Geologists use computers to track underground rock fractures, which helps prevent mine flooding. Yet, the way data is prepared can change how these geological lines connect on screen. We tested three data preparation methods using real and virtual well data from Poland. We found that only the standard approach keeps the fracture lines connected and stable when adding new data. This allows safer engineering without having to recalculate everything from the beginning.

Mapping rock fractures in high resolution aerial images is essential for understanding Earth processes and managing resources, but manual tracing is slow and inconsistent. We created FraXet, a large harmonized dataset of nearly nine thousand images, and compared standard image filters with modern deep learning models. The deep learning methods were far more accurate and produced smoother, more reliable maps, while also showing where results are uncertain.

The shape of the water filled cavity hidden beneath ice shelves is extremely important, but hard to measure. Gravity data collected from aircraft can be used to predict the shape. However, in the Dotson Crosson ice shelves in West Antarctica, new direct observations from autonomous platforms sent beneath the ice, and seismic observations from the ice surface showed gravity models were systematically too shallow. We explain why and provide a solution applicable her and in other ice shelves.