Integrating geophysics and hydrology for reducing the uncertainty of groundwater model predictions and improved prediction performance

Abstract:

A major purpose of groundwater modeling is to help decision-makers in efforts to manage the natural environment. Increasingly, it is recognized that both the predictions of interest and their associated uncertainties should be quantified to support robust decision making. In particular, decision makers want to have a full description of model prediction uncertainties so that they can consider costs and benefits of different model predictions. Limited data remains a major source of uncertainties in groundwater models and associated predictions.

Traditionally, groundwater models have been constructed from geological and hydrological data. However, geophysical data are increasingly used to inform hydrogeologic models because they are collected at lower cost and much higher density than geological and hydrological data. Despite increased use of geophysics, it is still unclear whether the integration of geophysical data in the construction of a groundwater model increases the prediction performance. We suggest that modelers should perform a “test-bench” analysis of the likely value of geophysics data for improving groundwater model prediction performance before actually collecting geophysical data. At a minimum, an analysis should be conducted assuming settings that are favorable for the chosen geophysical method. If the analysis suggests that data collected by the geophysical method is unlikely to improve model prediction performance under these favorable settings, it is unlikely that such data will be useful under more challenging settings.

We have developed a synthetic “test-bench environment” to test the benefits and limitations of alternative hydrogeophysical data and inversion approaches. The environment consists of multiple high resolution realizations of synthetic, "true", hydrogeological and geophysical systems. The two types of "true" systems can be used together with corresponding forward codes to generate hydrological and geophysical datasets, respectively. These synthetic datasets can be interpreted using any hydrogeophysical inversion scheme and the resulting predictions can be compared with predictions from the ‘true’ model. By performing this analysis we expect to give the modeler insight into how the uncertainty of model-based prediction can be reduced.