A 60-cm Aerial Photography-based Water Body Mapping: Application to the Tama and Tsurumi Rivers in Japan

The current global water body maps offer an approximate resolution of 30 meters, contingent on available remote sensing data. However, to address the needs of advanced applications like global carbon cycle analysis and real-time flood predictions, a water body map with higher spatial resolution becomes imperative, especially for resolving smaller rivers. Traditional water extraction methods rely on water indexes that combine visible and infrared spectra. State-of-the-art remote sensing data, including aerial photography with spatial resolutions in the order of a few meters, often includes only the visible spectrum.

In response to this challenge, we have developed a water extraction method at an impressive 60cm resolution utilizing Bayesian inference based solely on the visible spectrum from aerial photography, without using the infrared spectrum. To enhance our methodology, we integrated references of water existence from a Landsat-based dataset called G1WBM and Open Street Map (OSM), along with a hydrography dataset (J-FlwDir) presumed to be linked to water bodies.

Our method successfully detected the main streams of the Tsurumi River and Tama River in Japan, including their previously unrecognized tributaries in the Landsat-based dataset. Notably, this study identified rivers with a width exceeding 10 meters. Furthermore, it contributed valuable area information for 37% of small rivers represented as "line" features in the OSM.

These findings underscore the effectiveness of our Bayesian water detection approach, which leverages hydrography data and existing water body maps to improve the spatial resolution of large-scale water mapping significantly. Notably, this improvement is achieved using remote sensing data that lacks infrared spectra, showcasing the potential of our method in advancing the accuracy and precision of global water mapping efforts.