Abstract
Reliable and accurate high-resolution maps of surface waters are critical inputs to models that help understand the impacts and relationships between the environment and human activities. Advances in remote sensing technology have opened up the possibility of mapping very small bodies of water that are closely related to people's daily lives and are mostly affected by anthropogenic pressures. However, a robust and scalable method that works well for all types of water bodies located in diverse landscapes at high-resolution has yet to be developed. This paper presents a method that can accurately extract surface water bodies up to a very fine scale in a wide variety of landscapes. Unlike optical image-based methods, the proposed method exploits the robust assumption that surface water is flat as gravity always pulls liquid molecules down. Based on this natural law, the proposed method extracts accurate, high-resolution water bodies including their elevations in a fully automated manner using only airborne LiDAR data. Extensive experiments with large ($\approx$ 2,500$km^{2}$) and diverse landscapes (urban, coastal, and mountainous areas) confirmed that our method can generate accurate results without site-specific parameter tunings for varied types of surface water. The proposed method enables an automated, scalable high-resolution mapping of a full 3D topography that includes both water and terrain, using only point clouds for the first time. We will release the code to the public in the hope that our work would lead to more effective solutions to help build a sustainable and resilient environment.
Abstract (translated)
URL
https://arxiv.org/abs/2301.06567
