Document Type : پژوهشی
Authors
1 Department of Nature engineering, Agricultural Sciences and natural Resources University of Khuzestan, Mollasani, Khuzestan, Iran
2 Ph.D. Candidate, Department of Rangeland and watershed management, Yazd university, Iran
Abstract
Rainfall-runoff processes are among the most complex and nonlinear phenomena in hydrology. In water resources management, runoff forecasting faces challenges in ungauged watersheds.In this study, the efficiency of lumped models and machine learning methods was investigated in the Kabkian watershed, that is one of the principal branches of the Karun River. AWBM, Sacramento, SIMHYD, TANK, and SMAR Also, some algorithms in decision trees, artificial neural networks, and support vector regression were applied to simulate daily and monthly runoff in the Kabkian watershed. Monthly and daily discharge, precipitation, and potential evapotranspiration for the period between 1972 and 2022 were used,. The accuracy and efficiency of the methods were examined using R², the Nash-Sutcliffe coefficient, and RMSE. Results showed that SMAR and AWBM, in comparison to other lumped models, have the best efficiency in the simulation of daily discharge in the Kabkian watershed. The Nash-Sutcliffe coefficients for them in the test stage are 0.79 and 0.78, respectively, showing that these models have good efficiency in daily discharge simulation. Also, the SMAR and AWBM models' Nash-Sutcliffe coefficients are 0.71 and 0.72, respectively, and the R2 for the two models is 0.79 in the monthly time series. These values show that these models have good efficiency. In machine learning methods, in the daily series, the random forest algorithm's R2 is 0.61 and has the best efficiency in comparison to other methods. Also, in the monthly series, the random forest's R2 is 0.93, which illustrates good discharge simulation efficiency.
Keywords
Main Subjects
)