A novel study of SWAT and ANN models for runoff simulation with application on dataset of metrological stations

Abstract

Rainfall-runoff simulation is one of the most important processes in flood simulation, especially in the watersheds (Darake watershed) located upstream of large cities (populated city of Tehran) and exposed to floods. The study used SWAT and ANN models to simulate rainfall-runoff. The reasons for selecting these two models are A) SWAT is a physical and complex model that needs precipitation, temperature, wind speed, relative humidity, sundial, soil map, land use map and DEM to simulate, B) ANN model is a simple linear model that needs precipitation, runoff and temperature data. Moreover, SWAT model needs more time and cost than ANN model. In general, the purpose of the study is to evaluate the performance of two models with different structure in urban watershed. The results of this research showed that the performance of the artificial neural network is appropriate for predicting the maximum and minimum runoff values (R2 = 0.75, NSE = 0.74), while the inputs of the model is appropriate and in areas where information is scarce is very appropriate, while the performance of the SWAT model is also appropriate and has very good performance (R2 = 0.66, NSE = 0.65) in managerial and planning and economics studies, especially when there is no statistical station in the upper watershed. The SWAT model can properly simulate. It is better to use the SWAT model in the studies that are related to the flow trend.