PhD defence: Xiulan He

Xiulan He, Geology Section, defends her thesis

Geostatistical simulation of geological architecture and uncertainty propagation in groundwater modeling

Supervisors
Professor Karsten Høgh Jensen, IGN
Senior Researcher Torben Sonnenborg, GEUS

Assessment Committee
Professor Alain Dassargues, University of Liege - Belgium
Senior Scientist Steve Carle, Lawrence Livermore National Laboratory - CA, USA
Assistant Professor Majken Looms Zibar (chair), IGN

Abstract
Groundwater modeling plays an essential role in modern subsurface hydrology research. It’s generally recognized that simulations and predictions by groundwater models are associated with uncertainties that originate from various sources. The two major uncertainty sources are related to model parameters and model structures, which are the primary focuses of this PhD research. Parameter uncertainty was analyzed using an optimization tool (PEST: Parameter ESTimation) in combination with a random sampling method (LHS: Latin Hypercube Sampling). Model structure, namely geological architecture was analyzed using both a traditional two-point based geostatistical approach and multiple-point geostatistics (MPS). Our results documented that model structure is as important as model parameter regarding groundwater modeling uncertainty. Under certain circumstances the inaccuracy on model structure can be compensated by model parameters, e.g. when hydraulic heads are considered. However, geological structure is the primary source of uncertainty with respect to simulations of groundwater age and capture zone.
Operational MPS based software has been on stage for just around ten years; yet, issues regarding training image (TI) and secondary conditioning are currently active research topics. This study examined these two problems by introducing a new data source, SkyTEM (airborne transient electromagnetic), into the implementation of MPS. MPS was applied at three sites in western Denmark; the largely distinct geological structures of these three sites provided appropriate conditions for testing the methods. Our study documented that MPS is an efficient approach for simulating geological heterogeneity, especially for non-stationary system. The high resolution of geophysical data such as SkyTEM is valuable both for developing a training image and for soft conditioning.

The thesis is available from the PhD administration office 04.1.409