Coincident vortices in Antarctic wind fields and sea ice motion



[1] This study introduces a method to examine the coincidence of rotational ice drift and winds caused by the forcing of ice motion by Antarctic cyclones. Vortices are automatically detected using the algorithm of Murray and Simmonds (1991) from both ECMWF surface pressures and SSM/I sea ice motions. For compatibility with this algorithm sea ice motion vectors are transformed to a scalar stream function. During a seven-day test period positions of pressure minima and stream function maxima (SFM) of ice drift are within 300 km in 96% of the cases. Lowest pressure minima are related to highest stream function maxima. The results promise the method to provide a complementary tool of detecting and localizing low-pressure systems over sea ice, adding to numerical pressure analyses.

1. Introduction

[2] In-situ observations of meteorological and sea ice quantities in the Antarctic sea ice region are sparse and irregular, which limits the accuracy both of empirical climatologies and weather and sea ice forecasts. Satellite remote sensing data can improve the database and provide additional information for use in numerical analyses and forecasts. Important elements of both are synoptic scale and mesoscale storms, greatly impacting the weather conditions for human activities, and also affecting sea ice distribution. Infrared and visible satellite data provide surface information only under cloudless conditions. In this study, the focus is on the vortex motion of sea ice as an auxiliary means of detecting synoptic scale atmospheric vortices. Significant correlations of wind and sea ice motion have been found in the spectral band between one day and up to two weeks [Kottmeier and Sellmann, 1996]. Cyclonic winds thus can cause cyclonic patterns in the sea ice drift vector field both in the Arctic and Antarctic. The authors seek to develop a method of cyclone detection from ice motion patterns, and to evaluate the coincidence of atmospheric cyclones and vortices in sea ice motion.

2. Data

[3] SSM/I vectors employed in this study were derived from tracking SSM/I brightness temperature signatures, optimally interpolated with drifting buoy data of the International Program for Antarctic Buoys [Drinkwater, 1998; Kwok et al., 1998] (also C. Schmitt et al., Atlas of Antarctic sea ice drift, 2004, http:/ The sea ice motion vectors were calculated from pairs of daily composites of passive microwave radiometer images (SSM/I) as described by Liu and Cavalieri [1998] and Agnew et al. [1997]. One set of SSM/I motion vectors is derived from two days of satellite observations (48 hours from 0 to 0 UTC [Kwok et al., 1998]), named by the first contributing day, for example, 25.08.1997 for the period August 25, 0 UTC to August 27, 0 UTC. In this method, cross correlation techniques are employed to track significant brightness features in the satellite images. Errors result from clouds (at 85 GHz) and from modifications of the feature tracked due to ridging and melting. Details of the algorithm and data accuracy are given by Maslanik et al. [1998]. In this study the SSM/I optimally interpolated (OI) drift data are used on a 100 × 100 km polarstereographic grid. They are obtained as weighted sums of measurements of the 37 GHz and the 85 GHz channel of SSM/I and buoy drift data where available. Atmospheric reanalyses of the ECMWF are available on a 1.25 × 1.25 degree grid. The authors mainly use surface pressures for automatic localization of cyclone centers.

3. Method

[4] Under conditions of free drift the sea ice moves almost parallel to the isobars with a linear relation to surface wind velocities, and adjusts to changes of the wind field quickly [Kottmeier et al., 1992]. Although ocean currents and internal stresses can cause deviations from linear response to the wind forcing [Kottmeier et al., 1992], cyclonic wind fields can be expected to result in cyclonic drift patterns in many areas, in particular distant to the coasts. The SSM/I drift vectors of August 25 (Figure 1) and daily means of ECMWF surface pressure of August 25 and 26, 1997, as well as 10-m winds for August 25, 1997, demonstrate this kind of relationship. On that day an atmospheric vortex with a central pressure of less than 955 hPa is located approx. 30° west of the coincident ice vortex. It has filled up to 960 hPa on August 26 and is centered very close to the ice vortex. The SSM/I drift is consistently affected by the atmospheric depression on both days with an elongated vortex structure. Depending on temporal changes of the surface wind field the ice vortex can be under stronger influence of the cyclone on one of the two days. To reduce the possible effects of averaging, drift vortices are compared to the spatially nearest minima in the ECMWF pressure fields of the 48 contributing hours.

Figure 1.

Means of ECMWF isobars for August 25 (black) and the August 26, 1997 (grey), that contributed to sea ice motion vectors (black arrows). The center of ice drift vortex arises about 30° east of the earlier depression, and is very close to the later one.

[5] The vortex centers in atmosphere and in sea ice are compared by applying an automatic analysis scheme. The cyclone localization algorithm (CLA) of Murray and Simmonds [1991] and Simmonds et al. [1999] is a common technique localizing low pressure systems from ECMWF pressure fields. That method also interpolates cyclone centers between grid points. It is not restricted to depressions with closed isobars. The algorithm is optimized to atmospheric pressure fields, but can also be employed to search for minima and maxima in any atmospheric variable and at any level [Murray and Simmonds, 1991].

[6] Basically the same CLA is applied to the ice motion data. As a vector field is not compatible with the CLA, a scalar function with properties similar to mean sea level pressure is required. According to the Helmholtz theorem, each vector field can be divided into a non-rotational part and a divergence-free part:

equation image

[7] Atmospheric depressions usually show a rotational wind field. We therefore expect vorticity to be the dominant part and neglect the divergence, keeping in mind that this assumption may not be applicable in dynamic studies and could change the results in a minor way. Thus we obtain

equation image

[8] The Helmholtz theorem can be converted into the Poisson equation of a scalar stream function (SF)

equation image

Krishnamourti and Bounoua [1996] propose two different methods to solve equation (3) in order to derive a SF from surface geostrophic winds. Both methods are applied to calculate SF from ice drift. The Overrelaxation Method (RM) with Dirichlet and Neumann boundary conditions accounts for the conservation of horizontal mass fluxes across lateral boundaries. It iterates the calculation of differences between a forcing function and the actual magnitude of the stream function, and over-corrects the actual SF until the residuum is lower than a defined tolerance factor. The spectral Fourier Transform Method (FTM) assumes periodic boundary conditions. A differential equation

equation image

is obtained, where ∇x and Δx are forward and backward difference operators in x-direction. Wind components u, v, and the perturbation Ψ′ afterwards are expressed in terms of a double Fourier transformation. The SF is calculated from the resulting Fourier transformed of Ψ.

[9] Both methods are described more detailed by Krishnamourti and Bounoua [1996] and Stephens and Johnson [1978].

[10] Application shows that FTM streamlines tend to be orientated more longitudinally than RM streamlines (Figure 2), especially so near the boundaries, as was also found by Krishnamourti and Bounoua [1996]. Considering that sea ice drift is more likely to have zonal character in most Antarctic regions (except in parts of the Weddell and Ross Seas), and that spectral methods as FTM are preferable (R. Fitzpatrick, Introduction to computational physics, 2003,, FTM was chosen.

Figure 2.

Stream function by RM (above) and FTM (below) in 103equation image calculated for the Weddell Sea region.

[11] Since ice motion is also affected by mean ocean currents and internal stresses [Kottmeier et al., 1992], cyclonic winds often leave only blurred footprints in the sea ice drift. As ocean currents are expected to vary mainly at low frequencies on regional scales, their impact on sea ice drift patterns ought to be minimized by subtracting regional means (SRM) before SF is calculated. Elliptical averaging areas with axes of approx. 10° longitude and 3° latitude are used to account for dominant zonal drift. To obtain regular integration areas northern ice free regions are zero-filled, although that diminishes the quality of the SF. SRM is calculated without filled-up grid points with interpolated values of SF instead. Resulting SFs show maxima in regions with cyclonic ice motion, in contrast to atmospheric pressure. CLA is then used in searching for SF maxima without major modifications.

4. Comparison of Atmospheric and Sea Ice Vortices

[12] Rotational structures in sea ice motion were identified from SSM/I drift data after SRM over a period of seven days from August 2–8, 1992, when cyclone activity was high. Additional buoy data of the Winter Weddell Gyre study improve the quality of SSM/I data. As an example, two days of ECMWF pressure minima are compared to maxima of drift derived SFM on August 2, 1992 (Figure 3). All strong SFM (Ψ > 250 SU, 1 stream function unit = 1 SU = 103equation image) with closed streamlines (closed SFM, filled dots) are assigned to the closest of the 6-hourly ECMWF cyclone positions (pairs numbered by 1 to 6 in Figure 3). Considering the temporal resolution of the data base and a displacement velocity of atmospheric cyclones of approx. 0 km/h to 65 km/h, the vortex pairs coincide well with 114 km in the average, although the root mean square error (RMSE) of 124 km is quite high. The intensities of SFM and atmospheric depressions show no significant linear correlation (r2 = 0.1). SFMs with open streamlines (grey squares at marks A and B) often arise as footprints of fast moving systems or mesocyclones with a lifetime of less than 24 hours (Figure 3, marks A and B), where the influence of previous or following wind fields blurred the cyclonic ice motion.

Figure 3.

Positions of strong maxima of the FTM stream function calculated after SRM (filled dots) for August 2, 1992. Grey squares are weaker SFM without closed streamlines. Pressure minima from ECMWF data are marked by black circles and crosses in four sizes from 0 UTC (smallest) to 18 UTC (largest) for August 2, 1992. Unfilled squares mark ECMWF pressure minima of August 3, 1992, following the same principle. Triangles are 0 UTC of August 4, 1992. Grey arrows show the daily SSM/I drift vectors [m/s] used for stream function calculation. Black lines highlight storm tracks from ECMWF data, black arrows show assignment to SFM.

[13] Over the whole test period 96% of 25 strong SFM with closed streamlines after SRM are less than 300 km away from one of the 6-hourly ECMWF pressure minima (Figure 4), if SF extrema north of the ice edge are excluded. Coherent vortices are separated from each other slightly more than at August 2, 1992, with 143 km ± 136 km. A linear relation between the arising intensities was not found (r2 = 0.2). This maybe mainly due to the insufficient temporal resolution of the sea ice motion vectors. For six SFM there was no obvious atmospheric cyclone in the ECMWF data. Without additional data it cannot be clarified, if this is an inaccuracy of the ECMWF data due to sparse observations in this region or of the SFM calculation.

Figure 4.

Frequency distribution of distances between the maxima of the SRM stream function and pressure minima for August 2, 1992 (grey) and August 2–8, 1992 (black) (all based on the 6-hourly positions of ECMWF cyclones).

[14] Including secondary SFM (without closed streamlines, open SFM) and SFM north of the ice edge slightly diminishes the quality of our results (position differences of 165 km ± 153 km, r2 = 0.3). In 15 cases, SFM could not be assigned. 64% of 98 of the assigned SFM are closer than 300 km to the nearest ECMWF cyclone position. The vicinity of vortex locations suggests, that this method is a valuable tool to derive cyclone positions from sea ice motion data.

5. Conclusions

[15] We automatically extract information about positions and intensities of vortices from satellite derived sea ice motion vectors (SSM/I optimally interpolated data) and relate them to atmospheric low pressure systems. The newly implemented method for localizing vortices from observations of sea ice motion works well. Vortex localizations largely correspond to the results of ECMWF analyses of positions, but cannot be used to conclude on lowest pressure of depressions. Over a seven-day period of comparison, most automatically detected vortices in sea ice drift are related to strong cyclones within a distance of less than 600 km. Most of these systems correspond to a cyclone position from at least one (of four per day) ECMWF surface pressure fields and, therefore, are considered reasonable. Some of the SFM are not associated with visible vortices in ice drift and/or pressure minima in the ECMWF data. Whether this is due to excessive sensitivity of the new method or to the errors in the ECMWF analyses is not known. The different character of instantaneous ECMWF data compared to drift data that are derived from daily composites of SSM/I brightness temperatures also affects our results. Visual comparisons of ECMWF surface pressure with sea ice drift vectors suggest that the differences in positions are considerably influenced by the insufficient temporal resolution of the SSM/I data. Additionally the ECMWF reanalyses have earlier been found to be less reliable in the Antarctic than in other regions because of the poor availability of observations for data assimilation in the Antarctic [Simmonds et al., 1999]. Considering the above restrictions, our method works quite well, and provides a valuable opportunity to improve cyclone climatologies. Further it could help to avoid storms travelling from the ocean towards the continent to be missed, that endanger human activities on the Antarctic continent. For such operational purposes higher resolution sea ice motion data would be more suitable.


[16] SSM/I data have been provided by JPL, Polar Remote Sensing Group, courtesy Ron Kwok, and Mark Drinkwater of the European Space Agency, Oceans and Ice Unit, (formerly JPL Polar Oceans Group), respectively. That work was performed at JPL, California Institute of Technology, supported under contract to NASA Code Y. The ECMWF data were contributed by the German Weather Service, Offenbach. Antarctic Coastline is taken from Global Land One-Kilometer Base Elevation (GLOBE) data from the National Geophysical Data Center.