A Framework for Practical Factor Fixing Demonstrated with Uncertainty Quantification of a Water Quality Model
Abstract
Factor Fixing (FF) is a common approach adopted for dimension reduction, with Global sensitivity analysis (GSA) used to identify insensitive factors to be kept constant in further analysis. However, depending on the quantity of interest (QoI) and the values at which factors are fixed, FF may still lead to unacceptable errors. This means that the impacts of FF need to be evaluated at the least cost possible in terms of the specific QoIs of interest to end-users . We propose a framework containing four principles to guide the use of FF and increase confidence in the resulting dimension-reduced model. The four principles for practical FF consist of: 1) f ocus on decision or purpose relevance in selection of QoIs and measure of errors induced by FF, 2) re-parameterize first to achieve loss-less dimensionality reduction and mitigate impacts of parameter interactions on FF, 3) adaptive evaluation of impacts of FF recognising that information is often asymmetric and that factors that cannot be fixed can often be identified efficiently, and 4) re-consider whether FF is necessary given the insights gained through adaptive evaluation of its impacts.
The framework is demonstrated on a distributed water quality model, focusing on conducting FF for uncertainty quantification. The model we use for the O'Connell catchment is part of the GBR-Dynamic SedNet model which was developed to simulate water quality processes in the catchments of the Great Barrier Reef, Australia. Results indicate that varying a single factor may provide an acceptable estimate of model uncertainty in the average annual load of Total Suspended Solids (TSS). This is partly due to the large uncertainty in that factor (related to bank erosion), such that reducing that uncertainty is the preferred next step, with use of informative monitoring data to obtain an updated factor distribution, rather than proceeding with FF.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMH219...01W
- Keywords:
-
- 0430 Computational methods and data processing;
- BIOGEOSCIENCES;
- 1805 Computational hydrology;
- HYDROLOGY;
- 1846 Model calibration;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY