Satellite sea-ice concentration retrievals are commonly assimilated using a constant observational uncertainty of 0.10–0.15. This research instead estimates uncertainty within the assimilation system through an iterated Desroziers diagnostic constrained to the physical 0–1 concentration range. Two passive-microwave products and one optical product in both hemispheres show strongly state-dependent errors: roughly 0.02–0.10 near open water or complete cover, and 0.20–0.35 around half cover. Complete cover during melt remains difficult to constrain because the model rarely reproduces it. Comparisons among three collocated satellite products and published retrieval uncertainties support the pattern. The results suggest constant uncertainty overweights intermediate retrievals and underweights values near zero and one, with thin and ponded ice contributing errors beyond instrument noise.
Tue, Oct 20 · 20:00 UTC · Online