|
Abstract
|
This study focused on the relationship of river discharge to the water quality parameters of the Zarrineh River, the main tributary of Lake Urmia, for the period of 1966-2017. It has been shown in an analysis of 365 measurements that discharge versus TDS correlated very strongly and negatively, with R2 = 0.68 and p < 0.001, while discharge versus electrical conductivity also correlated very strongly and negatively, with R2 = 0.71 and p < 0.001. This relationship was described by the power function TDS = 523.4Q-0.235. The pH was weakly positively correlated with discharge R2 = 0.14, p < 0.05. Multiple regression models involving discharge, pH, and EC explained 76% of the TDS variance. Seasonal analysis pointed out a stronger discharge-TDS correlation in the high-flow periods, R2 = 0.79, compared to the low-flow periods, R2 = 0.56. The time series decomposition showed that the concentration of TDS was increasing 1.2 ppm per year with a 95% confidence interval of 0.8 to 1.6 ppm/year. In this study, the soft computing methods including ANN, ANFIS, and SVR performed better compared to traditional multiple linear regression. Among the soft computing models, ANFIS performed the best, with test set statistics showing an R2 of 0.879 and an RMSE of 25.2 ppm. These models performed much better in this study at different flow regimes and on extreme events. Bayesian hierarchical modeling revealed that there was spatial heterogeneity in the discharge-TDS relationship, with stronger correlations for the lower reaches than for the upper reaches: β = -0.72, 95% CI: -0.81 to -0.63, versus β = -0.54, 95% CI: -0.63 to -0.45.
|