A Hybrid Perturbation and Morris Approach for Identifying Sensitive Parameters in Surface Water Quality Models

Abstract

Surface water quality models (SWQM) are always developed as universal frameworks so that they can be flexibly employed to simulate a large variety of water bodies. These models are often over-parameterized (more parameters than needed are included in these models). As a result, it is necessary to identify sensitive parameters when these models are applied to the simulations of specific water bodies. Sensitivity analysis has been widely used as an effective tool to undertake the task. In this study, a hybrid approach was developed through integrating the parameter perturbation method and the Morris method into a general SWQM-parameter sensitivity analysis framework. The approach was applied to Lake Maumelle in Arkansas with its hydrodynamics and water quality being simulated by the model CE-QUAL-W2. The sensitivities of the 96 model parameters were firstly evaluated by the parameter perturbation method in the simulation of the variables including temperature, ammonium, nitrate-nitrite, dissolved oxygen, total phosphorus and chlorophyll a, and 51 of them were found sensitive. The sensitivities of the 51 parameters were further investigated using the Morris method. It was found that each output variable was strongly sensitive to a distinctive set of parameters. It is also observed that the highly sensitive parameters display nonlinear relationships with the model outputs or strong correlations with other parameters. The obtained results from this study could provide a scientific base and solid start for the calibration, validation and application of the model.