Fundamental Data for Environment-Aware Benchmarking of Spectral Water Indices Across Regions and Seasons

This globally representative benchmark dataset is designed for the robust evaluation of surface-water mapping algorithms using Sentinel-2 Level-2A surface reflectance imagery. At its core, the dataset comprises 24,000 manually annotated validation samples (water/non-water) systematically distributed across 12 eco-geographic regions and four seasonal variations, covering a total of 48 heterogeneous, independent scenes. Building upon these ground-truth samples, the dataset contains comprehensive, pre-calculated performance metrics for 12 representative spectral water indices, with a primary focus on their peak accuracy, robustness, and threshold stability. The data is structured to facilitate cross-index comparisons and provides a detailed analysis of spatial and seasonal performance variations. By highlighting the strengths and weaknesses of each index under diverse environmental conditions, this dataset serves as a foundational resource for environmental sensitivity analysis and the development of automated water extraction models in real-world settings.