Observations of River Ice Breakup Using GNSS-IR, SAR, and Machine Learning

Global navigation satellite system-interferometric reflectometry (GNSS-IR) is an emerging sensor technique that has become well-established for water-level monitoring. While GNSS-IR has previously been used for monitoring properties of lake ice and sea ice, it has not been applied for monitoring river ice. This article presents results from monitoring river ice breakup at three sites in Canada. GNSS-IR data were compared with co- located time-lapse camera imagery and it was found that GNSS-IR signal was sensitive to periods where there is rough or broken ice in view of the sensor. Using data from Sentinel-1 and the RADARSAT Constellation Mission (RCM), the first ever comparison of GNSS-IR with synthetic aperture radar (SAR) imagery is presented and a negative correlation of −0.8 is found between the GNSS-IR spectral power and SAR backscatter. Three classification algorithms of varying complexity (
$K$
-means clustering, neural network, and random forest) are explored for detecting river ice using GNSS-IR. Using a shallow neural network with two hidden layers, an optimal accuracy of up to 94% is achieved over all three sites, or 97% when mixed water–ice conditions are excluded from the analysis. In summary, GNSS-IR has strong potential for ice monitoring applications, including monitoring the formation of ice jams.