Author(s): Dylan Blaskey; Merritt Harlan; Keith N. Musselman

Linked Author(s):

Keywords: Remote sensing; River ice; Synthetic aperture radar; Ice thickness; Ice phenology

Abstract:

The rapid pace of climate change is affecting Arctic winter conditions, influencing the timing and thickness of river ice, thereby posing considerable risks to ice travel safety, and underscoring the urgent need for improved remote sensing of river ice conditions. Recently, synthetic aperture radar (SAR) backscatter data has been shown to accurately quantify ice presence and thickness. Using Sentinel-1 SAR data and in-situ river ice observations across Alaska, we developed random forest models for predicting river ice phenology and thickness. Our phenology retrieval method achieved a validation accuracy rate of 97% for ice presence, effectively identifying break-up events within the preceding repeat cycle (approximately 9 days) 90% of the time, resulting in a 5.3-day root mean square error (RMSE) for the break-up date. Additionally, we used SAR data to quantitatively evaluate river ice thickness. On a regional scale, the random forest retrieval method had a RMSE of 14.7 cm. However, leveraging at least four years of training data, our location-specific random forest retrieval of ice thickness demonstrated an average RMSE of 10.7 cm for years not included in the training dataset. Our adaptable approach exhibits promise for widespread application in various regions to monitor environmental changes and enhance river ice safety.

DOI:

Year: 2024