Uncertainty assessment of hydrogeological structures combining geophysical survey and geological knowledge: A stochastic level set optimization framework

Abstract:

Effective exploration for groundwater resources requires a comprehensive understanding of variability in hydrogeological structures. Geophysical investigations such as airborne electromagnetic survey (AEM), are widely used throughout the world to measure subsurface geophysical properties and detect structural changes. However, delineating hydrogeological structures efficiently and extensively remains challenging and often requires manual interpretation. In addition, geologists’ knowledge, such as formations, geometry, or connectivity in the subsurface needs to be integrated. Combing all this available information allows for more precise structural models.

In this study, we propose a statistical framework for modeling geological structures, given both inverted electrical conductivity and geological interpretations. An implicit level set function is proposed to model the 3D interface between two distinct lithologies. We optimize this level set using gradual deformation to meet two objectives: maximize the contrast from geophysical data and impose structural connectivity interpreted by geologists. Multiple optimizations, starting from different initial interfaces, will provide uncertainty assessment of hydrogeological structures.

This stochastic level set optimization framework is applied to the arid Musgrave region in South Australia, where groundwater is the only reliable water source. AEM data has been acquired for groundwater exploration over spatially discrete lines (spacing ~ 2km), targeting paleovalleys which have the greatest potential to contain productive aquifers. Paleovalley connectivity provided by geologists is subject to interpretation uncertainty. To represent such connectivity, we construct a graphical model. Uncertainty in interpretation is represented by perturbing this graph while keeping the same topology. In this way, we both incorporate interpretation uncertainty and preserve known connections. Our results provide geological realistic interfaces between paleovalley and bedrock with high horizontal resolution (~50m). We also assess the structural uncertainty between sounding lines. Our results show that we provide an efficient structural modeling framework from both AEM images and domain knowledge with realistic uncertainty estimation.