We present estimates of volume and heat transport through Fram Strait for the period 1997 to 2000 from data of moored instruments. The annual mean volume transports at 78°55′N were between 9 2 and 10 1 Sv northward and 13 2 and 12 1 Sv southward with a net transport between 4 2 and 2 2 Sv to the south. The temperature of the northward flow of Atlantic Water had a strong seasonality with a minimum in winter. Nevertheless, the northward heat transport was highest in winter caused by the winter maximum of northward volume transport. Between 1997 and 1999, the annual mean net heat transport across 78°55′N increased from 16 12 to 41 5 TW. This resulted from a very strong increase in heat transport in the West Spitsbergen Current (mean annual values from 28 5 to 44 6 TW and to 46 5 TW in 99/00) which was not compensated by an equivalent signal in the southward flow. The heat transport to the south remained constant within error limitations. Only half of the heat flux increase in the West Spitsbergen Current was due to a higher temperature; half of it was due to a stronger flow. A similar increase as observed between 1997 and 2000 would have been sufficient to explain the warming of intermediate layers in the Eurasian Arctic observed in the early 1990s. Consequently, we suggest that a warming signal from the late 1990s is presently spreading in the interior Arctic Ocean.
 Changes in the Arctic heat budget are of major interest since one is aware of the Arctic Ocean's possible link to global climate. Global models with coupled atmosphere, ice and ocean show a considerable impact on the climate whether or not the modeled Arctic ice cover is free to respond to atmospheric and oceanic variation [Rind et al., 1995]. The most dramatic changes observed recently in the Arctic Ocean are the reduction of sea ice thickness and extent and in the same time the extensive warming of intermediate layers [Quadfasel et al., 1991; Rothrock et al., 1999; Serreze et al., 2003]. These changes seem to correlate with the variability of large-scale atmospheric patterns. Warmer intermediate water and weaker ice cover are related to periods of a stronger cyclonic atmospheric circulation over the North Atlantic and the Arctic [Dickson et al., 2000; Proshutinsky et al., 1999].
 The dominant oceanic heat source for the Arctic Ocean is inflow of Atlantic Water combined with export of Polar Water and ice through Fram Strait. Northward flow of Atlantic Water forms a part of the Atlantic-wide meridional overturning cell which might be driven through thermohaline convection in subpolar and polar regions. Enhanced by the northward extent of the Atlantic circulation into the Arctic Ocean, the oceanic northward heat transport across 45°N is 600 TW in the Atlantic while it is orders of magnitude smaller in the Pacific [Ganachaud and Wunsch, 2000].
 Atlantic Water enters the Arctic Ocean through Fram Strait and through the Barents Sea. Both flows rejoin in the northern Kara Sea and continue in a boundary current along the Arctic Basin rim and ridges [Aagaard, 1989]. During their path in the Arctic Ocean, they are permanently modified. Atlantic Water is observed in the southern Makarov Basin since the late 1990s where in former times Pacific derived waters have dominated [McLaughlin et al., 1996; Smith et al., 1999]. While in large parts of the Arctic Ocean the Atlantic layer is shielded from sea ice and atmosphere by a strong cold halocline, a shift in the circulation of the fresh surface layer can interrupt the development of the halocline in certain areas and heat can be released from the Atlantic layer to the ice or atmosphere [Steele and Boyd, 1998; Martinson and Steele, 2001].
 It was suggested that the wider spreading of Atlantic Water and the warming in the Arctic Ocean would be most likely an advective feature [Grotefendt et al., 1998; Dickson et al., 2000]. Yet it is not clear to what extent variations in volume transport or in temperatures would be responsible. From a 26 year long time series of autumn temperature of Atlantic Water at 80°N, Saloranta and Haugan  conclude that the mere warming of the Atlantic Water can not explain the warming in the Arctic Ocean. Results from a coupled ice-ocean model [Karcher et al., 2003] show the warming in the 1990s as a result of increased volume transport from the Atlantic and reduced heat loss to the atmosphere in the Nordic Seas. Maslowski et al.  conclude that increased storminess over the Arctic in the late 1980s and early 1990s accelerated the barotropic slope currents especially in the Russian sector of the Arctic Ocean implying a stronger volume transport.
 The complicated topographic structure of the Fram Strait leads to a splitting of the West Spitsbergen Current carrying Atlantic Water northward in at least three branches [Quadfasel et al., 1987]. One branch follows the shelf edge and enters the Arctic Ocean north of Svalbard. This part has to cross the Yermak Plateau which limits the flow to a depth of approximately 600 m. A second branch flows northward along the northwestern slope of the Yermak Plateau and the third branch recirculates immediately in Fram Strait between 78° and 80°N [Perkin and Lewis, 1984; Bourke et al., 1988; Gascard et al., 1995]. Evidently, the size and strength of the different branches largely determine the input of oceanic heat to the Arctic Ocean. While part of the Atlantic Water flows to the central Arctic and is expected to be responsible for observed changes in heat content there, another part returns in a short loop within the northern Fram Strait. The latter can induce ice melt there and thus will affect the amount of fresh water entering the subpolar convection gyres in Nordic Seas as ice or as water.
 The East Greenland Current, carrying water and ice from the Arctic Ocean southward, is concentrated above the continental slope.
 Past estimates of heat transport through Fram Strait derived from observations were either based on inverse modeling or on velocity/temperature measurements at few data points which required considerable extrapolations. Simonsen and Haugan  compiled published oceanic heat transport estimates for the Arctic Ocean and the Nordic Seas. They reported values for oceanic heat transport through Fram Strait into the Arctic Ocean between 17.6 TW [Mosby, 1962] and 68.4 TW [Aagaard and Greisman, 1975], if referred to −0.1°C.
 In order to examine the total transport of mass, heat and fresh water through Fram Strait and the Barents Sea simultaneously, a long-term observation programme with moored instruments was started in summer 1997 in the frame of the European Union MAST III Programme VEINS (Variability of Exchanges in the Northern Seas). For the first time, an array with 14 moorings covering Fram Strait from the eastern to the western shelf edge was maintained for several years. Measurements from the first 2 years yielded for the Fram Strait yearly mean northward fluxes of 9 and 10 Sv respectively, and 12 and 11 Sv southward [Fahrbach et al., 2001]. Across the Barents Sea opening, a net inflow of 2 Sv to the Barents Sea was measured during the first year of parallel observations [Ingvaldsen et al., 2002].
 We present in this paper results for the heat transport from the first 3 years of measurements in Fram Strait. We determined the transport across a section at approximately 78°55′N, i.e., we calculated from our measurements the northward and southward transport and the net flow across this section. Obviously, our transport comprises flow that will return to Fram Strait in a relatively short loop as well as flow that takes a longer loop or that exits or enters the Arctic Ocean through other openings. In section 2 we describe the data set and methods used to obtain an interpolated data set, in section 3 we present the resulting transports and their regional distribution. In section 4 we reason about differences between our results and previously published transport numbers, and in section 5 we discuss how the heat transport variation measured in Fram Strait relates to observed temperature variations in the Arctic Ocean.
2. Data and Methods
2.1. Measurements of Time Series and CTD Sections
 From September 1997 to August 2000 moorings were deployed in Fram Strait (Figure 1). During the first 2 years 14 moorings with 47 instruments covered the entire cross section, while in the period 1999–2000 only 11 moorings were deployed and the central three positions from the earlier arrays (F7, F8, F9) were left out. The instruments were RCM7, RCM8 or DCM11 by Aanderaa Instruments and 3D-ACM by Falmouth Scientific Inc; all registered velocity and temperature at a 1-hour interval. The instruments covered the water column from 10 m above the seabed to approximately 60 m below the surface (Figure 2). The measurements extended from 6°51′W, the eastern Greenland shelf break, along 79°N until 0°E and then continued along 78°50′N to 8°40′E, the western shelf break off Spitsbergen. With a limited number of available instruments, we aimed at a good coverage of the strait by a relatively narrow horizontal spacing of the moorings over the continental slopes where strong horizontal gradients were expected, and a wider spacing in the interior. For more details, see the reports by Woodgate et al. , Fahrbach , and Schauer . For a description of the data processing we refer to Fahrbach et al. . For brevity we will refer to the periods 1997–1998 as 97/98, 1998–1999 as 98/99 and 1999–2000 as 99/00.
2.2. Filling of Data Gaps in the Time Series
 Because of instrumental failure several time series are incomplete or missing. In the period 1997 to 1998, one mooring (F11) was lost, and 3 other instruments provided neither velocity nor temperature. The lack of F11 increases the uncertainty of the estimate of the southward transport for this period considerably, since the measurements during the periods 98/99 and 99/00 show that often the core of the southward flow is at F11. In the period 98/99 the data coverage was much better than in 97/98. An overview of the cross-section distribution of the deployed instruments is given in Figure 2, where complete time series, incomplete (which means that for some period either temperature or velocity or both were not recorded) or completely missing time series are marked differently. The temperature distributions are displayed from the CTD surveys during the deployment in 1997 and the exchanges of the moorings in 1998 and 1999.
 The results of transport estimates depend on the ensemble of data grid points. Because of the failure of instruments the available data grid points vary from year to year. To exclude the generation of artificial interannual variability by this effect, one could use only the largest common grid point set. This would exclude the use of a lot of valuable data. Therefore we preferred to transfer information from existing time series to positions where data are missing. For this purpose we used the correlation of time series of neighbor positions and assumed that the correlations do not vary with time. Consequently they allow to derive time series at the missing grid points from the neighboring positions. For example, the velocity at level 250 m of F3 was missing for period 97/98. We performed a linear regression between the velocity at F3, level 250 m, of year 98/99 and its neighboring velocities (F3 above and below, F2 and F4 at 250 m) and checked which pair has the highest correlation. Assuming that the correlation between the velocities at the two locations remained constant, we calculated the missing velocity as a linear function of the respective neighbor velocity of year 97/98. The correlation and recalculation was done with detided 6-hour averages. The correlation coefficients are between 0.5 and 0.9 for used velocity pairs and between 0.4 and 0.8 for used temperature pairs. Only for the mooring F11 the coefficients are between 0.2 and 0.4 for both parameters.
 In order to provide an equal database for all records, all time series were truncated to start and stop at the same times which are given in Table 1. With this data set we base our calculations in the same way on monthly mean values of velocity and temperature as in the work of Fahrbach et al. .
Table 1. Common Duration of Time Series
19 Sept. 1997 to 30. Aug. 1998 (i.e., only 12 days in September)
17 Sept. 1998 to 12 Sept.1999 (i.e., only 14 days in September 1998; September 1999 omitted)
27 Sept. 1999 to 31 July 2000 (i.e., only 4 days in September)
2.3. Calculation of Transports
 We estimated the transports through a quasi-zonal section at about 78°55′N. In the part west of the Greenwich Meridian both the moorings and the CTD stations were along 79°N while the eastern moorings and CTD stations were aligned along 78° 50′N. To connect the two parts, we interpolated along a northwest-southeast running line between 1°W and 1°E and used the perpendicular components of F9 for this 50 km long part. The bottom depths for the transport calculations were composed from the depths of the CTD stations and the moorings.
 The monthly mean values of temperature and cross section velocity were then linearly interpolated in the vertical to 100 m intervals and from this data set a two-dimensional interpolation was carried out using the Surfer 7 software (Golden Software). Data fields were created on a regular grid with cell dimensions of 1 km by 5 m using linear interpolation with the kriging method which means that all neighbor values are included weighted by their respective distances. With regard to the different horizontal and vertical scales, an anisotropy ratio, scaling horizontal vs. the vertical weights, of 0.04 was applied for the velocity field and 0.3 for the temperature field. The values were chosen based on visual inspection of the velocity and temperature distribution plots so that extrema not supported by data points were excluded. The resulting monthly volume fluxes agree with those of Fahrbach et al.  within a difference which is an order of magnitude smaller than the error attributed to the interpolation itself (see below). The values published by Fahrbach et al.  differ slightly from those presented here. Fahrbach et al. base their computation on purely meridional components and discuss a small fraction of the recirculation separately while we present transports perpendicular to the mooring line.
 Examples of the interpolated fields are given in Figure 3. Comparing the temperature field with the CTD section (Figure 2) which has a much higher spatial resolution allows to judge the quality of the interpolation. Features like the warm core of the West Spitsbergen Current and the cold shallow layer of the East Greenland Current are well captured, as well as the Return Atlantic Current which extends almost to the East Greenland shelf edge. An eddy-like feature between moorings F8 and F7 might not be resolved because of our limited spatial distribution of instruments. To meet the difficulties caused by the low resolution, Fahrbach et al.  used monthly mean values of measured velocities to calculate transports under the assumption that structures of small space scales are related to short timescales and that their effect on the large-scale transports can be eliminated through time averaging. A long averaging period causes errors in the transport estimates due to nonlinearity if both velocity and cross-section area of northward or southward transport are subject to changes. Fahrbach et al.  suggest that monthly averages would be an acceptable compromise between spatial aliasing through eddies and the problem due to nonlinearity. Additionally, monthly mean transports still allow the seasonal signal to be resolved. The aliasing problem is obviously largest with slowly drifting or stationary eddies. According to the drift of floats [Gascard et al., 1995], east of F6 the flow is northward so that eddies would leave the area within days. West of F6, topographic features could give rise to stationary mesoscale features. However, F7-F9 show vigorous fluctuations on a timescale much smaller than a month (more in section 3.2).
 The transport estimates include uncertainties arising from the measurements of current velocities and temperatures and from the averaging and the interpolation method. Since the instruments were calibrated before and after the deployment and since the monthly mean values are averages over a large number of individual samples, the systematic and random instrumental error can be neglected.
 In this paper we consider monthly and yearly mean values only. The ratio between the yearly average covariation of temperature and cross-section velocity, based on 6-hourly values, and the mean temperature flow was generally below 10−2. The ratio was larger (up to 0.8) only at a few of the uppermost instruments in the western part. These instruments were located in the sharp thermocline and the high covariation results most likely from vertical displacement of the thermocline rather than from eddies transporting heat. This is underlined by the fact that not only the magnitude but also the sign of the covariation was changing from year to year at the same position.
 The largest error is due to the spatial instrumental resolution. Linear interpolation means that for a cross section, defined by, e.g., four instruments, the average of the four measured values is assumed to be valid throughout the cross section. This implies a possible underestimation or overestimation which is related to the difference between values measured at neighbor instruments. We take the standard deviation of the values at the corners defining a cross section (quads in the center, sometimes triangles at the border of the entire section) multiplied by 1.96 (for 95% significance) as a measure of the mean error due to spatial undersampling for the monthly mean interpolated values of velocity and temperature. From that we calculate the relative error and apply error propagation for the heat flow, yearly averages and differences between flows. Since in the period 97/98 the entire mooring F11 was missing we considered the error in the respective area to be twice as large.
 In addition, we can check the error deriving from low resolution of the mooring temperature field by comparison with the CTD sections. In the central part of the section, the layers between 800 and 1400 m are slightly too warm in the fields derived from the moored instruments due to the vertical linear interpolation (Figures 2b and 3b). In the interpolated section the 0°C isotherm is at about 1100 m instead of about 800 m. To check the sensitivity of the heat transport estimates to our temperature interpolation, we calculated heat transports from the CTD temperature fields and the velocity fields of the months next to the CTD surveys. The resulting values, given as crosses in Figure 4, show that the heat transport estimates from the moored instrument data alone are too large for the full water column by about 20%, but the difference can be up to 40% (when the transport was small as in September 1997). For the Atlantic Water the agreement is much better: the interpolated values are only up to 14% too high.
 We refer the heat transport to −0.1°C in order to allow comparison with the values compiled by Simonsen and Haugan . This value was found by Aagaard and Greisman  to be the mean temperature of the southward flow through Fram Strait. Because the net volume transport through Fram Strait is not zero the use of the term “heat transport” is physically not exact, instead we determine “enthalpy transports”.
 If the heat content of the Arctic Ocean is changed through variation of heat flow through Fram Strait this can be due to the change in temperature of the advected water or to the change in volume transport or a mixture of both. In the case of changing volume flow the mass of the Arctic Ocean has to be balanced by changing flow through one of the other passages. In that case the temperature of the flow through Fram Strait must be referred to that of the flow through the other passage. As long as we do not know through which passage(s) the flow is compensated, we have to estimate the heat transport change within the range of possible reference temperatures between −0.1°C (Fram Strait [Aagaard and Greisman, 1975]) and −0.7°C (Canadian Archipelago [Simonsen and Haugan, 1996]). In Table 2a we summarize the heat transport values discussed in section 3 for Tref = −0.1°C and Tref = −0.7°C. Some of the absolute heat flow values for a given year differ greatly, however, the year-to-year changes (“diff” in Table 2a) do not depend much on the reference temperature.
Table 2a. Yearly Mean Heat Flux Through the Entire Fram Strait Section for 2 Years a
Total Northward, TW
Total Southward, TW
Total Net, TW
AW, T > 1°C Northward, TW
AW, T > 1°C Southward, TW
During 99/00, no moorings were in the central part). AW means Atlantic Water; “diff” is the difference between values from 98/99 and 97/98. Heat transport is based on different reference temperatures Tref (the respective error limits are of similar magnitude). For the definition of the periods, see text.
3.1. Spatial Distribution of Volume Flow
 The spatial distribution of the volume flow follows generally the known pattern with a highly barotropic northward flow in the West Spitsbergen Current and a more baroclinic East Greenland Current in the western Fram Strait [Jonsson et al., 1992], the latter however having also a considerable barotropic part [Morison, 1991; Fahrbach et al., 2001]. The highest velocities were invariably found in the West Spitsbergen Current right at the shelf edge with maximum monthly values of more than 40 cm/s in February 1998. The East Greenland Current had two clearly separated cores (Figure 3), both with much weaker monthly mean velocities during all 3 years than the West Spitsbergen Current. The westernmost core of the East Greenland Current is situated at the shelf edge and transports mostly the cold (T < 0°C) Polar Water. The second East Greenland Current core at the foot of the continental slope over about 2500 m carries warmer modified Atlantic Water southward.
 In the central Fram Strait, the flow is mostly directed westward. The meridional component is weak and variant, but also mostly barotropic. The strongest westward flow occurs in winter (not shown). We cannot distinguish whether this reflects strengthening or a spatial shift of the Atlantic Water return flow. Also the northward flow is strongest in winter (Fahrbach et al.  and Figure 5)) while the southward flow of volume and heat does not show a clear seasonal signal confirming findings by Jonsson et al. .
3.2. Mesoscale Features
 Eddies are a common feature in Fram Strait. They evolve from both the West Spitsbergen Current and the East Greenland Current. The West Spitsbergen Current is very barotropic and therefore Gascard et al.  emphasized that its high variability may result from sensitivity to topographic irregularities. They conclude from drifter data that the Atlantic Water does not recirculate westward in continuous branches but rather in form of eddies. While Hanzlick  assumed eddies in the West Spitsbergen Current to be generated through baroclinic instability, Jonsson et al.  showed evidence that they are generated by wind fluctuations and that both wind and current fluctuations have a maximum in winter. Extensive measurements during MIZEX84 revealed small eddies that shed off from a frontal jet along the ice edge in the northern Fram Strait and grew in size while they were advected southward in the East Greenland Current [Johannessen et al., 1987]. Quadfasel et al.  note that recirculation of Atlantic Water is affected even through small topographic structures like the Molloy seamounts, although the cyclonic sense of rotation suggests the influence of the Molloy Deep rather than of the seamounts.
 We use isopleths of velocity in the Atlantic layer (i.e., from instruments at about 250 m depth) to demonstrate the occurrence and structure of mesoscale features (Figure 5). On the eastern side of the strait two types of mesoscale features appeared several times. One type was an anticyclonic eddy located at moorings F6 and F7 and the other type were westward propagating eddy- or meander-like features. On the western side, the mesoscale motion was only weak in accordance with findings of Jonsson et al. .
 The flow was steady northwestward only at moorings F1 and F2. Already at F3 (above 1035 m water depth) several flow reversals appeared mostly during winter, which lasted in the order of days. At F6, extended periods of flow reversal occurred so that the 2-years mean northward flow was only 0.5 cm/s. During most of these periods, southward flow at F6 was associated with northward flow at F7, implying an anticyclonic motion (noted by arrow pairs in Figure 5). These anticyclonic flow periods lasted between 15 and 20 days. Different from the barotropic flow of the West Spitsbergen Current along the slope, the nature of the flow at F7 is baroclinic with flow reversals toward the bottom.
 The two moorings F6 and F7 were located 21 km apart from each other at the northern end of the Knipovich Ridge. The ridge is visible as an elevation of 400 m at F7. Consequently that part of the West Spitsbergen Current that runs along the western side of the Knipovich Ridge is measured at F7. Atlantic Water is known to travel northward in different branches in the Nordic Seas. Orvik et al.  observed Atlantic Water entering into the Norwegian Sea as a barotropic branch at the upper continental slope and a separate baroclinic branch 150 km apart. At about 75°N, Van Aken et al.  describe Atlantic Water to extend westward to 7°E, the position of the Knipovich Ridge, and to form there a strongly meandering front toward the colder waters of the central Greenland Sea gyre. From several years of hydrographic measurements in the southern Fram Strait, Piechura (person communication) reports on a permanent baroclinic branch above the Knipovich Ridge. Our measurements show the continuation of this baroclinic branch west of the Knipovich Ridge to the sill at 79°N complementing the barotropic flow along the continental slope to form the entire West Spitsbergen Current. Actually, the baroclinic branch can also be seen in the current meter measurements made about 10 km east of F6 [Jonsson et al., 1992].
 At times, the baroclinic branch over the Knipovich Ridge turns eastward in a clockwise sense and then flows southward at the eastern rim of the ridge at F6. During such events the temperature at 250 m depth is similar at both moorings. This suggests that only little heat is released from that level to the atmosphere or ice between northward and southward flow which implies only a short excursion of the flow north of the mooring line.
 Besides this stationary eddy or meander, there are indications of westward propagating features. Correlated phase shifts occur at a few periods (noted by stippled lines in Figure 5) which show velocity extrema to progress westward at a speed of several cm/s.
3.3. Branches of the West Spitsbergen Current
 Besides the immediate return flow in Fram Strait, two branches of the West Spitsbergen Current are reported in the literature to proceed north and eastward. One is the Svalbard branch [Aagaard et al., 1987] which hugs to the shelf edge of Svalbard and transports the warmest water. Another branch, carrying less warm water, follows the western rim of the Yermak Plateau. Saloranta and Haugan  point out that this branch looses its Atlantic Water signal due to enhanced tidal mixing at the Yermak Plateau. This makes the core less obvious but does not necessarily imply reduced heat content due to heat loss to the atmosphere. Heavy ice cover prevents frequent hydrographic surveys of the northern Yermak Plateau, but the few measurements available show persistently a core of warm water. Rudels et al.  traced the Yermak Plateau branch up to 82°N. Johannessen et al.  refer to measurements made earlier which show a core of warm (T > 2°C) water at the northern rim of the Yermak Plateau and also a section from 1991 shows water warmer than 2°C at 82°50′N [Anderson et al., 1994].
 Some authors report a cyclonic circulation at the Yermak Plateau [Bourke et al., 1988; Muench et al., 1992] but these patterns were derived from the baroclinic field with unknown reference velocity or were restricted to the uppermost 100 m. Given the distinct barotropic nature of the West Spitsbergen Current, it is questionable whether the flow below the Ekman layers would circulate cyclonically around the plateau. In winter 1989, Muench et al.  observed an elongated filament of warm water along the northwestern rim of the Yermak Plateau. They interpreted the filament to be generated through tidal rectification rather than through advection; however, in the depth of the warm core their current meter (ship borne ADCP) measurements show northwesterly velocities. In all these measurements, the maximum temperature at the Yermak Plateau slope was the same or only slightly lower than that at the shelf edge north of Spitsbergen. Also a section taken in 1999 from Spitsbergen across the Yermak Plateau (Figure 6) cut a warm core northwest of the Yermak Plateau thus bringing again evidence that warm water is advected there.
 Drifters released at a water depth deeper than 1000 m in the West Spitsbergen Current moved northward along the western rim of the Yermak Plateau [Gascard et al., 1995]. Several drifters crossed the isobaths and turned eastward to cross the plateau north of a small shallow at about 80°40′N while two drifters continued further northward, one of them at least up to 82°N. We use this information in section 3.4 to refer the flow to various branches.
3.4. Section-wide Northward and Southward Flow
 In this section we discuss the heat transported with northward and southward flow separately. All values are summarized in Tables 2a–2d. Further we distinguish between transports over the whole water column and those in the upper layers, i.e., Atlantic and Polar waters, and the deep layers. In the Arctic Ocean, often 0°C is used as lower temperature boundary of Atlantic Water while usually higher values are used for the Nordic Seas, e.g., 3°C by Swift and Aagaard . Since our interpolated 0°C isotherm lies deeper than in reality we choose 1°C as lower boundary of Atlantic Water. The error in depth is only small for this isotherm and its maximum depth (700 m) is close to the water depth of the Yermak Plateau. By this definition of Atlantic Water, we separate it from the deep parts of the water column which cannot cross the Yermak Plateau and enter the Eurasian Basin along this pathway.
Table 2b. Yearly Mean Heat Flux Through Individual Branches in Fram Strait for 2 or 3 Years a
Yermak Plateau Branch
Knipovich Branch Northward
Knipovich Branch Southward
Deep Water Northward
Deep Water Southward
For the definition of the branches, see text. For details, see Table 2a. MAW means Modified Atlantic Water.
Table 2c. Yearly Mean Volume Transport Through Individual Branches in Fram Strait for 2 or 3 Years a
Svalbard Branch, Sv
Yermak Plateau Branch, Sv
Knipovich Branch Northward, Sv
Knipovich Branch Southward, Sv
MAW Northward, Sv
MAW Southward, Sv
Polar Water, Sv
Deep Water Northward, Sv
Deep Water Southward, Sv
For details see Table 2a. MAW means Modified Atlantic Water.
Table 2d. Yearly Mean Temperature of Individual Branches in Fram Strait for 2 or 3 Years a
Svalbard Branch, °C
Yermak Plateau Branch,°C
Knipovich Branch Northward,°C
Knipovich Branch Southward,°C
Deep Water Northward,°C
Deep Water Southward,°C
For details see Table 2a. MAW means Modified Atlantic Water.
 The total yearly mean northward volume flux in the periods 97/98 and 98/99 was 9 2 and 10 1 Sv while the southward transport decreased from 13 2 Sv to 12 1 Sv, and the net transport toward south decreased from 4 2 to 2 2 Sv [see also Fahrbach et al., 2001]. Consequently we have to assume a compensating transport change elsewhere, e.g., through the Canadian Archipelago or the Barents Sea.
 The northward heat transport over the full water column varied between 8 3 TW in June 1997 and 75 10 TW in February 1999. Notably, the maximum northward heat transport occurred in winter in both years when the temperature was at its minimum, though this was less pronounced in 97/98. The same pattern and very similar absolute values emerge for the flow of Atlantic Water which means that almost all the heat transport (referred to −0.1°C) takes place in that layer. While there is no seasonal correlation between heat transport and temperature, there is a strong correlation between the variation of heat transport and that of the volume transport (Figure 4). The strong dependence of heat transport on the volume transport is also reflected in the spatial distribution along the cross section (Figure 7). Despite the small cross-section area, typically the highest heat transport is found together with the strongest velocities on the West Spitsbergen slope.
 The yearly averaged heat advected with northward currents was 32 6 TW for the full water depth and 35 3 TW for the Atlantic Water in the period 97/98, increasing to 55 4 TW (57 3 TW for Atlantic Water) in 98/99. Part of the heat returned toward south with southward flowing water warmer than the reference value. However, the change of the total southward heat transport associated with southward currents, 2 10 TW, is not significant within the error limits. Restriction to Atlantic Water gave also similar results for both periods (21 3 TW). Consequently the total net heat flux (with Tref = −0.1°C) toward the Arctic Ocean more than doubled from 16 17 TW to 41 5 TW.
3.5. Distribution of Heat Transport in the Various Branches
 On the basis of water mass criteria as well as on the spatial structure of the flow described in section 3 we distinguish different branches and examine their individual contributions to the overall volume and heat transport (Tables 2a–2d). We divide the cross section at 700 m into deep water and upper water in order to encompass the water which would be able to flow across isobaths onto the Yermak Plateau. The upper water is subdivided along the scheme sketched in Figure 2b.
3.5.1. Deep Water
 All water below 700 m is summarized as deep water here. 700 m is approximately the depth of the 0°C isotherm, which is used to distinguish intermediate and deep water masses [e.g., Rudels et al., 2000]. The northward transport of deep water was about 4.6 2 Sv in both periods, 97/98 and 98/99, and the southward transport was 7 2 Sv. This is about half of the total transport in both directions respectively. The temperature of the northward flowing water increased by 0.1°C which can be partly caused by the increase of the bottom water temperature in the Greenland Sea. There is also influence from the upper layers the temperatures of which are used for the interpolation. The consequence was a slight increase in the net heat transport from 3 2 TW to 6 2 TW.
3.5.2. Svalbard and Yermak Plateau Branches
 At the eastern slope, we attribute flow over bottom depths less than 1000 m to the Svalbard branch and flow above bottom depths between 1000 m and 2400 m to the Yermak Plateau branch. This part of the strait was also covered by moorings in the period 99/00 so that the time series extends over 3 years. The western end of the Yermak Plateau branch is at 6°E.
 Together the two branches carried between 28 5 and 46 5 TW northward. Despite lower velocities the volume flux of the Yermak Plateau branch was similar to that of the Svalbard branch (Figure 8) due to its cross section being more than twice that of the Svalbard branch. Throughout the 3 years the volume flux of the Svalbard branch remained about the same (1.7 Sv) while the Yermak Plateau branch rose from 1.4 0.2 Sv in 97/98 to 2.5 0.3 Sv in 99/00 so that the sum of both increased from 3 to 4 1 Sv. The heat flux evolved similarly. In 97/98 the heat flow was 17 5 TW in the Svalbard branch and 11 1 TW in the Yermak Plateau branch. Until 99/00, the heat transport through the Svalbard branch increased only to 19 4 TW and that of the Yermak-Plateau branch to 27 2 TW. Over the 3 years, the average temperature rose by 0.6°C in both branches.
 The increase of temperature in the eastern Fram Strait between 97/98 and 99/00 was also found in the summer CTD surveys which had a better spatial resolution: the mean values of the 50–500 m layer east of 5°W increased by 1°C from 1998 to 1999. Incorporating these years into the 27 years long time series of summer temperatures in the upper layer of the West Spitsbergen Current by Saloranta and Haugan  (Figure 9) suggests 99/00 to be a fairly warm period in the long-term perspective comparable to the early 1970s and the mid 1980s but less warm than the early 1990s.
 The respective impact of the temperature versus the transport change emerges when a hypothetical heat transport is derived by keeping one or the other parameter constant (dashed lines in Figure 8). The strong rise of heat transport in the Yermak Plateau branch is mainly due to an increase of the volume transport, while the heat transport in the Svalbard branch increased only little despite a temperature rise similar to that in the Yermak Plateau Branch. The strong enhancement of heat transport in the West Spitsbergen Current in the late 1990s was therefore equally caused by higher temperatures of the entire West Spitsbergen Current and by increase in volume flux, which was composed of a strong increase in the Yermak Plateau branch volume flux and even a decrease in the Svalbard branch volume flux.
3.5.3. Knipovich Ridge
 To the west of the Yermak Plateau branch, between 3° and 6°E, we interpret the flow to be influenced by the Knipovich Ridge at least during its anticyclonic phases (Figure 5). This 60 km wide section was covered by moorings only for 2 years. The northward volume transport was about 1 0.3 Sv and half of it returned southward. The mean temperature was lower than that of the Yermak Plateau and Svalbard branches by 0.5°C but the rise by 0.5°C during the 2 years was about the same. The similarity of the temperatures in the northward and southward flow indicates that the returning water had made only a short excursion north of the section. The net heat transport was around 3 2 TW in both periods.
3.5.4. Atlantic Water West of 3°E
 Water west of 3°E which is warmer than 0°C is addressed as Modified Atlantic Water. The northward transport was below or close to the level of significance. The transport toward south was between 4.6 0.6 Sv and 3.4 0.4 Sv with temperatures rising slightly from 1.1°C to 1.3°C between 2 years. The temperature was lower and its increase was much smaller than that in the eastern northward flowing branches; hence even if part of the Atlantic Water might recirculate here southward after only a short excursion through northern Fram Strait, it has lost a considerable amount of heat there. However, not all moderately warm water in Fram Strait is from this short-loop recirculation branch. Rudels et al.  showed evidence of water returning from the central Arctic to Fram Strait which was warmer than 1.5°C. The modified water carried between 21 5 and 15 5 TW back to the Greenland Sea, i.e., again the change in volume transport made a stronger effect compared to the weak warming.
3.5.5. Polar Water
 Polar Water, defined as water west of 3°E which is colder than 0°C, was transported at a persistent rate of 1 0.2 Sv over 3 years flowing southward above 300 m. The persistency of the transport is remarkable since the monthly values varied by more than a factor of 2. Maximum southward volume and heat transports typically occur in late winter. The yearly averaged temperature decreased only slightly from −0.8 to −1°C which resulted in a northward heat transport between 3 and 4 0.5 TW.
4. Comparison to Other Published Transport Values
 Our transport estimates are higher than any published estimates which we are aware of. However, a closer look reveals less disagreement than a simple comparison of the numbers.
 For the eastern part of the West Spitsbergen Current our results confirm earlier findings that it is strongly barotropic [Hanzlick, 1983; Gascard et al., 1995] and therefore we do not refer to values obtained by baroclinic estimates (a comprehensive overview is given by Hanzlick ). Within the range of reported transports, the highest numbers stem from interpolation of direct current observations while all indirect estimates (i.e., through inverse modeling of hydrographic properties and the use of budget estimates) result in lower values, e.g., Schlichtholz and Houssais .
Aagaard and Greisman  report 7.1 Sv toward north for the West Spitsbergen Current from 4 current meters. Hanzlick  obtained 5.6 Sv at 79°N from 4 moorings. The moorings were located between 7°E and 8°30′E and contained 7 instruments (for comparison, these locations were covered by our moorings F1 to F4). Using hydrographic information, Hanzlick  extrapolated a northward flow up to 3°30′E which as a generalization is consistent with our findings despite the extended periods of southward flow which we found at F6 at the northern end of the Knipovich Ridge. However, Hanzlick  limited his transport calculations to 1200 m or even shallower depths at the western end, not taking into account the distinct barotropic character of the flow. Although by this exclusion he missed a considerable fraction of the flow his average northward flow of 5.6 Sv, is among the highest reported values; however, because of neglecting about 50% of the cross section it is still lower than our values.
Foldvik et al.  calculated transports for the upper waters of the East Greenland Current for the year 1984 to 1985 from three moorings which covered the range between 3°W and 5°30′W (for orientation: this range was covered by our moorings F11 to F13). Extrapolating the yearly mean velocity field to the area between 8°W and 1°W they obtained a southward flow of 3 Sv between the surface and 700 m divided into 1 Sv of Polar Water with T < 0°C and 2 Sv of warmer water. When integrating our data over the same domain, between 8°W and 1°W, the results are slightly higher than those of Foldvik et al. , but within an error of about 1 Sv, they compare well for all 3 years of our measurements. The value of about 1 Sv for Polar Water transport results also when integrating eastward up to 3°E (east of F8, see section 3 and Figure 2b; for 99/00 up to F9 at 0°15′W). However, the southward flow of warm Modified Atlantic Water often extends eastward beyond the Greenwich Meridian and thus is only partially captured in the Foldvik et al.  domain. Our results for the southward flow of Modified Atlantic Water are therefore about twice as large (Table 2c) as the 2 Sv of Foldvik et al. .
 The data of Foldvik et al.  as well as ours show that southward flow reaches down to the bottom during most of the time. Estimates from earlier direct current measurements which were limited to fractions of the flow are in reasonable agreement with our transport estimates.
5. Heat Supply to the Arctic Ocean
 The total heat transport through Fram Strait derived from our measurements between summers 1997 and 1999 increased by 25 13 TW (referred to −0.1°C) within this period. Most of this increase (17 6 TW) is attributed to strengthening and warming of the West Spitsbergen Current, comprising both the inshore Svalbard branch and the offshore Yermak-Plateau branch. Part of the oceanic heat input through this boundary current is carried to remote areas in the Arctic Ocean but a considerable amount of heat is released from the ocean to ice and to the atmosphere in Fram Strait itself and in the region north of Svalbard. The strong heat loss in Fram Strait and adjacent areas requires exact notation across which section any heat transport is calculated. This is inevitable when comparing different heat or volume transport estimates from the literature. Furthermore, there is no boundary beyond which water unambiguously will enter the “interior Arctic” which is not a well-defined term. However, though we know only how much heat passed the line along 79°N, we try a first-order estimate how the observed variation in heat transport relates to recently observed warming of the upper layers in the interior Arctic Ocean.
 Two high-resolution CTD sections across the Amundsen and Nansen basins, the Oden section in 1991 [Anderson et al., 1994] and a nearby Polarstern section from 1996, allow quantification of the heat change within 5 years. Anticipating little lateral or vertical exchange within the interior Eurasian Basin, subsurface water should pass from one section to the other without significant modification. Therefore any difference in intermediate water properties should reflect temporal rather than spatial variation. The difference in temperature averaged between 150 m and 400 m depth was approximately 0.5 °C along the entire cross section [Schauer et al., 2002]. From these and other reported temperature comparisons [Grotefendt et al., 1998; Morison et al., 1998] we feel safe to assume that an increase of about 0.5 °C had taken place at least all over the Eurasian Basin in the early 1990s. The Eurasian basin takes about 1/6 of the Arctic Ocean's total area of 6 × 106 km2. To increase the temperature of a 250 m thick layer over 106 km2 by 0.5°C an amount of heat of ΔH = 5 × 1020 J would have been required (using specific heat cp = 4 × 103 J/kg3/K and density ρ = 103 kg/m3). If this heat were provided through additional oceanic heat advection, an increase by 16 TW for 1 year would be needed. This is in the same order of magnitude as the increase of mean annual heat transport through Fram Strait observed between 97/98 and 98/99 and a similar excess of heat transport occurred in the West Spitsbergen Current also in the period 99/00.
 Some of the additional oceanic heat input might not remain in the ocean but be released to the atmosphere and ice in the northern Fram Strait before it spreads through the Eurasian Basin. The additional heat gained together in 98/99 and 99/00 as compared to 97/98 was about 10 × 1020 J. If half of the additional heat input would produce a temperature increase of 0.5°C in the Eurasian Basin, 5 × 1020 J would be released to the atmosphere and ice. Assuming this to occur in Whalers Bay north of Svalbard within an area of approximately 5 × 104 km2, the yearly mean surface heat loss would have to be about 300 W/m2 higher for 2 years than in 97/98. This might be an alarming value compared to estimates by Aagaard et al.  of a winter average of 200 W/m2 in that area.
 We do not intend to explain the warming of the Arctic Ocean intermediate waters in the early 1990s with our measurements in the late 1990s, but we want to emphasize that warming events of the Arctic Ocean which continued until 2000 can be explained by advection through Fram Strait due to increase in both temperature and flow strength.
6. Summary and Conclusions
 Data from instruments, moored in Fram Strait over 3 years with a high spatial coverage, were used for detailed estimates of volume and heat transport. Between September 1997 and August 1999, full depth volume transports at 78°55′N were in the order of 10 1 Sv both northward and southward with a yearly mean net flow between 2 2 and 4 2 Sv to the south. Because of the barotropic nature of the current about half of the transport is below 700 m which might explain discrepancies to otherwise reported lower transport numbers.
 The heat transport, of course, takes place mainly in the upper 700 m with a large part in the West Spitsbergen Current. Two current cores of the West Spitsbergen Current with similar transports exist, one along the shelf edge and the other one proceeding along the western Yermak Plateau slope. Although the water in both cores is warmest in summer, the northward heat transport is highest in winter because the volume transport is much stronger then. From 1997 to 2000, yearly averages in the West Spitsbergen Current varied between 28 5 and 46 5 TW with the strongest increase occurring in the Yermak-Plateau branch due to both rise in temperature and volume transport. Extension of decades-long observations of summer temperatures in the West Spitsbergen Current by Saloranta and Haugan  with our CTD surveys shows only moderate warming in the late 1990s as compared to earlier periods. We conclude that the bulk of heat flux changes in the West Spitsbergen Current cannot be captured by hydrographic summer surveys of the Svalbard branch.
 In contrast to the West Spitsbergen Current, the volume and heat transport of East Greenland Current remained fairly constant and an integration over a subsection showed even similar values of volume transport to that obtained by measurements in the 1980s [Foldvik et al., 1988]. Also the southward heat transport through Modified Atlantic Water varied only little, so that the net heat transport through the entire Fram Strait increased by a similar magnitude (from 16 17 to 41 5 TW) as in the West Spitsbergen Current. This implies that the additional heat input remains in the Arctic Ocean at least for several years or exits elsewhere. The magnitude of the heat transport amplification from 1997 to 2000 was sufficient to explain a similar temperature increase of the basin-wide intermediate layer as was detected in the early 1990s.
 This work was supported by the European Union MAST III Programme VEINS (Variability of Exchanges in the Northern Seas) contract number MAS3-CT96-0070, and the Naval Postgraduate School Office of Naval Research chair in Arctic marine science. We acknowledge the cooperation of Rebecca Woodgate, Ekkehard Schütt, and Andreas Wisotzki. We thank three reviewers for their very helpful comments.