Unsupervised Intelligent Framework for Earth Science Remote Sensing Applications Based on Clustering Algorithms
Journal of Environmental & Earth Sciences · 12.08.2026 · 355-381
DOI: 10.30564/jees.v8i8.13389 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Environmental & Earth Sciences, online erschienen am 12.08.2026. Verfasst von Hongyan Zhang und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
Archives of Earth observations have grown and continue to grow in volume, modality, and temporal density, providing unprecedented access for monitoring the Earth system and exacerbating the paucity and high cost of quality labels. This is a review of unsupervised intelligent methods for Earth science remote sensing based on clustering algorithms and capable of analyzing data in a scalable, interpretable manner under domain shift. We begin by summarizing the remote sensing data properties that make it difficult to learn unsupervised, and they are: high-dimensionality, mixed-pixels, sensor-specific noise, spatial autocorrelation, non-uniform time series, heterogeneous across sensors. We next list key clustering families that are applied in remote sensing, including centroid-based and medoid-based, probabilistic mixtures, density-based, hierarchical, graph clustering, spectral clustering, subspace/low-rank models of hyperspectral data, fuzzy clustering to ambiguity, and deep clustering using self-supervised representation learning. On these algorithmic mammoths, we introduce end-to-end patterns of framework design, i.e., preprocessing, harmonization, sampling, and tiling on a scale of archive computations, spatial-context synthesis with super pixels and regularization, multi-modal fusion with shared latent spaces and spatiotemporal clustering with consistency and drift management. Lastly, we also talk about application areas like stratifying land-covers, grouping of hyperspectral materials, discovering SAR regimes, hazard surveillance, and analyzing climate-hydrology regimes with the focus on not analyzing by labels but by stability tests, uncertainty reporting, and process-based validation. The review summarizes practical advice and presents open dilemmas to effective, operational unsupervised intelligence for Earth observation.
Zitieren
APA
Zhang, H., & Zhao, L. (2026). Unsupervised Intelligent Framework for Earth Science Remote Sensing Applications Based on Clustering Algorithms. Journal of Environmental & Earth Sciences, 355-381. https://doi.org/10.30564/jees.v8i8.13389
Vancouver
Zhang H, Zhao L. Unsupervised Intelligent Framework for Earth Science Remote Sensing Applications Based on Clustering Algorithms. Journal of Environmental & Earth Sciences. 2026:355-381. doi:10.30564/jees.v8i8.13389
BibTeX
@article{Zhang2026,
author = {Zhang, Hongyan and Zhao, Li},
title = {Unsupervised Intelligent Framework for Earth Science Remote Sensing Applications Based on Clustering Algorithms},
journal = {Journal of Environmental & Earth Sciences},
year = {2026},
pages = {355--381},
doi = {10.30564/jees.v8i8.13389}
}
EUVIMED