# On Evaluation of Recharge Model Uncertainty: a Priori and a Posteriori

## Abstract

Hydrologic environments are open and complex, rendering them prone to multiple interpretations and mathematical descriptions. Hydrologic analyses typically rely on a single conceptual-mathematical model, which ignores conceptual model uncertainty and may result in bias in predictions and under-estimation of predictive uncertainty. This study is to assess conceptual model uncertainty residing in five recharge models developed to date by different researchers based on different theories for Nevada and Death Valley area, CA. A recently developed statistical method, Maximum Likelihood Bayesian Model Averaging (MLBMA), is utilized for this analysis. In a Bayesian framework, the recharge model uncertainty is assessed, a priori, using expert judgments collected through an expert elicitation in the form of prior probabilities of the models. The uncertainty is then evaluated, a posteriori, by updating the prior probabilities to estimate posterior model probability. The updating is conducted through maximum likelihood inverse modeling by calibrating the Death Valley Regional Flow System (DVRFS) model corresponding to each recharge model against observations of head and flow. Calibration results of DVRFS for the five recharge models are used to estimate three information criteria (AIC, BIC, and KIC) used to rank and discriminate these models. Posterior probabilities of the five recharge models, evaluated using KIC,more »

- Authors:

- Publication Date:

- Research Org.:
- Desert Research Institute, Nevada System of Higher Education

- Sponsoring Org.:
- USDOE

- OSTI Identifier:
- 875590

- Report Number(s):
- Conf-2006-001

TRN: US0601151

- DOE Contract Number:
- AC52-00NV13609

- Resource Type:
- Conference

- Resource Relation:
- Conference: 2006 International High Level Radioactive Waste Management Conference, Las Vegas, NV, April 30 to May 4, 2006

- Country of Publication:
- United States

- Language:
- English

- Subject:
- 12 MANAGEMENT OF RADIOACTIVE WASTES, AND NON-RADIOACTIVE WASTES FROM NUCLEAR FACILITIES; CALIBRATION; EVALUATION; PROBABILITY; RADIOACTIVE WASTE MANAGEMENT; SIMULATION

### Citation Formats

```
Ming Ye, Karl Pohlmann, Jenny Chapman, and David Shafer.
```*On Evaluation of Recharge Model Uncertainty: a Priori and a Posteriori*. United States: N. p., 2006.
Web.

```
Ming Ye, Karl Pohlmann, Jenny Chapman, & David Shafer.
```*On Evaluation of Recharge Model Uncertainty: a Priori and a Posteriori*. United States.

```
Ming Ye, Karl Pohlmann, Jenny Chapman, and David Shafer. Mon .
"On Evaluation of Recharge Model Uncertainty: a Priori and a Posteriori". United States.
doi:. https://www.osti.gov/servlets/purl/875590.
```

```
@article{osti_875590,
```

title = {On Evaluation of Recharge Model Uncertainty: a Priori and a Posteriori},

author = {Ming Ye and Karl Pohlmann and Jenny Chapman and David Shafer},

abstractNote = {Hydrologic environments are open and complex, rendering them prone to multiple interpretations and mathematical descriptions. Hydrologic analyses typically rely on a single conceptual-mathematical model, which ignores conceptual model uncertainty and may result in bias in predictions and under-estimation of predictive uncertainty. This study is to assess conceptual model uncertainty residing in five recharge models developed to date by different researchers based on different theories for Nevada and Death Valley area, CA. A recently developed statistical method, Maximum Likelihood Bayesian Model Averaging (MLBMA), is utilized for this analysis. In a Bayesian framework, the recharge model uncertainty is assessed, a priori, using expert judgments collected through an expert elicitation in the form of prior probabilities of the models. The uncertainty is then evaluated, a posteriori, by updating the prior probabilities to estimate posterior model probability. The updating is conducted through maximum likelihood inverse modeling by calibrating the Death Valley Regional Flow System (DVRFS) model corresponding to each recharge model against observations of head and flow. Calibration results of DVRFS for the five recharge models are used to estimate three information criteria (AIC, BIC, and KIC) used to rank and discriminate these models. Posterior probabilities of the five recharge models, evaluated using KIC, are used as weights to average head predictions, which gives posterior mean and variance. The posterior quantities incorporate both parametric and conceptual model uncertainties.},

doi = {},

journal = {},

number = ,

volume = ,

place = {United States},

year = {Mon Jan 30 00:00:00 EST 2006},

month = {Mon Jan 30 00:00:00 EST 2006}

}