Advancing landscape characterisation: a comparative study of machine learning and manual classification methods
Huang, T., Griffiths, G., Huang, B., Zhu, J., Warnock, S. and Lukac, M.
It is advisable to refer to the publisher's version if you intend to cite from this work. See Guidance on citing. To link to this item DOI: 10.1016/j.ecoinf.2025.103349 Abstract/SummaryThis study evaluates and compares three automated classification methods for Landscape Character Assessment (LCA) to assess their suitability for consistent, objective, and scalable mapping. We applied One-pass Multi-view Clustering (OPMC), Self-Organising Feature Map clustering (SOFM), and Swin Transformer Segmentation Clustering (STSC) to classify Landscape Character Types in Bannau Brycheiniog National Park, Wales, UK. Their outputs were compared against an expert-based manual classification (EBMC) using pixel-by-pixel accuracy assessment. To interpret model outputs, we used SHapley Additive exPlanations analysis to quantify the influence of key landscape character elements on classification outcomes. STSC showed the highest agreement with EBMC, followed by SOFM and OPMC. Across all models, geology, historic landscape, and soil type were the most influential variables, while habitat and landform contributed less. The automated methods demonstrated strong spatial coherence and boundary delineation comparable to expert-based mapping. Our findings demonstrate the potential of automated approaches to improve the consistency, efficiency, and objectivity of LCA and support their integration into scalable landscape characterisation frameworks for planning and management applications. Source code and datasets are available on GitHub (https://github.com/TingtingHwang/BBNP_LCA).
Altmetric Deposit Details University Staff: Request a correction | Centaur Editors: Update this record |