Comparison of the SWAT Model and ANN for daily simulation of runoff in snowbound ungauged catchments

摘要

Snowy catchments are one of the main sources of fresh water. Reliable estimation of daily flows is needed for water resources management of such catchments. However, snowy catchments are often poorly gauged with short-term data. Thus models with less dependency on long-term and requiring few number of data are needed. The conceptual models and artificial neural networks are the two model structures that are more suitable to cope up with the short-term and scarce data. In this paper, the conceptual SWAT model with two different snow modules, one is the original snow module of the version 98.1 of this model and other is derived from SNOW17 Anderson method (SWAT_(SNOW17)) have been compared with a feed forward artificial neural networks model (ANNs) for simulation of daily flow on a snowbound terrain in Iran. Despite the differences in model structures, the comparison of results obtained from these models shows almost similar performance in general. However, it has been observed that the ANNs perform better for low flows simulation, whereas the SWAT_(SNOW17) performs better for high flows simulation. In case of ANNs, less dependency was found on types of data. It has been able to simulate daily flows with only observed daily maximum and minimum temperature and estimated daily solar radiation. On the other hand, in case of the SWAT model, after successful calibration of the model, there are many facilities that can be gained from it as well as applying the calibrated model for another homogenous catchments.