676 respectively