Giving gully detection a HAND: Testing the scalability and transferability of a semi-automated object-orientated approach to map permanent gullies

Gully erosion can incur on- and off-site impacts with severe environmental and socio-economic consequences. Semi-automated mapping provides a means to map gullies systematically and without bias, providing information on their location and extent. If used temporally, semi-automated mapping can be used to quantify soil loss and identify soil loss source areas. The information can be used to identify mitigation strategies and test the efficacy thereof. We develop, describe, and test a novel semi-automated mapping workflow, gHAND, based on the distinct topographic landform features of a gully to enhance transferability to different climatic regions. Firstly, topographic heights of a Digital Elevation Model are normalised with reference to the gully channel thalweg to extract gully floor elements, and secondly, slope are calculated along the direction of flow to determine gully wall elements. As the gHAND workflow eliminates the need to define kernel thresholds that are sensitive towards gully size, it is more scalable than kernel-based methods. The workflow is rigorously tested at different gully geomorphic scales, in contrasting geo-environments, and compared to benchmark methods explicitly developed for region-specific gullies. Performance is similar to benchmark methods (variance between 1.4 % and 14.8 %). Regarding scalability, gHAND produced under- and over-estimation errors below 30.6 % and 16.1 % for gullies with planimetric areas varying between 1421.6 m² and 355403.7 m², without editing the workflow. Although the gHAND workflow has limitations, most markedly the requirement of manually digitising gully headcuts, it shows potential to be further developed to reliably map gullies of small- to large-scales in different geo-environments.