Classifying watershed based on a semi-supervised approach under limited hydrological data conditions: A case study of the Chishui Watershed, China

文献类型： 外文期刊

第一作者： Wang, Yue

作者： Wang, Yue;Xia, Jihong;Cai, Wangwei;Zu, Jiayi;Wang, Qihua;Ji, Shuyi;Ren, Yifei;Huang, Yating;Huang, Siqi;Che, Xuan

作者机构：

关键词： Watershed classification; Physical features; Time-series classification; Semi-supervised Classification; Hydrological process

期刊名称：JOURNAL OF HYDROLOGY-REGIONAL STUDIES （ 影响因子：5.0； 五年影响因子：5.3 ）

ISSN：

年卷期： 2025 年 61 卷

页码：

收录情况： SCI

摘要： Study region: The Chishui Watershed, located in southwest China. Study focus: In data-limited regions, traditional frameworks combining deductive and inductive approaches to improve the hydrological relevance of clusters are often inapplicable. These frameworks rely on mapping relationships between physical and hydrological features for feature selection, which requires extensive hydrological data. Accordingly, this study proposes a framework for classifying watershed by semi-supervised method that integrates deductive and inductive approaches using limited data. Additionally, SHapley Additive exPlanations (SHAP) is then used to assess feature importance and explain performance improvements. New hydrological insights for the region: Validation results show that, in data-limited regions, the framework achieves better clustering performance and more effectively captures hydrological processes compared with commonly used unsupervised watershed classification and other semisupervised clustering methods. This improved accuracy stems from two aspects: (1) decomposing runoff time series to extract key hydrological information, and (2) identifying hydrologically relevant features while suppressing irrelevant or noisy ones. This study demonstrates the potential of semi-supervised methods for watershed classification in data-limited regions and offers new insights into the relationship between classification and feature selection. Unlike traditional frameworks that separate feature selection from clustering, this framework uses limited hydrological knowledge as constraints to directly and accurately adjust feature importance during clustering, thereby improving classification accuracy. Overall, watershed classifications with more hydrological significance can better support water resource management in data-limited regions.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序