Rainfall-Runoff Modeling of Aji Chai Basin Using Random Forest and Artificial Neural Network Models

Estimating runoff due to rainfall is a very important step in planning water resources, especially in watersheds without hydrometric stations. In this study, the rainfall-runoff simulation of Akhola station located in the Ajichai basin was discussed and the most suitable rainfall-runoff model was presented by using data mining methods and comparing their performance. For this purpose, the desired data (rainfall, discharge, temperature) for this study were received monthly from the water and meteorological organizations of the East and West Azerbaijan provinces. Random Forest and Artificial Neural Network data mining models were used for simulation. The comparison of observed monthly runoff values with monthly runoff estimated by models was done using evaluation criteria. In this study, CC values ​​(correlation coefficient) for test sets in the random forest model and artificial neural network were determined as 0.77 and 0.86, respectively. The analysis of the results showed that the ANN model has a higher performance and efficiency than the RF model for the Akhola station. Among other results of this research, we can mention the time series of rainfall and runoff of the station during the last 20 years. According to the obtained graphs, during these 20 years, there was no clear trend for precipitation in the Ajichai basin, and the time series graphs showed that the precipitation in these areas was fluctuating. But the time series for Ajichai discharge at Akhula station showed that a completely downward trend was recorded for the river water flow, which is the reason for the decrease of the discharge entering Lake Urmia and the lowering of the lake water level.