We present CobWeb 1.0, a graphical user interface for analysing tomographic images of geomaterials. CobWeb offers different machine learning techniques for accurate multiphase image segmentation and visualizing material specific parameters such as pore size distribution, relative porosity and volume fraction. We demonstrate a novel approach of dual filtration and dual segmentation to eliminate edge enhancement artefact in synchrotron-tomographic datasets and provide the computational code.

Machine learning techniques are a promising alternative for processing (phase segmentation) of 3-D X-ray computer tomographic rock images. Here the performance and accuracy of different machine learning techniques are tested. The aim is to classify pore space, rock grains and matrix of four distinct rock samples. The porosity obtained based on the segmented XCT images is cross-validated with laboratory measurements. Accuracies of the different methods are discussed and recommendations proposed.

The objective of this study is to improve the reliability of large-scale hydrological models. In environmental policies, many decisions are based on prognosis simulated by models. The parameterisation of these models is challenging due to the scarcity of available data. Particularly, parameters of soil properties are rare, but have a strong influence on model results. To account for the heterogeneity of soil properties, we developed a methodology that does not need additional field data.