The UH SMOS product by design (and by the soil satellite's limitations) specializes in measurement of thinner ice. That's an important contribution because thin ice is an area of weakness for other ice thickness products.
The flip side is that UH SMOS cannot measure thick ice, maxing out at about 1.2 m. One of the UH files (below) shows the ratio of measured thickness to maximal possible measurable thickness, ie where SMOS starts to lump all the thicker ice into one bin. The map below shows in pink where this ratio is 97% or greater. That is of interest but the best part of a SMOS thickness map is elsewhere, on the periphery.
UH also provides a most welcome uncertainty estimate map in the daily netCDF bundle. This may diminish later in the season as Oct 1st is at the limits of usefulness. Their recent paper needs to be consulted as to exactly what uncertainty means here and how it is calculated.
The uncertainty data, like the thickness data, is shown in meters whereas it might have been in percentages. That is, when the ice is 1.00 m thick and the uncertainty is 0.25 m, that is a different situation from the ice being 0.5 m thick and the uncertainty still 0.25 m.
These are formal uncertainties; they do not mean that the thickness posted is necessarily wrong by that much. In fact, combining successive days in their time series might very much reduce uncertainty, which originates in part from variable radio frequency interference from ground stations and other sources.
'Combine Arrays' in Panoply can draw the percent error map from the two data sets (3rd image, looks too much like 2nd?). It would be trickier to work out a rolling average and the diminishing uncertainty.
The SMOS-derived snow surface temperature, 4th image, is quite relevant to the ability of cold air to induce ice thickness growth. The UH product may be better than modeled products in this regard.
The palette names are shown to allow replication by others or to revisit later for a time series. Panoply provides over a hundred of these so trial and error duplication is slow. The original color tables can be found in palette galleries by searching on the name provided. Some of these like .cpt are dynamic scripts rather than an explicit layout.