Impact of input data uncertainties on stormwater model parameters

The use of urban drainage models requires careful calibration, where model parameters are selected in order to minimize the difference between measured and simulated results. It has been recognized that often more than one set of calibration parameters can achieve similar model accuracy. A probability distribution of model parameters should therefore be constructed to examine the model’s sensitivity to its parameters. With increasing complexity of models, it also becomes important to analyze the model parameter sensitivity while taking into account uncertainties in input and calibration data. In this study a Bayesian approach (including the Metropolis-algorithm) was used to develop a framework for quantification of impacts of uncertainties in the model inputs (i.e. systematic and random errors in rainfall time series) on the parameters of a simple integrated stormwater model. The model uses KAREN to simulate flow, and a very simple flow-based regression to calculate total suspended solids and total nitrogen loads. The framework was applied to two catchments in Australia. It was found that only systematic rainfall errors have a significant impact on flow model parameters. The most sensitive flow parameter was the effective impervious area, which interestingly can be calibrated to completely compensate for the input data uncertainties. On the other hand, the pollution model parameters were influenced by both systematic and random rainfall errors. It was also interesting that the parameters of both flow and water quality models behaved differently depending on circumstances (e.g. catchment type, data availability, data accuracy).