downscaling ipcc control run and future scenario with focus on the barents sea

23
Ocean Dynamics (2014) 64:927–949 DOI 10.1007/s10236-014-0731-8 Downscaling IPCC control run and future scenario with focus on the Barents Sea Anne Britt Sandø · Arne Melsom · William Paul Budgell Received: 26 June 2013 / Accepted: 28 April 2014 / Published online: 8 June 2014 © Springer-Verlag Berlin Heidelberg 2014 Abstract Global atmosphere-ocean general circulation models are the tool by which projections for climate changes due to radiative forcing scenarios have been pro- duced. Further, regional atmospheric downscaling of the global models may be applied in order to evaluate the details in, e.g., temperature and precipitation patterns. Similarly, detailed regional information is needed in order to assess the implications of future climate change for the marine ecosys- tems. However, regional results for climate change in the ocean are sparse. We present the results for the circulation and hydrography of the Barents Sea from the ocean compo- nent of two global models and from a corresponding pair of regional model configurations. The global models used are the GISS AOM and the NCAR CCSM3. The ROMS ocean model is used for the regional downscaling of these results (ROMS-G and ROMS-N configurations, respectively). This investigation was undertaken in order to shed light on two questions that are essential in the context of regional ocean projections: (1) How should a regional model be set up in order to take advantage of the results from global projec- tions; (2) What limits to quality in the results of regional models are imposed by the quality of global models? We approached the first question by initializing the ocean model in the control simulation by a realistic ocean analysis and specifying air-sea fluxes according to the results from the Responsible Editor: Jin-Song von Storch A. B. Sandø () · W. P. Budgell Institute of Marine Research and Bjerknes Centre for Climate Research, Bergen, Norway e-mail: [email protected] A. Melsom Norwegian Meteorological Institute, Oslo, Norway global models. For the projection simulation, the global models’ oceanic anomalies from their control simulation results were added upon initialization. Regarding the second question, the present set of simulations includes regional downscalings of the present-day climate as well as projected climate change. Thus, we study separately how downscal- ing changes the results in the control climate case, and how scenario results are changed. For the present-day climate, we find that downscaling reduces the differences in the Bar- ents Sea between the original global models. Furthermore, the downscaled results are closer to observations. On the other hand, the downscaled results from the scenario simula- tions are significantly different: while the heat transport into the Barents Sea and the salinity distribution change mod- estly from control to scenario with ROMS-G, in ROMS-N the heat transport is much larger in the scenario simula- tion, and the water masses become much less saline. The lack of robustness in the results from the scenario simula- tions leads us to conclude that the results for the regional oceanic response to changes in the radiative forcing depend on the choice of AOGCM and is not settled. Consequently, the effect of climate change on the marine ecosystem of the Barents Sea is anything but certain. Keywords Barents Sea · Downscaling · Future climate · ROMS 1 Introduction The atmosphere-ocean global circulation models (AOGCMs) used in the fourth assessment report from the Intergovernmental Panel on Climate Change (IPCCAR4 IPCC 2007) are run with different radiative forcing depend- ing on the climate scenario of interest. The scenarios are

Upload: william-paul

Post on 23-Jan-2017

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949DOI 10.1007/s10236-014-0731-8

Downscaling IPCC control run and future scenariowith focus on the Barents Sea

Anne Britt Sandø · Arne Melsom · William Paul Budgell

Received: 26 June 2013 / Accepted: 28 April 2014 / Published online: 8 June 2014© Springer-Verlag Berlin Heidelberg 2014

Abstract Global atmosphere-ocean general circulationmodels are the tool by which projections for climatechanges due to radiative forcing scenarios have been pro-duced. Further, regional atmospheric downscaling of theglobal models may be applied in order to evaluate the detailsin, e.g., temperature and precipitation patterns. Similarly,detailed regional information is needed in order to assess theimplications of future climate change for the marine ecosys-tems. However, regional results for climate change in theocean are sparse. We present the results for the circulationand hydrography of the Barents Sea from the ocean compo-nent of two global models and from a corresponding pair ofregional model configurations. The global models used arethe GISS AOM and the NCAR CCSM3. The ROMS oceanmodel is used for the regional downscaling of these results(ROMS-G and ROMS-N configurations, respectively). Thisinvestigation was undertaken in order to shed light on twoquestions that are essential in the context of regional oceanprojections: (1) How should a regional model be set up inorder to take advantage of the results from global projec-tions; (2) What limits to quality in the results of regionalmodels are imposed by the quality of global models? Weapproached the first question by initializing the ocean modelin the control simulation by a realistic ocean analysis andspecifying air-sea fluxes according to the results from the

Responsible Editor: Jin-Song von Storch

A. B. Sandø (�) · W. P. BudgellInstitute of Marine Research and Bjerknes Centre for ClimateResearch, Bergen, Norwaye-mail: [email protected]

A. MelsomNorwegian Meteorological Institute, Oslo, Norway

global models. For the projection simulation, the globalmodels’ oceanic anomalies from their control simulationresults were added upon initialization. Regarding the secondquestion, the present set of simulations includes regionaldownscalings of the present-day climate as well as projectedclimate change. Thus, we study separately how downscal-ing changes the results in the control climate case, and howscenario results are changed. For the present-day climate,we find that downscaling reduces the differences in the Bar-ents Sea between the original global models. Furthermore,the downscaled results are closer to observations. On theother hand, the downscaled results from the scenario simula-tions are significantly different: while the heat transport intothe Barents Sea and the salinity distribution change mod-estly from control to scenario with ROMS-G, in ROMS-Nthe heat transport is much larger in the scenario simula-tion, and the water masses become much less saline. Thelack of robustness in the results from the scenario simula-tions leads us to conclude that the results for the regionaloceanic response to changes in the radiative forcing dependon the choice of AOGCM and is not settled. Consequently,the effect of climate change on the marine ecosystem of theBarents Sea is anything but certain.

Keywords Barents Sea · Downscaling · Future climate ·ROMS

1 Introduction

The atmosphere-ocean global circulation models(AOGCMs) used in the fourth assessment report from theIntergovernmental Panel on Climate Change (IPCCAR4IPCC 2007) are run with different radiative forcing depend-ing on the climate scenario of interest. The scenarios are

Page 2: Downscaling IPCC control run and future scenario with focus on the Barents Sea

928 Ocean Dynamics (2014) 64:927–949

plausible representations of the future climate that are con-sistent with prescribed emissions of greenhouse gases andother pollutants, and with our understanding of the effectof increased atmospheric concentrations of these gases onglobal climate.

Arzel et al. (2006) found that over half of the AOGCMsused in IPCC-AR4 overestimates sea ice in the southernBarents Sea. They also note that the models that performwell when evaluated against the present-day sea ice climatedo not necessarily simulate realistic poleward oceanic heattransport. Thus, to end up with a sea ice cover close to whatis observed, there must be other unrealistic, compensatingheat fluxes. It is, therefore, important to simulate both thepoleward oceanic heat transport and the sea ice cover real-istically, in order to improve the understanding of BarentsSeas processes in the present climate and to increase ourconfidence in the resulting scenarios.

Regional ocean modeling is only briefly described inIPCC-AR4. The results that are presented in this contextare for the present-day climate only, no results from oceanicdownscaling of climate projections are discussed.

Adlandsvik (2008) used the ROMS ocean model fordownscaling results from a set of atmosphere-ocean gen-eral circulation model (AOGCM) simulations (control runand SRES A1B scenario), for the North Sea. He found thatdownscaling leads to a general warming of the North Sea,with an increasing vertical temperature gradient with highersea surface temperatures (SSTs) in the results from theregional model. These differences were largely attributedto an increasing Atlantic inflow in the regional model.The changes in ocean heat from control to scenario werealso stronger in the regional model, and this warming wasattributed to changes in the atmospheric forcing.

Adlandsvik (2008) suggests that it might be beneficialto apply atmospheric forcing from an atmospheric down-scaling simulation. This was done by Holt et al. (2010),who forced their North Sea configuration of the POLCOMSocean model by the results from an atmospheric downscal-ing of a global control and scenario (SRES A1B) simulation.Their results regarding the warming trend were in generalagreement with Adlandsvik (2008), but a somewhat largerchange in ocean temperatures from the control run to thescenario was reported by Holt et al. (2010). They suggestthat this difference could be attributed to differences inthe atmospheric forcing, which were derived from differentAOGCMs in the two studies.

Meier et al. (2012) used a regional, high-resolutioncoupled atmosphere-ice-ocean-land surface model (RCAO;Doscher et al. (2002)) with lateral boundary conditionsfrom the two GCMs for physical downscaling of the BalticSea for past variations and future projections. Evaluationof the past showed that the simulated sea surface salinity(SSS) and SST both were slightly overestimated. At the

end of the twenty-first century, temperatures were projectedto increase and salinities to decrease as a consequence ofincreased air temperature and freshwater supply, and theA1B changes were larger than all decadal variations everobserved or reconstructed after 1850.

The physical oceanography in the Barents Sea is stronglyinfluenced by the inflow of warm, high-salinity Atlanticwaters from the south and cold, low-salinity waters fromthe Arctic (Helland-Hansen and Nansen 1909; Loeng 1991;Ingvaldsen et al. 2002; Loeng and Drinkwater 2007), as wellas air-sea heat fluxes therein (Hakkinen and Cavalieri 1989;Simonsen and Haugan 1996; Smedsrud et al. 2010; Sandøet al. 2010; Arthun et al. 2011; Arthun et al. 2012). It hasbeen suggested that the observed warming in the high lat-itude northern regions is a manifestation of anthropogenic-induced global warming (IPCC 2007). While this may betrue, part of the observed signal is also likely to be due tolocal natural variability (Drinkwater et al. 2009). Marineecosystems seem vulnerable to climate change, especiallywhen key species are affected. Thus, one objective of thisstudy is to investigate how future climate change will affectthe physical conditions which are important for the lowertrophical levels in the Barents Sea.

In order to address the implications of future climatechange on assessments of marine resources, model resultsare needed which resolve the relevant circulation featuresand constraints such as the bottom topography. Presently,global climate models do not have a horizontal resolutionwhich is needed in order to properly resolve the relevantfeatures in the Barents Sea (Melsom et al. 2009).

Melsom et al. (2009) tried to overcome some of theseproblems by applying forcing from the atmospheric mod-ule of a global AOGCM from NASA Goddard (GISS AOM,hereafter GISS) to a basin-scale, high resolution, coupledocean-sea ice model (ROMS). This configuration of ROMSis referred to as ROMS-G in the present study (Table 1).Important processes associated with bottom topography onfiner scales, shelf-ocean interactions, tides, and mixing arenot properly resolved or included in the coarse climate mod-els. Inaccuracies due to these deficiencies were amelioratedby the application of ROMS-G. The evaluation in Melsomet al. (2009) of the model results for the present-day climaterevealed that the hydrography and its variability were repro-duced with an encouraging quality by ROMS-G, despite thefact that the atmospheric forcing for the regional model hada substantial (regional cold) bias associated with the tooextensive ice cover in the Barents Sea in the global model.

In this study, the analysis of Melsom et al. (2009), inwhich the ROMS-G model was integrated with forcing fromthe present climate (20C3M experiment), is extended byemploying the same regional model but with forcing fromthe coupled NCAR CCSM model (Collins et al. 2006),hereafter called NCAR. This configuration of ROMS will

Page 3: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 929

be referred to as ROMS-N (Table 1). The extended analy-sis is done in order to study the effects of using differentglobal models for atmospheric forcing, as well as differ-ent resolution in the global ocean models. These choicesof global AOGCMs, from which the atmospheric forcingis collected, are based on the study by Overland and Wang(2007) and their assessment of results for Arctic sea ice fromAOGCMs. They found that out of 20 AOGCMs, only threemodels were able to reproduce Arctic Sea ice area within20 % and seasonal ice zone within 30 %. GISS and NCARwere among these, and their respective March sea ice con-centration together with the NCEP reanalysis are shown inFig. 1. From this figure, it is clear that NCAR is closer tothe observations than GISS, which is probably related to thedifferent resolutions in the two models. How downscalingand increased resolution improve model results is thereforea strong motivation for this study.

In addition, the regional simulations are run with forc-ing from one of the future scenarios (the A1B scenario). Wemay then quantify how much the model results differ andcompare these differences with the results from the two con-trol climate simulations in order to assess the reliability ofthe climate projections for the Barents Sea.

The particular model system used for this study is pre-sented in Section 2. In Section 3, a brief description ofthe observations used for model evaluations is provided,and results from these evaluations are presented. Then, inSection 4, the scenario simulations are discussed. Implica-tions of the results in this study are discussed in Section 5,and our conclusions are finally given in Section 6.

2 Model description

The coupled ice-ocean numerical model used for theregional simulations is the same as in Melsom et al. (2009),namely the Regional Ocean Modeling System (ROMS),described in Shchepetkin and McWilliams (2005). It is run

on a stretched orthogonal curvilinear grid with an averageresolution of 10 km, covering the Arctic and the AtlanticOcean down to about 20◦S. The domain and the variableresolution are displayed in Fig. 2. In the vertical, 40 gen-eralized sigma (s-coordinate) levels are applied using thescheme of Song and Haidvogel (1994), with stretching thatenhances the resolution towards the surface and the bottom.

Lateral motions and diffusive energy losses induced bysmall-scale processes are related to the gradients of themean velocities and tracers by eddy viscosity and diffu-sivity coefficients. In this study, the downscalings use thedeformation-dependent horizontal viscosity as proposed bySmagorinsky (1963), but no explicit horizontal diffusivity isspecified. We apply the third-order upwind biased advectionscheme of Shchepetkin and McWilliams (2008). With thepresent horizontal resolution, the locally adaptive diffusionis sufficient to account for any subgrid-scale variability.

ROMS employs split-mode explicit time stepping. In thisstudy, the baroclinic (internal) mode time step is 200 s,while the barotropic (external) mode time step is 10 s. Thebarotropic numerical stability parameter, the Courant num-ber (Cr ), is 0.35, where for stability Cr ≤ 1. The propertiesof the temporal averaging filter applied to the fast externalmode results for coupling to the internal mode are describedin Shchepetkin and McWilliams (2005).

The ROMS simulations have been performed for twoalmost identical periods, 1981–2000 (ROMS-G) and 1980–1999 (ROMS-N) representing the present climate (20C3Mexperiment), and 2046–2065 representing a time slice ofa future scenario (A1B). The A1B scenario represents theemissions resulting from a balance of different technolog-ical changes in the energy system, both fossil-intensiveand non-fossil energy sources (IPCC 2007). For the futureperiod used here, the differences between the different sce-narios with respect to global surface warming are not verylarge compared to the period 2080–2099 (IPCC 2007). Thefirst 5 years are considered as spin-up, and the remaining15 years are used in the analysis.

Table 1 Overview of acronyms, parameterizations, and boundary conditions for the different models used in this study

Init. and lat. bound. cond.

Acronym Model Domain Par. of flux. 20C3M A1B

ROMS-G ROMS Regional Fairall et al. (2003) with SODA SODA +(GISSA1B -GISS20C3M )

GISS atm. and ROMS-G

ocean surf. states

GISS GISS AOM Global – – –

ROMS-N ROMS Regional Bentsen and Drange (2000) SODA SODA +(NCARA1B -NCAR20C3M )

with NCAR atm. fluxes

and ROMS-N ocean surf. state

NCAR NCAR CCSM3 Global – – –

Page 4: Downscaling IPCC control run and future scenario with focus on the Barents Sea

930 Ocean Dynamics (2014) 64:927–949

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 o

25o

E 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

Fig. 1 Ice concentration in GISS (upper left), ROMS-G (upper right), NCAR (mid left), ROMS-N (mid right), NCEP observations (lower left),and difference between ROMS-N and ROMS-G (lower right) in March in the control simulations (20C3M)

Page 5: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 931

Fig. 2 Model domain and horizontal resolution for ROMS-G andROMS-N. Color shading corresponds to the resolution in kilometer asgiven by the colorbar to the right

Daily mean sea level pressure, surface winds, surface airtemperatures, surface-specific humidity, downward long-and short-wave radiations at the surface, and precipitationvalues from GISS are combined with the appropriate seasurface values from ROMS-G in accordance with Fairallet al. (2003) to provide atmospheric forcing for the ROMS-G model. In NCAR, atmospheric fluxes are available, butif the sea surface temperature and/or ice concentration inROMS-N differ from these NCAR fields, then the sensi-ble and latent heat fluxes, and wind stress are modifiedas described in the Appendix. Basically, the bulk fluxalgorithms are inverted to compute winds, air tempera-ture, and specific humidity at a reference level from theNCAR fluxes, then the sensible and latent heat fluxes andwind stresses are computed for open water and over seaice, and recombined with the ROMS-N ice concentrationas a weighting factor. The net long-wave radiation fluxis adjusted for the ROMS-N ice concentration differencesusing the Berliand and Berliand (1952) algorithm. If theROMS-N and NCAR sea surface temperature and ice con-centration do not differ, then the NCAR fluxes are appliedas is (see the Appendix for more details).

As for initial and lateral boundary conditions for theregional ocean model, the intuitive approach would be toapply the results from the respective AOGCMs that areinterpolated to the ROMS grid. However, the horizontal res-olution of the ocean modules in GISS is 4◦ in longitude by3◦ in latitude. This is much too coarse to reliably repro-duce the regional ocean circulation in the Barents Sea andadjacent seas. The resolution in NCAR’s ocean module isvariable, about 50–70 km in the present region, which isalmost an order of magnitude higher than in GISS. Hence,

this approach would impose contrasting constraints on thepair of regional simulations due to a direct effect of theglobal model’s differences in resolution. Regional devia-tions in the global model results from observations are likelyalso impacted by other shortcomings upstream. In order toavoid these contrasting constraints to dominate the regionalsimulations, an alternative approach is adopted by the appli-cation of results from the Simple Ocean Data Assimilation(SODA) data set (Carton et al. 2000a; Carton et al. 2000b).

In the control run, monthly mean climatological valuesfor the period 1981–2000 from SODA are applied at the lat-eral open boundaries. These open boundaries in the NorthPacific and the South Atlantic are so far from the BarentsSea entrance that any impact of this simplification (clima-tological forcing) is negligible over a few decades. Thisis shown by an examination of inter-annual variability inSection 3.4. Along the open boundary in the South Atlantic,SODA has a horizontal resolution of 50–55 km.

Further, ice concentration, thickness, and velocity lat-eral boundary conditions for the sea ice module were takenfrom an annual monthly mean climatology constructed fromthe respective global model fields for the period 1981–2000 (from Fig. 2, we note that with the exception of thenortheast Pacific Ocean, sea ice is not expected to occurat the open boundaries). The lateral boundary conditionsare radiation conditions with nudging (Marchesiello et al.2001), where the incoming components have 1/120 the

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

BW

FB VN

SB FJL

NS

Siberia

Norway

0

200

400

600

800

1000

1200

1400

1600

1800

2000

Fig. 3 Bottom topography in the ROMS simulations for the BarentsSea region. The cruise tracks where the observations are taken from arethe Bjørnøya West (BW), the Fugløya-Bjørnøya (FB), and the VardøNorth (VN). Positions of the CTD stations are indicated by green stars.Volume and heat transports are calculated in the Barents Sea Opening(BSO) between Norway and Spitsbergen (SB), between Spitsbergenand Franz Josefs Land (FJL), in the Barents Sea Exit (BSX) betweenFranz Josefs Land and Novaya Zemlya (NS), and between NovayaZemlya and Siberia

Page 6: Downscaling IPCC control run and future scenario with focus on the Barents Sea

932 Ocean Dynamics (2014) 64:927–949

nudging time scale of the out-going. Initial conditions weretaken from January values from SODA climatology andthe global model climatologies for ocean and ice variables,respectively.

In the scenario simulations, the SODA data set is alsoused as initial and lateral boundary conditions, but this timeadjusted by the difference between the A1B and 20C3M val-ues for GISS and NCAR, respectively. Thus, the ROMS-Gand ROMS-N control climate simulations provide down-scaled results for the Barents Sea that are due to theatmospheric forcing from the respective AOGCMs. Theregional A1B scenario simulations provide additional infor-mation that is attributed to the oceanic climate change in theAOGCMs by means of their anomaly from the present-dayclimate.

To avoid long-term drift in the model salinity, theSSS is restored to the monthly SODA climatology. Therelaxation time is 360 days and allows for some inter-annual variability in the model salinity. This reduces themodel sensitivity to regional biases in the climate modelatmosphere.

Surface freshwater runoff forcing is obtained fromCORE (CORE; see http://data1.gfdl.noaa.gov/nomads/forms/mom4/COREv2.html) (Large and Yeager 2009).These data are the annual mean river runoff distributedglobally with a resolution of 1◦ in longitude and latitude.The data are interpolated to the model grid and take thesame form as precipitation input, i.e., the freshwater sup-ply alters the surface salt flux, but no mass or momentum isadded to the system. The diffuse nature of this runoff forc-ing does therefore not allow for the evolution of vigorouscoastal currents and salinity fronts close to the coastline,even though the resolution in ROMS is sufficiently highto describe such currents. For the A1B scenario, the SSSrelaxation and CORE runoff forcing are modified by the dif-ference between the A1B and the 20C3M values for GISSand NCAR, respectively. Adding T and S anomalies to theSODA data set could lead to ocean instabilities due to thenonlinear equation of state, which might cause different cir-culation changes in the two model systems. Any densityinversions coming from the initialization procedure are dealtwith by the vertical mixing scheme.

Table 2 Salinity evaluation results for fixed cruise tracks. Positive bias values correspond to higher salinities in the model results. Biases in boldand italic have absolute values ≤ 0.2 and ≥ 0.5, respectively. Capitalized letters in the Season column denote months. DJF results for the BWsection are omitted due to very sparse data coverage

ROMS-G GISS ROMS-N NCAR

Depth Section Season Bias Std Bias Std Bias Std Bias Std

0–50 BW MAM −0.04 0.10 −0.26 0.14 0.09 0.12 0.33 0.14

JJA −0.02 0.21 −0.28 0.17 0.09 0.23 0.36 0.22

SON 0.10 0.26 −0.19 0.33 0.32 0.40 0.42 0.33

FB DJF 0.03 0.24 −0.05 0.33 0.17 0.26 −0.50 0.81

MAM 0.04 0.64 −0.09 0.67 0.17 0.65 −0.57 1.09

JJA 0.12 0.23 −0.10 0.38 0.30 0.27 −0.52 0.79

SON 0.18 0.26 0.00 0.37 0.34 0.31 −0.43 0.81

VN DJF 0.06 0.17 −0.79 0.35 0.18 0.20 −0.61 0.54

MAM 0.05 0.16 −0.76 0.31 0.19 0.19 −0.62 0.53

JJA 0.15 0.34 −1.33 0.67 0.26 0.36 −0.77 0.66

SON 0.20 0.34 −1.43 0.80 0.24 0.32 −0.44 0.86

50–300 BW MAM −0.06 0.06 −0.29 0.08 0.07 0.07 0.30 0.09

JJA −0.10 0.10 −0.29 0.08 0.06 0.08 0.31 0.09

SON −0.07 0.10 −0.31 0.12 0.08 0.11 0.30 0.12

FB DJF −0.08 0.12 −0.20 0.18 0.03 0.14 −0.06 0.46

MAM −0.07 0.10 −0.23 0.16 0.05 0.13 −0.10 0.48

JJA −0.08 0.11 −0.24 0.15 0.07 0.14 −0.01 0.33

SON −0.10 0.13 −0.26 0.17 0.04 0.15 0.01 0.32

VN DJF −0.05 0.12 −0.47 0.23 0.06 0.13 −0.08 0.37

MAM −0.05 0.11 −0.50 0.24 0.08 0.12 −0.14 0.42

JJA −0.08 0.11 −0.46 0.22 0.07 0.11 −0.08 0.41

SON −0.06 0.12 −0.42 0.23 0.08 0.14 0.04 0.33

Page 7: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 933

Table 3 As Table 2, for temperature. Biases in bold and italic have absolute values ≤0.5 and ≥ 1.5◦C, respectively

ROMS-G GISS ROMS-N NCAR

Depth Section Season Bias Std Bias Std Bias Std Bias Std

0–50 BW MAM 0.3 1.2 −1.0 1.8 −0.5 1.2 0.9 1.9

JJA 0.1 1.1 −1.2 1.8 −0.5 1.6 0.8 2.2

SON 0.2 1.2 −0.5 1.8 0.1 1.7 1.9 1.9

FB DJF 0.6 1.1 −1.2 1.6 −0.2 1.1 −0.5 2.6

MAM 0.6 1.0 −1.6 1.6 −0.5 1.2 −1.0 2.9

JJA 0.2 1.2 −1.8 1.9 −0.5 1.6 −0.6 3.1

SON −0.2 1.2 −1.2 1.7 −0.5 1.4 0.5 2.5

VN DJF 0.0 0.8 −4.2 1.2 −0.4 1.0 −1.2 2.2

MAM −0.5 0.9 −4.2 1.0 −0.5 0.9 −1.4 2.1

JJA −0.8 1.1 −5.3 1.6 −0.9 1.6 −1.1 2.2

SON −1.0 1.1 −4.3 2.2 −0.8 1.7 −0.1 2.4

50–300 BW MAM 0.6 1.0 −0.9 1.4 −0.2 1.1 0.8 1.4

JJA 0.6 1.0 −0.6 1.5 −0.4 1.2 1.0 1.3

SON 0.3 1.1 −0.9 1.2 −0.2 1.3 1.0 1.3

FB DJF 0.6 0.9 −1.6 1.2 0.0 1.0 0.7 1.6

MAM 0.5 0.1 −1.7 1.1 −0.3 1.2 0.6 1.7

JJA 0.6 0.9 −1.4 1.2 −0.2 1.2 0.8 1.7

SON 0.3 0.9 −1.4 1.3 −0.2 1.3 0.7 1.8

VN DJF −0.1 0.9 −3.0 1.2 −0.3 0.1 0.9 1.4

MAM −0.5 1.1 −3.1 1.2 −0.4 1.2 0.7 1.7

JJA −0.3 1.1 −3.1 1.4 −0.3 1.4 0.2 1.9

SON −0.2 1.0 −2.8 1.7 −0.2 1.5 0.3 1.9

Fig. 4 Probability densityfunctions for salinity (left) andtemperature (right) during thefall (SON) season for theROMS-G (upper) and ROMS-N(lower) models in the VN track.Observations are shown inblack, model results from thecontrol simulation (20C3M) ingrey, and model results from thefuture scenario (A1B) in red

32.5 33 33.5 34 34.5 35 35.5

ObservationsROMSG 20C3MROMSG A1B

−1 1 3 5 7 9 11

ObservationsROMSG 20C3MROMSG A1B

32.5 33 33.5 34 34.5 35 35.5

ObservationsROMSN 20C3MROMSN A1B

−1 1 3 5 7 9 11

ObservationsROMSN 20C3MROMSN A1B

Page 8: Downscaling IPCC control run and future scenario with focus on the Barents Sea

934 Ocean Dynamics (2014) 64:927–949

3 Control simulations (20C3M)

3.1 Observations

Below, the model results are evaluated against hydrographicdata along fixed cruise tracks and cast positions that areavailable from the Institute of Marine Research as describedin Blindheim and Loeng (1981) and Melsom et al. (2009).These tracks are Bjørnøya West (BW), Fugløy-Bjørnøya(FB), and Vardø North (VN) and are shown in Fig. 3.

The main objective of this study is to extend the knowl-edge of consequences of global warming and projectedchanges in the Barents Sea hydrography. Most focus is,therefore, put on the VN section which extends from the

Norwegian coast into the central Barents Sea. The eval-uation of this section then reflects the respective models’ability to represent the Norwegian Coastal Current andtransport of Atlantic Water into the Barents Sea, as well asthe southward extent of Arctic Water. The results are alsorelevant in the context of air-sea heat fluxes in the sameregion.

Current meter data and temperature observations areavailable from moored observation platforms at selectedpositions in the FB section from August 1997 onwards.These observations are used to estimate the volume and heatfluxes of Atlantic Water into the Barents Sea, see Ingvaldsenet al. (2004) for details. Based on additional observations ofcurrents and hydrography, as well as the results from model

0100200300400500600

0

50

100

150

200

250

300

350

0100200300400500600

0

50

100

150

200

250

300

350

34 34.2 34.4 34.6 34.8 35 35.2

0100200300400500600

0

50

100

150

200

250

300

350

0100200300400500600

0

50

100

150

200

250

300

350

−1 0 1 2 3 4 5 6 7 8 9

0100200300400500600

0

50

100

150

200

250

300

350

0100200300400500600

0

50

100

150

200

250

300

350

34 34.2 34.4 34.6 34.8 35 35.2

0100200300400500600

0

50

100

150

200

250

300

350

0100200300400500600

0

50

100

150

200

250

300

350

Observations

−1 0 1 2 3 4 5 6 7 8 9

Fig. 5 Salinity (left) and temperature (right) from ROMS-G (upper)and ROMS-N (lower) for the fall (SON) season from the VN track.Observed values for the corresponding periods are shown on top of

each plot. Values along the x-axis are distances from the cast positionnearest the coast in Norway, in km

Page 9: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 935

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

32.5

33

33.5

34

34.5

35

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−2

−1

0

1

2

3

4

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

32.5

33

33.5

34

34.5

35

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−2

−1

0

1

2

3

4

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−4

−3

−2

−1

0

1

2

3

4

Fig. 6 Sea surface salinity (left) and temperature (right) in March from the control run for ROMS-G (upper), ROMS-N (mid), and the differencebetween ROMS-N and ROMS-G (lower)

Page 10: Downscaling IPCC control run and future scenario with focus on the Barents Sea

936 Ocean Dynamics (2014) 64:927–949

Table 4 ROMS-G and ROMS-N net volume transports in Sv. The reference direction is towards the Arctic Ocean. The sections are defined inFig. 3. In addition to errors due to rounding, there are also some inaccuracies in the diagnostic tool which calculates the transports. This problemoccurs specifically in regions with steep topography. The model itself conserves volume

ROMS-G ROMS-N

Section 20C3M A1B 20C3M A1B

BSO 2.4 2.2 2.3 3.1

Spitsbergen-Franz Josefs Land −0.2 −0.5 0.1 −0.2

BSX 2.2 2.3 2.0 3.1

Novaya Zemlya-Siberia 0.2 0.4 0.1 0.1

simulations, Gammelsrød et al. (2009) and Smedsrud et al.(2010) provide estimates for volume and heat fluxes acrossthe open boundaries of the Barents Sea. These estimates,which are primarily based on observations, are used as a ref-erence in the evaluation of model results in the present study.The transport estimates are subject to errors both in the mea-sured velocities and the calculation method, but errors inthe current measurements are small due to calibration and alarge number of individual samples (Ingvaldsen et al. 2004).The error of the mean BSO volume transport is calculated to±0.1Sv. Note that this estimate is the error of the mean, notan error for the actual transport at a given time (Ingvaldsenet al. 2004).

3.2 Processing of model results

Prior to the comparison with observations, all the modelresults are interpolated linearly in time to match the exactdates of the hydrographic data. The observations and thecorresponding interpolated model results are thereafterlumped together into different seasons to evaluate the mod-els on a seasonal basis. Bilinear interpolation is appliedin the horizontal; while in the vertical direction, resultsare extracted from the model layer that corresponds to theobservation depth. When the bottom depth of the modelis smaller than the observed bottom depth, the deepestmodel value is extrapolated downward. The evaluation ofthe model results is grouped on a seasonal basis, with win-ter in December, January, and February; spring in March,

April, and May; summer in June, July, and August; and fallin September, October, and November.

The coupled global IPCC models used here do notinclude assimilation of observations, and hence, the localatmospheric circulation in these models from a particularyear will generally not correspond to the observed state thatyear, due to the absence of phase locking with respect to theinternal variability of the climate system. A simulation thatis designed in a manner like this can be referred to as a freerun. Hence, we compare the observations to model resultsat the corresponding time of year, but the actual year is cho-sen randomly from the available ROMS results. In orderto shed light on this approach, we also consider how NAOvariability affects the validation results for the upper 50 m.For this, we first computed the NAO index for DJF for eachyear and for the results from each of the two global mod-els that we consider. Next, the validation was repeated, butthis time, observations from the year with the highest NAOindex value were compared with the results from the modelresults with the highest model NAO index value etc.

We find that the validation results barely change whencompared to the results for DJF from the random approachoutlined above. Biases and standard deviations change byabout 10 % or less from one approach to the other. To under-stand why this is the case, the correlations between yearlyDJF observations and model results on one hand, and NAOindex values on the other were computed. We found thatthe correlations between temperature observations from theupper 50 m of the FB and VN sections, and the NAO index

Table 5 As Table 4, for net heat transports in TW

ROMS-G ROMS-N

Section 20C3M A1B 20C3M A1B

BSO 69 69 56 82

Spitsbergen-Franz Josefs Land −2 −3 −1 −3

BSX −11 −2 0 31

Novaya Zemlya-Siberia −1 −1 0 1

Page 11: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 937

Table 6 Annual values of BSO heat transports (TW) from observations (Ingvaldsen et al. 2004), ROMS-G and ROMS-N. The first column givesobserved estimates of Atlantic Water only

Observations ROMS-G ROMS-N

Year Heat transport Year Heat transport Year Heat transport

1997 34 1981 62 1980 69

1998 52 1982 78 1981 67

1999 47 1983 75 1982 61

2000 40 1984 72 1983 49

2001 33 1985 70 1984 54

2002 56 1986 63 1985 49

2003 60 1987 76 1986 54

2004 48 1988 68 1987 57

2005 61 1989 68 1988 45

2006 61 1990 78 1989 62

2007 52 1991 67 1990 58

2008 42 1992 71 1991 57

2009 49 1993 66 1992 55

2010 54 1994 68 1993 48

2011 49 1995 68 1994 59

1996 74 1995 66

1997 75 1996 60

1998 77 1997 59

1999 67 1998 60

2000 62 1999 60

are 0.66 and 0.57, respectively. Then, only about 35 % of thevariance can be related to NAO variability. Moreover, thecorresponding correlations in the model results were muchlower (generally about 0.2).

The integration periods for the two models are shifted by1 year, and the plots of the observations which the modelresults are compared against therefore differ slightly. In the

context of free runs of a control climate scenario, an offsetof 1 year can safely be ignored.

3.3 Evaluation of model results

ROMS-G was thoroughly evaluated in Melsom et al. (2009),and the downscaling was demonstrated to be successful in

Table 7 Fraction of observations that falls outside, inside the span from the model-based 8 member ensemble. Optimally, 0.22 and 0.78 shouldfall outside and inside of the ensemble span, respectively. The observations and model results that are analyzed here are vertical averages betweendepths of 40 and 50 m of each of the casts from the VN transect inside the Barents Sea. See the text for further details

GISS ROMS-G

No. obs. Outside span Inside span No. obs. Outside span Inside span

Temperature 1204 0.37 0.63 1204 0.31 0.69

Salinity 1204 0.29 0.71 1204 0.61 0.39

Salinity >34.9 566 0.20 0.80 566 0.32 0.68

NCAR ROMS-N

No. obs. Outside span Inside span No. obs. Outside span Inside span

Temperature 1190 0.41 0.59 1208 0.19 0.81

Salinity 1190 0.56 0.44 1208 0.69 0.31

Salinity >34.9 559 0.57 0.43 559 0.26 0.74

Page 12: Downscaling IPCC control run and future scenario with focus on the Barents Sea

938 Ocean Dynamics (2014) 64:927–949

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−4

−3

−2

−1

0

1

2

3

4

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−4

−3

−2

−1

0

1

2

3

4

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−4

−3

−2

−1

0

1

2

3

4

Fig. 7 Sea surface salinity (left) and temperature (right) difference in March between the future scenario and the control run for ROMS-G (upper),ROMS-N (mid), and the difference between ROMS-N and ROMS-G future scenarios

Page 13: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 939

terms of the ocean circulation in the shelf sea. Despite thenoticeable sea ice bias in the global model, the regionalmodel was capable of a realistic representation of the oceancirculation. The role of advection and large-scale circu-lation is particularly important for the import of warmAtlantic Water to the Barents Sea. In other words, a cor-rect representation of volume and heat fluxes into andout of the area is important for regional climate. Here,results from both the ROMS-G and ROMS-N integrationsare evaluated against observations and compared to eachother. Given that the regional model (ROMS) is identicalin the two integrations, differences are mainly due to theatmospheric forcing from the GISS and NCAR AOGCMsimulations.

We first consider the model biases and standard devia-tions with respect to observations of salinity and temper-ature. This part of the evaluation is divided into an upperpart of the water column (0–50 m) which is significantlyimpacted by the ocean-atmosphere fluxes and a lower part(50–300 m) which is less influenced by surface fluxes, andmore by lateral fluxes. The biases and standard deviationsfrom the three fixed cruise tracks BW, FB, and VN are listedin Tables 2 and 3.

With very few exceptions, the quality of the model resultsimproves by downscaling the global models. The improve-ment from GISS to ROMS-G for the VN track is particularlynoteworthy. A comparison of the biases and standard devi-ations in the evaluation of the results from the regionalmodels shows that they are generally of similar quality inthese sections. The most noticeable differences from theobservations is that both ROMS-G and ROMS-N have coldbiases in the VN track during summer and fall, and thatROMS-N has a positive salinity bias in the fall season. Thereis also a slight increase in the salinity bias from west to eastin the regional models.

The probability density functions for salinity and temper-ature of the VN section based on observations and ROMSresults during fall are shown in Fig. 4. ROMS-G gives salin-ities in a range from 34.5 to 35.0 which is smaller than theobserved range of 34.0 to 35.1, but it shows most waters at34.9 which is in agreement with the observations. ROMS-N gives even saltier waters in a range from 34.7 to 35.3with a maximum probability at 35.0. Vertical section plotsof salinity and temperature for the VN section are shown inFig. 5, and give the spatial distribution of the high-salinityAtlantic waters and the relative fresh Norwegian CoastalCurrent waters. From these plots, it can be concluded thatthe lack of freshest waters in Fig. 4 is due to a misrepre-sentation of the coastal waters in both models. In addition,the salinities from ROMS-N are in general too high, also inthe interior basin where the Atlantic waters dominate, whileROMS-G is too fresh below 50 m. With regard to temper-ature, the probability density distributions show that both

model systems seem to lack the warmest waters (Fig. 4).From Fig. 5, we find that ROMS-N gives a fair representa-tion of the temperature above 50 m, while ROMS-G is toocold there. Below 50 m, both ROMS-G and ROMS-N aretoo cold in the southern part, while the ROMS-N is far toowarm in the northern part of the section.

The ROMS-G and ROMS-N surface distributions of themodeled salinity and temperature for March are shown inFig. 6. As already seen in the probability density functions(Fig. 4) and section plots (Fig. 5) for salinity, the salinityin the water masses near the coast is shifted towards highervalues in the ROMS-N results when compared to the resultsfrom ROMS-G. As discussed in Section 2, the diffuse natureof the runoff forcing leads to an abatement of the coastalcurrent. Using a ROMS simulation which was configuredon a 4 km grid and forced with atmospheric reanalysis, Lienet al. (2013) obtained a coastal current which was strongerthan in our study, but still weaker than observed (see, e.g.,their Figure 4.11). As indicated by Fig. 5, the bottom tem-peratures in ROMS-N are too high in the northern part.There are also much higher bottom temperatures in ROMS-N in the eastern Barents Sea compared to ROMS-G (notshown). According to observations (Loeng 1991), bottomwaters in the eastern Barents Sea have a temperature around-1.5◦C. ROMS-G has values close to this, while ROMS-Ntemperatures are around 0◦C in this area.

The respective modeled sea ice concentrations for Marchare shown and compared to observations in Fig. 1, andit is evident that the ROMS-G ice extent has improvedconsiderably from GISS. The differences between the twoROMS simulations are mainly in the central Barents Seaand to some extent north of Svalbard. Model results fromclimate models show that there is a negative correlationbetween heat transports in the Barents Sea Opening (BSO)and the Fram Strait (Sandø et al. 2014), which may lead toa flip-flop in the ice distribution in the central Barents Seaand the area north of Svalbard on inter-annual to decadaltime scales. This might explain some of the differences inthe respective downscalings in Fig. 1, depending on howmuch is masked out by 15-years averaging in the figure.All simulations have too high concentrations in the east-ernmost parts of the Barents Sea, off the west coast ofNovaya Zemlya.

Tables 4 and 5 present net volume and heat transportstowards the Arctic through the sections between Norwayand Spitsbergen (BSO), Spitsbergen and Franz Josefs Land,Franz Josefs Land and Novaya Zemlya (BSX), and NovayaZemlya and Siberia (Fig. 3). Reference temperature forheat transport calculations is set to -0.1◦C for the gatewaysbetween the Barents Sea and the Arctic Ocean as recom-mended by Simonsen and Haugan (1996) for heat budgetsin that area. For consistency, this is also applied for theBSO.

Page 14: Downscaling IPCC control run and future scenario with focus on the Barents Sea

940 Ocean Dynamics (2014) 64:927–949

The mean net BSO inflow in the ROMS simulations is2.4 and 2.3 Sv (Table 4). When comparing these values withthe observational based estimate of 2 Sv (Smedsrud et al.2010), we find that the modeled volume transport is real-istic. The modeled volume transports northwards throughBSX for ROMS-G and ROMS-N are almost identical to theBSO transports, 2.2 and 2.0 Sv (Table 4), which is some-what higher than the 1.6 Sv calculated from current meterdata (Gammelsrød et al. 2009).

The corresponding average net BSO heat transport inthe two models differ: ROMS-G has 69 TW, ROMS-N56 TW (Table 5). (Changing the reference temperature to0◦C would lead to a reduction of 1 TW.) Thus, ROMS-N underestimates the observed (Smedsrud et al. 2010) andmodeled (Harms et al. 2005; Sandø et al. 2010) net heattransport of about 73 TW. As in the BSO, the heat trans-ports differ also at BSX: -11 TW in ROMS-G and -0.2 TWin ROMS-N (Table 5). Thus, the total heat import into theBarents Sea in ROMS-G exceeds the ROMS-N import bymore than 20 TW. Table 6 gives the annual BSO heat trans-ports from observations, ROMS-G and ROMS-N, and thestandard deviations for these time series are 8.9, 5.0, and6.4, respectively. The observations show large inter-annualto decadal variability and suggest that forcing from differentperiods of natural variability can influence the mean overa random decade. That said, the simulated heat transportsshow less variability than the observations and are not influ-enced by the choice of period to the same degree as for theobservations.

3.4 Inter-annual variability

In addition to the seasonal cycle, natural variability inthe Atlantic sector of the world oceans takes placeover time scales ranging from inter-annual to multi-decadal (the Atlantic Multi-Decadal Oscillation has aperiod of around 70 years; Schlesinger and Ramankutty(1994)). In assessing how natural variability is repro-duced in the simulations, we are limited by the 15 yearspost-spin up period of the 20C3M experiments withROMS. Hence, we only analyze the inter-annual variabilityhere.

We assess the models’ capability of reproducing theobserved variability in the Barents Sea by comparing obser-vations from the VN transect at intermediate depths in thesurface layer with model results from the same month, fromevery second year. Model values from years in betweenwere discarded, since they were not statistically indepen-dent from the previous year, as discussed by Melsom et al.(2009).

For each observation, we are then left with eight inde-pendent realizations from the various 20C3M experiments.Since the focus here is on temporal variability, the biases

were discarded by subtracting the mean offset between theobservations and the corresponding model results.

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

0 o

25oE 50

oE

75o E

68 oN

72 oN

76 oN

80 oN

84 oN

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

Fig. 8 Same as Fig. 7, but for ice concentration

Page 15: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 941

Further, since observations are instantaneous measure-ments while the model results are monthly averages, wemust represent the discrepancy in the sampling frequency.This is achieved by adding Gaussian noise to the ensem-ble results, similarly to the manner in which observationalerrors are treated in the analysis of ensemble forecasts(Sætra et al. 2004). The standard deviations were set accord-ing to the intra-monthly variability in year-long records fortemperature and salinity at a depth of 50 m, from two moor-ings in the FB section. These values then became 0.4 and0.06 K for temperature and salinity, respectively.

A simple measure of variability is given by the fre-quency of observations that are inside and outside of therange of ensemble values. When the eight values are sortedaccording to their magnitude, the expected probability thata value drawn from the model’s probability distribution isbetween two neighboring values, smaller than the minimum,or larger than the maximum, are 1/9. If the observed vari-ability is larger than the model variability, the observationswill be outside of the ensemble range more frequently than2/9.

The results from the analysis of inter-annual variabil-ity are given in Table 7. Generally, the model variabilityis lower than the observed variability, albeit with vary-ing degrees of discrepancy. We note that for temperature,the regional models have an inter-annual variability that iscloser to the observations than the global models’ inter-annual variability. Since the temporal resolution of theatmospheric forcing in all simulations is daily, the improve-ment is likely due to an improved description in the ROMSconfigurations of the ocean circulation processes that giverise to such variability. The results for variability of salinityare more convoluted. While the GISS results have inter-annual variability that nearly matches the observations, thecorresponding variability in the remaining models is muchtoo low. As described in Section 2, the SSS has beenrestored towards the CORE data in the regional 20C3Msimulations.

Damping of inter-annual variability due to SSS restor-ing has been discussed by Ferry and Reverdin (2004). Here,we apply SSS restoring to SODA annual climatology witha restoring (e-folding) time of 360 days. Further, the COREclimatology provides runoff forcing. These configurationfeatures have two effects on variability: (1) restoring in itselfdampens inter-annual variability and (2) coarse resolutionimpedes the development and variability of the NorwegianCoastal Current.

The Norwegian Coastal Current (NCC) is a persistentoceanic feature of the region, defined as a low-salinity baro-clinic coastal jet that extends from the Skagerrak to theBarents Sea (Sætre R 2007). Skagseth et al. (2011) havenoted that with regard to climate change “the effect on theNCC would critically depend on changes to the mean and

seasonal cycle of both the freshwater and wind forcing”.The 1 degree CORE runoff forcing is much coarser than the0.5 degree width of the region of freshwater influence ofthe NCC and coarser still than the 0.2 degree width of theNCC jet observed by Skagseth et al. (2011). Restoring ofSSS to SODA climatology does supply a freshwater flux tothe continental shelf, but the SSS NCC signal is very diffuse(1–2 degree spatial scale) and does not extend to the BarentsSea, presumably because of lack of resolution of the narrowNCC ( 20 km) on the northern Norwegian shelf.

In order to examine how these effects are represented inthe regional simulations, the analysis of variability in salin-ity was repeated, but only observations with salinities above34.9 were considered. (The standard deviation for the Gaus-sian noise was halved to 0.03, since the restriction in salinityspace reduces the expected intra-monthly variability.) Asexpected, this had a modest impact on the results from theglobal models. However, the interpretation of results fromthe regional models is substantially different, as the inter-annual variability in these salty water masses is representedwith a quality similar to temperature variability.

4 Scenario simulations (A1B)

As observations for the scenario simulations are obviouslynot yet available, these experiments are evaluated by inspec-tion of the horizontal distributions of salinity, temperature,and ice concentration. Furthermore, differences between thetwo A1B downscaled models are shown, and changes fromthe control runs to the scenarios are displayed to illustratepotential future changes.

When inspecting the probability density function for tem-perature in the VN section (panels to the right in Fig. 4),we note that while the change from 20C3M to A1B isminuscule in the ROMS-G simulations, ROMS-N becomesconsiderably warmer. Further, due to the absence of waterscolder than 4◦C, the ROMS-N results have no Arctic Waterremaining in the VN section in the scenario simulation. Thecorresponding results for salinity also reveal little change inthe ROMS-G simulations. However, the reduction in salin-ity in ROMS-N is dramatic, since no waters with salinityexceeding 34.7 is present. Thus, by today’s terminology, theVN section is completely filled with coastal waters in theROMS-N scenario.

Next, consider the surface changes in the Barents Sea,which follow from the depictions in Fig. 7. We find thatROMS-G has a relative weak surface salinity reduction inthe western Barents Sea and a strong increase in the east,while ROMS-N has an overall reduction in the Barents Sea,particularly in the southern and eastern regions. Surfacetemperature has increased in the southern to central partsin both models, but also in the northern parts in ROMS-N,

Page 16: Downscaling IPCC control run and future scenario with focus on the Barents Sea

942 Ocean Dynamics (2014) 64:927–949

in areas which today are mostly occupied by Arctic Water.Corresponding changes are seen at the bottom (not shown).

The results in Fig. 7 are for March conditions, when thesea ice extent is at its maximum. Unsurprisingly, the generalsurface warming in the scenario results leads to a reductionin sea ice. The changes in sea ice concentrations are shownin Fig. 8. Considerable losses are evident in the central andeastern parts in ROMS-G, while the largest reductions inROMS-N are seen in the northern parts from Spitsbergento Franz Josefs Land. Strong reductions in ROMS-N seaice concentration have therefore taken place in areas north-west of the topographically steered Barents Sea Polar Frontwhich today defines the maximum extension of the sea icecover (Vinje and Kvambekk 1991).

Finally, consider the results for lateral fluxes into and outof the Barents Sea in the scenario simulations, as given inTables 4 and 5. The volume flux in the BSO from ROMS-G remains at about the same transport level in the scenario.However, this transport increases to 3.1 Sv in the ROMS-N scenario. The ROMS-G heat transport remains at 69 TW,while the ROMS-N increases to 82 TW, in response to theincreased volume transport of warmer waters. The ROMS-G volume transports through the BSX increases slightlyto 2.3 Sv. There, the ROMS-N volume transport increasesto 3.1 Sv. The heat transport through BSX into the Arcticincreases in both models. However, the rise in BSX heattransport is larger in ROMS-N than in ROMS-G by a factorof three.

5 Discussion

5.1 Present-day climate in global and regional models

The representation of hydrography in the cross sections ofthree cruise tracks is given in Table 2 and 3 for all four con-trol climate simulations. The main conclusion that can bedrawn from these tables is that downscaling reduces biasesin AOGCM results for the Barents Sea, a polar shelf sea,considerably. To a large degree, the improvement can beattributed to better representation of fluxes into the BarentsSea, as discussed for GISS vs. ROMS-G by Melsom et al.(2009).

From Table 2, we note that the salinity in the ROMS-Gresults for the upper 50 m is somewhat closer to obser-vations than in ROMS-N. From the probability densityfunctions in Fig. 4, as well as from the cross sections inFig. 5, we find it likely that this contrast is related to themodels’ representation of the Norwegian Coastal Current.While this current is obviously too weak in the ROMS-Gmodel, it is almost absent in ROMS-N. The misrepresenta-tion of the fresh coastal waters along the Norwegian coastis likely impacted by the use of a 1◦ product for the runoff

forcing. Nevertheless, since the implementation of runoffis the same in ROMS-G and ROMS-N, the differences insalinity must rather be caused by differences in circulationor differences in the parameterizations of air-sea fluxes.

ROMS-G and ROMS-N both have a cold bias in the VNsection (Table 3) with only a shallow layer of waters exceed-ing 7◦C (Fig. 5). Further, ROMS-N has too warm deep waterin northern parts of the VN section as well as in the easternBarents Sea (not shown), whereas the ROMS-G results aremore realistic there. Except for internal recirculation, thisdifference can hardly be explained by horizontal advectionsince transport of heat through all sections into the BarentsSea is lower in ROMS-N (Table 5).

The much improved description of topography in theROMS simulations and its effect on the regional circula-tion features also likely contribute to the improvements inspatially integrated properties (e.g., Tables 2 and 3).

A lack of correlation between the NAO index fromAOGCM simulations and the corresponding model resultsfor upper layer temperature was found in Section 3. Thismay be due to a weak impact of NAO in the Barents Searegion in the coarse resolution AOGCMs. So, like in theNorth Sea case (Adlandsvik 2008; Holt et al. 2010), appli-cation of forcing from regional atmospheric downscaling

Fig. 9 Sea ice concentration in the central and northern Bar-ents Sea (20C3M May climatology). Model results are presentedfor GISS (upper left), ROMS-G (upper right), NCAR (lowerleft), and ROMS-N (lower right). The 200 m isobath is indi-cated by the black contour line (for GISS, grids with depths<200 m are boxed)

Page 17: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 943

experiments might prove beneficial: The local impact ofNAO variability in the Barents Sea might be affected if sucha configuration was adopted.

The sea ice extents in the two regional models are shownin Fig. 1. The extent is overestimated by GISS and, thoughby much less, it is also overestimated by ROMS-G. On

60oW 30oW 0o 30oE 60oE 60oS

30oS

0o

30oN

60oN

33.5 34 34.5 35 35.5 36 36.5

60oW 30oW 0o 30oE 60oE 60oS

30oS

0o

30oN

60oN

−1 −0.5 0 0.5 1

Fig. 10 Sea surface salinity in March from the control run (1981–1999) (left) and the change in sea surface salinity ((2046–2061)–(1981–1999))(right) in GISS (upper) and NCAR (lower)

Page 18: Downscaling IPCC control run and future scenario with focus on the Barents Sea

944 Ocean Dynamics (2014) 64:927–949

the other hand, the results from NCAR and ROMS-N aremore realistic. The ice concentration largely depends onthe sea surface temperature, which in turn is a function ofheat advection and mixing in the ocean, and air-sea heatfluxes (Parkinson et al. 2006; Smedsrud et al. 2010; Sandøet al. 2010; Arthun and Schrum C 2010). Downscaling withincreased horizontal resolution and more realistic heat trans-ports and mixing should therefore lead to more realistic icecover. This is also what we find in the downscaling of GISS.

However, the heat flux into the Barents Sea is lower inROMS-N than in ROMS-G (Table 5). A possible explana-tion of the higher ice concentrations in ROMS-G is thatthe different air-sea heat flux parameterizations describedin Section 2 will imply an unrealistic high ice concentra-tion in ROMS-G where ice is present in GISS: Heat fluxesin ROMS-G are calculated as a function of ROMS-G seasurface temperature and GISS air temperature (Fairall et al.2003). In case of ice cover in GISS and open ocean inROMS-G, the GISS atmosphere temperature used in the cal-culations will likely be biased cold due to the insulationby the ice cover in the global run. Thus, the implementa-tion of air-sea fluxes in ROMS-G may lead to an unrealistic

and strong heat flux to the atmosphere. The correspond-ing excessive oceanic heat loss may then give rise to localfreezing.

When it comes to ROMS-N, heat fluxes are directlyavailable from NCAR. In case of ice cover in NCARand open ocean in ROMS-N, the heat fluxes will be lessthan in ROMS-G since the fluxes are based on too coldtemperatures both in the atmosphere and in the ocean.

The average sea ice concentration for May from the var-ious 20C3M experiments is displayed in Fig. 9. We notethat in the ROMS experiments, sea ice remains over shallowregions, which are mostly occupied by cold, Arctic water.Over the deeper regions in the southwest, the presence ofrelatively warm Atlantic water leads to ice-free conditions.Similar behavior is also seen in the global models, but theregional description here suffers from the lack of resolutionin bottom topography.

5.2 Future perspectives

First, recall that anomalies from the GISS and NCAR A1Bresults relative to those from the corresponding 20C3M

Fig. 11 Results for salinity in azonal cross section across theNordic Seas at 70◦ N. Theclimatology from the WorldOcean Atlas (WOA) is displayedinn the top left panel. The otherpanels in the left column displayclimatologies based on the20C3M simulations from GISS(middle) and NCAR (bottom).The color codes for salinityvalues are displayed by the labelbar to the left. The correspondingchanges from the 20C3M to theA1B scenario results from2045–2060 are displayed by thepanels in the right column. Here,salinity differences are given bythe color bar to the right

Page 19: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 945

simulations were added to the SODA fields upon initial-ization of the scenario simulations in the ROMS-G andROMS-N configurations. Anomalies were also added toSODA conditions at the open boundaries.

The surface salinities in the 20C3M control runs withGISS and NCAR, and their changes in the results from theA1B scenario simulations are shown in Fig. 10. We find thatinside the Barents Sea, the GISS results reveal a change tosaltier and warmer (not shown) conditions in the southeast,while the change in NCAR is to warmer conditions (notshown), with a substantial decrease in salinities in the south-ern and central Barents Sea. Although we discard the first5 years in the two scenario simulations with ROMS due tothe estimated Barents Sea flushing time (3 years; Harms etal. 2005), there may be regions where the initial conditionshave an impact beyond 5 years.

However, the main impact from the imposed anoma-lies in the scenario simulations is the conditions upstream.The conditions in the central and southern Barents Sea inROMS-G and ROMS-N will change in response to changesimposed by initial conditions in the Norwegian Sea and

along the northern subtropical gyre in the Atlantic Ocean.From the panels to the right in Fig. 10, we observe thatwhile the upstream surface salinity only changes modestlyin GISS in these regions, the waters become much less saltyin NCAR. Furthermore, the surface temperatures in the Nor-wegian Sea change little in GISS, while NCAR becomeswarmer (not shown).

The most striking feature in the anomaly fields fromthe future scenario ROMS simulations is the ROMS-Nfreshening in large parts of the Barents Sea (Fig. 7). Thisstrong reduction in salinity is almost certainly due to con-ditions upstream in the global model (NCAR) anomaliesas described above. The increased salinities in the easternBarents Sea in ROMS-G are likely influenced by the largeregional initial anomalies derived from GISS (Fig. 10). Dueto the large SSS anomalies, restoring will probably act aswell.

In agreement with the IPCC multi-model mean (IPCC2007) and the expectations for a future climate, both ROMS-G and ROMS-N show warming of the Barents Sea (Fig. 7).However, the model differences in the bottom waters are

Fig. 12 As Fig. 11, but fortemperature climatologies andtemperature changes

Page 20: Downscaling IPCC control run and future scenario with focus on the Barents Sea

946 Ocean Dynamics (2014) 64:927–949

considerably larger than the surface changes (not shown).The ROMS-G warming of 1–2◦C in the bottom waters isabout the same as in the surface waters, while the ROMS-Nwarming of the bottom waters exceeds 4◦C, more than twicethat of the surface waters. This is probably caused by a cor-responding strong temperature anomaly in the deep Nordicand Barents Seas in NCAR (not shown), which will affectthe bottom waters in the Barents Sea through initial condi-tions and waters being advected steadily from the NordicSeas to the Barents Sea.

A detailed investigation of the results from the globalA1B simulations for the ocean circulation in the NorthAtlantic sector is beyond the scope of the present study.However, since the differences between the relevant resultsfrom GISS and NCAR are substantial, and since we findthat these differences to a large degree dominate the dif-ferences in results from the regional scenario simulations,a general evaluation of the results from GISS and NCARis needed. From Fig. 11, we find almost no Atlantic Waterin the coarser resolution GISS 20C3M simulation, andalso that salinities decrease everywhere in the A1B projec-tion. The salinities in NCAR 20C3M exhibit a pattern thatclosely resembles observations, but with somewhat too largeextremes. This means too much water of Atlantic origin,and too much fresh water in the pathways into the Arctic(NCC) and out from the Arctic (East Greenland Current). Inthe scenario simulation with NCAR, the water masses above500 m become saltier in the interior, while closer to the coastthe opposite change occurs. These reduced salinities mayat least in part be due to advection of fresher water massesfrom the Atlantic Ocean by the Norwegian Atlantic Current(see Fig. 10). The shift towards lower salinities in this part ofthe Nordic Seas in NCAR is reflected by the negative shift inthe salinity probability distribution for ROMS-N in the VNsection (Fig. 4). Figure 12 reveals that the results for tem-perature are in most ways similar to the salinity results. Thecross section pattern from the NCAR 20C3M simulation ismuch more realistic than the corresponding GISS results.The warming seen in the scenario results is larger in NCAR,and a corresponding shift is shown in the probability distri-bution for temperature from NCAR in the VN section forROMS-N (Fig. 4).

The reduced surface salinity in the eastern North Atlanticand the increased surface salinity in the Subpolar Gyre in theNCAR results are not supported by the evaporation and pre-cipitation patterns from the IPCC multi-model mean for theA1B scenario for the same period (IPCC 2007, p. 769). Fur-ther, we also note that the change expected from the NCARsimulation over the coming 40 years is of considerablemagnitude.

This analysis demonstrates the strong influence of ini-tial conditions from global models. If the initial conditionsare reasonable, as they are in the control run where the

initial conditions are based on the SODA data set, the down-scalings show a high potential for improving the hydrogra-phy, and then especially temperature (Table 3), which areimproved due to more realistic topography, more small-scale features, and more realistic heat transports through theBSO. On the other hand, if the initial conditions are char-acterized by big anomalies upstream of the area of interest,realistic or not, this can influence the results for a long time.

6 Conclusions

We have shown that with the configurations of ROMS thatwere described in Section 2, the results for water massesand sea ice in the Barents Sea improve for the present-dayclimate simulations. In contrast to this convergence in thecontrol simulations with ROMS, the scenario simulationsdiverge.

These dissimilarities can be attributed to the configu-ration of the regional experiments. In the control climatesimulations, the initial ocean conditions and lateral bound-ary conditions in the ROMS simulations are identical, so theresults only reflect the differences in the atmospheric forc-ing and the implementation of air-sea fluxes. Advection ofheat and salt from the Norwegian Sea, which is essential forthe conditions in the Barents Sea, ameliorates the discrepan-cies in the results from the global models; see Tables 2 and3 (and also Melsom et al. 2009, Table 5).

In order to capture changes that can be expected in thefuture conditions in the Barents Sea by regional modeling,it is obviously necessary to somehow incorporate changesin the ocean in the model configuration for scenario sim-ulations. Furthermore, to provide information about theuncertainties associated with differences in AOGCM sce-nario simulations, we choose also to adopt initial condi-tions that reflect such AOGCM differences. Due to thelong time scales and high heat capacity of the ocean, theresults from each of the regional simulations become dom-inated by the ocean anomalies in the respective AOGCMexperiments.

We find that the application of a high-resolution regionalocean model can improve the description of ocean advec-tion when compared to global models, thus improving thedescription of, e.g., local hydrography, even though theimplementation of the air-sea fluxes is no longer fully con-sistent due to decoupling. However, the benefit from down-scaling ocean circulation scenarios is limited since changeswill necessarily be dominated by the conditions imposedfrom the respective AOGCM scenarios. The much longertime scales of anomalies in the ocean as compared to atmo-spheric anomalies make oceanic downscaling significantlymore challenging than downscaling of the atmospheric com-ponent of AOGCMs.

Page 21: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 947

Results from downscaling AOGCM simulations revealthat changes from the control experiment to the A1B sce-nario in the water mass properties inside the Barents Seaare impacted substantially by changing properties of theocean fluxes across the open boundaries of this region, aswell as by SSS restoring. This is in contrast to the resultsin similar studies for the North Sea, which mainly attributeprojected changes in ocean temperature to regional atmo-spheric forcing (Adlandsvik 2008; Holt et al. 2010). Giventhe very large differences in the SSS anomalies from GISSand NCAR, it seems obvious that downscaling for the A1Btime slice cannot overcome the problems which seems toorigin from high latitude processes not well covered byGISS and NCAR.

Acknowledgments This work was supported by the NorwegianResearch Council projects NorClim and EarthClim, the Centre forClimate Dynamics at the Bjerknes Centre, and by the NorwegianSupercomputer Committee through a grant of computing time.

Appendix

The equations in this appendix are provided by Bentsen andDrange (2000). Bulk expressions relating turbulent fluxesof momentum and heat to measurable atmospheric variables(Smith et al. 1996)

τ = −ρaCDSu, (1)

QH = ρacpaCHS(Ts −�), (2)

QL = ρaLeCES(qs − q), (3)

where ρa is the air density; CD , CH , and CE are the transfercoefficients for momentum, sensible heat, and latent heat,respectively; S is the average wind speed; u is the meanwind vector; � is the potential temperature; and q is thespecific humidity, all measured at the reference height zr .Ts and qs are the temperature and specific humidity at seasurface, respectively.

If the turbulent fluxes of momentum, heat, and sea sur-face state are available from reanalysis or an atmosphere cir-culation model, the corresponding near surface atmosphericstate can be estimated. This atmospheric state together withthe sea surface state from the model is used to estimatethe final fluxes to be applied. An iterative scheme is usedto calculate the atmospheric state at height zr , and firstguesses must be made for the transfer coefficients, the gusti-ness, wG, and air density. In the following, the subscript ndenotes the iteration number, d denotes data from reanalysisor atmospheric model, and m denotes ocean model.

The mean wind speed vector un at zr is solved from thebulk expression:

τd = −ρn−1a Cn−1

D,d

√(un)2 + (wn−1

G,d )2un, (4)

and is given by

un =

√√√√√√1

2

⎛⎜⎝−(wn−1

G,d )2 +

√√√√(wn−1G,d )

4 + 4

(τd

ρn−1a Cn−1

D,d

)2⎞⎟⎠.

(5)

The average wind speed S =√u2 + w2

G at zr is updated as

Sn =√(un)2 + (wn−1

G,d )2. (6)

Thereafter, the potential temperature and specific humid-ity at zr are solved from the bulk expressions QH,d =ρn−1a cpaC

n−1H,d S

n(Ts,d −�n) and

QL,d = ρn−1a LeC

n−1E,d S

n(qs,d−qn), respectively, and resultin

�n = Ts,d − QH,d

ρn−1a cpaC

n−1H,d S

n, (7)

qn = qs,d − QL,d

ρn−1a LeC

n−1E,d S

n. (8)

The air density are finally updated by a standard equation ofstate for moist air (Gill 1982)

ρna = ρa(�

n, qn, ps,d) (9)

where ps,d is the surface pressure from the data. A newset of transfer coefficientsCD,m, CH,m, andCE,m and gusti-ness, wG,m, are computed with the atmospheric state, con-sistent with the sea surface state represented by Ts,m andqs,m. Using Eqs. 1–3, the turbulent fluxes to be used by theocean model are:

τm = −ρnaCD,mS

nun, (10)

QH,m = ρna cpaCH,mS

n(Ts,m −�n), (11)

QL,m = ρnaLeCE,mS

n(qs,m − qn). (12)

Page 22: Downscaling IPCC control run and future scenario with focus on the Barents Sea

948 Ocean Dynamics (2014) 64:927–949

For parameterization of net long-wave radiation at the seasurface, Ql,m, the approach by Berliand and Berliand (1952)is used

Ql,m = 4εσT 3a (Ts,m − Ta)+ εσT 4

a

(0.39 − 0.005

√e(Ta)

)(1 − χC2),(13)

where ε is the emissivity of water, σ is the Stefann-Boltzmann constant, e is the water vapor pressure, and(1−χC2) is a cloud correction term. The air temperature atzr =10 m is related to the estimated potential temperatureas

Ta = �n − 0.0098zr (14)

To get consistent heat fluxes, the net long-wave radiationsupplied by the reanalysis or atmospheric model, Ql,d , isused

Ql,d = 4εσT 3a (Ts,d − Ta)+ εσT 4

a

(0.39 − 0.005

√e(Ta )

)(1 − χC2) (15)

By subtracting Eq. 15 from Eq. 13, the following expres-sion for net long-wave radiation results

Ql,m = Ql,d + 4εσT 3a (Ts,m − Ts,d). (16)

References

Adlandsvik B (2008) Marine downscaling of a future cli-mate scenario for the North Sea. Tellus 60A:451–458.doi:10.1111/j.1600-0870.2008.00311.x

Arthun M, Schrum C (2010) Ocean surface heat flux variability in theBarents Sea 83(83):88–98. doi:10.1016/j.jmarsys.2010.07.003

Arthun M, Ingvaldsen RB, Smedsrud LH, Schrum C (2011) Densewater formation and circulation in the Barents Sea 58:801–817.doi:10.1016/j.dsr.2011.06.001

Arthun M, Eldevik T, Smedsrud LH, Skagseth Ø, Ingvaldsen R (2012)Quantifying the influence of Atlantic heat on Barents Sea icevariability and retreat 25:4736–4743

Arzel O, Fichefet T, Goosse H (2006) Sea ice evolution over the 20thand 21st centuries as simulated by current AOGCMs 12:401–415.doi:10.1016/j.ocemod.2005.08.002

Bentsen M, Drange H (2000) Parameterizing surface fluxes in oceanmodels using the NCEP/NCAR reanalysis data. Regclim GeneralTechnical Report 4, Norwegian Institute for Air Research, Kjeller,Norway

Berliand M, Berliand T (1952) Determining the net long-wave radia-tion of the earth with consideration of the effect of cloudiness. IsvAkad Nauk SSSR Ser Geofis (1)

Blindheim J, Loeng H (1981) On the variability of Atlantic influencein the Norwegian and Barents Seas. Fisk dir Skr Ser Havunders17:161–189

Carton JA, Chepurin G, Cao X (2000a) A simple ocean data assimila-tion analysis of the global upper ocean 1950–95. Part II: Results30:311–326

Carton JA, Chepurin G, Cao X, Giese B (2000b) A simple ocean dataassimilation analysis of the global upper ocean 1950-95. Part I:Methodology 30:294–309

Collins WD, Bitz CM, Blackmon ML, Bonan GB, Bretherton CS,Carton JA, Chang P, Doney SC, Hack JJ, Henderson TB, KiehlJT, Large WG, McKenna DS, Santer BD, Smith RD (2006) Thecommunity climate system model version 3 (CCSM3) 19:2112–2143

Doscher R, Willen U, Jones C, Rutgersson A, Meier HEM, HanssonU, Graham LP (2002) The development of the regional coupledocean-atmosphere model RCAO. Boreal Environ Res 7:183–192.doi:10.1088/1748-9326/7/3/034005

Drinkwater K, Mueter F, Friedland K, Taylor M, Hunt GL,Hare J, Melle W (2009) Recent climate forcing and physi-cal oceanographic changes in Northern Hemisphere regions: areview and comparison of four marine ecosystems 73:190–202.doi:10.1016/j.pocean.2007.02.002

Fairall CW, Bradley EF, Hare JE, Grachev AA, Edson JB (2003) Bulkparameterization of air-sea fluxes: Updates and verification for theCOARE algorithm 16:571–591

Ferry N, Reverdin G (2004) Sea surface salinity interannual variabil-ity in the western tropical Atlantic: an ocean general circulationmodel study 109(C05026)

Gammelsrød, Leikvin Ø, Lien V, Budgell WP, Loeng H, MaslowskiW (2009) Mass and heat transports in the NE Barents Sea: Obser-vations and models 75:56–69. doi:10.1016/j.jmarsys.2008.07.010

Gill AE (1982) Atmosphere–ocean dynamics. Academic, LondonHakkinen S, Cavalieri DJ (1989) A study of the oceanic heat fluxes

in the Greenland. Norwegian and Barents Seas 95(C5):6145–6157

Harms IH, Schrum C, Hatten K (2005) Numerical sensitivity studieson the variability of climate-relevant processes in the Barents Sea110(C06002)

Helland-Hansen B, Nansen F (1909) Report on Norwegian fish andmarine investigations. In the Norwegian Sea, vol. 2

Holt J, Wakelin S, Lowe J, Tinker J (2010) The potential impacts ofclimate change on the hydrography of the northwest Europeancontinental shelf 86:361–379. doi:10.1016/j.pocean.2010.05.003

Ingvaldsen RB, Loeng H, Asplin L (2002) Variability in the Atlanticinflow to the Barents Sea based on a one-year time series frommoored current meters 22:505–519

Ingvaldsen RB, Asplin L, Loeng H (2004) Velocity field ofthe western entrance to the Barents Sea 109(C03021):1–12.doi:10.1029/2003JC001811

IPCC (2007) Climate Change 2007: The Physical Science Basis. In:Solomon S., Qin D., Manning M., Chen Z., Marquis M., AverytK. B., Tignor M., Miller H. L. (eds) Contribution of working groupI to the fourth assessment report of the intergovernmental panel onclimate change. Cambridge University Press, Cambridge, pp 747–847

Large WG, Yeager SG (2009) The global climatology of aninterannually varying airsea flux data set 33:341–364.doi:10.1007/s00382-008-0441-3

Lien VS, Gusdal Y, Albretsen J, Melsom A, Vikebø F (2013) Eval-uation of a Nordic Seas 4 km numerical ocean model hindcastarchive (SVIM), 1960-2011. Fisken og Havet 7

Loeng H (1991) Features of the physical oceanographic conditions ofthe Barents Sea. In: Sakshaug CCEHE, Britsland NA (eds) Pro-ceedings of the pro mare symposium on polar marine ecology,pp 5–18

Loeng H, Drinkwater K (2007) An overview of the ecosystems ofthe Barents and Norwegian seas and their response to climatevariability 54:2478–2500

Marchesiello P, McWilliams JC, Shchepetkin A (2001) Open bound-ary conditions for long-term integration of regional oceanic mod-els 3:1–20

Meier HEM, Andersson HC, Arheimer B, Blenckner T, ChubarenkoB, Donnelly C, Eilola K, Gustafsson BG, Hansson A, Havenhand

Page 23: Downscaling IPCC control run and future scenario with focus on the Barents Sea

Ocean Dynamics (2014) 64:927–949 949

J, Hoglund A, Kuznetsov I, MacKenzie BR, Muller-Karulis B,Neumann T, Niiranen S, Piwowarczyk J, Raudsepp U,Reckermann M, Ruoho-Airola T, Savchuk OP, Schenk F,Schimanke S, Vali G, Weslawski JM, Zorita E (2012)Comparing reconstructed past variations and future projec-tions of the Baltic Sea ecosystem-first results from multi-model ensemble simulations. Environ Res Lett (034005):8.doi:10.1088/1748-9326/7/3/034005

Melsom A, Lien VS, Budgell WP (2009) Using the RegionalOcean Modeling System (ROMS) to improve the ocean cir-culation from a GCM 20th century simulation 59:969–981.doi:10.1007/s10236-009-0222-5

Overland JE, Wang M (2007) Future regional Arctic sea ice declines34(L17705):1–7. doi:10.1029/2007GL030808

Parkinson CL, Vinnikov KY, Cavalieri DJ (2006) Evaluation of thesimulation of annual cycle of Arctic and Antarctic sea ice cover-ages by 11 major global climate models 111(C07012)

Sætra Ø, Hersbach H, Bidlot JR, Richardson DS (2004)Effects of observation errors on the statistics for ensem-ble spread and reliability. Mon Weather Rev 132:1487–1501

Sætre R (2007) The Norwegian coastal current-oceanography andclimate. Tapir Academic Press, Trondheim

Sandø AB, Nilsen JEØ, Gao Y, Lohmann K (2010) The impor-tance of heat transports and local air-sea heat fluxes for theBarents Sea climate variability. J Geoph Res 115(C07013).doi:10.1029/2009JC005884

Sandø AB, Gao Y, Langehaug HR (2014) Poleward ocean heattransports, sea ice processes, and Arctic sea ice variability

in NorESM1-M simulations. J Geoph Res 119(3):2095–2108.doi:10.1002/2013JC009435

Schlesinger ME, Ramankutty N (1994) An oscillation in theglobal climate system of period 65-70 years 367:723–726

Shchepetkin AF, McWilliams JC (2005) The regional oceanic mod-eling system (ROMS): a split-explicit, free-surface, topography-following-coordinate oceanic model 9:347–404

Shchepetkin AF, McWilliams JC (2008) Quasi-monotone advectionschemes based on explicit locally adaptive dissipation 126:1541–1580

Simonsen K, Haugan PM (1996) Heat budgets of the Arctic Mediter-ranean and sea surface flux parameterizations of the Nordic Seas101(C3):6553–6576

Skagseth Ø, Drinkwater KF, Terrile E (2011) Wind-induced trans-port of the Norwegian Coastal Current in the Barents Sea116(C08007):1–15. doi:10.1029/2011JC006996

Smagorinsky J (1963) General circulation experiments with theprimitive equations 91:99–164

Smedsrud LH, Ingvaldsen R, Nilsen JEØ, Skagseth Ø (2010) Heat inthe Barents Sea: Transport, storage, and surface fluxes 6:219–234.URL www.ocean-sci.net/6/219/2010/

Smith SDCWF, Geemaert GL, Hasse L (1996) Air-sea fluxes: 25 yearsof progress. Boundary Layer Meteor 78:247–290

Song Y, Haidvogel D (1994) A semi-implicit ocean circulation modelusing a generalized topography-following coordinate system. JComput Phys 115:228–244

Vinje T, Kvambekk AS (1991) Barents Sea drift ice characteristics10(1):59–68. doi:10.1111/j.1751-8369.1991.tb00635.x