A New Method to Filter Out Radio-Frequency Interference (RFI) from SMOS Level 1C Data for Sea Ice Applications - Towards an Interdisciplinary Approach in Earth System Science - page 88

averaging method. In Fig. 3 a the SIT retrievals based on un ltered L1C data are

spatially inhomogeneous in the Kara and Barents seas on that day due to RFI.
Figure 3 b shows the result after applying the original RFI

filtering method of
discarding complete snapshots. The SIT in the Kara and Barents seas seem more
reasonable and more spatially consistent compared to the un ltered version.

However, this strict filters obviously useful data east of Greenland where

the SIT looks undistorted in the un

filter also

ltered data. Figure 3 c shows the results after

applying the new ight where the

standard deviation is higher than 10 K within each single incidence angle bin.
The SIT in the Kara and Barents seas look reasonable and very similar to the SIT in
the original

filter masking out all data from a single over fl

filter in Fig. 3 b. In addition the SIT east of Greenland is preserved and
show reasonable values as in the un ltered version before. The difference between

the two filters mostly agree well.

The highest values in the difference map are from SIT values of 30

filters is shown in Fig. 3 d. Results after the two

50 cm, where
the retrieval is more sensitive to small changes in the brightness temperatures.
The small SIT offset of a few cm in the thin ice area in the Kara sea stem from the
averaging over several over

-

fl ights, i.e., a later over fl ight of that day contains RFI

somewhere in the snapshot. In the old filter these data are discarded, where the new

filter uses these data as long as they show only small variability within the few
seconds of acquisition. An obvious interpretation is that this SIT difference, which
is positive almost everywhere, originates from the ice thickness increase by freezing
within that day.

4 Conclusion

A new data filter and averaging method of SMOS brightness temperatures for Level

1C data have been introduced, filtering for unphysical changes of brightness tem-

peratures within small incidence angle and temporal variations. The filtered, inci-

dence angle binned data may serve as stable comparison to model data or as input
for a retrievals of geophysical parameters like ice thickness. However, also retri-
evals of other parameters might pro

t from this technique.
Thresholds of 3 observations at least with a standard deviations of 10 K or less
per over fl ight of one grid cell per incidence angle bin are suggested as filter criteria.

This way, the in uence of RFI on an example of sea ice thickness retrieval was

eliminated preserving most of the data while achieving better data coverage than
with the previously used RFI

fl

filter discarding complete snapshots. The given
conditions leave about 3 to 20 valid observations for a single average. The proposed
method might miss RFI in case of small RFI effects, which could introduce an offset
to the brightness temperatures. This effect can be reduced further by tuning the
parameters of minimum number of observations and of standard deviation. Since
globally the RFI reduced since the beginning of the SMOS mission, for more recent
data a more strict criteria for the standard deviation is suggested.