Satellite Sea Ice Concentration Observation Uncertainty Is State-Dependent and Exceeds Prescribed Values in Data Assimilation

Joey Rotondo

University of Washington

Tuesday, Oct 20, 2026, 2:00 pm MT

JOIN THE WEBINAR:

Abstract

Earth system data assimilation systems commonly assimilate satellite retrievals of sea ice concentration (SIC). How much weight each retrieval receives relative to the model is set by its observation uncertainty, which is poorly constrained and is almost always prescribed as a constant, typically 0.10 to 0.15. We instead measure this uncertainty from the assimilation system itself. An iterated Desroziers diagnostic, computed in a bounded framework that keeps SIC between 0 and 1, gives the observation uncertainty for two passive microwave products and one optical product in both hemispheres. That uncertainty is far from constant: it is 0.02 to 0.10 for a retrieval of open water (0) or complete cover (1) and 0.2 to 0.35 for a retrieval of one half. The exception is complete cover in the melt season, which the model rarely simulates and the method therefore cannot constrain properly. A similar curve emerges from three collocated satellite products, with no model involved, and from the uncertainty one product publishes with its retrievals. Most of the uncertainty is not instrument noise; its pattern is consistent with known passive microwave retrieval errors over thin and ponded ice. A constant prescription therefore gives mid-range retrievals too much weight and retrievals near 0 and 1 too little.

Bio