Subsurface data handling in infrastructure planning: A geological perspective

A substantial part in planning for linear infrastructure like road and railways are dependent on the conditions of the ground. As such, much data is collected to categorise the subsurface in order to adequately design the planned structure, estimate volumes in earth works and apply reinforcements. This data, pertaining to the subterranean structure and composition, is usually not plentiful in early stages of planning during the desktop study. The work presented in this thesis demonstrates the usefulness and drawbacks of already existing data by applying both experience based methods and machine learning methods in ground classifications. By means of multi criteria analysis and analytical hierarchy process, existing surface maps can be used not only to categorise project specific ground suitability but also point to the uncertainties in the pairwise comparison of criteria. While input data in the form of surface information does provide value in early stage ground classification, subsurface information is needed to better understand the total geology of the investigated area. A Swedish database of well-logs were used to train a lithological complexity classifier in order to evaluate the datasets readiness for machine learning purposes. The dataset was contrasted with an archaeological dataset of gravefield and settlements. While sufficiently performant models can be built using the national well-log database, labelling of lithology suffers from non standard notations. Recommendations for upcoming national initiatives are made with data-centricity in mind. In preparation of the the upcoming national geotechnical database, a survey was sent out to the Swedish municipalities. The survey shows that Swedish municipalities, in general, are not able to seamlessly integrate their data into a national database and that data formats, softwares and routines is significantly varied between municipalities. Larger municipalities are in general more technologically mature with their subsurface data management and recommendations for pilot-integrations with interested parties are made.