Arctic sea ice surface roughness at micro- and macro-scales plays a critical role in modulating skin drag and form drag, respectively, thereby influencing the air-ice momentum and heat exchange across the Arctic Ocean. However, the potential for mapping surface roughness at both scales remains challenging for microwave remote sensing. In this study, we integrate ASCAT backscatter observations with ICESat-2 (IS2)–derived macro-scale roughness to estimate winter Arctic sea ice surface roughness across scales. We first simulate the ASCAT backscatter coefficients and their dependence on incidence angle (IAD) for the winter of 2018-2019 and 2019-2020 using the Snow Microwave Radiative Transfer model and inverse the micro-scale roughness for level sea ice (IS2 macro-scale roughness < 0.05 m). Furthermore, we develop a Piecewise Embedding Neural Network (PENN) to predict macro-scale roughness across the Arctic during 2007–2022 using four predictors: ASCAT backscatter and IAD, PIOMAS sea ice thickness, and day of winter. The simulated backscatter coefficients and IAD show good agreement with ASCAT measurements over level ice, with RMSEs < 1.3 dB and < 0.01 dB/°, respectively. The retrieved micro-scale ice surface roughness ranges from 0.2 to 1.5 cm across the Arctic, with means of 0.9 cm. PEN shows good accuracy against independent IS2 and OIB macro-scale roughness observations with RMSEs of 0.075 m and 0.076 m, respectively, corresponding to reductions of 35% and 40% compared to empirical regression model. The maximum winter mean roughness occurs north of Greenland and Canadian Arctic Archipelago (> 0.5 m), whereas the lower values are found in the Eurasia seas (< 0.2 m). The Arctic-wide average macro-scale roughness increases from 0.13-0.17 m in November to 0.27-0.30 m in April during 2007–2022. The maximum roughness appears in 2014–2015 and minimum in 2012–2013.
 

Arctic, sea ice, roughness, remote sensing