Linking winter and spring thermodynamic sea-ice states at critical scales using an object-based image analysis of Sentinel-1

Changing Arctic sea-ice extent and melt season duration, and increasing economic interest in the Arctic have prompted the need for enhanced marine ecosystem studies and improvements to dynamical and forecast models. Sea-ice melt pond fraction f(p) has been shown to be correlated with the September minimum ice extent due to its impact on ice albedo and heat uptake. Ice forecasts should benefit from knowledge of f(p) as melt ponds form several months in advance of ice retreat. This study goes further back by examining the potential to predict f(p) during winter using backscatter data from the commonly available Sentinel-1 synthetic aperture radar. An object-based image analysis links the winter and spring thermodynamic states of first-year and multiyear sea-ice types. Strong correlations between winter backscatter and spring f(p), detected from high-resolution visible to near infrared imagery, are observed, and models for the retrieval of f(p) from Sentinel-1 data are provided (r(2) >= 0.72). The models utilize HH polarization channel backscatter that is routinely acquired over the Arctic from the two-satellite Sentinel-1 constellation mission, as well as other past, current and future SAR missions operating in the same C-band frequency. Predicted f(p) is generally representative of major ice types firstyear ice and multiyear ice during the stage in seasonal melt pond evolution where f(p) is closely related to spatial variations in ice topography.