Assessing the Relative Performance of GNSS-R Flood Extent Observations: Case Study in South Sudan

Flooding is one of the deadliest and costliest natural disasters. Climate change-induced flooding events are increasing worldwide, disproportionately impacting low-income and developing communities. While early warning systems save lives, satellite-based observation systems are vital for the disaster relief and recovery phases. Current satellite-based operational flood products are largely based on either optical remote-sensing methods, which exhibit a limited ability to detect water through clouds and vegetation, or microwave remote sensing, which provides relatively low spatial and temporal resolution. New small satellite constellations using radar or GNSS reflectometry (GNSS-R) have been shown to enhance our ability to overcome these deficiencies. In this work, we quantify the performance of using GNSS-R measurements from the NASA Cyclone Global Navigation Satellite Systems (CYGNSS) satellite constellation to map surface water in South Sudan and the Sudd wetland in comparison with a set of representative operational products. We make quantitative comparisons of our results with operational flood products based on Visible Infrared Imaging Radiometer Suite (VIIRS) and MODIS and with C-band Sentinel-1 synthetic aperture radar. We find that our method detects 35.4% more surface water than Sentinel-1, while the VIIRS- and MODIS-based products underestimate by 4.8% and 83.7%, respectively. We use several metrics commonly used to evaluate classification performance: precision, true positive rate (TPR), true negative rate (TNR), F2-score, and the Matthews correlation coefficient (MCC) and assess the comparisons in this statistical framework. We discuss the consequences of our findings, including ways CYGNSS data may enhance current flood products and assist decision-makers and emergency managers.