Abstract
The most common expression of dissolution in karst landscapes are sinkholes, often termed ‘dolines’. Although remote sensing can be used to map dolines, the utility of this technology is limited when the karstified surface is obscured, as is often the case, by soil, vegetation, and urbanization. Additionally, in the subsurface, processing and resolving karst in 3D is challenging because of limited seismic resolution, horizon-based migration, noise, and diffraction of wavelets. Mapping karst is important, however, because it promotes stratigraphic heterogeneity and can represent a critical drilling hazard. For these reasons, a shortcut to predicting karst in the subsurface from sparse data would speed the pathway to a more refined understanding of karst-afflicted reservoirs and aquifers. The overall goal of this research is to understand the varied controls on the development of a karst landscape in Australia. We plan to attain the overall goal by pursuing the following specific aims. First, to use remote sensing to quantify the spatial distribution of surficial karst in an environment where a full population of dolines can be resolved, and where the tectonic, physical, and environmental controls on their distribution can be measured. For this purpose, we have selected the Nullarbor Plain in Southern Australia – the largest subaerially-exposed limestone terrace on Earth. Our second specific aim is to take a comparative sedimentological approach and use the patterning of the Nullarbor Plain karst landscape as conditioning data to drive an ensemble-learning simulation of doline distribution for three real-world case studies. The object-based simulation operates on prior knowledge of the distribution of faults, terrain, and seismic-scale dissolution features within the model domain. We anticipate the proposed modelling workflow to have considerable utility in subsurface carbonate strata where accurate identification of karst is important for understanding reservoir heterogeneity and drilling hazards.