Predictions and their uncertainty are key aspects of any modelling effort. The prediction uncertainty can be significant when the predictions depend on uncertain system parameters. We analyse prediction uncertainties through constrained nonlinear second-order optimization of an inverse model. The optimized objective function is the weighted squared difference between observed and simulated system quantities (flux and time-dependent head data). The constraints are defined by the maximization/minimization of the prediction within a given objective function range. The method is applied in capture-zone analyses of groundwater supply systems using a three-dimensional numerical model of the Espanola basin aquifer. We use the finite-element simulator, FEHM, coupled with parameter-estimation/predictive-analysis code, PEST. The model is run in parallel on a multi-processor supercomputer. We estimate sensitivity and uncertainty of model predictions, such as capture zone identification and travel times. While the methodology is extremely powerful, it is numerically intensive.
[1]
D. George,et al.
The X3D grid generation system
,
1996
.
[2]
Richard L. Cooley,et al.
Simultaneous confidence and prediction intervals for nonlinear regression models with application to a groundwater flow model
,
1987
.
[3]
B. Robinson,et al.
Summary of the Models and Methods for the FEHM Application-A Finite-Element Heat- and Mass-Transfer Code
,
1997
.
[4]
S. P. Neuman,et al.
Estimation of aquifer parameters under transient and steady-state conditions: 2
,
1986
.
[5]
Yonathan Bard,et al.
Nonlinear parameter estimation
,
1974
.