2024 Volume 18 Issue 1 Pages 1-6
The current global water body maps have a resolution of approximately 30 m depending on the available remote sensing data. A water body map with higher spatial resolution is required to distinguish smaller rivers for advanced applications involving global carbon cycle and real-time flood predictions. Conventional water extraction methods use water indices that combine visible and infrared spectra. State-of-the-art remote sensing data including aerial photography, offers a few-meter spatial resolution but only contains the visible spectrum. Here, we established a water extraction method at 60-cm resolution via Bayesian inference, using aerial photography’s visible spectrum combining the Landsat-based Global 1-second Water Body Map and the Open Street Map (OSM). The Japan Flow Direction Map (J-FlwDir) was used to link the water bodies. Our method detected the main streams of the Tsurumi and Tama rivers and their tributaries, which were not resolved by the Landsat-based dataset. With our approach, rivers with widths >10 m were detected and water extent was obtained for 37% of small rivers, represented as lines on the OSM. These findings indicate that our method, solely based on visible spectra with hydrography data and existing water body maps, improves the spatial resolution of water mapping without requiring infrared.