Harnessing machine learning for landscape character management in a shallow relief region of China

Huang, T. ORCID: https://orcid.org/0000-0001-5308-0479, Zhang, Y. ORCID: https://orcid.org/0009-0007-6550-4954, Li, S. ORCID: https://orcid.org/0000-0003-4429-8673, Griffiths, G. ORCID: https://orcid.org/0000-0003-0714-6689, Lukac, M. ORCID: https://orcid.org/0000-0002-8535-6334, Zhao, H. ORCID: https://orcid.org/0009-0004-5837-079X, Yang, X. ORCID: https://orcid.org/0000-0003-0788-1480, Wang, J. ORCID: https://orcid.org/0009-0008-2448-945X, Liu, W. ORCID: https://orcid.org/0009-0003-6162-5370 and Zhu, J. ORCID: https://orcid.org/0000-0003-2001-797X (2023) Harnessing machine learning for landscape character management in a shallow relief region of China. Landscape Research, 48 (8). pp. 1019-1040. ISSN 1469-9710 doi: 10.1080/01426397.2023.2241390

Abstract/Summary

Due to China’s rapid human activity expansion, landscapes have lost their distinctive and typical characteristics. This paper addresses this issue by proposing a landscape character management framework for the Beijing shallow relief area. The framework utilises machine learning techniques to assess and enhance landscape integrity. The process involves landscape character identification through Principal Component Analysis, Gaussian Mixture Model clustering, and Canny Edge Detection. Additionally, a comprehensive landscape sensitivity evaluation considers both landscape character and visual sensitivity. The study develops five landscape management strategies based on field surveys and employs a Transformer Matrix Process and a multi-expert decision-making mechanism. Extensive validation confirms the framework’s effectiveness in improving the recognition accuracy of Landscape Character Types. The findings reveal that over 30% of the landscape characters in the study area require improvement. Importantly, the machine learning techniques employed in this study can be transferred to other regions, facilitating landscape characterisation, evaluation, and management.