Deriving a frozen area fraction from Metop ASCAT backscatter based on Sentinel-1

Surface state data derived from space-born microwave sensors with suitable temporal sampling are to date only available in low spatial resolution (25-50km). Current approaches do not adequately resolve spatial heterogeneity in landscape scale freeze/thaw processes. We propose to derive a frozen fraction instead of binary freeze/thaw information. This introduces the possibility to monitor gradual freezing and thawing of complex landscapes. Frozen fractions were retrieved from Advanced Scatterometer (ASCAT; C-band) backscatter on a 12.5km grid for three sites in non-continuous permafrost areas in northern Finland and the Austrian Alps. To calibrate the retrieval approach, frozen fractions based on Sentinel-1 Synthetic Aperture Radar (SAR, C-band) were derived for all sites and compared to ASCAT backscatter. We found strong relationships for ASCAT backscatter with Sentinel-1 derived frozen fractions (Pearson correlations of -0.85 to -0.96) for the sites in northern Finland and less strong relationships for the alpine site (Pearson correlations -0.579 and -0.611, including and excluding forested areas). Applying the derived linear relationships, predicted frozen fractions using ASCAT backscatter values showed root mean square error (RMSE) values between 7.26% and 16.87% when compared with Sentinel-1 frozen fractions. Validation of the Sentinel-1 derived freeze/thaw classifications showed high accuracy when compared to in situ near-surface soil temperature (84.7% to 94%). Results are discussed with regard to landscape type, differences between spring and autumn and gridding. This study serves as a proof of concept, showcasing the possibility to derive frozen fraction from coarse spatial resolution scatterometer time series to improve the representation of spatial heterogeneity in landscape scale surface state.