Mechanisms Determining the Variability of Arctic Sea Ice Conditions and Export

Similar documents
Arctic decadal and interdecadal variability

Arctic Sea Ice and Freshwater Changes Driven by the Atmospheric Leading Mode in a Coupled Sea Ice Ocean Model

The Arctic Ocean's response to the NAM

Arctic sea ice response to wind stress variations

Arctic sea ice in IPCC climate scenarios in view of the 2007 record low sea ice event A comment by Ralf Döscher, Michael Karcher and Frank Kauker

The Thinning of Arctic Sea Ice, : Have We Passed a Tipping Point?

Effect of the large-scale atmospheric circulation on the variability of the Arctic Ocean freshwater export

Spectral Albedos. a: dry snow. b: wet new snow. c: melting old snow. a: cold MY ice. b: melting MY ice. d: frozen pond. c: melting FY white ice

The arctic ice thickness anomaly of the 1990s: A consistent view from observations and models

On Modeling the Oceanic Heat Fluxes from the North Pacific / Atlantic into the Arctic Ocean

Arctic sea ice falls below 4 million square kilometers

The Arctic Energy Budget

Recent Changes in Arctic Sea Ice: The Interplay between Ice Dynamics and Thermodynamics

The Northern Hemisphere Sea ice Trends: Regional Features and the Late 1990s Change. Renguang Wu

APPENDIX B PHYSICAL BASELINE STUDY: NORTHEAST BAFFIN BAY 1

The North Atlantic Oscillation: Climatic Significance and Environmental Impact

Estimate for sea ice extent for September, 2009 is comparable to the 2008 minimum in sea ice extent, or ~ km 2.

An Introduction to Coupled Models of the Atmosphere Ocean System

Possible Feedback of Winter Sea Ice in the Greenland and Barents Seas on the Local Atmosphere

The impact of an intense summer cyclone on 2012 Arctic sea ice retreat. Jinlun Zhang*, Ron Lindsay, Axel Schweiger, and Michael Steele

Outline: 1) Extremes were triggered by anomalous synoptic patterns 2) Cloud-Radiation-PWV positive feedback on 2007 low SIE

Ice and Ocean Mooring Data Statistics from Barrow Strait, the Central Section of the NW Passage in the Canadian Arctic Archipelago

10.2 AN ENERGY-DIAGNOSTICS INTERCOMPARISON OF COUPLED ICE-OCEAN ARCTIC MODELS

Don't let your PBL scheme be rejected by brine: Parameterization of salt plumes under sea ice in climate models

Accelerated decline in the Arctic sea ice cover

Arctic Ocean-Sea Ice-Climate Interactions

Atmospheric forcing of Fram Strait sea ice export: A closer look

What makes the Arctic hot?

Origins of the SHEBA freshwater anomaly in the Mackenzie River delta

Recent changes in the dynamic properties of declining Arctic sea ice: A model study

Eurasian Snow Cover Variability and Links with Stratosphere-Troposphere Coupling and Their Potential Use in Seasonal to Decadal Climate Predictions

Comparison of Arctic sea ice thickness variability in IPCC Climate of the 20th Century experiments and in ocean sea ice hindcasts

MODELLING THE EVOLUTION OF DRAFT DISTRIBUTION IN THE SEA ICE PACK OF THE BEAUFORT SEA

Advancements and Limitations in Understanding and Predicting Arctic Climate Change

Whither Arctic sea ice? A clear signal of decline regionally, seasonally and extending beyond the satellite record

Polar Portal Season Report 2013

North Atlantic response to the above-normal export of sea ice from the Arctic

Arctic sea ice response to atmospheric forcings with varying levels of anthropogenic warming and climate variability

Global Atmospheric Circulation

Modeling the Arctic Climate System

Arctic Sea Ice Variability in the Context of Recent Atmospheric Circulation Trends

Changes in Frequency of Extreme Wind Events in the Arctic

The Atmospheric Circulation

THE RELATION AMONG SEA ICE, SURFACE TEMPERATURE, AND ATMOSPHERIC CIRCULATION IN SIMULATIONS OF FUTURE CLIMATE

Special blog on winter 2016/2017 retrospective can be found here -

Storm-driven mixing and potential impact on the Arctic Ocean

North Pacific Climate Overview N. Bond (UW/JISAO), J. Overland (NOAA/PMEL) Contact: Last updated: August 2009

Regional Sea Ice Outlook for Greenland Sea and Barents Sea - based on data until the end of May 2013

The Arctic Ocean Climate a balance between local radiation, advected heat and freshwater

North Pacific Climate Overview N. Bond (UW/JISAO), J. Overland (NOAA/PMEL) Contact: Last updated: September 2008

On the dynamics of Atlantic Water circulation in the Arctic Ocean

Preface. Helsinki, 22 April Annu Oikkonen Department of Physics University of Helsinki

Improving numerical sea ice predictions in the Arctic Ocean by data assimilation using satellite observations

Sea Ice Motion: Physics and Observations Ron Kwok Jet Propulsion Laboratory California Institute of Technology, Pasadena, CA

Lecture 1. Amplitude of the seasonal cycle in temperature

Observed rate of loss of Arctic ice extent is faster than IPCC AR4 predictions

Regional Outlook for the Bering-Chukchi-Beaufort Seas Contribution to the 2018 Sea Ice Outlook

Impacts of Climate Change on Autumn North Atlantic Wave Climate

( ) = 1005 J kg 1 K 1 ;

Recent anomalously cold Central Eurasian winters forced by Arctic sea ice retreat in an atmospheric model

Observing Arctic Sea Ice Change. Christian Haas

On the Circulation of Atlantic Water in the Arctic Ocean

Simulated Response of the Arctic Freshwater Budget to Extreme NAO Wind Forcing

Modeling the Formation and Offshore Transport of Dense Water from High-Latitude Coastal Polynyas

What drove the dramatic retreat of arctic sea ice during summer 2007?

Variability of the Northern Annular Mode s signature in winter sea ice concentration

Ocean Mixing and Climate Change

The forcings and feedbacks of rapid Arctic sea ice loss

Current status and plans for developing sea ice forecast services and products for the WMO Arctic Regional Climate Centre Sea Ice Outlook

Changes in the thickness distribution of Arctic sea ice between and

Arctic climate projections and progress towards a new CCSM. Marika Holland NCAR

Sea Ice Observations: Where Would We Be Without the Arctic Observing Network? Jackie Richter-Menge ERDC-CRREL

Bugs in JRA-55 snow depth analysis

SIMULATION OF ARCTIC STORMS 7B.3. Zhenxia Long 1, Will Perrie 1, 2 and Lujun Zhang 2

2013 ATLANTIC HURRICANE SEASON OUTLOOK. June RMS Cat Response

Mechanisms of Decadal Arctic Climate Variability in the Community Climate System Model, Version 2 (CCSM2)

Impact of sea ice. Rüdiger Gerdes. Alfred Wegener Institute for Polar and Marine Research Bremerhaven, Germany

Arctic oceanography; the path of North Atlantic Deep Water

The impact of polar mesoscale storms on northeast Atlantic Ocean circulation

New perspectives of climate change impacts on marine anthropogenic radioactivity in Arctic regions

Comparison of the Siberian shelf seas in the Arctic Ocean

The Planetary Circulation System

Office of Naval Research Arctic Observing Activities

EFFECTS OF DATA ASSIMILATION OF ICE MOTION IN A BASIN-SCALE SEA ICE MODEL

ARCTIC SEA ICE ALBEDO VARIABILITY AND TRENDS,

Climatic Conditions Around Greenland 1995

June Report: Outlook Based on May Data Regional Outlook: Beaufort and Chuckchi Seas, High Arctic, and Northwest Passage

Trends in Climate Teleconnections and Effects on the Midwest

NSIDC Sea Ice Outlook Contribution, 31 May 2012

ICE DRIFT IN THE FRAM STRAIT FROM ENVISAT ASAR DATA

Causes of Changes in Arctic Sea Ice

GEOCHEMICAL TRACERS OF ARCTIC OCEAN CIRCULATION

Sea-ice change around Alaska & Impacts on Human Activities

Arctic Ocean simulation in the CCSM4

SIO 210 Final examination Answer Key for all questions except Daisyworld. Wednesday, December 10, PM Name:

Temperature and salinity fluctuations in the Norwegian Sea in relation to wind

introduction National Council of Teachers of Mathematics.

Uncertainty in Ocean Surface Winds over the Nordic Seas

MERIDIONAL OVERTURNING CIRCULATION: SOME BASICS AND ITS MULTI-DECADAL VARIABILITY

The fieldwork during the Polarstern cruise ANT XVI/2 as a contribution to the study of bottom water formation and sea ice transport in the Weddell Sea

Transcription:

1SEPTEMBER 2003 KÖBERLE AND GERDES 2843 Mechanisms Determining the Variability of Arctic Sea Ice Conditions and Export CORNELIA KÖBERLE AND RÜDIGER GERDES Alfred-Wegener-Institut für Polar- und Meeresforschung, Bremerhaven, Germany (Manuscript received 1 July 2002, in final form 5 March 2003) ABSTRACT In an ocean sea ice model of the Arctic and the northern North Atlantic driven with 50-yr NCEP NCAR reanalysis data, no appreciable trend in sea ice volume is found for the period 1948 98. However, rather long subperiods, for example, 1965 95, exhibit a large decline in Arctic sea ice volume. These results and the current data situation make connecting global warming to Arctic ice thinning very difficult because the large decadal and multidecadal variability masks any trend. Thermal and wind effects linearly contribute to the total sea ice volume variability. Wind stress forcing significantly contributes to the decadal variability in the Arctic ice volume, affecting both thermodynamic growth and the ice export rate. Ice export events are triggered by enhanced cyclonic wind stress over the eastern Arctic. However, large ice export events depend to a similar degree on the presence of thick ice that is generated in a preceding accumulation phase and do not depend on the local wind conditions around Fram Strait. 1. Introduction The Arctic is an important freshwater source for the North Atlantic (Aagaard and Carmack 1989). Freshwater in the form of sea ice can be transported quickly over long distances and become dynamically active far away from the ice formation region. Freshwater that leaves the Arctic can influence the stratification in the deep water formation areas of the North Atlantic and thus affect the variability of the global thermohaline circulation (Mauritzen and Häkkinen 1997). One example of an eventlike change of deep convection in the Labrador Sea caused by enhanced ice export from the Arctic is the Great Salinity Anomaly of the 1970s (Lazier 1980; Dickson et al. 1988; Häkkinen 1993). Recent observations in Fram Strait (Vinje et al. 1998; Vinje 2001) not only confirm Aagaard and Carmack s transport estimates, but also establish the ice export s large and rapid changes. The time series include at least one ice export event that was associated with the high-index North Atlantic Oscillation (NAO) forcing of the 1994/ 95 winter. A high correlation between ice export through Fram Strait and the NAO index apparently exists since the mid-1970s but not earlier (Hilmer and Jung 2000). Observations of Arctic sea ice and conditions that lead to export events could help to predict part of the large-scale oceanic circulation variability. Unfortunately, Arctic ice properties are difficult to observe. Sea ice Corresponding author address: Dr. Rüdiger Gerdes, Alfred-Wegener-Institut für Polar- und Meeresforschung, Bussestr. 24, Bremerhaven D-27570, Germany. E-mail: rqerdes@awi-bremerhaven.de drift is available from buoys drifting on the pack ice of the Arctic Ocean since 1979 [International Arctic Buoy Programme (IABP)]. Rigor et al. (2002) have used these data to establish a link between the Arctic oscillation (AO) and sea ice motion. They find a slight increase in ice advection out of the Arctic through Fram Strait associated with high-index AO states and relate observed thinning of sea ice to the trend in the AO toward higher index polarity. From the late 1970s onward, satellite observations provide ice cover information. Comprehensive ice thickness data will not be available until the launch of the CryoSat satellite. A limited amount of thickness data from upward-looking sonar on board British and U.S. submarines are available from the National Snow and Ice Data Center, University of Colorado in Boulder. While those data do not completely cover the Arctic in space and time, they provide information about longterm changes along repeated sections. Rothrock et al. (1999) estimated a 40% decrease in Arctic ice volume from the 1960s to the 1990s. Their findings were confirmed by Wadhams and Davis (2000) who compared data from two cruises in the Eurasian basin, which were taken in 1976 and 1996. For the 1990s (using data from submarine cruises from 1991 to 1997), Winsor (2001) found no thinning of the ice in the Beaufort Sea, the North Pole area, and along several submarine tracks connecting the two regions. To distinguish forced long-term trends like those to be expected by global climate change from natural variability, we need to know the mean state and the variability of the Arctic sea ice system for as long as 2003 American Meteorological Society

2844 JOURNAL OF CLIMATE VOLUME 16 possible. In light of the present data situation, model simulations are the only feasible means to achieve this kind of information. Here, we have applied a coupled ocean sea ice model to this problem. The length of the simulation is still confined by the atmospheric forcing data that we took from the National Centers for Environmental Prediction National Center for Atmospheric Research (NCEP NCAR) reanalysis starting at 1948. After a short description of model components and forcing data, we show variability of several sea ice properties and compare them to available data. We discuss possible mechanisms for the modeled variability and present results from sensitivity experiments designed to clarify the role of different forcing components. 2. Model description a. Ocean model The simulations were performed with a coupled ocean sea ice model that was forced by prescribed surface wind stresses and by surface heat and salt fluxes that are calculated using given atmospheric fields and predicted SST, sea ice thickness, and concentration. The ocean model derives from the Geophysical Fluid Dynamics Laboratory modular ocean model (MOM-2; Pacanowski 1995). It solves the primitive equations for the horizontal velocity components, temperature, and salinity. Vertical velocity, density, and pressure are calculated from diagnostic equations. The advection of tracers is handled by the flux-corrected transport (FCT) scheme (Zalesak 1979; Gerdes et al. 1991), which is distinguished by its low implicit diffusion while still avoiding false extrema ( overshooting ) in advected quantities. The implicit diffusion associated with the advection scheme is the only diffusion acting on the tracers in these experiments. Friction is implemented as Laplacian diffusion of momentum with horizontal and vertical viscosities of A MH 2.5 10 4 m 2 s 1 and A MV 10 3 m 2 s 1, respectively. The model domain encompasses the Atlantic north of approximately 20 S, including the whole Arctic Ocean. Open boundary conditions following Stevens (1991) have been implemented at the southern boundary. Monthly mean values for the streamfunction of the vertically integrated flow for the southern boundary have been taken from the Family of Linked Atlantic Model Experiments (FLAME) of Kiel University (C. Dieterich 2001, personal communication) while temperatures and salinities at inflow points are taken from climatology. Linear interpolation to the time step of the model is done for all monthly input fields. To avoid the singularity of geographical spherical coordinates at the pole, the model is formulated on a rotated spherical grid where the equator coincides with the geographical 30 W meridian. The pole of this grid lies at 60 E on the geographical equator. The horizontal resolution is 1 1 in the rotated grid, resulting in FIG. 1. Detail of the model domain with bottom topography (depth in m) and names of features mentioned in the text. Arctic ice export has been calculated perpendicular to the thick dashed lines, Fram Strait, Canadian Archipelago, and between Svalbard and the Kola peninsula. A further thick dashed line marks the southern boundary of the sea ice component of the model. The boxes in the Arctic proper were used to compare average ice thickness from the model and submarine measurements shown in Fig. 2. nearly equal spacing of about 100 km in both horizontal directions in the whole Arctic Ocean. In the vertical, the model contains 19 unevenly spaced levels. Bottom topography is derived by horizontal averaging of the Etopo5 dataset (NGDC 1988). Modifications were made in the Denmark Strait and the Faeroe Bank Channel area to retain the sill depths of the passages. Grid points are regarded as land when more than 40% of the original Etopo5 data points within a grid box are land. The Canadian Archipelago is opened by increasing the requirement to 80% land points within the grid box (Fig. 1). Thus, a passage of at least two active tracer boxes wide is introduced that connects the Canada Basin with Baffin Bay. While the oceanic transport through the shallow Canadian Archipelago is small, this rather wide opening allows a transport of sea ice that appears too large. Sea ice transports in a higher-resolution version of the model (Kauker et al. 2003) are very similar in all key passages. Only in the transport through the Canadian Archipelago do we see a large discrepancy (200 km 3 yr 1 versus 375 km 3 yr 1 ). However, as we will show below, the effect on the overall ice balance and its variability remains small. Compared to a simulation with a closed Canadian Archipelago, the freshwater balance of Baffin Bay was improved. The sea ice thickness distribution north of the Canadian Archipelago was also improved as the southward transport avoids excessive accumulation of sea ice.

1SEPTEMBER 2003 KÖBERLE AND GERDES 2845 TABLE 1. Parameters of the sea ice model (see Hibler 1979). Model term Value Atmospheric drag coefficient 2.5 10 3 Oceanic drag coefficient 5.5 10 3 Atmospheric turning angle 0 Oceanic turning angle 0 Ice strength 15 000 N m 2 Eccentricity 2 Ice concentration parameter 20 Regime parameter 5.0 10 9 s 1 Transfer coefficient for sensible heat Transfer coefficient for latent heat Specific heat capacity of air Emissivity for longwave radiation Lead closing parameter Thermal conductivity of sea ice Thermal conductivity of snow Specific latent heat of sea ice Specific latent heat for evaporation Specific latent heat for sublimation Surface air pressure Density of sea ice Density of snow Density of sea Density of air Stefan Boltzmann constant Freezing temperature of freshwater Freezing temperature of seawater Solar constant Albedo of frozen snow Albedo of melting snow Albedo of frozen ice Albedo of melting ice Albedo of open water 1.75 10 3 1.75 10 3 1004 J kg 1 0.99 1m 2.1656 W m 1 K 1 0.31 W m 1 K 1 3.34 10 5 Jkg 1 2.5 10 6 Jkg 1 2.834 10 6 Jkg 1 1013 hpa 910 kg m 3 300 kg m 3 1025 kg m 3 1.3 kg m 3 5.67 10 8 Wm 2 K 4 0 C 1.86 C 1368 m 2 0.80 0.77 0.70 0.68 0.10 b. Sea ice model The ocean model is coupled with a dynamic thermodynamic sea ice model that has been developed by Harder (1996) from the original Hibler (1979) model. The prognostic variables are sea ice and snow thickness, ice concentration, and ice age, while sea ice drift is diagnosed from the momentum balance where the explicit time dependence has been neglected. This model has been used as a stand-alone model in an otherwise identical configuration and identical parameters in the Sea Ice Model Intercomparison Project (SIMIP). Sea ice model parameters are summarized in Table 1. In the framework of SIMIP this model with viscous plastic rheology gave the best reproduction of observed ice drift statistics and other measured parameters describing sea ice behavior in the Arctic (Lemke et al. 1997). c. Coupling of ocean and sea ice components Sea ice and ocean models use the same time step and the same horizontal grid except for the southern boundary of the ice model, which is located at approximately 50 N. Outflow of ice from the domain is allowed at the southern boundary and at the Bering Strait. Ice that is advected across these boundaries vanishes immediately, with no freshwater being transferred to the ocean. Ice transport across the southern boundary was monitored and remained negligible during the whole integration. The models are coupled following the procedure devised by Hibler and Bryan (1987). The most important feature of their approach is that the stress acting on the uppermost level containing sea ice, as well as liquid water, consists of the sum of wind stress and internal ice stresses. This ensures that in an ice-covered ocean the convergence of ice is considered in the Ekman convergence. The sea ice is forced by wind stress, internal ice stress, a quadratic ocean ice drag, Coriolis force, and surface tilt. The latter is estimated from the second ocean level s velocities that are assumed to be in geostrophic balance. The sea ice component calculates the surface heat fluxes from standard bulk formulas using prescribed atmospheric data and SST predicted by the ocean model. d. Initial data and atmospheric forcing Ice and ocean start from rest. Annual mean potential temperatures and salinities in the ocean were taken from the Polar Science Center Hydrographic Climatology (Steele et al. 2001). As a first guess, ice concentration was taken from climatology derived from Scanning Multichannel Microwave Radiometer (SMMR) and Special Sensor Microwave Imager (SSM/I) data for the period 1986 92. Ice thickness was then interpolated linearly between 0 m for ice-free grid cells and 4 m for completely ice-covered grid cells; points with ice concentration below 0.15 were then reset to zero ice thickness. Initial ice conditions incompatible with the atmospheric forcing adjust very fast during the model integration. In simulations with climatological forcing we found adjustment times for the Arctic sea ice thickness distribution of about 3 yr. The resulting surface freshwater flux could, however, change the ocean s surface salinity to an unrealistic distribution. To avoid this, a preliminary 3-yr run has been performed with the above-mentioned first-guess initial values. The preliminary December mean Arctic ice volume was compared to all December means from a previous 40-yr (1958 97) NCEP forced run. Ice thickness as well as concentration from the most similar December (as judged from the total Arctic ice volume), which was the one of 1982, were then chosen as initial conditions for the experiment. Forcing data are derived from the NCEP NCAR reanalysis dataset of 1948 97 (available online at www. cdc.noaa.gov). We use daily wind stress and monthly means of surface (2 m) air temperature, dewpoint temperature, cloud cover, and scalar wind. Surface freshwater fluxes are due to melting and freezing of snow and sea ice. Precipitation is taken from a climatology (Röske 2001) of the European Centre for Medium- Range Weather Forecasts (ECMWF) reanalysis data while evaporation is calculated using a bulk formula. Continental runoff and the flow through the Bering Strait are not included in the current model. To com-

2846 JOURNAL OF CLIMATE VOLUME 16 pensate for this and to prevent the model salinity from drifting too far from observed distributions, the first model level contains a restoring toward prescribed surface salinity. The time constant for the restoring is 180 days. The reference salinities are climatological winter mean data from Steele et al. (2001). The model has been integrated for the 50 yr for which atmospheric reanalysis data were available. 3. Comparison with measurements Ocean (e.g., Gerdes and Schauer 1997) and ice model (Lemke et al. 1997; Kreyscher et al. 2000; Hilmer 2001) components have been validated in stand-alone mode. For the ice model the validation mainly used ice extent and sea ice drift data with satisfactory results. For the coupled system, in this paper we concentrate on ice thickness and ice export through Fram Strait. a. Ice thickness The main source of ice thickness data are British and American submarine cruises, although there are some additional data from drilling and electromagnetic methods (e.g., Haas and Eicken 2001). Ice draft data from submarine cruises for several years from 1976 onward are publicly available at the National Snow and Ice Data Center (NSIDC; see the Submarine upward-looking sonar ice draft profile data and statistics, available from nsdic@kryos.colorado.edu). The data points are confined to the interior Arctic, defined by the U.S. Navy approved data release area. It excludes the shelf seas of the eastern Arctic as well as a band north of Greenland and Canada. Rothrock et al. (1999) compared data from 1958, 1960, 1962, 1970, and 1976 with recent measurements (1993, 1996, 1997) and found a decrease of 42% in averaged ice thickness at the crossover points of the submarine tracks. Wadhams and Davis (2000) found a similar decrease analyzing submarine data from 1976 and 1996 cruises between 81 N in Fram Strait and the North Pole. Winsor (2001) examined the development of ice thickness in an area from the Beaufort Sea to the North Pole. He states that the 1991 97 data show no trend in sea ice thickness in this area. Tucker et al. (2001) analyzed data from a similar area for the period 1976 94 and found a drop of 1.5 m between the mid- 1980s and the early 1990s in a narrow band from offshore Alaska to near the North Pole. The North Pole region itself is distinguished by rather constant ice thicknesses over this period. Overall, the available observations paint a picture of strong interannual variability with pronounced regional differences and a large-scale downward trend over the last three to four decades. It is very difficult to compare individual submarine tracks and corresponding model results because both underlying fields vary strongly in space and time. Here for model validation purposes, we compare modeled and observed sea ice thickness averaged over subdomains FIG. 2. Average ice thickness in (a) the Beaufort Sea, (b) the central Arctic, and the (c) North Pole area. The exact locations of these areas are shown in Fig. 1. The stars indicate averages of submarine measurements that were taken in these areas. of the Arctic (Fig. 1), namely, the Beaufort Sea (Fig. 2a), the central Arctic (Fig. 2b), and the vicinity of the North Pole (Fig. 2c). Overall, the model results agree well with the observations. The correlation between the thus prepared observations and corresponding model results varies between 0.56 and 0.78. The model reproduces most of the observed long-term changes in ice thickness. The observations show a pronounced thinning of Beaufort Sea ice from a maximum in the mid- 1980s while the model puts the maximum ice thickness a few years later. This discrepancy is partly due to the choice of the averaging region. During the early 1990s, a positive anomaly develops in the model s southern Beaufort Sea where no observations are available. Central Arctic ice undergoes a similar thinning, with observations and model in better phase agreement. The model seems to overestimate the thinning somewhat, with summer values below 1 m in several years during the 1990s. Along the 150 W meridian [analyzed in Tucker et al. (2001)] the submarine data show an average decrease of the ice thickness of around 1.6 m between 1986 and 1992 and rather constant ice thickness in the first half of the 1990s. Here, the model underestimates the decrease with an average of 0.7 m between 72 N and the North Pole. Especially in the southern

1SEPTEMBER 2003 KÖBERLE AND GERDES 2847 part, the model predicts almost no decrease in ice thickness between the two years. The different behavior (under- versus overestimate of ice thickness changes) for the 150 W track and the larger central Arctic area indicates the importance of spatial variability in the ice distribution, both in the model (cf. Fig. 5 below) and in observations. Contrary to Winsor (2001), our analysis shows a downward trend in the North Pole area both in the model and in the observations over the last 15 yr. Most of the changes apparently occur in the 5 yr period since 1988 while ice thickness seems to stay rather constant in the second half of the 1990s. These developments are hard to estimate from the observations alone because the seasonal cycle masks some of the information. Note that the earlier data are mostly from April May while the later observations were predominantly made in September October. A correction of the seasonal data with a single, model-derived annual cycle as in Rothrock et al. (1999) and Winsor (2001) is problematic because the amplitude of the seasonal cycle varies interannually. b. Transport through Fram Strait Vinje et al. (1998) published ice thickness data derived from upward-looking sonar (ULS) measured ice draft in Fram Strait. They have used these measurements together with ice drift velocity calculated from ECMWF wind data and some assumptions about the ice velocity profile near the coast to estimate Fram Strait ice export for the period 1990 96. The position of the instruments is not the same for all months and years and so the ice thickness has been interpolated to 79 N, 5 W. A linear dependence of ice thickness on longitude has been assumed based on results from seven months in 1992/ 1993 when at least four instruments were deployed simultaneously in Fram Strait [for details, see Table 7 in Vinje et al. (1998)]. The model results are in excellent agreement with these estimates after mid-1993 (Fig. 3a). For the period after mid-1993, the correlation of Vinje s estimate and the model result is 0.87 while it is 0.68 for the whole period shown in Fig. 3a. Even on a monthly basis, the model reproduces the winter 1994/95 ice export event (with ice transports of up to 700 km 3 month 1 early in 1995) and the following drop of the transport. It is curious that the high correlation between model- and observation-based estimates breaks down for the earlier years of the observational record. Here the model consistently predicts higher ice transports in winter. We cannot distinguish whether these differences are due to model shortcomings or due to uncertainties in Vinje et al. s estimate. It is noteworthy, however, that before 1993 the position of most of the ULSs in Fram Strait was farther east than in the following years and that they were situated outside of the longitude range of the 1993 array. The assumed linear dependence of ice thickness on longitude might actually not be fulfilled. FIG. 3. (a) Monthly ice transport through Fram Strait in the model (thick line) and estimated by Vinje et al. (1998; thin line). (b) Ice area flux through Fram Strait averaged over winter (Oct May) in the model (thick line) and from Table 3a of Kwok and Rothrock (1999; thin line). Kwok and Rothrock (1999) obtained a time series of ice area transport through Fram Strait from satellite observations for the period 1979 96. Because the satellite observations have relatively large uncertainties for summer, only winter means [October May; Table 3a of Kwok and Rothrock (1999)] are compared in Fig. 3b. Model and satellite data derived time series are highly correlated (0.84). However, the model yields a somewhat higher mean transport over this period, hinting at systematically higher ice velocities or denser ice cover in Fram Strait. However, Kwok and Rothrock state that their estimates might be biased toward lower values due to the assumption of vanishing ice velocity near the coasts. 4. Sea ice variability in the NCEP hindcast experiment Arctic sea ice volume shown in Fig. 4 summarizes ice conditions over the whole Arctic. Over the years 1948 97, it shows pronounced interannual and decadal variability with ice volume maxima at the end of the 1960s, 1970s, and 1980s. Anomalies almost reach the amplitude of the climatological annual cycle. On a longer timescale, we see a 15-yr-long growth of the ice volume toward the maximum in 1965, followed by a

2848 JOURNAL OF CLIMATE VOLUME 16 FIG. 4. (a) Simulated Arctic ice volume; dashed horizontal lines denote upper and lower limits of climatological annual cycle. (b) Monthly mean anomalies; the solid horizontal lines denotes one standard deviation to either side from the 50-yr mean. 35-yr-long decline. Sea ice volume in the 1990s is similarly low as during the 1950s. We compare the sea ice thickness composites for winters where the ice volume deviates from the 50-yr mean by more than one standard deviation to either side in Fig. 5. Relatively small differences exist at the North Pole where the thickness is between 3 and 3.5 m. The high ice volume composite contains thick ice of relatively uniform thickness in an arc from Greenland through the Canada Basin to the East Siberian Sea. The thickness gradient in the Eurasian basin is much larger than in the thin ice composite, lines of constant thickness are aligned with the general direction of the Lomonosov Ridge. Over the full simulation period there is no appreciable trend in total Arctic ice volume. Many reports of decreasing ice amount (e.g., Wadhams 1990; Rothrock et al. 1999; Wadhams and Davis 2000) refer only to the period after 1958 or to even shorter periods (see the following section for details). Thus, the model results indicate that the observed decrease might not be a result of a persisting trend in the atmospheric forcing but could equally well be part of a multidecadal cycle in Arctic sea ice. Taken over subperiods of the simulation, the spatial distribution of a trend is of interest as it describes the longest term variability in Arctic sea ice resolved in the experiment. Here, we present the trend in form of slopes of the linear regression of annual mean ice thickness for each grid point for the period 1965 95 (Fig. 6), that is, the period characterized by a long-term decline in ice thickness (cf. Hilmer and Lemke 2000). For this period, ice thickness decreases almost over the whole Arctic. Only in the Canadian Archipelago, north of Canada and Greenland, and in parts of the Laptev Sea does ice thickness increase. Fastest decrease occurs in the East Siberian Sea with more than 0.5 m decade 1. A secondary minimum (fast decrease) is located immediately north of Fram Strait. The spatial pattern of the trend is very similar to that of Hilmer and Lemke (2000) for the period 1961 98. Namely, the results agree in the area of strong decrease extending from the East Siberian Sea to the North Pole and increasing trend north of the Canadian archipelago. The magnitude of the trend is slightly larger in the present model. The slightly different time period or a positive feedback due to oceanic processes are likely causes for this difference. There is also good agreement between the model results and the estimates by Rothrock et al. (1999) derived from submarine data from a similar period. Holloway and Sou (2002) simulate ice thickness in a coupled sea ice ocean model of the Arctic and find a much smaller decrease than estimated by Rothrock et al. (1999). They attribute this to a redistribution of ice between the central Arctic and peripheral regions, especially the Canadian sector. The submarine surveys included the central Arctic but did not cover the Siberian shelf seas and the near coastal regions of North America. Like Holloway and Sou, we find large changes in the East Siberian Sea that are not included in the submarine surveys. Both results point to a redistribution between the Siberian shelf seas and the Canadian Arctic. However, largest variability in the Canadian Arctic occurs somewhat farther into the Canadian Archipelago in our model, while the Arctic proper shows a smaller variability, possibly an artifact of our model where the passages in the Canadian archipelago are unrealistically wide to allow an oceanic transport through the archipelago. 5. Causes of ice volume variability a. Thermodynamic growth and ice export Changes in Arctic ice volume V are the result of net ice production (freezing minus melting) and the net export of ice. Integrated over the Arctic, in our case defined as the area north of 65 N except for Baffin Bay and the Nordic Seas to Fram Strait and the connection North Cape Svalbaard (Fig. 1), the balance reads dv G E, dt where G is the areal integral over the thermodynamic growth rate and E describes the combined export through Fram Strait, from the Barents Sea into the Nordic Seas, and through the Canadian Archipelago. It

1SEPTEMBER 2003 KÖBERLE AND GERDES 2849 FIG. 5. Composite of all winter (Nov Apr) sea ice thickness distributions where the ice volume deviates from the 50-yr mean by more than one standard deviation: (a) below average, (b) above average, and (c) difference between above and below average composites. Contour interval in (a), (b) is 0.25 m and in (c) is 0.10 m. should be noted that thermodynamic growth not only depends on the thermal forcing as represented, for instance, by the surface air temperature. Thermal growth of sea ice depends crucially on the existence of open water. This, in turn, is governed by the wind-forced divergence of sea ice transport as well as the opening of leads through ridging. The variables G and E show pronounced variability on interannual timescales (Fig. 7a). The export term E is dominated by the contribution from the Fram Strait. However, the three export components are not coherent (Fig. 7b). The time series of ice volume anomaly north of Fram Strait from the 50-yr monthly means (Fig. 4) shows two periods of major ice accumulation in the Arctic, one in the early 1960s and one in the second half of the 1980s. Each growth phase takes several years to reach maximum ice volume. Positive ice volume anomalies develop according to a distinct pattern that is most pronounced around the mid-1960s ice accumulation that preceded the Great Salinity Anomaly. An

2850 JOURNAL OF CLIMATE VOLUME 16 FIG. 6. Trend in sea ice thickness derived from the slopes of the linear regressions of annual mean ice thickness for each grid point for the period 1965 95. Contour interval is 0.1 m decade 1. initially strong thermodynamic growth phase is followed by reduced growth when the ice has gained sufficient thickness to impede further ice formation. Ice volume is typically growing during this phase because the ice export is diminished compared to the long-term mean export rate. In a period of only one to two winters the ice volume is then dramatically reduced by an ice export event that is triggered by a change in the predominant wind forcing and the corresponding changes in ice drift patterns. A compilation of winter-centered annual mean Fram Strait ice export broken down in its components (Fig. 8) shows that the large ice export event of the late 1960s was due to both enhanced ice thickness and southward ice velocity. Later events, like in the early and late 1980s, were weaker because the ice thickness was not significantly larger than average. This is especially the case for the 1994/95 ice export event that was restricted to one winter and thus of rather minor importance for the Arctic Ocean freshwater balance. Arfeuille et al. (2000) describe the relationship between wind forcing and sea ice thickness distribution in the Arctic Ocean and the export of sea ice through Fram Strait and the Canadian Archipelago in their sea ice model. They point out the important role of ice thickness in Fram Strait in several large ice export events between 1958 and 1998. Arfeuille et al. also stress that these export events were preceded by accumulation of sea ice in the East Siberian Sea over periods of several years. This is remarkable because their model was only driven by interannually varying winds while surface air temperature followed a climatological seasonal cycle. Hilmer and Jung (2000) found that the correlation of NAO and ice export increased after 1977 due to an eastward shift in the position of the northward extension of the Iceland low. This shift led to pronounced anomalous southerly winds in Fram Strait and to an in-phase variability of the NAO index and the Arctic ice export through Fram Strait. Tremblay (2001) describes this shift as a change in the phase relationship between the Barents oscillation and the Arctic Oscillation. In Hilmer and Jung s study, as well as in Hilmer et al. (1998), the main role of the NAO is in affecting the local wind and thus the ice velocity in Fram Strait. Here, we see that ice thickness plays an equally important role for the ice export. Ice thickness is built up over several years and depends on remote processes like the thermodynamic conditions in the main formation areas of the East Siberian, Laptev, and Kara Seas. Furthermore, the source region of the ice in Fram Strait (and other export passageways) is not always the same. Figures 9a,b show the ice transport (ice velocity times ice thickness) for the situation before and after the ice export events of the late 1960s and the late 1980s. For several years before the ice volume maximum in 1966, the anticyclonic gyre of ice transport is confined to the inner Arctic. Sea ice drift to the Canadian Archipelago and Fram Strait is only from areas of relatively thin ice; divergence of the ice transport occurs mainly in the Kara and Laptev Seas. During the two years of strong ice export the pressure difference between eastern and western Arctic increases, mainly related to a drop in sea level pressure (SLP) over the Barents Sea. The cyclonic circulation in the whole Eurasian basin and especially between the North Pole and Fram Strait intensifies and diverts a substantial part of the ice that previously circulated in the Beaufort gyre toward Fram Strait. Since the cyclonic motion sets in close to Greenland, very thick ice turns toward Fram Strait. Divergence of sea ice transport now exists in an arc from the Beaufort Sea to the Kara Sea and beyond to the northeastern corner of Greenland. Most remarkably, divergent transport is also present in the interior, from the northern end of the East Siberian and Laptev Seas to the North Pole. The pre-1988 situation (Fig. 9c) is similar to the ice transport pattern during the 1960s accumulation, corresponding to a very similar wind stress field averaged over the accumulation phases. The ice export event is then associated with a retreat of the anticyclonic gyre into the Beaufort Sea and an expansion of the cyclonic sea ice motion from the Barents and Kara Seas into the Eurasian basin. These changes led to an enhanced ice transport through the Canadian Archipelago and to a supply of thick ice from north of Greenland to Fram Strait (Fig. 9d). The divergence of the ice transport before and during the event shows weak convergence over most of the Arctic during the accumulation phase. Especially large convergence occurs near the coast of the Chukchi and East Siberian Seas, as well as south of Fram and Davis Straits. Kara and Laptev Seas are the only important source regions of ice transport. The region of divergent ice transport increases dramatically in the export event, when ice moves from the East Siberian Sea and large parts of the interior Arctic into the subpolar seas. Accumulation through converging ice transport still prevails in the Chukchi Sea and is actually stronger than during the accumulation phase in the Beaufort Sea.

1SEPTEMBER 2003 KÖBERLE AND GERDES 2851 FIG. 7. (a) Annual rate of change of Arctic ice volume (solid black line), annual Arctic ice export (magenta bars), and annual net thermodynamic ice growth integrated over the Arctic (yellow bars); horizontal lines denote 50-yr means. (b) Annual net total ice export rates from the Arctic (black bars) and the contributions through Fram Strait (green bars), through the Canadian Archipelago (white bars), and east of Spitzbergen (yellow bars). All bars start from the zero line. The accumulation phases are similar in both periods. Also, the following ice export events are both due to the expansion of the cyclonic atmospheric circulation regime into the interior Arctic. In the mid-1960s case, the cyclonic circulation is centered in the Barents Sea while it appears as an extension of the large-scale low pressure belt of the Icelandic low in the late-1980s event. The different sea ice motion regimes, a closed anticyclonic gyre during the accumulation phases with little communication with the exits of the Arctic Basin, and a retreat of the anticyclonic gyre into the Beaufort Sea with a strong cyclonic sea ice motion in the eastern Arctic, have been shown from buoy data collected within the International Arctic Buoy Program (Rigor et al. 2002). They have been associated with different phases of the Arctic Oscillation. Ice export events triggered by enhanced cyclonic wind stresses over the eastern Arctic, however, go back beyond the recent decades of strong positive phases of the Arctic Oscillation as the case of the 1960s event demonstrates.

2852 JOURNAL OF CLIMATE VOLUME 16 FIG. 8. Annual mean (from Aug of the previous year through Jul) of (a) Arctic ice volume and (b) Fram Strait ice export. Solid horizontal lines denote the respective 50-yr means; (c) the components # h dl, #h dl (thick and thin lines, respectively) of the ice transport, where the overbar denotes an average over 50 yr and the prime denotes deviations from that mean. The integral is taken across Fram Strait. The correlation of primed quantities is negligible. The removal of substantial parts of the Arctic ice volume leads to reduced average thickness of Arctic sea ice, which enables a period of enhanced ice growth and reestablishment of the depleted ice volume. This recovery phase can again be best observed after the 1967/68 ice export event (Fig. 7). The spatial distribution of ice thickness anomaly before this event (Fig. 10a) is characterized by anomalously thick ice extending from the East Siberian Sea far into the interior Arctic. After the export event, anomalously thick ice is present in the northern Barents Sea, north of the Canadian Archipelago and Greenland, and south of Fram Strait (Fig. 10b). The area of previously thick ice in the interior Arctic is now depleted of ice; the ice edge has retreated far from the coast. The retreat of ice is associated with an increase in surface air temperature (SAT), a connection that has been noted by Hilmer and Lemke (2000) and taken as an indication for SAT-induced sea ice thinning. In our analysis, on the other hand, we see a wind-induced divergence of the ice transport in the East Siberian and Laptev Seas associated with the large sea ice export events. The increase in SAT appears to be secondary, as a result of reduced sea ice thickness and cover and the correspondingly higher ocean atmosphere heat fluxes. Based on observed buoy drift, sea ice cover, and SAT, Rigor et al. (2002) come to similar conclusions

1SEPTEMBER 2003 KÖBERLE AND GERDES 2853 FIG. 9. Mean ice transport (ice velocity times ice thickness) averaged over the periods (a) Sep 1959 Aug 1966, (b) Sep 1966 Aug 1968, (c) Sep 1984 Aug 1988, and (d) Sep 1988 Aug 1990. The color map shows the corresponding negative divergence of the ice transport (in myr 1 ).

2854 JOURNAL OF CLIMATE VOLUME 16 FIG. 10. Anomaly of ice thickness in (a) Jul 1966 and (b) Jul 1968 from the 50-yr Jul mean. The contour interval is 0.25 m. regarding the impact of ice export on the heating of the atmosphere over the eastern Arctic. b. Sensitivity experiments To discern the relative importance of wind and thermal forcing, additional experiments have been carried FIG. 11. (a) Arctic ice volume anomaly for the NCEP hindcast (thick line) and the sum of the Tbar and Wbar experiments (thin line); (b) Arctic ice volume anomalies in the Tbar (thin line) and Wbar (thick line) experiments. out that differ only in the atmospheric forcing. The first experiment, denoted Tbar, is forced with a climatological seasonal cycle (derived from the 50 yr of NCEP reanalysis data) of all variables that determine the thermohaline fluxes (air temperature, dewpoint temperature, scalar wind, cloudiness, precipitation). For short, we refer to thermal forcing, although the scalar wind can in no way be neglected. Otherwise, everything is the same as in the reference experiment. In another experiment, denoted Wbar, each year s actual monthly means for wind stress have been substituted by the climatological monthly means. The daily variability is thus preserved. All forcing variables except wind stress are the same as in the reference experiment. The Arctic ice volume anomalies from these two experiments are shown in Fig. 11. Total ice volume, thermodynamic ice growth integrated over the Arctic domain, and Fram Strait ice export are put together for these experiments and the NCEP hindcast run in Fig. 12 and in Table 2. A first surprising result of these experiments is that the Arctic ice volume anomalies linearly sum up to the value of the reference run (Fig. 11). Thus, we can attribute partial sea ice volume changes to wind and thermal forcing, respectively. The large positive ice volume anomaly of the mid-1960s is mostly due to thermal forcing anomalies that act over more than a decade to increase the ice volume. The wind forcing contributes to the growth in the early phase but remains neutral from 1955 onward. The following decline of the ice volume is, in equal parts, due to wind and thermal forcing. The positive ice volume anomaly in the late 1970s is due to the wind forcing, consistent with a reduction of ice velocity and ice export in Fram Strait (Fig. 7). The thermal forcing tends to decrease the ice volume near the end of the accumulation phase; both wind and thermal forcing reduce the ice volume until the 1982 minimum is

1SEPTEMBER 2003 KÖBERLE AND GERDES 2855 FIG. 12. Table of (left) thermodynamic growth rate integrated over the Arctic (middle) net ice export through Fram Strait, and (right) Arctic ice volume for the experiments (top) Tbar, (middle) NCEP, and (bottom) Wbar. The straight lines depict the linear trend. reached. The following large positive ice volume anomaly in the late 1980s is due both to wind and thermal forcing. This is also true for the following rapid decrease of ice volume into the 1990s. The discussion of thermodynamic growth and ice export through Fram Strait (Fig. 7) above revealed a similar combination of both components. It seems that the wind effect can at least qualitatively be equated with the ice export term in the ice volume balance (cf. Figs. 12b and 12e). At the end of the integration period, ice volume stabilizes as the thermal forcing still tends to decrease ice volume and the wind forcing contributes to increase the ice volume. The latter is again an expression of decreasing Fram Strait ice export as the NAO switches to its negative phase in 1995/96. Considering only the ice volume, the temperature forcing is far more important than the wind forcing, with a correlation of c 0.91 between Wbar and the reference experiment (Table 2). However, the change in volume is achieved in an unrealistic way with the variability in the export term strongly reduced. With variable wind and climatological thermal forcing (Tbar) the

2856 JOURNAL OF CLIMATE VOLUME 16 TABLE 2. Characteristics of the response experiments Tbar and Wbar (section 6) in comparison to the reference experiment NCEP. Shown are correlations and standard deviations for the Arctic ice volume, the thermodynamic growth (growth), and Fram Strait ice export (FSexport) for the time series shown in Fig. 12. The correlations are between the corresponding time series of the response experiment and the reference run. The standard deviations are given in km 3 yr 1 for growth and FSexport, and in km 3 for volume. Tbar Wbar NCEP c Growth 0.61 0.86 c FSexport 0.95 0.17 c Volume 0.63 0.91 Growth 433 637 760 FSexport 703 293 716 Volume 1075 1742 2190 correlation is only c 0.63. Although the major ice accumulations are present in Wbar, much of the higherfrequency variability in ice volume is missing in this experiment (the standard deviation is reduced to 1742 km 3 from 2190 km 3 in the reference experiment). Fram Strait export is clearly better represented with the correct wind variability (c 0.95 versus c 0.17). The thermal forcing affects Fram Strait ice export mainly on the longest timescales through variations in the thickness of the ice exiting through Fram Strait. The thermodynamic ice growth does not offer such a clear connection to the components of the forcing, although the thermal forcing has somewhat more influence (c 0.86 versus c 0.61) and provides for higher variability. The above results indicate that on long timescales, the ice volume is to a good approximation given by the (variable) thermodynamic growth and the mean ice export that is mostly given by the mean wind stress. A slight modification is due to the effect of ice thickness variations on the ice export. This conclusion can be substantiated by an experiment in which the wind stress is taken from the accumulation phase of the mid-1980s and applied to the model together with the realistic temporal development of the thermal forcing after the accumulation phase. Here we use the continuously repeated 1985 wind stress and perform a simulation beginning at the maximum of the ice volume in September 1988. The artificially lowered ice export (not shown) is roughly in balance with the net thermodynamic growth such that the annual mean ice thickness remains almost constant over the duration of the experiment (Fig. 13). In the reference run, the ice volume strongly decreases over the same period. This appears to be in contrast to Zhang et al. (2000) who simulated Arctic sea ice from 1979 through 1996 with a mixed-layer ocean sea ice model and forcing derived from SLP and SAT observations of the International Arctic Buoy Program (IABP). They find that the simulated Arctic ice volume reduction is mostly caused by increased export to the Nordic Seas from the eastern Arctic. Due to reduced advection of ice from the western to the eastern Arctic, there is also an internal redistribution of sea ice volume in the Arctic. It is claimed that ice mass therefore responded more strongly FIG. 13. Arctic ice volume simulated with continuously repeated 1985 wind stress after Sep 1988 (thick line). The ice volume for the NCEP hindcast experiment (thin line) is included for reference. to changes in ice dynamics than to thermal forcing. The total Arctic ice volume and its interannual changes in their model are very similar to our results for the period considered. For this period, the results of Tbar and our reference run are also similar although the Tbar experiment yields somewhat higher ice volumes. However, before 1980 and also after 1995, the ice volume in Tbar and the reference run differ substantially in ice volume. Conclusions about the predominance of thermal or wind forcing can apparently not be drawn from observations or model simulations that cover only one or two decades over which fluctuations in the wind stress do not average out to yield the effective mean wind stress. In the case of Zhang et al. (2000), only one phase of the Arctic oscillation with reduced ice export is captured while two phases of the Arctic oscillation with increased ice export fall into their integration period of 1979 96. 6. Summary and conclusions The amount of measurements of sea ice thickness is still limited and fluctuations of Arctic sea ice volume are thus insufficiently known. We have here explored the time variability of Arctic sea ice properties in a coupled ocean sea ice model of the Arctic and the northern North Atlantic driven with 50-yr NCEP NCAR reanalysis data and additional experiments with modified surface forcing. The hindcast simulation reproduces observed variability in Arctic sea ice with reasonable accuracy. No appreciable trend in sea ice volume was found for the period 1948 98 although rather long subperiods, for example, 1965 95, exhibit a large decline in sea ice vol-

1SEPTEMBER 2003 KÖBERLE AND GERDES 2857 ume. This decline is consistent with estimates based on observations (Rothrock et al. 1999; Wadhams and Davis 2000; Tucker et al. 2001). We have looked into the respective roles of thermal and wind forcing. The Arctic ice volume anomalies in experiments with interannual variability restricted to thermal (Wbar) or to wind (Tbar) forcing linearly sum up to the value of the hindcast experiment. The ice volume in Wbar is highly correlated with the hindcast result. However, the wind forcing significantly contributes to the decadal variability in the Arctic ice volume, affecting both thermodynamic growth (by opening and closing of areas of open water due to divergent sea ice transport and ridging) and the ice export rate. Ice export events are triggered by enhanced cyclonic wind stresses over the eastern Arctic. However, large ice export events depend to a similar degree on the presence of thick ice that is generated in a previous accumulation phase. These results make connecting global warming to Arctic ice thinning very difficult for two reasons. First, large decadal and longer-term variability masks any trend. Restricted time series that are available from observations (limited by the availability of submarine observations after 1958) and model results (limited by the available forcing data) produce trends that are more or less arbitrary. To exclude further artificial effects due to the chosen time interval, experiments must be conducted with forcing data that reach into the preindustrial period. Second, the wind stress strongly affects the longterm development of ice volume. A long-term change in the wind stress over the Arctic, possibly by an increase in the number of atmospheric circulation states that favor ice export, would affect the ice volume in a similar manner as a temperature increase. However, it is not clear if more of those have occurred after the onset of global temperature increase. Acknowledgments. NCEP reanalysis data were provided by the NOAA CIRES Climate Diagnostics Center, Boulder, Colorado, from their Web site at www.cdc.nooa.gov. Submarine ice draft data were obtained from the National Snow and Ice Data Center, University of Colorado at Boulder. We thank Christian Dieterich for providing boundary values from the FLAME model of Kiel University. Two anonymous reviewers provided valuable suggestions to improve the manuscript. This work was partly funded by the projects VEINS (EU Contract MAS3-CT96-0070), DEKLIM (BMBF Contract 01 LD 0047), and CONVECTION (EU Contract EVK2-CT-2000-00058). REFERENCES Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and other fresh water in the Arctic circulation. J. Geophys. Res., 94, 14 485 14 498. Arfeuille, G., L. A. Mysak, and L.-B. Tremblay, 2000: Simulation of the interannual variability of the wind driven Arctic sea ice cover during 1958 1998. Climate Dyn., 16, 107 121. Dickson, R. R., J. Meincke, S.-A. Malmberg, and A. J. Lee, 1988: The Great Salinity Anomaly in the northern North Atlantic 1968 1982. Progress in Oceanography, Vol. 20, Pergamon Press, 103 151. Gerdes, R., and U. Schauer, 1997: Large scale circulation and water mass distribution in the Arctic Ocean from model results and observations. J. Geophys. Res., 102, 8467 8483., C. Köberle, and J. Willebrand, 1991: The influence of numerical advection schemes on the results of ocean general circulation models. Climate Dyn., 5, 211 226. Haas, C., and H. Eicken, 2001: Interannual variability of summer sea ice thickness in the Siberian and central Arctic under different atmospheric circulation regimes. J. Geophys. Res., 106, 4449 4462. Häkkinen, S., 1993: An Arctic source for the Great Salinity Anomaly: A simulation of the Arctic ice ocean system for 1955 1975. J. Geophys. Res., 98, 16 397 16 410. Harder, M., 1996: Rauhigkeit und Alter des Meereises in der Arktis Numerische Untersuchungen mit einem großskaligen Modell (with English summary). Ph.D. thesis, Alfred-Wegener-Institut, 127 pp. Hibler, W. D., 1979: A dynamic thermodynamic sea ice model. J. Phys. Oceanogr., 9, 815 846., and K. Bryan, 1987: A diagnostic ice ocean model. J. Phys. Oceanogr., 17, 987 1015. Hilmer, M., 2001: A model study of Arctic sea ice variability. Ph.D. thesis, Institut für Meereskunde an der Universität Kiel, 157 pp., and T. Jung, 2000: Evidence for a recent change in the link between the North Atlantic Oscillation and Arctic sea ice. Geophys. Res. Lett., 27, 989 992., and P. Lemke, 2000: On the decrease of Arctic sea ice volume. Geophys. Res. Lett., 27, 3751 3754., M. Harder, and P. Lemke, 1998: Sea ice transport: A highly variable link between Arctic and North Atlantic. Geophys. Res. Lett., 25, 3359 3362. Holloway, G., and T. Sou, 2002: Has Arctic sea ice rapidly thinned? J. Climate, 15, 1691 1701. Kauker, F., R. Gerdes, M. Karcher, C. Köberle, and J. Lieser, 2003: Variability of Arctic and North Atlantic sea ice: A combined analysis of model results and observations from 1978 to 2001. J. Geophys. Res., 108, 3182, doi:10.1029/2002jc001573. Kreyscher, M., M. Harder, and P. Lemke, 2000: Results of the Sea Ice Model Intercomparison Project: Evolution of sea ice rheology schemes for use in climate simulations. J. Geophys. Res., 105, 11 299 11 320. Kwok, R., and D. A. Rothrock, 1999: Variability of Fram Strait ice flux and North Atlantic Oscillation. J. Geophys. Res., 104, 5177 5189. Lazier, J. R. N., 1980: Oceanic conditions at Ocean Weathership Bravo, 1964 1974. Atmos. Ocean, 18, 227 238. Lemke, P., W. D. Hibler, G. Flato, M. Harder, and M. Kreyscher, 1997: On the improvement of sea ice models for climate simulations: The Sea Ice Model Intercomparison Project. Ann. Glaciol., 25, 183 187. Mauritzen, C., and S. Häkkinen, 1997: Influence of sea ice on the thermohaline circulation in the Arctic North Atlantic Ocean. Geophys. Res. Lett., 24, 3257 3261. NGDC, 1988: ETOPO5 data. Data Announcement 88-M66-02, Digital relief of the surface of the Earth, NOAA/National Geophysical Cata Center, Boulder, CO. Pacanowski, R. C., 1995: MOM 2 documentation, user s guide and reference manual. GFDL Ocean Group Tech. Rep. 3, Geophysical Fluid Dynamics Laboratory, Princeton University, Princeton, NJ, 232 pp. Rigor, I. G., J. M. Wallace, and R. L. Colony, 2002: On the response of sea ice to the Arctic Oscillation. J. Climate, 15, 2648 2663. Röske, F., 2001: An atlas of surface fluxes based on the ECMWF reanalysis A climatological database to force global ocean general circulation models. Rep. 323, Max-Planck-Institut für Meteorologie, Hamburg, Germany, 31 pp.