Reducing uncertainty in the subsurface interpretation of fold-thrust structures - machine learning from outcrop

点击翻译按钮获取中文摘要

Complex structures are difficult to image seismically making their subsurface interpretation uncertain. Consequently, idealised models of structures are commonly used when building subsurface interpretations, to quantify trapped hydrocarbon volumes and to assess fault behaviour. But how good are these models? What are the implications of using particular idealised models, and how well do they reduce risk in subsurface interpretation? This project addresses these issues by examining fold-thrust complexes, structures that form important traps in many frontier hydrocarbon provinces. Rather than 'picking' classic examples of single structures, which likely bias our understanding of fold-thrust geometries, well-exposed transects of outcropping fold-thrust systems will be mapped digitally to collect fully-representative suites of structures, plus associated digital data. Digital mapping will employ UAV and ground-based photogrammetry to obtain accurate structural geometries. Fieldwork will consider examples in Pembrokeshire, Devon, the SubAlpine chains. This database of geological structures: inter-limb angles, thrust displacements, curvature etc. will be compared with idealised models of fold-thrust structures and interpretations of a range of seismically imaged subsurface fold-thrusts. Mismatch between models, seismic interpretations and digital analogues will quantify and identify which parts of the structural interpretation carry the greatest risk, and will inform how decisions using large digital datasets of analogue data can inform sub-surface interpretation and risk analysis. The research will yield much better understanding of the structural evolution of fold-thrust complexes and improve risk assessment for sub-surface exploration through the use of large digital datasets that can statistically analyse structural form. The outcomes can inform the development of machine learnt interpretation. The student will gain excellent training in 3D structural interpretation, model building, hypothesis testing and visualization during the completion of the PhD. 3D thinking will be developed in the field and augmented whilst building and interpreting 3D virtual outcrop models, and through the interpretation of seismic data. We will train the student in the analysis and manipulation of big datasets and the use of industry standard and academic software. Training in software use (e.g. Petrel, Move, Agisoft) will be through training modules, self-teaching and peer-support in the University of Aberdeen's SeisLab suite. The student will become an expert in cross-section construction, structural modelling and analysis including the use of restoration and forward modelling tools, and digital data handling and analysis.