How to find the baddies - a borehole quality management and outlier detection software for 3D-model data selection

Exploration datasets such as borehole logs and geophysical profiles form the fundamental basis of geological modeling. Among these, borehole records are particularly influential, as they typically include detailed descriptions and interpretations of petrography and stratigraphy. Such information is essential for constructing three-dimensional representations of lithostratigraphic units, which can be affected by inconsistencies or errors skewing borehole interpretations. Distinguishing reliable borehole data from problematic records is therefore critical, but becomes increasingly challenging when dealing with large datasets. Although visual assessment of the resulting geological models can help identify questionable boreholes, this approach typically requires many iterative modeling steps, making the process inefficient and costly.
To improve the efficiency of borehole data quality assessment, we developed B-QualMT, a Python-based borehole quality management tool with a GUI interface that enables automated filtering of borehole records using both a user-defined quality check as well as a purely data-driven approach. The software applies a suite of deterministic tests that incorporate auxiliary information such as existing 3D geological models and regional geological knowledge, including expected stratigraphic successions, to identify anomalous borehole logs within geologically similar areas. Furthermore, spatial outliers can be identified using a combination of borehole similarity analysis, various clustering techniques, and a Bayesian-based novelty detection system. To evaluate the functionalities and edge cases of these methods, synthetic borehole data besides real borehole data were used. Different test scenarios were utilized to systematically control and test the outlier detection approaches, enabling workflow optimization and a detailed assessment of their performance, limitations, and sensitivity under controlled synthetic conditions. The limitations identified during testing with synthetic data are subsequently used to inform and improve the interpretation of results derived from more complex real borehole logs.