Quantifying uncertainty within an ecologically-coupled lake hydrodynamic model: Lake Wairarapa case study

(Inc. GST)
(Ex. GST)
Write a Review

Allan, M. 2018 Quantifying uncertainty within an ecologically-coupled lake hydrodynamic model: Lake Wairarapa case study. Lower Hutt, N.Z.: GNS Science. GNS Science report 2018/47. 23 p.; doi: 10.21420/J482-WF31

The present study investigated uncertainty within an ecologically coupled 1-D hydrodynamic model of Lake Wairarapa and uncertainty of total chlorophyll a (TCHLA) estimation. This uncertainty was compared and resulted from both model history matching or model parametrization (within lake) and uncertainty generated from inflowing nitrate concentrations derived from prior and posterior catchment modelling. The study indicated that catchment model calibration effort has significant implications for lake modelling uncertainty. The posterior catchment model scenario led to more feasible within lake estimations of TCHLA reductions under GOLD2040 (nitrate reduction) scenarios, where nutrient loading to lake Wairarapa was reduced. The lake modelling model calibration also resulted in significant reduction in posterior uncertainty in estimation of TCHLA when compared to prior uncertainty. Post-calibration uncertainty analysis should become part of everyday modelling, especially when auto-calibration routines are applied. These auto-calibration routines allow uncertainty analysis to be applied simultaneously alongside model calibration. Post-calibration uncertainty was also likely reduced in the catchment model. The applied Monte Carlo based calibration and uncertainty analysis allowed quantification and comparison of uncertainty derived from catchment models and lake models. Lake modelling realizations (for the estimation of absolute TCHLA) using Monte Carlo derived parameter ranges had a much larger uncertainty than catchment nitrate realizations (for the estimation of absolute TCHLA). Inputs to lake models are often comprised of interpolated daily flows and concentrations, sometimes from monthly or lower resolution field data, often contributing to uncertainty. However, the present study shows that while uncertainty derived from inflows is significant, uncertainty arising from lake model parameter error is larger, due to the larger range of mean TCHLA displayed in the realisations. This can potentially be exacerbated in shallow lakes, where nutrient release and resuspension from the sediment can dominate nutrient budgets, with model parametrization having a large influence on these nutrient fluxes. This finding would vary greatly between lake systems depending on a multitude of factors including residence time and trophic state. (auth)