Sensitivity of the air-sea CO2 exchange in the Baltic Sea and Danish inner waters to atmospheric short-term variability

. Minimising the uncertainties in estimates of air– sea CO 2 exchange is an important step toward increasing the conﬁdence in assessments of the CO 2 cycle. Using an atmospheric transport model makes it possible to investigate the direct impact of atmospheric parameters on the air–sea CO 2 ﬂux along with its sensitivity to, for example, short-term temporal variability in wind speed, atmospheric mixing height and atmospheric CO 2 concentration. With this study, the importance of high spatiotemporal resolution of atmospheric parameters for the air–sea CO 2 ﬂux is assessed for six sub-basins within the Baltic Sea and Danish inner waters. A new climatology of surface water partial pressure of CO 2 ( p CO w2 ) has been developed for this coastal


Introduction
The capacity of ocean and land to take up and re-emit atmospheric CO 2 has a dominating effect on the greenhouse gas balance, and hence changes in climate.Currently, land areas and global oceans are estimated to take up about 27 and 28 %, respectively, of the CO 2 emitted by anthropogenic sources (Le Quéré et al., 2013).
In recent years, biogeochemically active coastal seas have been given increased attention (Borges et al., 2006;Chen et al., 2013;Mørk et al., 2014).Although such coastal waters only amount to 7 % of global oceans, high inputs, production, degradation and export of organic matter might result in coastal air-sea CO 2 fluxes contributing a great deal more than 7 % to the global air-sea flux (Gattuso et al., 1998).Due to the high heterogeneity of these areas, coastal CO 2 fluxes are prone to large uncertainties.Several studies agree that continental shelves, in general, act as sinks, while estuaries act as sources of CO 2 to the atmosphere.However, global estimates vary in size according to applied methodology, with oceanic uptake in shelf areas between 0.21 and 0.40 Pg C yr −1 , and release from estuaries in the range of 0.10 to 0.50 Pg C yr −1 (Cai, 2011;Chen et al., 2013;Chen and Borges, 2009;Laruelle et al., 2010).The poor coverage of observations in both space and time makes validation of these global estimates difficult.
In order to better quantify the impact of coastal regions on the global carbon budget, detailed studies of the processes at the regional scale are necessary (Kulinski and Pempkowiak, 2011).A coastal region that has been well studied is the Baltic Sea.The Baltic Sea is a high-latitude inner shelf sea connected to the North Sea through the shallow transition zone of the Danish straits, and enclosed by land with various terrestrial ecosystems and densely populated areas.Seasonal amplitudes of up to 400 µatm are observed in the partial pressure of CO 2 (pCO w 2 ) in the Baltic Sea (Thomas and Schneider, 1999) with maximum values of pCO w 2 found in winter and minimum during summer.Since the difference between the pCO 2 level in the ocean and the atmosphere controls the direction of the air-sea CO 2 flux, this is an indication of the pronounced seasonal variation of the flux in the Baltic Sea, with outgassing of CO 2 to the atmosphere during winter and uptake during summer (Thomas et al., 2004;Thomas and Schneider, 1999).Despite numerous studies, it is still uncertain whether the Baltic Sea currently acts as a net sink or source of atmospheric CO 2 , as previous studies have given ambiguous results varying from −4.3 to 2.7 g C m −2 yr −1 for the entire Baltic Sea region (Gustafsson et al., 2014;Kulinski and Pempkowiak, 2011;Norman et al., 2013).Therefore, it is also difficult to project how the Baltic Sea will contribute to the global carbon budget in the future.Moreover, the region may possibly have changed from being a net source to a net sink of atmospheric CO 2 , due to industrialisation and the enormous input of nutrients (Omstedt et al., 2009).These inputs will, however, likely change in the future due to changes in climate and anthropogenic activities (Geels et al., 2012;Langner et al., 2009).
As the Baltic Sea is bordered by land areas, the atmospheric CO 2 concentration found here will be directly affected by continental air leading to greater temporal and spatial variability in the CO 2 level than what is found over open oceans.The impact of temporal variations in atmospheric CO 2 on the air-sea CO 2 exchange has been discussed by Rutgersson et al. (2008) and(2009).They show an overestimation in the amplitude of the seasonal cycle for calculated air-sea CO 2 fluxes, when using a constant annual mean value of atmospheric CO 2 concentration instead of daily levels of atmospheric concentration.Annually, the difference was less than 10 % between the two cases, but weekly flux deviations of 20 % were obtained.This indicates how synoptic variability in the atmosphere cannot always be ignored (Rutgersson et al., 2009).Further, Rutgersson et al. (2008) note that the uncertainties connected with the transfer velocity are much greater than uncertainties related to temporal variations in atmospheric CO 2 .However, it is still worthwhile to minimise the bias in the estimation of the flux by including detailed information on atmospheric CO 2 concentrations.The short-term variability (hourly) of both meteorology and atmospheric CO 2 concentrations is not always accounted for or has not been discussed in previous estimates of the air-sea CO 2 fluxes in the Baltic Sea (Algesten et al., 2006;Gustafsson et al., 2014;Kulinski and Pempkowiak, 2011;Löffler et al., 2012;Norman et al., 2013;Wesslander et al., 2010) .
The present study aims to determine the importance of the short-term variability in atmospheric CO 2 concentrations on the net air-sea CO 2 flux of the Baltic Sea and Danish inner waters (which consists of Kattegat and the Danish straits; Øresund and the Belt Seas).A modelling approach is applied, which includes both short-term (hourly to synoptic) and long-term (seasonal to interannual) variability in the atmospheric CO 2 concentrations.The analysis is carried out by constructing a mesoscale model framework based on an atmospheric transport model covering the study region in high resolution in both space and time.The model includes a new spatial pCO w 2 climatology developed especially for the investigated marine area, as existing climatologies do not cover this area.The advantages of the present study are that the same and consistent method is applied to the entire Baltic Sea and Danish inner waters, and that the impact of spatial and temporal short-term variability in atmospheric parameters will be investigated in more detail than in the previous studies of this region.
Recently, national CO 2 budgets that include both anthropogenic and natural components have been estimated for various countries (Meesters et al., 2012;Smallman et al., 2014).The present study is likewise part of a national project, Ecosystem Surface Exchange of Greenhouse Gases in an Environment of Changing Anthropogenic and Climate forcing (ECOCLIM), which is determining the CO 2 budget for Denmark.For that reason, the present study will also estimate the marine component of the Danish CO 2 budget.
In Sect. 2 the study area, the applied surface fields of pCO w 2 and the model framework are described.Results are presented in Sect.3, leading to a discussion in Sect. 4 and with concluding remarks in Sect. 5.

Study area
The marine areas investigated in this study are shown in Fig. 1.In the following, a short introduction to these hetero- geneous marine areas is given as well as a description of the overall atmospheric CO 2 field in the region.
The Baltic Sea is a semi-enclosed continental shelf sea area with a large volume of river runoff adding a substantial amount of nutrients and terrestrial carbon to the Baltic Sea (Kulinski and Pempkowiak, 2011).The circulation in the Baltic Sea is influenced by a relatively large runoff from the surrounding drainage areas, and this causes a low-salinity outflowing surface water mass from the area.The Baltic Sea can, therefore, be considered a large estuary.Inflow of highsalinity water from the North Sea ventilates the bottom waters of the Baltic Sea, and the exchange between these water masses occurs through the shallow North Sea/Baltic Sea transition zone centred around the Danish straits (Bendtsen et al., 2009).Ice coverage is observed in the northern part of Baltic Sea during winter (Löffler et al., 2012), which has implications for the air-sea exchange of CO 2 .The ice extent in the Baltic Sea during 2005-2010 fluctuated between average conditions in the winter 2005-2006 (ice cover of 210 000 km 2 ), a general mild period in the winters between 2007-2009 (with a minimum ice cover of 49 000 km 2 in 2007-2008) and a severe winter condition in 2010-2011 where the sea ice extent reached a maximum value of 309 000 km 2 (Vainio et al., 2011).Thus, there was no apparent trend of the sea ice extent in the simulation period.
Atmospheric concentrations of CO 2 in the Baltic region have a greater seasonal amplitude than at, for example, Mauna Loa, Hawaii, which often is referred to as a global reference for the atmospheric CO 2 background, due to the remoteness of the site.The larger seasonal amplitude over the Baltic can be explained by the difference in latitude between the studied area (54-66 • N) and Mauna Loa (20 • N), and the undisturbed air at the high altitude site of Mauna Loa compared to the semi-enclosed Baltic Sea (Rutgersson et al., 2009).The study by Rutgersson et al. also showed that the atmospheric CO 2 concentration in the southern part of the Baltic Sea is more affected by regional anthropogenic and terrestrial sources and sinks than the more remote northern part of the Baltic Sea area.

Surface water pCO w 2 climatology
Model calculations of the surface air-sea gas exchange of CO 2 are parameterised in terms of the difference in partial pressure of CO 2 (i.e.pCO 2 ) between the atmosphere and the ocean surface.The global climatology of oceanic surface pCO w 2 by Takahashi et al. ( 2009) is commonly used in atmospheric transport models of CO 2 (e.g.Geels et al., 2007;Sarrat et al., 2009) and is also applied here for areas outside the Baltic Sea and Danish inner waters.However, this climatology does not cover the Baltic Sea area, and therefore, a new Baltic Sea climatology has been created and merged with the climatology of Takahashi et al. (2009) in the model domain towards the North Sea and the northern North Atlantic.
Available pCO w 2 surface measurements and water chemistry data from the Baltic Sea and Danish inner waters are combined in six sub-domains of the Baltic Sea to provide monthly averaged pCO w 2 values for this new climatology.The sub-domains cover Skagerrak, Kattegat and the Belt Sea (henceforth referred to just as Kattegat), the western Baltic Sea, the Baltic proper, the Gulf of Finland, the Bothnian Sea and the Bay of Bothnia.Two data sets are analysed: one from marine stations (stationary) and the other obtained from ships (on-board).All available data collected since the year 2000 is included in the analysis (Fig. 1).Hence, measurements from a depth of 5 m from all stations were averaged for the period 2000-2012, and on-board pCO w 2 measurements from the surface layer (surface intake approximately 5 m) were averaged for the period 2000-2011.From the two data sets monthly mean values for each sub-domain are determined.
Surface measurements of salinity, temperature, alkalinity and pH from six marine measuring stations (operated by the Swedish Meteorological and Hydrological Institute, SMHI; SHARK database, 2013) are applied to calculate the surface pCO w 2 values by a similar approach as described in Wesslander et al. (2010).The six stations are located from the central Skagerrak to the Bay of Bothnia (Fig. 1), but no measurements are available from the Gulf of Finland.A relatively high frequency of observations is obtained at the six monitoring stations with the number of observations in each month ranging between 4-8 at station A17, 15-36 at station Anholt E, 6-18 at station BY5, 7-17 at station BY15, 1-5 at station C3 (but no data representing November) and F9 (but no data representing January, February and November).Surface levels of pCO w 2 from the central Baltic Sea (Schneider and Sadkowiak, 2012) have been measured by on-board pCO w 2 systems (Körtzinger et al., 1996;Schneider et al., 2006) from cargo and research ships.In particular, a route between Germany (Kiel) and Finland (Helsinki) has regularly been monitored from cargo ships, whereas no measurements are available in the northern part of the Baltic Sea, the Danish straits, Kattegat and Skagerrak.Good data coverage of on-board pCO w 2 measurements is obtained in the sub-domain of the western Baltic Sea, with the number of observations in each month ranging between 9000 and 55 000, and in the Baltic proper, where the corresponding number of observations ranges from 20 000 to 116 000.In the Bothnian Sea the number of observations ranges from 2000 to 77 000, but there are no observations in December.Only a single month (March) is represented in the Bay of Bothnia with about 5000 observations.The Gulf of Finland is represented with observations ranging from 3000 to 18 000 each month.
The stationary data from the monitoring stations and the on-board data have been combined in such a manner that if on-board data exists for a sub-domain, these data is used for the pCO w 2 fields in the given subdomain.Otherwise, measurements from the monitoring stations are used to calculate the pCO w 2 fields.Thus, pCO w 2 fields for Skagerrak, Kattegat, and the Bay of Bothnia are calculated solely based on data from the SMHI stations.The pCO w 2 fields for the western Baltic Sea, the Baltic proper, the Gulf of Finland and the Bothnian Sea are obtained from the on-board measurements of pCO w 2 , except for December in the Bothnian Sea, which is represented by the monitoring station C3.The data used to obtain the monthly averages of surface pCO w 2 in each subdomain have all been normalised to the year 2000 using an annual increase in CO 2 of 1.9 µatm yr −1 found for the central Baltic Sea (Wesslander et al., 2010).
The resulting pCO w 2 climatology for the Baltic Sea and Danish inner waters is combined with the global open ocean pCO w 2 climatology from Takahashi et al. (2009).This climatology is calculated for a global oceanic grid with a horizontal resolution of 5 • × 4 • in longitude and latitude, respectively.Consequently, this field has an even coarser spatial resolution than the sub-domains defined in the Baltic Sea area.The global climatology is by Takahashi and colleagues, referenced to the year 2000 with an annual trend of 1.5 µatm yr −1 .This trend is also used to extrapolate the global data for the year 2000 to the proceeding years covered in this study.Note that the trend used for the Baltic Sea and Danish inner waters is 1.9 µatm yr −1 , as this trend is shown to match this particular area.However, the difference in annual trends between the two climatologies is so small compared to the absolute pCO w 2 values, and thus it is reasonable to assume that the impact on the current results will be insignificant.
The monthly averaged pCO w 2 values show a characteristic seasonal pattern at all monitoring stations and for the on-board pCO w 2 data (Fig. 2, and Table S1 and Fig. S1 in the Supplement).The surface pCO w 2 is under-saturated during spring and summer and super-saturated during autumn and winter (Fig. 3a).However, there is a large spatial gradient in the seasonal amplitude from Skagerrak to the Baltic Sea.A seasonal amplitude of about 140 µatm characterises the variation in Skagerrak and Kattegat, where the pCO w 2 varies between 275 and 420 µatm, and the surface water is only slightly super-saturated during the winter months.In the Baltic Sea, a relatively large seasonal amplitude of up to 400 µatm is observed, as primary production during the growing season, i.e. spring and summer, causes a large uptake of total dissolved inorganic carbon in the surface layer and contributes to lowering the surface pCO w 2 values.The data shows how biological uptake causes a reduction of surface pCO w 2 , despite the general warming during the summer months, which normally tends to increase the pCO w 2 in the surface water.During autumn and winter, the surface pCO w 2 values increase because sub-surface waters enriched in total dissolved inorganic carbon from remineralisation of organic matter during the summer are mixed into the surface layer.In the areas north-east of the western Baltic Sea, in particular, this allows for high monthly averaged surface pCO w 2 values of 460-530 µatm during winter with the largest average winter values observed in the Gulf of Finland.
The calculated pCO w 2 values at the monitoring stations agree well the on-board pCO w 2 data.The on-board pCO w 2 data includes both temporal and spatial variability within each sub-domain during the period since 2000.Therefore, their standard deviations (SD) are larger than the SDs from the monitoring stations, which mainly arise due to interannual variability in the period.Two sub-domains, the western Baltic Sea and the Baltic proper, have good data coverage from both the monitoring stations and on-board pCO w 2 data.The stations, BY5 and BY15, that represent the western Baltic Sea and the Baltic proper, respectively, have lower surface pCO w 2 values during the summer period than the onboard pCO w 2 data, but the difference between the two data sets are within their SD.

Model framework
The model framework is based upon the Danish Eulerian Hemispheric Model (DEHM) -a well validated three-dimensional large-scale atmospheric chemical transport model (Brandt et al., 2012;Christensen, 1997).DEHM is based on the equation of continuity and uses terrain following sigma levels as vertical coordinates.Here, 29 vertical levels are distributed between the surface and 100 hPa with a higher density of vertical levels in the lower part of the atmosphere.The main domain of DEHM covers the Northern Hemisphere with a horizontal grid resolution of 150 km × 150 km using a polar stereographic projection Biogeosciences, 12, 2753-2772,  true at 60 • N. Furthermore, DEHM has nesting capabilities allowing for a nest over Europe with a resolution of 50 km × 50 km, a nest of northern Europe with an approximate resolution of 16.7 km × 16.7 km, and a 5.6 km × 5.6 km nest covering Denmark.In order to cover the Baltic Sea and Danish marine areas in focus, a setup with two nests is applied in the current study (the European and the northern European nests).The main domain and the nests each comprise of 96 × 96 grid points.This study uses a modified version of DEHM solely simulating transport and exchange of CO 2 (Geels et al., 2002(Geels et al., , 2004(Geels et al., , 2007)), but with an updated description of the surface exchange of CO 2 (described in Sect.2.2.1).DEHM is driven by meteorological data from the meteorological model MM5v3.7 (Grell et al., 1995) using National Centers for Environmental Prediction, NCEP, data as input.

Model inputs
To accurately simulate the atmospheric content of CO 2 , a number of CO 2 sources and sinks within the model domain as well as inflow at the lateral boundaries are required together with a background concentration.The atmospheric concentration of CO 2 (X atm ) can be described by where X ff is the contribution of CO 2 from fossil fuel emissions, X fire from vegetation fires and X bio and X ocn are the contribution to the atmospheric concentration from exchange of CO 2 with the terrestrial biosphere and ocean, respectively.X background is the atmospheric background of CO 2.

Fossil fuel (X ff )
Fossil fuel emissions for the domain covering the Northern Hemisphere are implemented in DEHM from the Carbon-Tracker (hereafter referred to as CT) simulation system (Car-bonTracker CT2011_oi, 2013;Peters et al., 2007).This emission map has a 3-hourly temporal resolution on a 1 • × 1 • grid.
For the European area, the CT values are replaced by a fossil fuel emission inventory with a higher spatiotemporal resolution (hourly, 10 km × 10 km) developed by the Institute of Energy Economics and the Rational Use of Energy (Pregger et al., 2007).
For the area of Denmark, emissions with an even finer spatial resolution of 1 km × 1 km are applied obtained from the Department of Environmental Science, Aarhus University.These are based on the Danish national inventory submitted yearly to UNFCCC (United Nations Framework Convention on Climate Change) and constructed from energy statistics, point source and statistic sub-models (Plejdrup and Gyldenkaerne, 2011).
As scaled to total yearly national estimates of carbon emissions from fossil fuel consumption conducted by the Carbon Dioxide Information Analysis Center, CDIAC, in order to account for the year-to-year change in emissions (Boden et al., 2013).

Biosphere (X bio )
Terrestrial biosphere fluxes from the CT system, with a spatial resolution of 1 • × 1 • and a temporal resolution of 3 h, are applied in DEHM.In the CT assimilation system, the Carnegie-Ames-Stanford Approach (CASA) biogeochemical model is used for prior fluxes (Giglio et al., 2006;van der Werf et al., 2006).The prior terrestrial biosphere fluxes are optimised in the CT assimilation system by atmospheric observations of CO 2 .Via this atmospheric inversion a best guess of surface fluxes is obtained, and the optimised fluxes are implemented in DEHM.

Fires (X fire )
CO 2 emissions due to vegetation fires are obtained from the CT fire module and applied in DEHM.The CT fire module is based on the Global Fire Emission Database, GFEDv3.1, and CASA, while the burned area from GFED is based on MODIS satellite observations of fire counts.The resolution is likewise 3-hourly on a 1 • × 1 • grid.

Ocean (X ocn )
The CO 2 flux (F ) at the air-sea interface is calculated using the relationship: F = kα pCO 2 , where, k is the exchange coefficient, α is the gas solubility and pCO 2 is the difference in partial pressure of CO 2 between the surface water and the overlying air.The gas solubility of CO 2 is determined from Weiss (1974) and depends on the water temperature and salinity.A 0.25 • × 0.25 • salinity map is implemented in DEHM for the calculation of CO 2 solubility (Boyer et al., 2005).To calculate pCO 2 , the surface pCO w 2 fields described in Sect.2.2 are applied together with the concentration of CO 2 in the lowest atmospheric layer in DEHM.
No standardised parameterisation of the transfer velocity, k, exists, but k is most often parameterised as a power function of the wind speed (Garbe et al., 2014;Rutgersson et al., 2008) normalised to the Schmidt number (Sc) according to Wanninkhof (1992).In the present study we use the parameterisation of Wanninkhof (1992; hereafter referred to as W92).This parameterisation has been used in many previous studies within the study area (Löffler et al., 2012;Rutgersson et al., 2008;Wesslander er al., 2010), and by using W92 this allows for a direct comparison of the estimated fluxes.W92 is a function of the wind speed at 10 m above the surface (u 10 ) and when normalised to Sc at 20 However, a few additional parameterisations that could be more representative of the study area are also tested.One is from Nightingale et al. (2000), who estimate a transfer velocity based on tracer gas measurements in the North Sea of Another is by Weiss et al. (2007), who carried out measurements using eddy covariance techniques in the Arkona basin located within the Baltic Sea to estimate an accurate k for this particular area.This parameterisation takes the form The parameterisation by Weiss et al. (2007) often yields greater values than other transfer velocity parameterisations; however, it will be applied here, as the experiment was conducted within the study area.Sea ice coverage is in DEHM obtained from NCEP.The sea ice coverage is implemented in the calculations of the air-sea CO 2 exchange, such that the flux in a grid cell is reduced by the fraction of sea ice.If the fraction of sea ice coverage is 1, the entire grid cell will be covered with ice, and no exchange of CO 2 will take place between the ocean and atmosphere.Recent studies have shown that CO 2 exchange between ice-covered sea and the atmosphere does take place, but to what extent has not yet been quantified (Parmentier et al., 2013;Sørensen et al., 2014).For that reason, the exchange over sea ice is not accounted for here.
k 660 , α and pCO 2 are calculated at each time step of the model simulation (the time step of the model varies between ca. 3 and 20 min depending on, for example, the nest).Consequently, the air-sea CO 2 flux has the same temporal resolution as the simulated atmospheric CO 2 .

Atmospheric background (X background )
The level of atmospheric CO 2 has been increasing since pre-industrial times.It is not feasible to simulate this entire time period with the model system to replicate this buildup.Therefore, an atmospheric background of CO 2 is needed.The atmospheric background of CO 2 is established on the basis of the NOAA ESRL GLOBALVIEW-CO2 data product using observations from the Baltic station, BAL (55 • 35 N, 17 • 22 E; GLOBALVIEW-CO2, 2013).BAL lies within the area of interest, but far from local sources and sinks.It can, therefore, be assumed to represent the atmospheric background level in the study area.The atmospheric background of CO 2 is calculated based on the following equation: Here, X CO 2 2000 = 370.15ppm is the mean CO 2 concentration at the station in 2000, year and month is the simulated year and month, and 1.91 and 0.16 represent the yearly and monthly trend of atmospheric CO 2 .The trends are based on the times series at BAL for the period 2000-2010, in order to get a representative overall trend for the period in focus here (2005)(2006)(2007)(2008)(2009)(2010).

Boundary conditions
DEHM only covers the Northern Hemisphere; hence, boundary conditions for the main domain are needed at the lateral boundaries towards the Southern Hemisphere to account for inflow from the Southern Hemisphere.Three-dimensional atmospheric mole fractions of CO 2 from the CT system are applied at these boundaries.

Model evaluation
The period 2005-2010 is simulated by DEHM with setup and fluxes as described in Sect. 2. The performance of the model for this period is evaluated by comparing simulated atmospheric CO 2 concentrations against observed.The comparison is made at six stations within the study area where both remote continental (PAL), marine (F3, MHD, OST, WES) and anthropogenic (LUT) influenced stations are represented.
Measured and simulated atmospheric CO 2 from the marine site Östergarnsholm, Sweden (OST, 57 • 27 N, 18 • 59 E) and the anthropogenic continental site Lutjewad, the Netherlands (LUT, 53 • 40 N, 6 • 31 E; van der Laan et al., 2009) are shown for the year 2007 in Fig. 4. The Östergarnsholm marine micrometeorological field station has been running semi-continuously since 1995, measuring atmospheric CO 2 since 2005.The site has been shown to represent marine conditions and is described further in Rutgersson et al. (2008) and Högström et al. (2008).Hourly mean concentrations are plotted for simulated and measured atmospheric CO 2 , and at both sites a large diurnal variability is seen in the observations.The model is not able to capture the large amplitude in the diurnal cycle, but correlations of 0.75 and 0.71 are obtained for LUT and OST, respectively.The root mean square errors, RMSE, are 9.6 and 8.8 ppm, respectively.These high RMSEs are linked to the underestimation of the diurnal cycle in the model.Earlier model studies have shown the same tendency to underestimate the observed variability (e.g.Geels et al., 2007).The underestimation of the diurnal cycle by DEHM is most likely caused by the coarse spatial resolution of the biosphere fluxes.Further, weekly averages are made for both observed and modelled concentrations of atmospheric CO 2 (see Fig.   • 07 E; FMI, 2013) for the 6-year period (Fig. 5).In general, a reasonable correspondence between model and observations is seen during this period with correlations of 0.96, 0.98 and 0.89, and RMSEs of 1.8, 1.9 and 3.8 ppm for MHD, PAL and WES, respectively.The ability of the model to capture the seasonal cycle contributes to the very high correlation, but the model is also capable of capturing weekly variability and transport events especially during winter.
To conclude, this evaluation shows that the DEHM model captures the overall atmospheric CO 2 pattern across the marine region in focus in the current study.

Air-sea CO 2 fluxes
In order to investigate the effect of short variability in atmospheric CO 2 concentrations on the air-sea CO 2 flux, two different model simulations are conducted.One model simulation has atmospheric CO 2 concentrations that vary from time step to time step according to the fluxes and atmospheric transport in DEHM.This is in the following referred to as the VAT ("Variable ATmosphere") simulation.The other simulation contains at each time step and grid cell the monthly mean Biogeosciences, 12, 2753-2772, 2015 www.biogeosciences.net/12/2753/2015/CO 2 concentration for the given month.This is in the following referred to as CAT ("Constant ATmosphere").All other settings are identical in the two simulations.The simulations are made for the period 2005 to 2010 using the transfer velocity parameterisation by W92.First, the results of atmospheric CO 2 concentrations and air-sea CO 2 fluxes from the VAT simulation will be presented.These results can be used to get an understanding of how the atmospheric CO 2 concentrations vary, and of how the air-sea CO 2 fluxes behave in terms of size and direction in the different sub-basins of the Baltic Sea and Danish inner waters.This will be followed by the comparison of the VAT and CAT simulation.

Variable atmospheric CO 2 concentration
The variability of atmospheric CO 2 in the Baltic area is illustrated in Fig. 6, which shows a few examples of the hourly simulated surface concentration.The top panels show the variability in February 2007, where synoptic-scale variability influence transport of CO 2 , and hence the surface concentrations.On 1 February 2007 at 04:00 GMT, a low pressure system had during the past few days moved through southern Scandinavia and was then located over Poland.This system has rotated continental air with high levels of CO 2 from the east towards the Baltic Sea.On 3 February 2007, the prevailing winds were then westerly, where marine air masses with lower CO 2 concentrations were transported towards the Baltic Sea.The lower panels of Fig. 6 show the diurnal variability on 14 July 2007.At 02:00 GMT, air masses with high CO 2 concentrations were transported from land towards the marine areas -most evident in near-coastal areas.The same is the case at 14:00 GMT, but with lower concentrations due to extensive atmospheric mixing (a deep atmospheric boundary layer) and the uptake of CO 2 by the terrestrial biosphere at this time of the day.These examples show that large spatial gradients of up to 20 ppm can develop across the Baltic Sea during summer.
The seasonal averaged air-sea CO 2 fluxes estimated by DEHM in the VAT simulation are shown in Fig. 3b.In winter, a gradient is seen from the North Sea through the Danish inner straits towards the Baltic Sea, indicating a large release of CO 2 to the atmosphere in the Baltic, and uptake in the North Sea.Progressing to spring, the gradient towards the Baltic ceases and all areas now have marine uptake of atmospheric CO 2 , which continues throughout the summer.In autumn, the gradient starts to build up again, and the Baltic Sea becomes a source of CO 2 to the atmosphere.
The monthly mean 2005-2010 sub-basin averaged fluxes likewise depict this seasonality (Fig. 7).The highest seasonal amplitudes are found in the Baltic Sea area stretching from the Baltic proper and northwards with the greatest seasonal amplitude of 12 g C m −2 month −1 found in the Bothnian Sea.Less seasonal variation in the CO 2 flux is obtained for Kat-tegat and the Danish straits, which experience a yearly variability of just 4.3 g C m −2 month −1 .
The total sub-basin monthly mean fluxes of CO 2 between the atmosphere and ocean show a seasonal variation for all areas with release in winter and uptake of atmospheric CO 2 in summer (Table 1).The entire area comprising of the six sub-basins has for the period 2005-2010 an average annual net uptake of atmospheric CO 2 of 287 Gg C yr −1 .However, the net exchange varies greatly from sub-basin to sub-basin.Kattegat, the western Baltic Sea and the Baltic proper all have annual net uptake of atmospheric CO 2 averaged over 2005 to 2010, while the remaining three sub-basins release CO 2 to the atmosphere.The Baltic proper contributes the most to the total annual averaged flux with an uptake of 254 g C yr −1 , but during some individual months the fluxes in the Baltic proper are even larger (up to 900 g C month −1 ).Monthly fluxes of this considerable size are not obtained in any of the other sub-domains.This is of course related to the fact that the Baltic proper has the greatest spatial extent of all the six sub-basins.
To estimate the marine contribution in the Danish national CO 2 budget, the air-sea CO 2 flux in the Danish exclusive economic zone (EEZ) is calculated.The EEZ is a zone adjacent to the territorial waters extending up to 200 nautical miles offshore, and in the EEZ the coastal state has the right to explore, exploit and manage all resources within this area (United Nations Chapter XXI Law of the Sea, 1984).The Danish EEZ has an area of approximately 105 000 km 2 (Fig. S2).During the 6 years simulated, an average annual uptake in the Danish EEZ of 2613 Gg C yr −1 is obtained.Here, the annual average of 2616 Gg C yr −1 is reported.The interannual variability of the estimated flux will solely be a result of the interannual variations in the atmospheric CO 2 , as a climatology is used for the surface water pCO w 2 , due to the limited amount of data.The main part of the uptake in the Danish EEZ occurs in the North Sea.The North Sea has the largest extent in the Danish EEZ and combined with a small seasonal amplitude in pCO w 2 , this results in a constant uptake throughout the year.The other sub-basins within the Danish EEZ all release CO 2 in winter and take up CO 2 during summer.The marine uptake in the Danish EEZ corresponds to 18 % of the yearly Danish national emissions of anthropogenic CO 2 (Table 2).For the 6-year period investigated, the annual mean inventory in CO 2 excluding land use and land use change is 14.6 Tg C (Nielsen et al., 2013).

Constant atmospheric CO 2 concentration
The impact of variations in the atmospheric CO 2 concentration is analysed in the following by comparing the results of the air-sea CO 2 fluxes for the VAT and CAT simulations in the six sub-basins.A total annual difference of 184 Gg C yr −1 is obtained, which corresponds to a 64 % difference (calculated with VAT as the reference).CAT gives a total annual uptake of 471 Gg C yr −1 , while VAT only has an annual up-   2007, respectively.This site is chosen as it can be influenced by air masses from both land and sea depending on the wind direction.
February represents a case of outgassing, and July a case of marine uptake of atmospheric CO 2 .Time series of wind velocity at 10 m, u 10 , and the atmospheric mixing height, h mix , are also plotted to get indications of horizontal transport and vertical mixing.In addition, the differences in the atmospheric partial pressure of CO 2 ( pCO a 2 ) and in the air-sea CO 2 flux ( F CO 2 ) between the two simulations are shown (calculated as VAT -CAT).Differences in the pCO a 2 in the two simulations determine the difference in pCO 2 between the two simulations as the partial pressure of CO 2 in the water is the same in the two simulations.pCO a 2 is the only variable allowed to vary in the air-sea CO 2 flux calculations between VAT and CAT, and is thus responsible for the obtained flux difference.
For both months, pCO a 2 _VAT fluctuates around the constant pCO a 2 _CAT.During the first half of February, a period of anti-correlation between pCO a 2 _VAT and u 10 is seen.This anti-correlation is greatest during the second week with a weekly correlation coefficient (r) equal to −0.69.Thus, for this period the episodes of high wind speed tend to dilute the pCO a 2 levels allowing for a greater pCO 2 in the VAT simulation than in the CAT simulation.During the last week of February, a positive correlation of r = 0.62 between the two parameters is obtained with wind speeds above 10 m s −1 and high pCO a 2 levels in the atmosphere.This gives smaller pCO 2 in the VAT simulation than in the CAT simulation, which results in greater fluxes in the CAT simulation.In February, no clear diurnal cycle is seen in the mixing height, but the mixing height seems to follow the pattern of the wind speed with decreases in h mix during periods with low wind speeds and increases in h mix during high wind speeds.The correlation between these two parameters in February is r = 0.72.Hence, in February the pCO a 2 _VAT levels are dominated by horizontal transport.
In July, a clear diurnal variability is seen in pCO a 2 _VAT, and an anti-correlation between h mix and pCO a 2 _VAT is evident throughout the month with the highest anti-correlation during the last week (with r = −0.72).During July, the socalled diurnal rectifier effect is modelled by the VAT simulation.The rectifier effect is most apparent during the growing season and can be described as the collaboration between terrestrial ecosystems and boundary layer dynamics that act towards lowering pCO a 2 during the day and increase it during night (Denning et al., 1996).Due to the constant level of atmospheric CO 2 in the CAT simulation, the rectifier effect is absent here.This results in a greater uptake of atmospheric CO 2 in the CAT simulation than the VAT simulation during the growing season.
An anti-correlation between pCO a 2 and F CO 2 is seen in both February and July.During winter, the largest difference in the air-sea CO 2 flux between VAT and CAT coincides with high wind speeds or large differences in the atmospheric CO 2 concentrations (hence large pCO a 2 values).In summer, the diurnal cycle in the atmospheric CO 2 levels are translated into the flux difference.
Vertical profiles of atmospheric CO 2 at the site south of Sweden have been plotted together with h mix in Fig. 11.Note that the unit in Fig. 11 is parts per million and not microatmosphere.The variability of CO 2 is also evident in the vertical profile, where air masses with low or high CO 2 concentrations are being transported to and from the site (55   In winter, the fluxes in both VAT and CAT are positive, but larger in VAT than CAT, and thus the difference is positive.In summer, both the fluxes in VAT and CAT are negative, but CAT is numerical larger than VAT, and thus the difference is also positive. 13 • 55 E).Continental air is represented by high levels of CO 2 that extend up to 2 km into the atmosphere, while marine air masses have lower levels of CO 2 corresponding to the levels above 2 km.The shift between the two types of air masses is clearly seen in the vertical profile; e.g on 2 February.Here, higher wind speed leads to transport of marine air masses to the site (see Fig. 9).Like Fig. 9, the vertical profile in February shows no clear connection between surface concentrations of CO 2 and h mix .In July, the vertical profile depicts the rectifier effect.Low surface values of CO 2 coincide with the greatest boundary layer heights found during the daytime, and high surface levels of CO A representative map of surface pCO w 2 has been created for Skagerrak and six sub-domains in the Baltic using two data sets: one obtained from monitoring stations and one using on-board measurements of surface pCO w 2 (see Sect. 2.2).Previous estimates of pCO w 2 at two positions within the Baltic Sea have shown interannual variability of up to 25 % in winter and almost 140 % in summer (Wesslander et al., 2010).Likewise, large short-term variability has been measured in different coastal systems (Dai et al., 2009;Leinweber et al., 2009;Wesslander et al., 2011).
The representation of surface pCO w 2 values in the subdomains by a monthly averaged value does not account for the temporal variability during each month and the spatial variability in the relatively large areas.The estimated surface fields of pCO w 2 are based on all available data; however, the amount of available observations can be considered to be relatively small compared to the large study area, although on-board pCO w 2 measurements (Schneider and Sadkowiak, 2012) have increased the data coverage in the central Baltic Sea significantly in the past few years.
The choice of applying a surface map of pCO w 2 for six domains in the Baltic of course introduces some biases on the flux estimates as mechanisms, such as upwelling and algae blooms that act on a smaller spatial scale than the subdivision, are not specifically accounted for.It was essential for the present study to obtain a surface map of pCO w 2 that covered the entire region to be able to study the effect of short-term variability in atmospheric CO 2 on the air-sea CO 2 flux within the Baltic Sea region.Despite the possible biases of ignoring short-term and small-scale variability in ocean pCO w 2 , the simplified description of the conditions in the Baltic Sea in a number of sub-domains was evaluated to be the best solution to obtain a surface field of pCO w 2 that spatially covers the whole model domain for the present study.

Air-sea CO 2 fluxes
The atmospheric CO 2 concentration is seen to vary greatly within the study area (Figs. 6 and 11).The dynamics of the fluxes and the atmospheric transport and mixing lead to short-term variations and spatial gradients in the atmospheric CO 2 level across the study area.Pressure systems move through the region transporting air masses with different characteristics and CO 2 levels to and from the Baltic and the Danish inner waters.In the growing season, the effect from the terrestrial biosphere is apparent, with a clear diurnal cycle in the atmospheric CO 2 caused by respiration during night-time and photosynthesis during the day, complemented by boundary layer dynamics.Even these shortterm variations in atmospheric CO 2 concentrations over land can be transported to marine areas, indicating why it is important to include atmospheric short-term variability in the air-sea flux estimations.
For the 6-year period, an annual average uptake of 287 Gg C yr −1 is obtained with the VAT setup as a total for the six sub-basins.A statistical analysis of the simulated fluxes shows that Kattegat and the western Baltic Sea are annual sinks (at a significance level of 0.05), while the Gulf of Finland and the Bay of Bothnia are annual sources of atmospheric CO 2 .In the transition zone between these areas, i.e. the Baltic proper and the Bothnian Sea, large variations in the annual flux are seen in this study.During the 6 years simulated, these sub-domains change annually between being sources and sinks of CO 2 to the atmosphere.This also affects the total flux for the entire investigated area, which also shifts between being an annual source (376 Gg C yr −1 ) and sink (−1100 Gg C yr −1 ).A significant test (Student's t test with a significance level of 0.05) show that the variability from year to year during the 6 years simulated is so large that we cannot conclude that the area is a net sink, despite the estimated averaged uptake of 287 Gg C yr −1 .
The air-sea CO 2 fluxes obtained from the VAT simulation for six sub-basins are compared to previous results from the area to assess consistency.Previous studies of the airsea CO 2 flux in the Baltic Sea area are ambiguous on the Baltic Sea's role in the carbon cycle (see Table 3).This is partly caused by the various techniques used, ranging from in situ measurements using the eddy covariance method to model simulations (Kulinski and Pempkowiak, 2011;Rutgersson et al., 2009;Weiss et al., 2007;Wesslander et al., 2010), and partly by the different spatial areas investigated.Some of the previous studies are site specific (Algesten et al., 2006;Kuss et al., 2006;Löffler et al., 2012;Rutgersson et al., 2008;Wesslander et al., 2010) and other studies cover the entire area (Gustafsson et al., 2014;Kulinski and Pempkowiak, 2011;Norman et al., 2013).None of the previous regional studies have based their estimates of the air-sea CO 2 flux on results from an atmospheric transport model capable of combining large spatial coverage with high spatiotemporal resolution of the entire Baltic region as in the present study.Results from previous studies and the present study have been converted to the same unit of g C m −2 yr −1 to allow for a direct comparison (Table 3).
Table 3 reveals that in terms of the direction of the flux, the present study corresponds well with some of the previous studies and contradicts others.As the results obtained from the VAT simulation lie within the range of previous estimates, it seems reasonable to use the current model setup for sensitivity analysis of the air-sea CO 2 flux in the region.Additionally, it can be concluded that the obtained results from the VAT simulation together with recent studies converge towards the Baltic Sea and Danish inner waters being annual sinks of atmospheric CO 2 .

Impact of atmospheric short-term variability
The difference of 184 Gg C yr −1 between the annual air-sea flux in the CAT and VAT simulations was tested to be significantly different from zero at a 0.05 significance level.Therefore, it can be concluded that using a constant level of atmospheric CO 2 has a significant impact on the estimated annual air-sea CO 2 flux in this region.The greatest differences are found in winter and autumn in the Baltic Sea area (Fig. 8).But large differences are also found over open water areas in spite of a less variable atmospheric CO 2 concentration here, i.e. a smaller difference in the atmospheric CO 2 concentration between the two simulations.Despite the small concentration difference, the tendency towards higher wind speeds over open oceans leads to the large flux difference here.The same wind fields are applied in both simulations.

Biogeosciences
The deviation between the two simulations in the study region is mainly caused by a reduction in the winter uptake in the CAT simulation.The winter outgassing is reduced in CAT, when the pCO a 2 of the CAT is greater than the pCO a 2 of the VAT simulation.Thereby, pCO 2 is smaller in the CAT simulation than the VAT simulation, and the flux will be reduced.Furthermore, the nonlinearity of the wind speed in the parameterisation of the transfer velocity can amplify this reduction, in particular, when high wind speeds coincide with greater pCO 2 in the VAT simulation than in CAT simulation (e.g. as seen in Fig. 9 for the first week of February 2007).This mechanism must have a significant influence, as it results in a greater winter uptake in the VAT simulation then in the CAT simulation.The higher marine CO 2 uptake in summer by the CAT simulation is a result of diurnal boundary layer dynamics and the diurnal cycle or lack of it in the atmospheric CO 2 concentrations.The rectifier effect is not accounted for in the CAT simulation, and the constant pCO a 2 in CAT is higher during the day and lower during the night than in the VAT simulation.This allows for a greater air-sea pCO 2 in the CAT simulation during day, which together with a tendency of higher wind speeds during daytime increases the oceanic uptake in CAT.This is illustrated by F CO 2 , where positive values indicate how the flux is numerical larger in CAT than VAT (see Fig. 10).As described in Sect.3.1, the diurnal cycle of atmospheric CO 2 is generally underestimated by the DEHM model in near-coastal areas.This could indicate that the difference between the VAT and CAT simulations found during the growing season is a conservative estimate for the fluxes at the near-coastal areas in the Baltic Sea region.
While Rutgersson et al. (2008) found a slightly overestimated seasonal amplitude, when using a constant atmospheric CO 2 concentration, the present study finds that the seasonal cycle of the CAT simulation is displaced downwards as compared to the VAT simulation.This displacement results in a greater annual uptake in the CAT simulation.

Uncertainties
The estimated air-sea CO 2 flux is controlled by several parameters in the applied model setup: choice of transfer velocity parameterisation, wind speed, temperature, salinity, atmospheric CO 2 concentration and marine pCO w 2 surface values.Each of these is connected with some uncertainty and errors.Takahashi et al. (2009) estimate the combined precision on the global air-sea flux to be on the order of ±60 % when including a possible climatology bias due to interpolation and under-sampling.The uncertainty might be higher in the current study as the climatology for the pCO w 2 in surface waters used here covers areas where the spatiotemporal variability in the measured pCO w 2 is higher than in open waters.The natural variability within the sub-domains is represented by the standard deviations in Fig. 2, and it reflects both the spatial and temporal variation in the domains during the period of sampling, i.e. the last decade.The Baltic Sea domains (i.e.excepting the Kattegat sub-domain) are all characterised by a significant under-saturation of the surface water during spring and summer.During winter, these stations are in general supersaturated with respect to the atmospheric pCO a 2 .Thus, the sign of the CO 2 flux during the seasons is assumed to be well-determined in the Baltic Sea sub-domains due to the large seasonal amplitudes.However, during the seasonal change between summer and winter, where typical standard deviations in the climatology of 50 ppm are seen, we estimate that the uncertainty due to the ocean surface pCO w 2 values is on the order of 50 % in the Baltic Sea.The uncertainty in the Kattegat sub-domain is estimated to be up to 50-100 % because of the relatively small seasonal amplitude.
Atmospheric CO 2 , wind speed and temperature all vary in each model time step and grid cell.The uncertainties of wind speed and temperature are small compared to the uncertainties of the pCO w 2 fields.Short-term variability does not only exist in the atmospheric concentration of CO 2 , it has also been detected in the pCO w 2 of surface water (Dai et al., 2009;Leinweber et al., 2009;Rutgersson et al., 2008;Wesslander et al., 2011).The magnitude of the short-term variability is site dependent with the smallest variability found in open oceans (Dai et al., 2009) and greatest at near-coastal sites (Leinweber et al., 2009;Wesslander et al., 2011).Off the Californian coast, Leinweber et al. (2009) found a diurnal cycle of pCO w 2 with an average amplitude of 20 µatm -a diurnal amplitude double of what they found in the atmosphere.Short-term variability of marine pCO w 2 , could potentially alter the annual estimate of the coastal air-sea CO 2 flux.Thus, in the present study the fluxes at the near-coastal areas within the sub-domain could be affected by this short-term variability, and as a result possibly modify the total flux for these sub-domains.However, the short-term variability in marine pCO w 2 is not included in this study, and it is, therefore, difficult to estimate how this might affect the estimated flux.Additionally, the short-term variability in the air and water might be correlated, thus it is not possible to make a deduction of the combined effect in the present model study.
To assess the uncertainty connected to the choice of transfer velocity on the estimated air-sea flux model, simulations using parameterisations of Nightingale et al. (2000) and Weiss et al. (2007) have also been conducted.Throughout the seasons, the parameterisation by Weiss et al. (2007) gives more extreme values than that of Nightingale et al. (2000), but the annual sum for the study area results in −667 and −858 Gg C yr −1 for Nightingale et al. (2000) and Weiss et al. (2007), respectively.Other transfer velocity parameterisations could also have been interesting to use in the present study.An example is the parameterisation by Sweeney et al. (2007), which is based on an updated and improved version of the radiocarbon method used in W92.Here, the two different parameterisations by Weiss et al. (2007) and Nightingale et al. (2000) were chosen, as these experiments were conducted within and close to the study area, respectively.
The present study supports the findings briefly touched upon by Rutgersson et al. (2009), who concluded that the uncertainty due to the value of atmospheric CO 2 is small compared to uncertainty in transfer velocity.Introducing a surface pCO w 2 climatology in six sub-basins adds substantially to the uncertainty, as short-term variability in both space and time is ignored in this parameter.However, we have chosen to use this surface pCO w 2 climatology to get full spatial and temporal coverage of surface pCO

Conclusions
Using an atmospheric CO 2 model with a relative high spatial and temporal resolution, we have estimated the air-sea flux of CO 2 in the Danish inner waters and the Baltic Sea region.More specifically we have made a detailed analysis of the sensitivity to temporal variability in atmospheric CO 2 and the related impact of driving parameters like wind speed and atmospheric mixing height.
In the process of this study new monthly marine pCO w 2 fields have been developed for the region combining existing data from monitoring stations and measurements from ships.Due to the sparseness of these data, only seasonal variations are included in the pCO w 2 fields.The atmospheric concentration of CO 2 is often assumed to be constant or only vary by season in many marine model studies, but according to this novel sensitivity analysis, neglecting, for example, the diurnal and synoptic variability in atmospheric CO 2 concentrations could lead to a systematic bias in the annual net air-sea flux.Previous studies have looked at the entire Baltic region (Gustafsson et al., 2014;Kulinski and Pempkowiak, 2011;Norman et al., 2013;Thomas and Schneider, 1999), but not with the same approach as in the present study.
In all the included sub-basins, a seasonal cycle was detected in the air-sea CO 2 flux with release of CO 2 during winter and autumn, and uptake of atmospheric CO 2 in the remaining months.An annual flux for the study area of −287 Gg C yr −1 (−0.7 g C m −2 yr −1 ) was obtained for the 6 years simulated.This agrees with the previous findings of Norman et al. (2013) and Gustafsson et al. (2014), who estimated annual air-sea CO 2 fluxes of −2.6 and −4.3 g C m −2 yr −1 , respectively.
The importance of short-term variations in the atmospheric CO 2 in relation to the yearly air-sea flux was tested with two different model simulations.One simulation includes the short-term variations (the VAT simulation), while the other simulation includes a monthly constant atmospheric CO 2 concentration (the CAT simulation).A significant difference of 184 Gg C yr −1 (corresponding to 67 %) was obtained for the air-sea CO 2 flux for the Baltic Sea and Danish inner waters between the two model simulations.The seasonal amplitude of the air-sea CO 2 flux was shifted downwards in the CAT simulation as compared to the VAT simulation, resulting in a reduced winter release of CO 2 in the CAT simulation and an increased summer uptake.The difference occurs solely due to the difference in the atmospheric CO 2 concentrations.
As a part of the Danish project ECOCLIM with a focus on the Danish CO 2 budget, the natural marine annual flux of CO 2 was estimated for the first time in the present study.
The Danish waters -in this context defined as the Danish exclusive economic zone -is according to our simulations taking up 2613 Gg C yr −1 with the majority taken up in the North Sea.This is comparable to approximately 18 % of the Danish anthropogenic CO 2 emissions.
Uncertainties are bound to the results, particularly in connection with transfer velocity parameterisation and the applied surface pCO w 2 climatology.However, in the present study, with two model simulations that only differ in atmospheric CO 2 concentrations, a distinguishable difference in the air-sea CO 2 flux is obtained.This, therefore, stresses the importance of including short-term variability in the atmospheric CO 2 in order to minimise the uncertainties in the airsea CO 2 flux.Moreover, this deduce that also short-term variability in pCO w 2 of the water, in particular of coastal areas, needs to be included, as short-term variability in near-coastal surface water pCO w 2 potentially is greater than in the atmosphere.
To conclude, we recommend that future studies of the airsea CO 2 exchange include short-term variability of CO 2 in the atmosphere.Thereby, the uncertainty related to estimating the marine part of the carbon budget at regional to global scales can be reduced.
The Supplement related to this article is available online at doi:10.5194/bg-12-2753-2015-supplement.

Figure 1 .
Figure 1.The locations of the six monitoring stations in the Baltic Sea, where surface pCO w 2 values are calculated (SHARK database,2013).The stations are located in Skagerrak (A17), Kattegat and the Danish straits (Anholt E), the western Baltic Sea (BY5), Baltic proper (BY15), the Bothnian Sea (C3) and the Bay of Bothnia (F9).Data from on-board measurements of surface pCO w 2 are shown in yellow and cover, in particular, the area between Kiel and Helsinki.The division of the six sub-domains is indicated with black lines.

Figure 2 . 2 from
Figure 2. Monthly averaged surface pCO w 2 values from the six monitoring stations and from on-board pCO w 2 data in the subdomains in the study region.Monthly averaged values are shown with bullets together with the standard deviations.(a) Values from monitoring stations in Skagerrak (A17, blue) and Kattegat (Anholt E, green), (b) station BY5 (blue) and on-board pCO w 2 in the western Baltic Sea (black), (c) station BY15 (blue) and on-board pCO w 2 from the Baltic proper (black), (d) on-board pCO w 2 data from the Gulf of Finland and (e) station F9 (blue), C3 (green) and on-board pCO w 2 data from the Bothnian Sea (black) and on-board pCO w 2

Figure 3 .
Figure 3. (a) pCO 2 for selected months during 2005.For the calculations of pCO 2 , the combined surface map of the global pCO w 2 climatology by Takahashi et al. (2009) and the climatology for the Baltic Sea constructed in this study are used.The coarse resolution of the global climatology is clearly visible along the west coast of Norway.Periods of under and over-saturation are seen which indicate the direction of the air-sea CO 2 flux (positive pCO 2 indicates release of CO 2 to atmosphere, negative values indicate uptake).(b) The mean seasonal air-sea CO 2 flux for the years 2005 to 2010 in g C m −2 month −1 for the VAT simulation.Positive sign indicates release of CO 2 from the ocean to the atmosphere, negative sign indicates uptake of atmospheric CO 2 by the ocean.This sign convention is used throughout the paper.
4).Improvements are obtained in www

Figure 4 .
Figure 4. One-hour averages of modelled and continuously measured atmospheric CO 2 concentrations in 2007.Also, weekly averages of both modelled and measured CO 2 concentrations are shown.

Figure 5 .
Figure 5.The top panel shows hourly averages of modelled atmospheric CO 2 concentrations compared to flask measurements at F3.The three panels below include comparisons of weekly averages of modelled and continuous measurements of CO 2 at MHD, PAL and WES for the period 2005-2010.

Figure 6 .
Figure 6.Examples of the simulated variability of atmospheric CO 2 in the study area shown here as extracted from the European domain in DEHM with a 50 km × 50 km resolution.Top panels: two sets of conditions for February 2007.Continental air masses cover the Baltic region on 1 February, while marine air masses are dominating on 3 February.Bottom panels: the diurnal variability on 14 July 2007 (night-time on the left and daytime, right).

Figure 7 .
Figure 7.The monthly mean air-sea CO 2 flux for the years 2005 to 2010 in the six sub-basins in g C m −2 month −1 for the VAT simulation.

Figure 8 .
Figure 8.The seasonal flux difference between the VAT and CAT simulations for the period 2005 to 2010 in g C m −2 month −1 calculated as VAT−CAT.In winter, the fluxes in both VAT and CAT are positive, but larger in VAT than CAT, and thus the difference is positive.In summer, both the fluxes in VAT and CAT are negative, but CAT is numerical larger than VAT, and thus the difference is also positive.
2 concur during nighttime with the nocturnal boundary layer.It is remarkable how the vertical profile during July 2007 represents a much more mixed atmosphere as compared to February 2007, where the marine and continental air masses clearly are distinguished from each other.

Figure 9 .
Figure 9.Time series of driving parameters as extracted from the simulations at the site south of Sweden (55 • 18 N, 13 • 55 E) in February 2007.Top panel: pCO a 2 for VAT and CAT together with u 10 .Middle panel: pCO a 2 for VAT and CAT together with h mix .Bottom panel: difference in pCO a 2 ( pCO a 2 ) and F CO 2 ( F CO 2 ) between VAT and CAT.

Figure 10 .
Figure 10.Simulated parameters as in Fig. 9 at the site south of Sweden (55 • 18 N, 13 • 55 E), but for July, 2007.Top panel: pCO a 2 for VAT and CAT together with u 10 .Middle panel: pCO a 2 for VAT and CAT together with h mix .Bottom panel: difference in pCO a 2 ( pCO a 2 ) and F CO 2 ( F CO 2 ) between VAT and CAT.

Figure 11 .
Figure 11.Simulated vertical profiles of atmospheric CO 2 at the site south of Sweden (55 • 18 N, 13 • 55 E) in units of ppm.Top panel: 1-10 February 2007.Bottom panel: 11-20 July 2007.The black line represents the mixing height in kilometres.Note the different scales used in the two plots.
Figures 4 and 5  show how well the DEHM model captures the weekly and seasonal variability in the atmospheric CO 2 concentrations.However, some prob-capturing the variability on shorter timescales (e.g.diurnal).The diurnal cycle is under-estimated in this model setup, which is related to the coarse resolution of the biosphere fluxes, and of the model itself.

w 2 .
This allows us to inves-of short-term variability in atmospheric CO 2 concentration on the air-sea CO 2 flux.

Table 1 .
Monthly mean fluxes for the period 2005-2010 in the VAT simulation depicting seasonal variation of the air-sea CO 2 exchange.Values are given in gigagrams of carbon per sub-basin.Positive sign indicates release of CO 2 from the ocean to the atmosphere, negative sign indicates uptake of atmospheric CO 2 by the ocean.This sign convention is used throughout the paper.

Table 2 .
Annual Danish CO 2 emissions as reported to UNFCCC.The middle row contains the annual uptake of CO 2 in the marine area defined as the Danish exclusive economic zone as estimated in this study.The bottom rows give this uptake as a percentage of the Danish anthropogenic CO 2 emissions.