Improving hydrogeological models using the results of calibrated groundwater flow models : A probabilistic approach using piecewise linear probability density functions and Bayesian networks

Abstract

The subsurface of the Netherlands is an important source of water that is used for drinking water and plant growth, among other applications. This groundwater is stored in sediments like clay, sand and peat. Therefore, it is important to know how much groundwater can be used and how it flows. Groundwater models can, for instance, be used to calculate the effects of groundwater abstractions. A crucial part of a groundwater model is a description of the subsurface, which is a model by itself. An important subsurface model in the Netherlands is the REGIS model, which is developed and maintained at the Geological Survey of the Netherlands-TNO. REGIS is a general purpose subsurface model that is often used to build groundwater models. A common characteristic of models is that they are not perfect and to some degree subject to uncertainty. To increase the value of a model and its use, one generally aspires to decrease this uncertainty. If a groundwater model that is based on the REGIS model is improved (calibrated), then the subsurface description of the groundwater model is often altered in the process. However, these alterations in the groundwater model do not lead to an improvement of the REGIS model that was used as source. In this research, we developed two different procedures to improve the REGIS model based on the improvements made in groundwater models. These procedures account for the uncertainty of the data, where data with a higher uncertainty receive a lower weight than data which are more certain. The uncertainty of data is described with probability distributions. An important additional result of this research is the development of a method to work with probability distributions that can have arbitrary shapes.