Hemmings B, Knowling MJ, Moore CR. 2019. Assessing the uncertainty of water quality and lake influx predictions made using complex regional models: Ruamāhanga South case study. Lower Hutt (NZ): GNS Science. 72 p. (GNS Science report; 2019/30). doi:10.21420/9N73-QE35.
The Smart models for Aquifer Management research programme aims to explore the utility and implications of using simple numerical models for decision support applications. This report details a case study example of the application of complex numerical groundwater model as a tool of assessing the effectiveness of land-use resource management scenarios. The case study focuses on the lower Ruamāhanga catchment, Greater Wellington, NZ (Ruamāhanga South). Scenarios relate to nitrate loading change related to distributed change in land-use. The effectiveness of scenarios is assessed relative to desired ecological and societal outcomes related to predicted water quality (nitrate concentrations) in surface- and ground-water. An essential component of the demonstrated use of the numerical model for decision support is the application of probabilistic uncertainty analysis. We use ensemble based calibration and uncertainty analysis techniques (iterative ensemble smoother, iES) to quantify probabilistic uncertainty for a highly parametrised numerical model. iES allows the conditioning of the parameter uncertainty through assimilation of system observation date and propagation of conditioned (posterior) parameter uncertainty through to model predictions. Comparison of prior and posterior parameter uncertainty provides insights into the influence of system observations in conditioning of model parameters, and therefore the ability of data assimilation to reduce predictive uncertainty. Analysis of prediction uncertainty provides a probabilistic description of the likely effectiveness of land-use change in leading to improvement in water quality in the region. Comparison of prior and posterior prediction distributions relative to desired management decision thresholds highlights value (or otherwise) of model calibration efforts in improving the predictive utility of the model. (auth)