Investigating the impacts of parameter fixing on the performance of a spatially distributed water quality model with an adaptive emulation approach

Fixing model parameters deemed unimportant is a common practice in solving inverse problems, one that is often determined by Global Sensitivity Analysis. Yet little attention has been paid to understanding the variations in the model parameter-objective function (ObF) map caused by parameter fixing (PF), and the accompanying implications for model calibration and uncertainty assessment. We develop an approach to explore variations induced by PF in the parameter-ObF map of a spatially-distributed water quality model (the GBR Dynamic SedNet model), using an adaptive Gaussian Process emulator to enhance the computational efficiency. This study shows that characteristics of an ObF can be substantially altered by PF in ways that are quantified by a range of metrics. The approach warrants further application to other case study catchments and hydrologic models, and indeed to over-parameterized models more widely, to improve the use of parameter fixing and as a contributor to good modeling practice. • Characteristics of an objective function are substantially altered by parameter fixing • Five metrics are selected to quantify impacts of parameter fixing • Adaptive Gaussian Process enables computationally-efficient parameter fixing experiments