Mapping surface rock exposures to enhance geohazard susceptibility assessment: a random forest approach

Williams, Chris ORCID: https://orcid.org/0000-0003-1436-6479; Finlayson, Andrew; Palamakumbura, Romesh; Kearsey, Tim; Cornillon, Severine; Whitbread, Katie. 2020 Mapping surface rock exposures to enhance geohazard susceptibility assessment: a random forest approach. [Speech] In: EGU General Assembly 2020, Online, 4–8 May 2020. European Geosciences Union.

Before downloading, please read NORA policies.

Preview

Text (Open Access)
EGU2020-5017_presentation-1.pdf - Published Version
Available under License Creative Commons Attribution 4.0.
Download (1MB) | Preview

Official URL: https://doi.org/10.5194/egusphere-egu2020-5017

Abstract/Summary

We present the approach taken to map surface rock exposures in upland areas of Scotland. This has been carried out as a means of enhancing the mapping of superficial sediment thickness which has important applications including the assessment of potential geohazard susceptibility. The presented study includes selected test cases that have been constructed prior to scaling up the approach to upland areas across Great Britain (GB). The presence of rock at surface acts as a marker of locations with minimal superficial sediment cover (essentially a zero depth). The thickness of superficial sediments across GB are currently estimated based on borehole records which range in both quality and coverage, with limited data particularly for upland regions. Superficial sediment thickness is an integral factor for assessing geohazard processes including landslides. Therefore, by improving datasets detailing rock at surface, we can enhance superficial sediment thickness estimates and enhance the variable inputs to the models used to assess geohazard susceptibility. The GB landscape has been subject to a range of different environmental processes through time with its current topography being the subject of glacial erosion through to marine incursions. However, these patterns are not uniform and this results in a range of landscapes. The resulting domains are an important consideration when attempting to model the relationship between the presence and absence of natural rock exposures. With a wealth of information available across GB including high resolution topography, the resulting (often scale-dependent) geomorphometric derivatives, geological datasets as well as satellite imagery, we are able to consider a range of possible relationships that might exist. We combine these datasets coupled with field validation of rock absence/presence to train a random forest classifier for specific domains with the aim being to identify a way of modelling rock exposure in areas of limited data availability as is the case for many upland areas. The methodology and results of the approach for specific process domains will be presented with a specific focus on the Glen Gyle catchment, at the head of Loch Katrine (the primary water reservoir for the city of Glasgow) in the Trossachs National Park, Scotland. This is an area that has been subject to recent landslides which have affected local properties and infrastructure.