Atmospheric cyclones in the Fram Strait affect the sea ice transport from the Arctic Ocean into the Atlantic Ocean. During the field experiment FRAMZY in April 1999 a Fram Strait cyclone and its impact on the ice drift was measured using a research aircraft and an array of 15 ice buoys. The synoptic-scale cyclone moved from the south into the area. It was discernible up to 500 hPa in the pressure field, but the horizontal temperature contrast of up to 16 K between the warm and cold sides was confined to the lowest 500 m. The average ice drift was 0.21 ms−1 toward 200° but increased to 0.6 ms−1 during the cyclone passage. The ice drift amounted to 1.6% of the geostrophic wind with a turning angle of 51° on the average. Comparisons between the aircraft measurements and operational weather model analyses show an insufficient representation of the temperature inversion and indicate an underestimate of wind speed and, thus, momentum transfer to the sea ice.
 The ice drift in the Arctic Ocean is dominated by two large-scale transport regimes: the Beaufort gyre located in the American-Canadian sector of the Arctic and the Transpolar Drift Stream extending from the Siberian coast to the Fram Strait. Per annum, an amount of about 10–20% of the Arctic sea ice is transported through the Fram Strait into the Atlantic. Concerning the associated freshwater inflow into the world ocean, the ice export via the Fram Strait is only surpassed by the Amazon River. The Arctic freshwater export has an impact on the oceanic density stratification and hence on the conditions for deep convection in the Greenland Sea and thus for one main driving mechanism of the global oceanic circulation (conveyor belt).
 The ice export through Fram Strait has been estimated using model simulations (e.g., Häkkinen  for the period 1955–1975, Harder et al.  for the period 1986–1992, Hilmer et al.  for the period 1958–1997, Brümmer et al.  for the period 1979–1995), upward looking sonars for the period 1991–1995 [Vinje et al., 1998], satellite SSM/I observations for the period 1991–1995 [Kwok and Rothrock, 1999], and the cross-Fram Strait pressure difference for the period 1950–2000 [Vinje, 2001]. Taking all methods together, the long-term annual mean of the ice export is about 0.085 Sv (1 Sv (Sverdrup) = 106 m3 s−1). The interannual variability is large and ranges from minimum values around 0.040 Sv to maximum values around 0.140 Sv, i.e., the variability is about ±50% of the mean value. The differences between the averages based on the various methods amount to ±0.020 Sv, which can also be regarded as a kind of uncertainty of the ice export estimate. Calculations by Harder et al.  indicate an average amplitude of about 0.065 Sv for the annual cycle with the maximum in March and the minimum in August. Long-term monthly values of the ice export reveal variations whose magnitudes are in the order of the long-term averages.
 The variability of the ice export through the Fram Strait depends on both the regional and larger-scale atmospheric and oceanic conditions. On the one hand, the ice export is influenced by the general atmospheric circulation over the entire Arctic [Proshutinsky and Johnson, 1997]. On the other hand, the ice export is affected locally. Here, cyclones are of primary importance because of the inhomogeneous and time-dependent wind field and thus drag forces on the sea ice caused by a moving cyclone. Statistical studies indicate that the ice drift through the Fram Strait is increased in the presence of cyclones [Brümmer et al., 2001].
 Systematic observations of Fram Strait cyclones and their impact on ice drift have not been performed so far. Observational data were obtained more or less accidentally. Only three Fram Strait cyclone cases are reported in the literature so far. One case is reported by Guest et al. . Another case observed on 11 April 1989 was published by Rasmussen et al. , and a third case was observed on 13 March 1993 by Brümmer and Hoeber . In the latter two cases, a south-north moving cyclone passed close to or over a research vessel located about 100 km north of the Fram Strait ice edge. The meteorological conditions were relatively similar: the near-surface temperature contrast between the air on the warm and cold side of the cyclone was 20 to 25 K and low-level winds reached up to 20 ms−1. In the case of Brümmer and Hoeber  ice motion was measured. The passing cyclone caused a complete loop in the ice drift trajectory and the wind factor, i.e., the ratio of ice drift speed to wind speed, increased from 0.05 before to 0.09 after the cyclone's passage. This was accompanied by large variations of the ice strain-rate tensor suggesting that the internal ice stress was reduced.
 A specifically aimed field experiment, called FRAMZY 1999, on Fram Strait cyclones and their impact on sea ice took place in April 1999 [Brümmer, 2000]. An array of 15 ice buoys was deployed on 3 April and, for the first time, aircraft measurements within a Fram Strait cyclone were conducted. In this paper, a case study of a cyclone is presented which passed the Fram Strait area from south to north on 18 and 19 April 1999. Figures 1a and 1b show the surface pressure fields at noon on both days as analyzed by the model of the European Centre for Medium-Range Weather Forecasts (ECMWF) together with the aircraft flight patterns and the positions of the still remaining 11 ice buoys. In contrast to the case reported by Brümmer and Hoeber  the cyclone did not develop at the ice edge but entered the Fram Strait area from the Norwegian Sea and weakened on its way north.
 The objectives of this paper are: (a) to present the analysis of the atmospheric properties of the cyclone (e.g., temperature contrast, wind field and maxima wind, vertical depth) based on the aircraft measurements (section 2), (b) to present the impact of the cyclone on the ice drift based on the buoy measurements (section 3), and (c) to compare the observed atmospheric structure with the ECMWF and DWD (Deutscher Wetterdienst) model analyses (section 4). The latter objective is motivated by the fact that a simulation of the impact of the cyclone on the ice drift can only be successful if the atmospheric forcing, i.e., the atmospheric properties of the cyclone, is prescribed (in an ice model forced by analysis data) or simulated (in a fully coupled atmosphere-ice model) correctly.
2. Cyclone Properties as Analyzed From Aircraft Measurements
 The cyclone properties are analyzed from the measurements of the German Falcon research aircraft, which operated from the airfield in Longyearbyen on Spitsbergen. The Falcon aircraft was similarly equipped as in many preceding meteorological campaigns and as described, e.g., by Brümmer [1999, 2000] and Brümmer and Thiemann . To summarize shortly here, the equipment comprises instruments to measure mean and turbulent (100 Hz) values of three-dimensional wind vector, temperature (two sensors), humidity (three sensors), pressure and radar height and with 10 Hz resolution values of infrared surface temperature, shortwave and longwave radiation from both above and below.
 The cyclone moved slowly (about 5 m/s) through the experimental region so that two aircraft missions could be flown in it: on 18 April 1999 between 1030 and 1230 UTC northwest of the cyclone center over the array of ice buoys and on 19 April 1999 between 1000 and 1200 UTC more or less through the cyclone center. To measure the horizontal and vertical structure of the cyclone the flight missions consisted of a sequence of horizontal runs at different levels and vertical profiles.
2.1. Horizontal Structure
Figures 2a and 2b show for both days the horizontal distribution of horizontal wind vector , pressure p reduced to sea level, and air temperature T. The maps are based only on flight sections flown at 70 m altitude. On 18 April, the long axis of the slightly elliptically shaped low (Figure 1) was located at about 0°–1°E. East of the axis, the air was warm (0–2°C) with moderate southerly winds. West of the axis, a stronger pressure gradient was present and northerly winds of up to 20 ms−1 were observed. Here, air temperatures were clearly below zero and decreased to −16°C at 80 km distance from the ice edge. On 19 April, when the cyclone center nearly reached the ice edge at 80.2 N, 2 E, the mesoscale , p, and T distributions were basically the same as on the day before with two exceptions. (1) The central pressure has increased so that pressure gradient and wind on the cold side of the cyclone were weaker. (2) The zones of strong temperature contrast at the cyclone axis and at the ice edge now coincide locally.
 The low-level horizontal structure across the cyclone is further detailed in Figures 3a and 3b for the southern part of the flight pattern on 18 April and the northern part on 19 April, respectively. On 19 April, a continuous low-level leg of 210 km length was flown, whereas on 18 April only three short sub-sections are available because vertical profiles were flown in between. Thus, on 18 April, the contrasts of the meteorological conditions right at the cyclone axis are missing. To the east of the cyclone axis, the air-surface temperature difference is about neutral, whereas to the west of the axis the stratification is unstable, only weakly over closed ice, but strongly over the open water and over openings in the ice. The ice edge is clearly detectable both in the surface temperature Tsfc and the albedo record. A stronger than average air temperature gradient occurs at the ice edge, which is situated on the cold side of the cyclone.
 On 19 April, the stratification conditions are the same as on the preceding day in the warm air over the water (neutral to stable) and in the cold air over the ice (unstable). But the warm air has now entered the ice. Here, stratification is weakly stable and ice surface and water surface have the same temperature of about −1.8°C (the freezing temperature of the sea water) so that neither openings in the ice nor the ice edge are detectable by Tsfc but only by the albedo. As a consequence of this, there is no temperature gradient at the ice edge. Here (at the northern leg AB), the ice edge is oriented east-west and the air flow is from south. At the southern leg on 19 April, the conditions at the ice edge are completely different (Figure 2b). Thus, the temperature contrast at the ice edge and consequently its potential for baroclinic development depends on the angle between air flow and ice edge. The largest temperature contrast at the ice edge occurs for an ice edge parallel airflow.
 In accordance with the air-surface stratification conditions mentioned above, the sensible heat fluxes H on the warm side of the cyclone axis over both water and ice are close to zero or downward on the order of less 5 Wm−2 (latent heat fluxes E (not shown) are upward and between 5 and 40 Wm−2). On the cold side of the cyclone axis, H is only a few Wm−2 over the ice, but between 20 and 160 Wm−2 over water. The H flux appears to increase with the width of the openings. The E flux (not shown) on the cold side follows the variations of the H flux, but the E amplitude is only about half of the H amplitude. It is well known [e.g., Smith et al., 1990], that the openings in the sea ice dominate the heat flux value (H + E) over the ice region under cold-air conditions.
 The openings also affect the momentum flux τ. Figure 3b shows that τ is larger over the openings than over the ice although the wind speed at the flight level of 70 m is about the same. Following classic Monin-Obukhov surface layer theory, the drag coefficient CD would be larger (by about a factor of two) over the openings than over the ice. However, the stability dependence of CD would explain only a factor of 1.1 to 1.2 in the present case. Also a surface roughness difference (in this case rougher over the opening than over the ice) is very unlikely to be the reason for the τ difference. It may be possible that, in addition to turbulent motions for which the classic Monin-Obukhov theory is valid, convective motions may have contributed to the momentum flux.
 Since we have not measured the momentum flux profiles, it is not clear if the measured momentum flux at 70 m height can be regarded as to represent the flux conditions at the surface. But, if it is true that the surface flux over the openings is larger than over the ice, then this may indicate a problem in sea ice models. They generally do not distinguish, in case of fractured sea ice, between the momentum flux over ice and over openings, but use an area-averaged momentum flux. As a consequence, the area-averaged drag force applied to the ice-covered part is too high and, thus, causes too high ice drift speeds.
2.2. Vertical Structure
 In Figure 4 vertical profiles are presented which were measured on both days on the warm side of the cyclone over water and on the cold side over sea ice. Except for the wind direction (southeasterly wind compared to northeasterly wind), temperature, humidity, and wind speed showed no substantial differences between the warm side and the cold side at levels higher than 700 m. This suggests that above this level the warm and moist air has already circled around the cyclone center. The contrast between the warm and the cold side of the cyclone is only manifested below 700 m, and particularly strong below 300 m height. While the warm air is well mixed from the surface to 1500 m or higher, the cold air is capped by a strong inversion of up to 11 K. The air below 700 m has not circled around the cyclone center. Due to surface friction the air masses on the warm and cold side have different origins.
 The low-level horizontal temperature gradient or thermal wind that occurs on the cold side of the cyclone (Figures 2a and 2b) causes an additional pressure difference and thus downward increasing magnitude of the geostrophic wind. To quantify this, e.g., in case of 18 April, we can use Figure 2a to estimate the thermal wind at the surface (utherm (sfc) = 0.034 s−1 from αtherm (sfc) = 190°) and Figure 4 to estimate the vertical profile of the thermal wind (utherm (z) = utherm (sfc) up to 300 m and then a linear decrease to zero at 700 m height and assuming αtherm (z) = αtherm (sfc)). If we assume that the measured wind at 700 m height is geostrophic (ug (700 m) = 12.5 m s−1 from αg (700 m) = 50°) and integrate the thermal wind downward to the surface, we result in a geostrophic wind at the surface of ug (sfc) = 28 m s−1 from αg (sfc) = 26°. Due to the surface friction and to the restriction of the frictional influence to the shallow layer below the inversion, the actual wind speed decreases downward. As a consequence a wind maximum (low-level jet) is generated at the base of the inversion. Due to the strong winds of up to 25 ms−1 on 18 April and 15 ms−1 on 19 April the lowest 300 or 200 m of the cold air are also well mixed. A clear veering of the wind of 20 to 30° is observed in the inversion. This is the result of both the thermal wind and the surface friction. Both effects operate in the opposite direction.
 Overcast cloud conditions were present all over the experimental area on both days. On the warm side, several cloud layers with cloud-free layers between them were discernible in the profiles P12 on 18 April and P1 on 19 April (Figure 4). The main layers were: 8 octas of stratus between the surface and 1500 m, 8 octas of altocumulus between 2000 and 2500 m, and 8 octas of altostratus between 3000 and 4000 m height. On the cold side, 8 octas of cloud were measured everywhere below the top level of 2500 m in the profiles P6 on 18 April and P2 on 19 April, i.e., clouds were present in the boundary layer, in the inversion layer, and above the inversion layer.
 The dynamic and thermodynamic structures of the cyclone in vertical and horizontal direction are principally the same on both days. Only the differences between the warm and cold sides below 700 m height have decreased from 18 to 19 April.
3. Cyclone Impact on Ice Drift as Analyzed From Buoy Measurements
 On 3 April 15 ice buoys of the CALIB type manufactured by Metocean Data Systems Ltd. in Dartmouth, Canada, were deployed on parachutes from a transport aircraft. The initial buoy array was about 200 km times 200 km large and centered at about 81.4°N, 4.0°W, i.e., more or less in the middle of the Fram Strait. On 30 April, the array still consisted of 11 buoys and had drifted about 500 km south- to southwestward with its center then at 77.2°N, 11.0°W. All buoys were equipped with the Argos system to determine the position and with an ice temperature sensor. Four of the buoys were additionally equipped with a pressure sensor. From those four pressure stations the geostrophic wind was calculated (see further below in this section). Data were transmitted about every hour via the Argos satellite system.
 In Figure 5 the drift speed of five ice buoys that are representative for the conditions in the array is shown. The longer-term variations in the array are very similar for all buoys. The average drift speed was 0.21 ms−1 and the general drift direction was to 200° indicating the prevailing southerly drift in the East-Greenland Current (EGC). The cyclone event is clearly marked by the largest drift speed reaching values between 0.6 and 0.7 ms−1 on 17/18 April when the cyclone passed east of the array at a distance of about 200 km. A striking feature of the drift speed are periodic variations with periods around 12.5 hours which can be attributed to the tidal and inertial motions of the sea and could be resolved due to the “high-frequency” position measurements by the Argos system.
 The periodic variations also show up in the sea ice kinematics, which is represented in Figure 6 by the divergence, vorticity, and deformation of the ice motion field. The computation procedure was to select five ice buoys and to fit a first-order plane to the velocity components, uice and vice, according to
were calculated. Computations were made for several groups of five ice buoys. The results are basically the same and are therefore only shown for one array (buoys 1-2-6-10-15) representative for a scale of about 120 km.
 There is a high degree of variability in the divergence and deformation but not in the vorticity time series. Vorticity is close to zero on the long-term average but changes its sign with the tidal motion (positive vorticity when drift speed is low and vice versa). The variations of divergence and deformation are particularly interesting because they determine the elements of the strain-rate tensor
from which the stress tensor and finally the internal stress force in the momentum balance equation of the sea ice is derived. Unlike the drift speed, there are no extreme values of divergence and deformation during the cyclone event. The time series of drift speed shows two episodes of higher speeds centered around 10/11 April and 17/18 April connected with anticyclonic and cyclonic air flow conditions, respectively. In both cases, there are no remarkable values of divergence and deformation during the event itself but it is interesting that extreme values occurred with a time delay of about two days.
from which it follows that is reduced by a factor of ∣A∣ against and turned counter-clockwise against by an angle of θ. The intercept vector, , results from the forcing of the ice drift by ocean current and internal stress. Since the latter is irregular and may be assumed to cancel when long-term averages are considered, is essentially equal to the current, in this case the EGC.
 The geostrophic wind was computed by fitting a plane surface to the pressure data of four buoys and the tidal oscillation in the ice drift was filtered out since obviously it cannot be related to the wind forcing. The geostrophic wind and the complex factor are shown in Figure 7. For four days, when Falcon aircraft measurements took place over the buoy array, the aircraft-based values are added. Geostrophic wind speed varies between 3 and 30 ms−1 and the direction αg is predominantly from the north/northeast. The highest occurs with the cyclone passage on 18 April. The correlation between geostrophic wind speed and drift speed is 0.80; the tilt is 0.0102 and the intersect is . The geostrophic wind factor, i.e., the ratio varies irregularly around an average value of 1.6% and has a range between 0.6 and 4%. The turning angle, θ = αg − αice, varies between 25° and 80° with an average value of 51°. Both A and were derived from the data by taking the regression between geostrophic wind and ice drift over periods of 48 hours overlapping by 24 hours. The intercept vector is, on most of the days of the period, qualitatively a measure of the EGC with speeds between 0.1 and 0.25 ms−1 and directions from north.
 Unfortunately, one of the four “pressure” buoys was lost on 16 April and the remaining three buoys formed a very flat triangle that was not suited to calculate a reliable geostrophic wind. To estimate the relation between and during the cyclone event, though, we use the aircraft-based pressure field on 18 April (Figure 2a) when the flight pattern took place over the buoy array. The result is given in Figure 7 and indicates a high geostrophic wind speed on 18 April. The aircraft-based geostrophic wind estimate appears to be reliable as can be concluded from the comparison between aircraft-based and buoy-based geostrophic wind calculation on three additional days (10, 12, 14 April) when the Falcon research aircraft flew over the buoy array.
 To get a picture of the large-scale ice motion forced by the cyclone event, SSM/I (Special Sensor Microwave Imager) satellite-based ice drift estimates are presented in Figure 8. The large-scale ice motion fields clearly show the effects of the approaching (18 April) and passing (19 April) cyclone. To derive a reliable ice drift vector from SSM/I data a long time step Δt has to be taken (here Δt = 3 days), so that temporal details of the cyclone passages are, unfortunately, not available.
4. Comparison With Previous Observations and With Model Analyses
4.1. Comparison With the ARKTIS 1993 Cyclone Case
 Comparing the FRAMZY 1999 cyclone with the previous ARKTIS 1993 Fram Strait cyclone observation [Brümmer and Hoeber, 1999] shows similarities and differences. In both cases the vertical thermodynamic structure of the cyclone was similar: the horizontal warm-cold contrast was confined to a shallow layer. The aircraft measurements in the present case document the shallow layer to be even more flat than the radiosonde measurements in the ARKTIS 1993 case. The horizontal temperature contrast was ΔT = 15 K in FRAMZY 1999 compared to ΔT = 25 K in ARKTIS 1993.
 In both cases the measured impact of the cyclone on the ice drift was different because of a different relative position of the ice buoy array with respect to the cyclone's track. Although in both cases the cyclone moved from south to north, it passed the buoy array on the west side in ARKTIS 1993 and the east side in FRAMZY 1999. The consequence was that the ice drift trajectory showed a complete loop in ARKTIS 1993 while it only showed a linear acceleration of the mean ice drift from northeast toward southwest in FRAMZY 1999.
 The two cyclone cases suggest that a cyclone's impact on the ice drift in the Fram Strait may be twofold: a direct and an indirect impact. The direct impact occurs when a cyclone moves from south to north on a relatively easterly track, for example, over the open water of the West-Spitsbergen Current. On the rear side (western side) of the cyclone the increased northerly wind causes an increased southward ice drift through the Fram Strait. An indirect impact on the ice drift occurs when a cyclone moves over the ice in the Fram Strait. The variable (in space and time) wind field around the moving cyclone causes ice convergence/divergence or compression/dilatation and trends to reduce the inherent ice stress. The ice pack may break up and is loosened. As a result, the outflow of the loosened ice through the “bottleneck” of the Fram Strait is facilitated and increases compared to the same wind and current conditions. While the direct impact is felt immediately, the indirect impact is felt with some time delay.
4.2. Comparison With ECMWF and DWD Model Analyses
 In this section we compare the FRAMZY 1999 cyclone observation with model analysis of the European Centre of Medium-Range Weather Forecasts (ECMWF) and the Deutscher Wetterdienst (DWD). The horizontal grid is 0.5° for ECMWF and 50 km for DWD (Global Model GM). In the latter case 12 hours prognoses instead of analyses are used below.
Figure 9 shows cyclone position and central pressure every 12 hours between 18 April 00 UTC and 20 April 00 UTC. Both models agree well for 18 April at 00 and 12 UTC when the cyclone is over the open water. Beginning with 19 April 00 UTC and thereafter, when the cyclone approaches or is over the sea ice, cyclone positions disagree by up to 250 km (19 April 12 UTC). The Falcon aircraft measurements on 19 April (see Figure 2) support the cyclone position in the ECMWF analysis. The central pressure po in both models is not much different (maximum difference Δpo = 2 hPa). The reason for the diverging cyclone positions over the sea ice is not clear. Compared to the Falcon ice edge observations on 18 and 19 April the ice edge is too far east and south (nearly 100 km), however, in both models. Thus, the wrong ice edge is unlikely to be the reason.
 The vertical thermodynamic and kinematic structure on the cold and the warm side of the cyclone as analyzed by the two models is compared with the observations in Figure 10. The conditions on the warm side and close to the surface on the cold side are analyzed quite well by both models: The differences amount to ΔT ≤ 2 K, Δm ≤ 0.2 g/kg, ΔFF ≤ 2.5 ms−1 and ΔDD ≤ 15° for temperature, water vapor mixing ratio, wind speed, and wind direction, respectively. An exception is the large wind speed offset by about 7 m/s on the warm side in the ECMWF model. The reason for this is unknown. Remarkable differences occur on the cold side. The modeled boundary layer is too shallow (100–200 m instead of 300 m) and the overlying inversion is too weak (temperature increase 6–8 K instead of 11 K) and smeared over a too deep layer (from 100–200 to 1100–1200 m instead from 300 to 900 m). The wind below the inversion is about 5–8 ms−1 too weak, with the consequence that the wind shear in the boundary layer is too weak and, thus, causes the above-mentioned too shallow boundary layer. Due to these model deficiencies, it has to be supposed that the surface momentum flux, and thus the atmospheric forcing of the ice drift, is too weak in the two models.
 In contrast to the ARKTIS 1993 cyclone, which was generated in the baroclinic zone at the ice edge, the FRAMZY 1999 cyclone existed already several days and approached from the Norwegian Sea. According to the DWD analysis valid for 18 April 12 UTC (when both model analyses still agree and when the cyclone was centered at 76.8°N, 2.0°E over the open water) the cyclone was clearly visible with closed isohypses up to the 500 hPa level. There, the difference of the geopotential height Δϕ over a 200 km distance from the cyclone center was about half as large (Δϕ500 = 15 m) than at the 1000 hPa level (Δϕ1000 = 29 m). Thus, the FRAMZY 1999 cyclone extended to much higher levels than the ARKTIS 1993 cyclone (which reached to about 2 km height).
5. Concluding Remarks
 The Fram Strait is a data-sparse region. Apart from ships of opportunity, occasionally drifting ice buoys of the International Arctic Buoy Program, and underwater-anchored upward looking sonars, in situ data for the investigation of cyclones and their impact on the ice drift through the Fram Strait are absent. Satellite data from various satellites (NOAA-AVHRR, SSM/I, SAR and others) are available but either do not fulfill the requirements of the time resolution (about one hour) or the spatial resolution (about 10 km in the horizontal direction and about 100 m in the vertical direction) or the accuracy requirements (e.g., for air-surface temperature difference or for wind speed) in order to be used for process studies. However, satellite data are a useful supplement.
 For the first time, a Fram Strait cyclone experiment was conducted applying aircraft and an array of ice buoys. A Fram Strait cyclone and its impact on the sea ice could be quantified successfully. Unfortunately, only one cyclone event occurred during the two-week expedition (restricted availability of the aircraft). According to cyclone frequency statistics for the Fram Strait [Brümmer et al., 2000], which show between zero and 12 cyclones per month on the basis of the ECMWF analyses for the period 1988–1993, this was close to the lower end.
 The comparison of the atmospheric cyclone properties between the in situ measurements on the one hand and the ECMWF and DWD analyses on the other hand may indicate a general problem of the models: the insufficient simulation of the winter temperature inversion over sea ice and, thus, the consequences for the vertical wind profile and for the surface momentum transfer. This item has to be studied further, since it can have systematic consequences on the atmospheric forcing of the ice drift both in uncoupled and coupled models.
 In the meantime, a second FRAMZY experiment took place in March 2002. During a three-week period, three cyclones passed the Fram Strait. Having learned from FRAMZY 1999, all ice buoys were equipped with a pressure sensor, and the aircraft measurements were taken at levels as low as 15 m, so that the problem of geostrophic wind forcing and surface fluxes over ice and openings can be studied with a better data background. In addition, the Finnish research vessel Aranda operated in the area and monitored continuously the atmospheric and oceanic conditions and the conditions at the air-ice/sea interface. FRAMZY 2002 results will be presented in a future paper.
 We thank Robert Baumann (German Space Agency, Oberpfaffenhofen, Germany) for making the latest version of the Falcon aircraft data processing program available to the authors. We are grateful to Thomas Martin (Institut für Meereskunde, Kiel, Germany) for the satellite-derived SSM/I ice drift fields in Figure 8.