Impact of Integrated Versus Sequential Data Fusion on Hydrologic Predictions
Abstract
Fusing data from a wide variety of sources into a hydrologic model is one of the most promising approaches we have for improving hydrologic predictions. How to merge disparate types of data into a consistent hydrologic model is, however, a critical question that has not been fully resolved. One issue in particular is how different strategies for data integration may ultimately impact hydrologic predictions. We have been investigating integrated versus sequential data fusion methods as alternate strategies for integrating geophysical data into hydrologic estimation problems. The more traditional approach, sequential data fusion, relies on geophysical inversion and rock physics to produce a map of the hydrologic properties of the subsurface that can subsequently be used to constrain a hydrologic inverse problem. In contrast, integrated data fusion uses coupled hydrologic and geophysical models to directly constrain hydrologic parameters of interest with geophysical measurements. Sequential data fusion is easier to implement since it leverages existing methods and codes that often, in practice, allows geophysicists and hydrologists to work independently. Limited spatial resolution, reliance on prior knowledge of the subsurface, and the questionable existence of rock physics relationships between geophysical and hydrologic parameters at the field scale are all critical limitations of the approach. These limitations lead to bias in the hydrologic inverse problem resulting in poor hydrologic models and unreliable predictions. Integrated data fusion overcomes some of these issues by focusing on estimating parameters that control processes, rather than imaging state variables. For example, in a homogeneous soil an infiltration plume can be fully characterized by just a few variables (assuming boundary conditions are known) though the distribution of water content is highly variable through space and time and would require thousands of parameters to characterize directly. Because the parameters controlling processes are better constrained in integrated data fusion, more reliable predictions can ultimately be obtained. As a result, integrated data fusion is the most appropriate choice for data integration if the objective is to produce models capable of hydrologic predictions. We demonstrate how the choice of data fusion method impacts predictions using examples of solute transport monitored by electrical resistivity and infiltration monitored by borehole groundpenetrating radar.
 Publication:

AGU Fall Meeting Abstracts
 Pub Date:
 December 2007
 Bibcode:
 2007AGUFM.H41H..03M
 Keywords:

 1816 Estimation and forecasting;
 1835 Hydrogeophysics;
 1848 Monitoring networks;
 3238 Prediction (3245;
 4263);
 3260 Inverse theory