# Merged sea ice drift validation

#### 1. The Merged product

Since 1991, SSM/I data are available at the frequency of 85,5 GHz; these data are available at horizontal and vertical polarizations with a pixel size of 12,5 x 12,5 km (Ezraty et al., 2004). SeaWinds/QuikSCAT provides backscatter maps at a pixel resolution of 12,5 km from which sea ice drifts are also derived (Ezraty and Piollé, 2004a).

IFREMER makes available a 'Merged' product since 1999, based on the combination of SSM/I and SeaWinds/QuikSCAT drifts at 3 and 6-day lags. Drift vectors are estimated every five pixels of the polar stereo projection NSIDC grid (drift grid resolution is 62,5 x 62,5 km). The benefits to combine these two drifts are in three points :

- The geometry ground swath and orbit enable backscatter measurement very near the North Pole, the data gap is only 40 km for QuikSCAT compared with 254 km for SSM/I.
- The combination increases by 12 to 15% the drift data density, in particular at early fall and early spring (see Figure 7).
- Merging process enables discrimination of the few vector outliers remaining in the individual products (Ezraty and Piollé, 2004b).

#### 2. Comparison with buoys

At 3 day-lag, distribution of buoys and Merged drifts show mean drift of 17,7 km for buoys and 16,9 km for Merged, and similar standard deviation of 14,4 km. This is highlighted by figure 1 where Merged and buoys drifts are presented with probability in logarithmic scale, showing a slight dissymmetry towards the strong values for the buoys drifts.

* *

The mean bias (obtained from the difference between buoys and Merged drifts) is 0,30+/-1 km in a 95% confident interval, which is compatible with the large samples (13660 comparisons). The standard deviation of the difference is 7,50+/-1 km of which 5,1 km are uncorrelated noise. The quantification effect accounts for *x*²/6 with *x* the pixel size (in variances). In this case, this corresponds to 5,1 km, the uncorrelated noise is thus totally explained by the quantification effect.

Drifts data are converted into ice speed in order to compare drifts at 3 and 6 day-lags. The standard deviation of the difference of ice speeds is 2,91+/-0,04 cm/s, which is comparable with standard deviation of 2,96 cm/s with SSM/I and 2,80 cm/s with NSCAT at 4 day-lag on two winter months (Liu et al., 1999), 2,6 to 2,9 cm/s at 1 day-lag (Liu and Cavalieri, 1998), 6 cm/s for SSM/I at 1 day-lag (Emery et al., 1997). Kwok et al. (1998) and Zhao et al. (2002) compare buoys and satellite drifts with and without small drifts. Their results show smaller standard deviation when small drifts are dismissed (lower than one pixel) : 2,32 cm/s for QuikSCAT and 2,27 cm/s for SSM/I for Zhao et al. (2002). Kwok et al. (1998) improve their comparison with SSM/I at 3 day-lag : from 4,28 cm/s down to 2,58 cm/s with vertical polarization and 4,52 cm/s down to 2,89 cm/s with horizontal polarization. Our results are not comparable : there is no improvement excluding small drifts. One problem to exclude small drifts is that drifts lower than one pixel may have a strong weight (24% of the Merged drifts in our study), statistics on the samples are then strongly modified, but what are their meaning ?

Angles of Merged drift vectors present a strong uncertainty : for small drifts (lower than two pixels in component) the angle is 0, 45 or 90° whereas angles of buoys drift vectors are more accurate. Mean angle difference between buoys and Merged is 3,9+/-0.8° and the standard deviation is 39,2+/-0.5° of which 33.5° are uncorrelated noise. If drifts less than one pixel are excluded, the standard deviation of the difference decreases down to 29,40+/-0.4°. Excluding small drifts seems to improve significantly the angular accuracy.

These results are in agreement with the comparison of buoys vector angles with SSM/I by Liu and Cavalieri (1998) with standard deviation about 25,9° and 18° over two months, 34,4° with SSM/I and 28,6° with NSCAT at 4 day-lag on two winter months (Liu et al., 1999), and 29,8° with QuikSCAT and 35,5° with SSM/I for Zhao et al. (2002) without small drifts. Kwok et al. (1998) improve also strongly the angle comparison between buoys and SSM/I at 3 day-lag excluding small drifts : from 46,5° down to 30,6° for vertical polarization and from 50,4° down to 32,8° for horizontal polarization. Figure 2 shows the angle difference between the Merged product and buoys as a function of Merged drifts : the angle difference sharply decreases (smaller than 45°) for drifts higher than 40 km (about 3 pixels), this was also noticed by Liu et al. (1999).

Using 6 day-lag is more adequate to small drifts. Mean drifts are 28,5 km for buoys and 27,0 km for Merged, with a standard deviation of 20,4 km for buoys and 19,6 km for Merged. Mean ice speed at 6 day-lag is lower than that at 3 day-lag due to the better accuracy of small drifts measured at this day-lag. The standard deviation of drifts difference is 8,9+/-0,1 km of which 6,5 km are uncorrelated noise. In order to compare this value with that of 3 day-lag, they must be converted in ice speeds. With 6 day-lag, the standard deviation of the difference is lower than with 3 day-lag (2,91+/-0,04 cm/s at 3 day-lag and 1,72+/-0,02 cm/s at 6 day-lag). This delay is better for small drifts, consequently, the angle data have a better resolution with a standard deviation of the difference of 29,6+/-0,4°.

Comparing buoys and Merged drifts in North/East components frames enables independent estimate of uncertainties. North and East components standard deviation of the difference are 7,0+/-0,1 km of which 4,0 km and 4,3 km respectively are uncorrelated noise. For each component, the quantification effect accounts for x²/12, with x the pixel size (Ezraty and Piollé, 2004b) : thus 3,6 km of the uncorrelated noise are explained by this effect.

The Merged drift is validated with buoys drifts over five winters. Merged drifts at 3 day-lag are constraint by the ability to measure small drifts, the 6 day-lag is more adapted to small drifts but is constraint by the 6 pixels maximum drift set in the algorithm. The day-lag must be chose in function of the value of the drifts.