A study of Arctic sea-ice surface albedo and its uncertainty: impact of varying insolation and instrument characteristics

Abstract

The Arctic sea-ice surface albedo is observed by ground-based, airborne, and satellite instruments. However, the observed albedo is influenced by varying insolation, instrument characteristics, and associated retrieval processes. This thesis quantifies these influences, develops and improves corrective schemes, and gives recommendations to increase the accuracy of the surface broadband albedo products. The analysis is based on simulations with the radiative transfer model SCIATRAN. A single-scattering property database of nine ice crystal habits is implemented into SCIATRAN to enable a realistic snow surface reflection. The associated far-field assumption of snow grains is justified. A comparison of simulated and measured reflectance factors and albedo spectra reveals discrepancies of usually less than 0.05, legitimating the utilized SCIATRAN set-up. The atmosphere decreases the black-sky surface broadband albedo by a RMSD of less than 0.04. An optically thin (thick) cloud additionally influences the albedo of up to a RMSD of 0.02 (0.06). The albedo from irradiance measuring devices suffers from the instrument’s cosine error (RMSD<0.15) and has to be corrected for by calibration factors (RMSD<0.03). The total uncertainty of the satellite broadband albedo (RMSD<0.26) is mainly controlled by the anisotropic correction (RMSD<0.25) and the narrow-to-broadband conversion (NTBC) (RMSD<0.09). New, most accurate NTBCs are developed (RMSD≤0.02). Applying those and the RossThick-LiSparseReciprocal or the modified Walthall model as angular correction almost halve the total satellite albedo uncertainty. The new NTBC improves the MERIS derived sea-ice surface broadband albedo by up to one third. Surface inhomogeneities also significantly affect the observed surface albedo. A seasonal comparison of the Arctic sea-ice surface broadband albedo from MERIS, CLARA-A2, and ERA5 reanalysis reveals RMSDs exceeding 0.10 which is partly due to the above-mentioned uncertainties.