Evaluating the performance of pixel-based and object-based multidimensional clustering algorithms for automated surface water mapping

作者：Bohao Li, Kai Liu, Ming Wang, Yanfang Wang, Linmei Zhuang, Weihua Zhu, Chenxia Li, Linhao Zhang, Yanan Chen · 发表于：Geo-spatial Information Science · 年份：2025 · DOI：10.1080/10095020.2025.2523993 · 被引用次数：3 · 研究领域：Flood Risk Assessment and Management、Hydrology and Watershed Management Studies、Hydrological Forecasting Using AI

Remote sensing observations of surface water are vital for effective water resource management and sustainable development. Unsupervised classification holds promise for automating large-scale surface water detection, and it helps solve the difficult problem of sample collection in supervised classification. Here, we refined the water identification rule for automated surface water extraction on the basis of multidimensional clustering and a supervised classifier. We subsequently comprehensively investigated the classification performance of k-means, hierarchical clustering, and spectral clustering with 57 different feature combinations (features consisting of two bands, B8 and B12, along with four water indices: the automated water extraction index (AWEI), multiband water index (MBWI), normalized difference water index (NDWI), and modified normalized difference water index (MNDWI)) in eight challenging scenarios in China. These comparative experiments were performed from both pixel-based and object-based perspectives. The results show that pixel-based hierarchical clustering, which uses the optimal feature combination of B8, the NDWI, and the MBWI, is the algorithm with the best overall performance, with kappa coefficients exceeding 0.9 in each scene. The object-based hierarchical clustering using the optimal feature combination B8, B12, MNDWI, and MBWI achieves a kappa coefficient exceeding 0.85 in almost all scenes. This algorithm is suitable for scenes without small water...