Prediction of TDS and EC Values of River Using Machine Learning Methods

In this study, while examining the performance of random forest model and Gaussian process regression, simulating the values of electrical conductivity and total dissolved solids in the western stations of Lake Urmia (Chehriq-Olya, Dizj and Tepik) according to the values of flow discharge and The HCO3 of the river flow has been studied in the period of 1971-2020. According to the RMSE and NSE statistics, the simulation results of EC values in the studied stations showed that the error rate of the RF model is lower than the GPR model and the efficiency of the model is also higher. The error rate of simulating EC values using the RF model in the test phase in Chihriq-Olya, Dizj and Tepik stations is about 356, 36 and 47% less than the GPR model. In general, the results showed that according to the simulated confidence intervals of EC and TDS parameters, the performance of the two investigated models is acceptable, but in the case of TDS values, the behavior of the two investigated models in the two stations of Dizj and Tepik is different.