Investigation of the performance and accuracy of multivariate timeseries models in predicting EC and TDS values of the rivers of Urmia Lake Basin

Considering the complexity of hydrological processes, it seems that multivariate methods may enhance the accuracy of time series models and the results obtained from them by taking more influential factors into account. Indeed, the results of multivariate models can improve the results of description, modeling, and prediction of different parameters by involving other influential factors. In this study, univariate models (ARMA) and auto-correlated multivariate models with the simultaneous autoregressive moving average model (CARMA) were evaluated for modeling Electrical Conductivity and Total Dissolved Solid parameters of the western stations of Urmia Lake Basin. To use the CARMA models, annual flow rate time series, EC, TDS, SAR, and pH values measured across seventeen hydrometric stations between 1992 and 2013 were used. In the studied statistical period, the river flow in the west of Urmia Lake Basin decreased and experienced an incremental increase compared to the EC and TDS values in river flow. By applying influential parameters in CARMA models, the mean error value of the model in training and experimental stages reduces by 32% and 44% for EC values and 34% and 36% for TDS values, respectively.