Impact of Arctic Sea Ice Heterogeneity on Surface Reflectance Evaluated with Airborne Imagery

The spectral reflectance of the snow-covered or bare sea ice in the Arctic is a critical parameter for determining the surface energy budget and for developing satellite passive remote sensing of clouds and aerosol particles. For static land surfaces, the bidirectional reflectance distribution function (BRDF) is acquired by sampling reflectance over multiple overpasses. The aggregated reflectance data are then fitted by a kernel-based approach. While the kernels were originally developed for vegetated surfaces, they have been extended to other surface types and the algorithm has been operationally implemented for imagery by Moderate Resolution Imaging Spectroradiometer (MODIS) and Visible Infrared Imaging Radiometer Suite (VIIRS). However, applying these kernels to the snow surface results in poor fitting due to its highly anisotropic reflectance. Even recently developed snow kernels, whose performance is yet to be validated against field observations, do not consider processes unique to sea ice such as surface roughness or floe drift. They are also confronted with sparse temporal and viewing angle sampling that stems from the sun-synchronous satellite orbits of MODIS and VIIRS. As a first step toward development of a kernel-based sea-ice BRDF retrieval, this study focuses on the packed, homogeneous sea ice surface before the melt onset and evaluates the performance of snow kernels. Specifically, we examine the impact of surface roughness on the directional and hemispherical reflectance using novel aircraft datasets.

We used data from the NASA Arctic Radiation-Cloud-Aerosol-Surface Interaction Experiment (ARCSIX), an aircraft campaign that took place near the northern coast of Greenland from May to August 2024. Particularly, we used airborne nadir-looking all-sky camera imagery, spectral irradiance by the Solar Spectral Flux Radiometer (SSFR) and laser altimetry data by the Land, Vegetation and Ice Sensor (LVIS). The camera imagery was radiometrically and geometrically calibrated to derive the directional reflectance. The imagery was then collocated against surface roughness derived by camera imagery and by LVIS when available, as well as against hemispherical albedo obtained from SSFR. We found that the snow kernels adequately capture the anisotropy of the camera-derived reflectance within the observed range of roughness. The kernel fit coefficients and predicted albedo showed high sensitivity to roughness, which modulated albedo by up to 10% in the shortwave infrared wavelength range. However, when limiting the viewing angle to the subset of angles that are accessible to satellite imagers, there is not enough information for the kernels to accurately predict the anisotropic reflectance and albedo perturbed by roughness. Additional constraints would likely be needed as a next step toward the retrieval of the sea ice BRDF influenced by surface roughness, leveraging other data such as multi-angle imagery (e.g., PACE, 3MI) and laser altimetry (e.g., ICESat-2, CryoSat).