Before we build a predictive groundwater flow model, we first identify candidate conceptual models. To cull our collection of candidate models, we ask the seemingly simple but key question: Where is the water coming from? We argue for the use of a Bayesian framework to address this question. We start with an uninformed prior distribution that says all directions are equally likely. We then incorporate the available information (usually error-prone, noisy information of varied quality) and appropriately update the characterization of the uncertain direction. When the added information is extensive and internally consistent, a clear flow direction emerges. On the other hand, if the added information is minimal or internally inconsistent, the uninformed prior is only slightly modified.
|Original language||English (US)|
|Number of pages||9|
|Journal||Geotechnical Special Publication|
|Issue number||GSP 321|
|State||Published - Jan 1 2020|
|Event||Geo-Congress 2020: University of Minnesota 68th Annual Geotechnical Engineering Conference - Minneapolis, United States|
Duration: Feb 25 2020 → Feb 28 2020