Hou, Wenzhu1; Tsui, Wing Sum Regine1; Hart, Jonathan Roy1
1GeoRisk Solutions Ltd., Sheung Wan, Hong Kong
Knowledge of regolith types and distribution forms a key component of a geomorphological model. For engineering projects in Hong Kong, conventional methods widely used for regolith mapping include aerial photograph interpretation (API) together with ground ‘truthing’ by direct field observations. However, these techniques can be time consuming, and information obtained in the field is constrained by cost and accessibility. Regolith mapping from API is highly subjective, being dependent on the interpreter’s experience, and is constrained by photo quality, vegetation cover, cloud cover and shadow. Furthermore, although there are significant aerial photograph resources available in Hong Kong, in many countries this is not the case.
This paper describes an alternative method of regolith mapping for a landslide hazard assessment project in Hong Kong, using the latest available digital resources on an ArcGIS platform. A number of digital resources, particularly a territory-wide airborne LiDAR survey, are now available in Hong Kong and are being used increasingly to enhance conventional methods. However, the DEM derived from LiDAR only provides spatial information and lacks spectral information needed for ground modelling. Therefore, an integrated approach combining spectral and spatial data is used to establish a preliminary geomorphological model, which can be done quickly and at minimal cost at an early stage of a project. This digital method maps regolith automatically using image classification tools in ArcGIS, and is based on a multiband raster, including a raster from greyscale aerial photos, and rasters from the DEM.
The Hong Kong Government recently completed high-accuracy georeferencing of photos taken in 1963 when the vegetation cover in many parts of the territory was extremely low, providing an excellent record of the surface conditions. This monochromatic imagery was used as one band in the multiband raster. A territory-wide, airborne LiDAR survey was carried out in 2010 and a DEM of 0.5 m resolution or better is available from this survey. Under the alternative method described in this paper, three spatial rasters (slope gradient, surface roughness  and landform classification), were generated from the DEM, and combined with the first band to form the multiband raster. The slope gradient raster is used as an indicator of regolith type given that steep slopes reflect transportational processes whereas gentle slopes reflect depositional processes. An enhanced topographic position index method was adopted for landform classification . Five regolith categories (rock, saprolite, slope colluvium, valley colluvium and taluvium), based on limited field knowledge, were annotated on the multiband raster for training the classifier. The multiband raster was then classified to automatically map regolith using this trained classifier. The results were extremely encouraging despite relatively limited training samples. By combining spectral and processed spatial rasters, the alternative method produced a good regolith model closely matching that developed using conventional methods. The method can be tested elsewhere with different digital resources to develop good quality, cost-effective, early-stage geomorphological models.
 Garriss, R. N. (2019). Modeling Surface Roughness as an Indicator of Age and Landslide Susceptibility, and the Spatial Inventory of Prehistoric Landslides: Green River Valley, Washington. DOI: 10.15760/etd.7051
 De Reu, J., Bourgeois, J., Bats, M., Zwertvaegher, A., Gelorini, V., De Smedt, P., … & Van Meirvenne, M. (2013). Application of the topographic position index to heterogeneous landscapes. Geomorphology, 186, 39-49.
Dr Hou is an engineering geologist based in Hong Kong. She obtained her doctorate in geology from The University of Hong Kong and currently works on natural terrain hazard related projects.