科研之友
微信
新浪微博
Facebook
分享链接
摘要
The use of the generalized likelihood uncertainty estimation (GLUE) methodology in analyzing the results of stochastic groundwater models is evaluated. The ability of the GLUE methodology to mitigate the effect of the selection of the input parameter prior distributions on the modeling results is investigated. This is important when no prior information is available or when significantly different priors come from different sources or experts. The different approaches that can be used to implement the GLUE methodology in analyzing the stochastic results of such models and quantifying the uncertainty in model prediction are evaluated. Recent debates about the GLUE methodology and the problem of using "less formal likelihood" functions are discussed in terms of the applicability of such issues to groundwater studies in general and a given field site specifically. These issues are investigated using a density-driven groundwater flow model, of a nuclear testing site (Milrow) on Amchitka Island, Alaska. Results of the analysis highlight the subjectivity of the choice of the shape factor associated with the GLUE likelihood measures. However, the arbitrary choice of this factor can be tied to the level of confidence one can place on the available observations. While traditional GLUE applications focus on displaying prediction quantiles, GLUE can be used to develop uncertainty bounds that are qualitatively similar to predictive uncertainty. Interestingly, for the case study shown here the traditional GLUE quantiles and the uncertainty bounds are almost identical. Results also show that the GLUE-based ensemble averaging yields results that are controlled by the data more than by the prior distributions. The GLUE quantiles or GLUE-developed uncertainty bounds provide conditional predictions that are free from the artificial smoothing associated with ensemble averaging.
- 出版日期2008-11-30
