Detecting Inland Waterbodies Using GNSS-R Data: Intercomparison of Previous Methods and a New Machine Learning Approach

Abstract:

Inland waterbodies play a critical role in hydrological processes and ecosystems. Accurately capturing short-term dynamics of waterbodies is challenging because the extent and location are sensitive to a variety of factors including seasonal variations, flooding, meteorological events, agricultural usage, etc. Monitoring waterbodies is necessary to improve understanding of the complex short-term dynamics of inland waterbodies. While field research, drones, aircraft, and optical sensors on board satellites can be utilized, these methods are limited by cloud cover, temporal scale, spatial coverage, and expensive budget requirements. Recently, NASA launched the Cyclone Global Navigation Satellite System (CYGNSS): an innovative constellation of eight microsatellites designed to monitor tropical storm intensification in oceans and improve cyclone path prediction by measuring reflected Global Positioning System (GPS) signals. The reduced revisit time of 2.8 (median) and 7.2 (mean) hours per day between CYGNSS microsatellites within latitudes ±38º, ability for GPS L-band signals to penetrate through clouds and vegetation, and capability to observe surface reflectivity differences over land have lead researchers to utilize CYGNSS data for a variety of land-based applications. A few methods have been proposed to detect waterbodies using CYGNSS. Specifically, methods based on thresholding binary prediction, forward modeling, and random walker algorithms have been proposed (Gerlein-Safdi and Ruf 2019; Morris et al., 2019; Wan et al., 2019; Chew and Small 2020). Our research aimed to compare previous research to identify strengths and limitations of each CYGNSS-based waterbody detection method. Additionally, this study proposes a new method to leverage machine learning for waterbody detection using CYGNSS data. The results from this study can be applied to improve monitoring of short-term inland waterbody dynamics using Global Navigation Satellite System Reflectometry (GNSS-R).