Assessment of flooding in future periods using the flow of the watershed (Case study: west and south of the Urmia watershed)

Abstract

Prediction of streamflow is a crucial tool in planning and managing water resources and preventing floods. Due to the recent drought in Urmia Lake, predicting streamflow has become necessary for its rehabilitation. Therefore, selecting the best-optimized model for research is of particular importance. In this study, we modeled and predicted the inlet flow of Urmia Lake from 2019 to 2049, using the inlet flow statistics of ten stations from 1989 to 2019. The two employed software packages demonstrated good correlation with values ranging between 0.7 and 0.92. The neural network method outperformed R software by predicting the future with less MSE error. Unlike R software, the neural network considers the future prediction variable in addition to observational streamflow, making it possible to examine the possibility of flood in case of noticeable increase or decrease in the stations and account for uncertainties such as climate change. The Tapik station showed the highest correlation rate of 0.86 in R software, while Bandeurmiye station had the highest correlation of 0.92 in the neural network, which was performed by selected predictor variables under RCP 2.6 scenario. The neural network forecasting graph results indicate an increasing trend of streamflow in Tapik, Babarood, and Mako stations located in the northwest of the basin in the next 30 years. Babarood station is expected to have the highest streamflow increase of about 15 cubic meters per second in 30 years.

Keywords

• Flow estimate
• Time series
• R software
• Artificial neural network
• Urmia lake