基于Sentinel遥感观测及多源地理空间数据的全球不透水面10米分辨率产品（GISA-10m）

提出融合多源地理空间数据的全球不透水面制图方法，从现有全球不透水面数据集和OpenStreetMap中自动提取6千万个训练样本点，并进行时空清洗和筛选，对270万景Sentinel-1/-2影像进行自适应区域分类，得到全球10米不透水面产品（GISA-10m）。经过多源独立测试样本，GISA-10m的整体精度超过86%，并在农村和干旱地区表现出更少的遗漏。GISA-10m 数据集可用作地球系统科学的基本参数，并将为城市规划和水循环研究提供宝贵支持。

This study proposes a global impervious surface mapping method integrating multi-source geospatial data. 60 million training sample points are automatically extracted from existing global impervious surface datasets and OpenStreetMap, followed by spatio-temporal cleaning and screening. Adaptive regional classification is then performed on 2.7 million scenes of Sentinel-1 and Sentinel-2 imagery, resulting in the global 10-meter impervious surface product (GISA-10m). Evaluated with multi-source independent test samples, the overall accuracy of GISA-10m exceeds 86%, and it exhibits fewer omission errors in rural and arid regions. The GISA-10m dataset can serve as a fundamental parameter for Earth System Science, and will provide valuable support for urban planning and water cycle research.

搜集汇总

背景与挑战

背景概述

该数据集是全球10米分辨率不透水面产品（GISA-10m），基于Sentinel遥感观测和多源地理空间数据制成，采用自适应区域分类方法，整体精度超过86%，尤其在农村和干旱地区减少遗漏。它适用于地球系统科学研究，并为城市规划和水循环分析提供支持，数据覆盖2016年全球范围，以开放获取方式共享。