Similarity Metrics-Based Uncertainty Analysis of River Water Quality Models

Uncertainty analysis (UA) is essential to reinforce the decisions made by water resource engineers and managers. In this study, the stepwise multiple linear regression procedure assessed the relationship between water quality parameters and physical characteristics of 48 catchments in the southwestern basin of the Caspian Sea, Iran. The results of the modeling showed that the coefficient of determination ranged between 0.47 and 0.68 and indicated a positive relationship between the area (%) of agricultural lands and the sodium adsorption ratio (SAR), potassium (K) and total dissolved solids (TDS). A negative relationship was also found between bicarbonate (HCO3−) and the area (%) of the intermediate-density forest. In contrast to previous studies focusing on analyzing the uncertainty of the model parameters, we addressed the uncertainty of the model variables. The results of the GLUE-based uncertainty analysis (UA) performed on the model’s variables indicated that the measures of the R-factor for all models were between 0.13 and 0.98. The lowest R-factor was obtained for the HCO3− model (0.13) suggesting it performed well when predicting HCO3−. To increase the degree of objectivity in the GLUE-UA method, a set of similarity metrics, including Czekanowski, Motyka, Ruzicka, Cosine, Kumar-Hassebrook, Jaccard and Dice was applied to determine the degree of proximity and or similarity between the probability density functions of the measured and simulated water quality parameters. The measures of the similarity metrics for the HCO3− model were generally close to 1, indicating good performance and low uncertainty, while it showed higher uncertainty (between 0.2487 and 0.897) for the other three models (SAR, K, and TDS).