Multi-objective calibration of a river water quality model—Information content of calibration data

Substantial uncertainties exist in the identification of river water quality models, which partially depend on the information content of the calibration data. To evaluate the dependencies between available calibration data and model predictions, investigations were conducted based on a 536km free-flowing reach of the German part of the River Elbe. Five extensive flow-time-related longitudinal surveys with 14 sampling locations were used. The multi-objective calibration of the deterministic river water quality model QSIM was carried out with the nonlinear parameter estimator PEST. At the investigated river reach, parameter sensitivities were highly variable depending mainly on the growth of algal biomass. Based on 30 multi-objective calibration runs considering different numbers and combinations of the data sets, we found that calibration was only slightly improved using more than three data sets. Uncertainties can be decreased by increasing the amount of calibration data. For the calibration data sets, the cumulative distribution functions of the Nash and Sutcliffe coefficient steepen progressively and the uncertainties of model parameters decreased with an increased number of data sets included in the calibration procedure. Also the combination of different calibration data sets had an effect on the goodness of the model validation. Most uncertainties were associated with the calculation of oxygen. These findings are restricted to cases where data sets of different conditions are available. The suggested methodology for model calibration including a cross validation is especially suited for cases where available data are limited, which is common for river water quality modelling investigations. The results of this study will help model users to define appropriate data collections and monitoring schemes.