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 85

However, if an acquisition of brightness temperatures for a smaller incidence
angle range is desired or sub-daily changes are to be observed, then the number of
available observations is much smaller and a more careful processing is suggested.
A single over

cient to observe an area from different incidence angles
as consecutive SMOS snapshot overlap considerably. We start converting the L1C
data into the earth reference frame (Huntemann et al. 2014 ), averaging the obser-
vations into single degree incidence angle bins and derive the standard deviation for
each bin, representing the variability of brightness temperatures for a given grid cell
within the few seconds the grid cell is observed by SMOS under an incidence angle
±

fl ight is suf

segment. The surface conditions will not change within this temporal scale.
Therefore, if this variation is high, a non-geophysical in

1 °

uence on the brightness
temperatures is detected and the data point is discarded. From the resulting data set,
also wider incidence angle range averages like the 40

fl

50 ° averages used in

-

Huntemann et al. ( 2014 ) for a SIT retrieval can be calculated.
Due to the aperture synthesis step of the SMOS image generation process, RFI
sources propagate in a ringing structure throughout the snapshot along speci c

lines. Therefore, the values from the different snapshots of one over ight falling

into an incidence angle bin of one grid cell may show a large variability in
brightness temperature. As compromise between data quality and loss of data due to

fl

filtering we use 10 K standard deviation as a threshold, i.e. bins with a standard
deviation above 10 K are discarded. In addition we require at least 3 data points for
a valid average. The detailed statistics of an example day, the 25 Dec 2010, are
shown in Fig. 1 . The upper plot (a) shows the 3 data point criterion in a histogram
as a red line. On the left axis the number of data points is shown while the right axis
shows the percentage of remaining data points when applying the data point
requirement as a lower threshold. The criterion of having at least 3 data points is
ful

lled in a 63.3 % of the cases as can be read from this axis. The hatched area
shows data where the criterion is not ful lled. The shape of the histogram originates

from the incidence angle distribution within a single snapshot, where some inci-
dence angles have more observations in one over fl ight than others (Martin-Neira

et al. 2002 ).
For incidence angle bins with more than 3 data points, the standard deviation is
calculated. The histogram of standard deviations of TBh per incidence angle is
shown in Fig. 1 b. The standard deviation seems relatively evenly distributed over
all incidence angles with a peak at about 1 °

incidence angle. However, allowing just a maximum standard deviation of 4 K
would discard about 50 % of all data. Since the radiometric accuracy of MIRAS
varies not with the incidence angle but with the position of the point within one
snapshot (Khazaal and Anterrieu 2009 ; Martin-Neira et al. 2002 ), we de

4 K standard deviation and 30 60

- -

ne for our
application a required standard deviation of less than 10 K. In the example day of
Fig. 1 b about 87 % of the data remain. Again, the red hatches mark the data regime
not ful

lling this criterion.
A brightness temperature example of thin sea ice of 1 Oct 2010 is shown in
Fig. 2 . The left (a) shows TBh and TBv of an over

fl ight at 7:36 UTC. The moderate