Work

Ecosystem-Oriented Hierarchical Classification with Multispectral Data in Heterogeneous Arid Regions: A Case Study in Kashi, Xinjiang, China

Year: 2026

Type: article

Abstract: Arid mountain–oasis–desert regions exhibit strong terrain and surface-cover heterogeneity that is difficult to represent using conventional land-cover classification schemes. This study developed an ecosystem-oriented hierarchical framework for the Kashi region by combining a terrain-constrained Mountain extraction method with Automatic Deep Forest Shrinkage model (ADeFS). Nine ecosystem elements were defined: Mountain, Water, Forest, Cropland, Lake, Grassland, Desert, Ice, and Human. Mountain was first delineated as an independent physiographic element using a locally derived baseline surface, relative relief, slope, and topographic position, thereby reducing semantic overlap between terrain units and spectrally similar surface-cover classes. ADeFS was then adapted to classify the seven non-mountain classes, and Lake was subsequently separated from the unified Water class through visual interpretation. The results show that ADeFS achieved the highest accuracy, with an overall accuracy of 88.7% and a Kappa coefficient of 0.868. Independent field validation of the 2026 map yielded an overall accuracy of 86.2% and a Kappa coefficient of 0.825. From 2015 to 2026, the mountain-oasis-desert structure remained broadly stable, while Desert and Ice decreased and Forest, Grassland, and Cropland expanded. Ecosystem-element transitions were concentrated before 2021 and weakened thereafter. Landscape metrics showed that Desert remained the dominant matrix, Grassland had the highest patch density and edge density, and Cropland became increasingly aggregated within oasis agricultural areas. The framework provides an ecologically interpretable approach for ecosystem-element mapping and long-term monitoring in arid heterogeneous regions.

Source: Land

DOI: 10.3390/land15091561

Authors: Long Jia, Wenjin Wu, Xinwu Li, Yuhan Xie, Guillermo Jose Martínez Pastur

Institutions: Chinese Academy of Sciences, Aerospace Information Research Institute, University of Chinese Academy of Sciences, International Research Center of Big Data for Sustainable Development Goals, Consejo Nacional de Investigaciones Científicas y Técnicas

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

SDGs: Life on land, Climate action

Open Access status: gold

Estimated APC paid: $3,137