| Summary: | Subcritical fracturing, or fracturing driven by stresses below the failure strength of a material, is increasingly recognized as one of the most important processes in rock weathering. Most environmental stresses, such as thermal stresses from solar heating, are well below the failure strength of rock, so most fracturing is subcritical. In order to understand rates and controls on subcritical fracture processes in nature, it is necessary to amass large datasets on natural microfractures. However, thoroughly characterizing fractures takes many hours of tedious work for each sample, which has limited data availability. We present a new workflow for fracture characterization using Ilastik machine learning image segmentation software. We have trained our model to identify open voids, obstructed voids, fractures, non-fracture lineations, quartz grains, and other mineral grains in sandstones by identifying regions of the image on a pixel-by-pixel categorization basis and further hone in by determining whether the shape of questionable areas is likely to be a fracture. The software can create predictive models of unlabeled areas and gets better in real-time with additional training data, massively improving the speed of data collection. The computer-generated data compare reasonably well with manually-identified fracture data from the same sample. The sandstones we are analyzing are from the McMurdo Dry Valleys in Antarctica, one of the coldest and driest places on earth, with long periods of daylight and darkness during the summer and winter. These factors combine to produce unusually slow weathering rates, making it an interesting end-member field site. Our samples have been dated for exposure using cosmogenic isotopes beryllium-10 and aluminum-26. Some of these rocks have been exposed at the surface for two million years. The chronosequence of fracture densities reveals a linear increase in the total number of fractures per unit area with time, but a nonlinear increase in the total fracture length per unit area, ...
|
|---|