Eutrophication in the Baltic Sea and shifts in nitrogen fixation analyzed with a 3D ecosystem model

Journal of Marine Systems 74 (2008) 592–602

Contents lists available at ScienceDirect

Journal of Marine Systems

j ourna l homepage: www.e lsev ie r.com/ locate / jmarsys

Eutrophication in the Baltic Sea and shifts in nitrogen fixation analyzed witha 3D ecosystem model

Thomas Neumann⁎, Gerald SchernewskiLeibnitz-Institute for Baltic Sea Research Warnemünde, Rostock 18119, Warnemünde Seestr. 15, Germany

a r t i c l e i n f o

⁎ Corresponding author.E-mail address: thomas.neumann@io-warnemuen

0924-7963/$ – see front matter © 2008 Elsevier B.V.doi:10.1016/j.jmarsys.2008.05.003

a b s t r a c t

Article history:Received 28 August 2007Received in revised form 6 May 2008Accepted 8 May 2008Available online 22 May 2008

Since themiddle of the last century the Baltic Sea ecosystem has undergone a strong change. Anobvious indicator is the increase of winter nutrient concentrations. This increase is attributed toincreased anthropogenic nutrient loads to the Baltic Sea.With a 3D ecosystem model we made a hindcast simulation of eutrophication from 1960 to2000. The model systemwas able to reproduce the main hydrographic and ecologic features ofthis period. However, the observed strong increase in winter nutrient concentrations wasunderestimated by the model.The simulated nitrogen fixation shows a pronounced interdecadal variability. Nitrogen fixationincreased in the early 1990 at the same time nutrient loads to the Baltic Sea were decreasing.The changes in nutrient loads cannot fully explain the increased nitrogen fixation; in fact, theprimary trigger for this increase is an intensified wind speed inwinter, which is correlated withchanges in the North Atlantic oscillation (NAO).

© 2008 Elsevier B.V. All rights reserved.

Keywords:Baltic SeaBiogeochemical cycleModelingNutrientsNitrogen fixation

1. Introduction

The Baltic Sea is one of the largest brackish water eco-systems in the world. Presently, it is changing on both short-and long-term time scales; in the short term, it faces changesdue to fluxes from the atmosphere and from the drainagearea, while in the very long-term, it faces perspectivecoastline changes due to changing sea levels. During the last50 years, the time span of regular observations, the systemobviously has changed (e.g. Larsson et al., 1985; Sanden andRahm, 1993; Omstedt et al., 2004). Previous studies havedocumented an increase in winter phosphate and nitrateconcentrations in sea surface waters (which, in turn, yields anincrease in net primary production) between 1965 and 1985(e.g. Nehring and Matthäus, 1991). Understanding the com-plex interrelations of changes is an indispensable prerequisitefor any prediction of possible future changes. Now we havethe opportunity to build model systems that can provide

de.de (T. Neumann).

All rights reserved.

predictive capabilities. Long-term observations serve as asound basis for model validation and calibration. The under-standing of single processes and even more complexsubsystems have reached a level which allows formulatingthe basic processes of the ecosystem in a quantitativemathematical form. Homogeneous atmospheric forcing dataare available from about 1960 (Uppala et al., 2005), and runoffand load data can be compiled from several sources (e.g.Schernewski and Neumann, 2005).

Hindcast simulations of the last century have beenpresented by Kauker andMeier (2003), andMeier and Kauker(2003) for the physical part of the Baltic Sea system. They showinter alia how climate variability influences the intensity ofsaltwater inflows via river runoff. Meier (2006) took thelogical next step, which was to force this model system withpredicted atmospheric forcing for the end of the 21st century.In addition to the development of predictive capability for thephysical environment there is an increasing need to extendthis effort to the whole ecosystem. Both changes of thephysical environment and changes in the catchment (and,consequently, in the nutrient loads of river runoff) will affectthe ecosystem. A first attempt was made by Schernewski andNeumann (2005) to hindcast the trophic state of the Baltic Sea

mailto:[email protected]

http://dx.doi.org/10.1016/j.jmarsys.2008.05.003

http://www.sciencedirect.com/science/journal/09247963

Fig. 1. Model topography with some geographic annotations for laterreference. BS, Bornholm Sea; EGS, Eastern Gotland Sea; WGS, WesternGotland Sea; GoF, Gulf of Finland. Oder refers to the River Oder, Gotland to theIsland Gotland, and Landsort to the location of the Landsort sea level gauge.(This figure is in color in the online version.)

593T. Neumann, G. Schernewski / Journal of Marine Systems 74 (2008) 592–602

one century ago. These simulationswere based on a repetitionof 15-year time slices; recent developments enable simula-tions using approximately 40-year time slices.

Scenario simulations for the Baltic Sea ecosystem underpredicted climatic and anthropogenic change are of increas-ing scientific interest. Among others are questions on waterquality, algae blooms and hypoxic conditions. Spatiallyintegrated models have been applied for several scenarios(Wulff et al., 2007). While these models can be operated on alow computational demand, they do not resolve the coast-to-open-sea gradients. Three-dimensional models, as describedin Neumann et al. (2002), resolve cross-shore gradientsrequire much higher computational power. Although apartial decoupling of coastal areas and the central part ofthe Baltic Sea can be expected (Voss et al., 2005), comple-mentary simulations with spatially resolving models aredesirable.

In this studywe introduce a hindcast simulation of theBalticSea from 1960 to 2000 with a 3D ecosystem model validatedagainst simulations. The hindcast simulation is a fundamentalstep toward scenario simulations. Moreover, the simulationgives some insight into this period of eutrophication in theBaltic Sea. We focus in this study on nitrogen fixation and itsdependence on eutrophication and external forcing.

2. The ecosystem model

As a starting point, we use the model system described inNeumannet al. (2002). Thismodel hasbeen successfully appliedto15-year timeslices to exploredifferentnutrient load scenarios(Neumann and Schernewski, 2005). The physical part of themodel is based on the circulationmodel MOM (Pacanowski andGriffies, 2000) and is adapted to the Baltic Sea with an explicitfree surface, an openboundary condition towards theNorth Sea,and riverine freshwater input. A thermodynamic ice modelsimulates ice cover thickness and extent.

The model grid is horizontally and vertically stretched.High horizontal resolution (3 nautical miles) was applied inthe southwestern Baltic Sea and the whole Baltic proper.Towards the north and east, the grid size gradually increasesup to 9 nautical miles. Vertically the model is resolved into 77layers. The layer thickness of the upper 100 m is 3 m andbelow this a constant thickness of 6 m is applied. Fig. 1 showsthe model topography with some geographic notations towhich we refer later in the text.

The biogeochemical model consists of nine state variables.The nutrient state variables are dissolved ammonium, nitrate,and phosphate. Primary production is provided by threefunctional phytoplankton groups: large cells, small cells, andcyanobacteria. Large cells grow rapidly in nutrient-richconditions, while small cells have an advantage at lowernutrient concentrations, especially during summer condi-tions. The cyanobacteria are able to fix and utilize atmosphericdinitrogen and therefore themodel assumes that phosphate isthe only limiting nutrient for cyanobacteria. Owing to theirability to fix nitrogen, cyanobacteria are a nitrogen source forthe system. A dynamically developing bulk zooplanktonvariable provides grazing pressure on the phytoplankton.Dead particles are accumulated in a detritus state variable. Thedetritus is mineralized into dissolved ammonium and phos-phate during the sedimentation process. A certain amount of

the detritus reaches the bottom, where it accumulates in thesedimentary detritus. Detritus in the sediment is either buriedin the sediment, mineralized, or re-suspended into the watercolumn, depending on the velocity of near-bottom currents.The development of oxygen in the model is coupled to thebiogeochemical processes via stoichiometric ratios. The oxy-gen concentration controls processes such as denitrificationand nitrification. A detailed description of the model can befound in Neumann (2000) and Neumann et al. (2002).

2.1. Modifications to the ecosystem model

Long-term simulations require a more sophisticateddescription of processes with longer time scales, particularlyprocesses connected with physical transport and sediments.In the model version of Neumann et al. (2002) we do notconsider iron–oxide-related fate of phosphate. In the revisedversion, we allow phosphate in the sediment layer to bindiron–oxide under oxic conditions. These complexes formparticles which sink out and accumulate in the sedimentlayer. Erosion events can re-suspend phosphate-rich particlesand currents transport them toward the deposition areas.Under anoxic conditions, iron–oxide becomes reduced andphosphate is liberated and available as dissolved phosphate.Depending on the sediment thickness, a part of theparticulate iron–phosphate-complexes is assumed to bediagenetically consolidated and, hence, buried in the sedi-ments. This is the final sink for phosphate in the model. Theequations for phosphate dynamics in the sediment layernow are:

dPdt jz¼zbot

¼ rfr rsed 1−f rð ÞDsed þ li IP ð1Þ

dIP ¼ rfr rsedfr Dsed−br IPDsed −li IP−res IP ð2Þ

dt Dsedmax

Here we only consider state variables phosphate P next tothe bottom, iron–phosphate-complexes IP , and detritus in the

Fig. 2. Simulated and observed dissolved inorganic phosphorus (DIP) in thecentral Eastern Gotland Sea. The solid line is a simulation with iron–phosphate and bioturbation, the dashed line is from a simulation with amodel without these extensions and the dotted line refers to observations. a)surface; b) near bottom (235 m).

Fig. 3. Change in dissolved inorganic phosphorus (DIP) pools plotted againstchanges in area of hypoxic zones (less than 2ml l−1 oxygen) in bottomwaters.Dots are from a model with explicitly simulated iron–phosphate-complexesand crosses are from a model without these complexes. The correlationcoefficient for the model with iron–phosphate-complexes is r2=0.32, whilefor the other model no correlation exists.

594 T. Neumann, G. Schernewski / Journal of Marine Systems 74 (2008) 592–602

sediment layer Dsed. rfr is the molar phosphorus–nitrogenRedfield ratio (Redfield et al., 1963) because Dsed is accountedfor in nitrogen units. rsed is the mineralization rate and fr =fr0Θ(O) is the fraction of mineralized phosphate that isconverted to iron–phosphate-complexes. Under anoxic condi-tions, phosphate is liberated at the rate li = li0Θ(–O) from IP. IP iseroded or deposited at the rate res and Θ() is the heavy-sidefunction with oxygen (O) as the argument. Eroded IP istransport as a passive tracer with a certain sinking velocity.The rate br describes the final diagenetic consolidation ofphosphate. Burial is dependenton the thicknessof the sedimentlayer Dsed.

The basic mechanism behind this modification is thatphosphate is transported toward deposition areas as particlesand becomes available under anoxic conditions. This idea isbased on the work of Laima et al. (2001).

Furthermore, we have taken into account the effect ofbioturbation. In an oxygenated environment, benthic animalspopulate the sediment. Their activity reduces the cohesiveforces in the sediment and injects sedimentary material intothe water column and, hence, contributes to re-suspension.We allow for this process by an additional term in the detritusequation:

dDdt jz¼zbot

¼ N þ bt Dsed: : : ð3Þ

where bt ¼ bt0 O2

O2þHO2is the bioturbation rate and depends on

the oxygen (O) concentration in the bottom layer (HO2 is aconstant). The importance of benthic fauna for sedimentdynamics is discussed in Meysman et al. (2006) and Graf andRosenberg (1997).

Fig. 2 illustrates the effect of the extensionson the simulatedphosphate concentration in the Eastern Gotland Sea (EGS). Theinclusion of phosphate storage in iron–phosphate-complexesand its transport considerably improves the reproduction ofphosphate dynamics in the model. The simulated data are nowmuch closer to the observations.

The phosphorus dynamics in the Baltic Sea are assumed tobe more dependent on redox conditions at the sea floor thanon phosphorus loads (Conley et al., 2002). During hypoxicsituations phosphate is liberated from the sediments and at abasin scale, changes in the phosphate pool are correlated withchanges in the area of hypoxic bottoms. To demonstrate thisrelationship we have reproduced Fig. 4a from Conley et al.(2002), butwith data fromour simulations. Againwe comparetwomodel versions: onewith the iron–phosphate-complexesextension; the other without this extension (our Fig. 3). Fordata from the newmodelwith the iron–phosphate-complexesextension (dots in Fig. 3) we found a correlation similar to thatin Conley et al. (2002). The older model version (crosses inFig. 3) does not show this dependency. This clearly shows thatan appropriate representation of phosphorus storage andrelease in models is essential to reproduce long-termdynamics of the Baltic Sea ecosystem.

2.2. Nutrient loads

In this section we briefly outline how nutrient load datahave been compiled. In the compiled data set we consideronly fractions of nitrogen and phosphorus which are availablefor primary production.


2.2.1. Riverine loadsIn an earlier approach (e.g. Neumann et al., 2002), we

attributed the total riverine loads into the Baltic Sea to the 15most important rivers. The River Oder e.g. represented notonly the real River Oder load but also the load of smaller riversalong at the German coast. This simplificationwas reasonablefor simulations with exclusive focus on the open Baltic Sea. Toreceive more realistic information about coastal waters,existing gradients and the impact of river plumes, wemodified this approach. The estimated load of the 15 majorBaltic rivers and of 5 additional rivers along the German coasthas been applied to 20 model rivers. The load of all smallerrivers, point sources, and diffusive loads has been calculatedfor 9 regions and we have assumed an evenly distributeddiffuse entry of these loads into the coastal waters of everyregion. In the model this is implemented by according anatmospheric nutrient flux to each grid point of the coast line.

The basic data set we used to estimate riverine loads isfrom the Department of Systems Ecology, Stockholm Uni-versity (2006). From this data set the period 1970 to 1990was available to us. To extend the data we used informationfrom HELCOM compilations (HELCOM, 1996; Lääne et al.,2002; HELCOM, 2003). For River Oder there was a separatedata set from hydrological modeling available (Behrendt andDannowski, 2005). All these data sources were compiled intoa homogenous data set which covers the period 1960 to2000.

Fig. 4. Compiled nutrient loading for the ecosystemmodel. The loads comprise riveriP ratio, and d) runoff.

2.2.2. Atmospheric loadsAtmospheric deposition is a major nitrogen source for the

Baltic Sea. For estimation of atmospheric loads, we again useddata sets from the Department of Systems Ecology, StockholmUniversity (2006). Additionally, we based the data compila-tion on detailed annual data in HELCOM (2005) and Lääneet al. (2002). Atmospheric phosphorus deposition plays only aminor role. Referring to HELCOM (2005) we assumed it to be0.03 % of the nitrogen load.

In Fig. 4 the compiled nutrient loading to the Baltic Sea isshown. The loads contain riverine, diffusive, and atmosphericsources. Nitrogen and phosphate display a similar tendency(Fig. 4a andb),witha continuous increase until 1988 and a rapiddecrease afterwards. After 1995 loads stabilize at a level whichis similar to loads in the1970. ThemolarN:P ratio (Fig. 4c) of theloads decreases in the 1980 and increases with the loadreduction around 1990. However, the N:P ratio of the loads ismuch higher than the classical Redfield ratio 16 (Redfield et al.,1963). The runoff to the Baltic Sea (Fig. 4d) decreased in the late1980, which also contributes to the reduced nutrient loads.

2.3. Model setup

For atmospheric forcing, we used data from the ERA40project (Uppala et al., 2005). River runoff and loads werecompiled from several sources, as outlined in the previoussections. At the open boundary to the North Sea we applied

ne, diffusive, and atmospheric sources. a) nitrogen, b) phosphorus, c) molar N:

Fig. 6. Bottom salinity anomalies at a central Bornholm Sea station. solid line:observation; dashed line: simulation.


sea levels reconstructed from sea gauge data at Smøgen(Sweden). We prescribed climatological data derived fromJanssen et al. (1999) for temperature and salinity, and frommodel simulations for the biogeochemical variables. Themodel then was spun up by several 5-year time slices for thefirst simulation years. Physical fields (temperature, salinity,sea level, currents) were restored to the initial values at thebeginning of each time slice, while the biogeochemical fieldswere taken from the end of the previous time slice. Thisprocedure allowed the biogeochemical variables, which arebased on much less observations than the physical variables,to adjust to the model dynamics.

3. Model performance

In this section we do not intend to present a comprehen-sive validation. In fact, we present some selected modelproperties and discuss some strengths and weaknesses.

3.1. Physical variables

The sea level at Landsort gauge is commonly accepted torepresent the filling level of the Baltic Sea. In Fig. 5 daily meanvalues of observed and simulated sea level at Landsort areshown. From both time series, themean value is removed. Thesimulated sea level reasonably represents the observed sealevel and, hence, we can assume that the volume fluxesbetween North Sea and Baltic Sea are reproduced by themodel. However, it is also obvious that in some cases thesimulated sea level deviates from the observations, inparticular for negative levels. Potential reasons are uncertain-ties in the meteorological forcing and in the reconstructed sealevel at the open boundary.

A characteristic of the Baltic Sea are irregular inflows ofdense, saline and, in most cases, oxygen rich water. This is theonlyway to ventilate the deepwaters below the halocline and,hence, an important process contributing to the peculiarity ofthe Baltic Sea ecosystem. Consequently, the performance of anecosystem model depends inter alia on the capability to

Fig. 5. Observed and simulated sea level at Landsort.

Fig. 7. Comparison of simulated data to observed data from the BornholmSea. The parameters are temperature (T), salinity (S), dissolved inorganicnitrogen (DIN) and dissolved inorganic phosphorus (DIP). s and b indicatesurface and bottom values, respectively. Observations are represented by 1 atthe x axis. (This figure is in color in the online version.)

reproduce the inflow dynamics. In the following subsectionwe will demonstrate the strengths of our model with respectto representing inflow events. First we present bottomsalinities from a central station in the Bornholm Sea. Salinityanomalies of the near-bottom water are shown in Fig. 6. Theseasonal cycle is removed from the data and a 3-year runningmean filter was applied in order to emphasize the biggerinflow events. Observed data are represented by a solid lineand simulated by a dashed line. Simulated variability andevents follow the observations. However, some events areover- or underestimated. The most pronounced effect arisesfrom the overestimation of an inflow event in 1970. In 1970,the simulated salinity strongly increases due to an inflowevent and as a consequence the 1976 inflowdoes not reach thedeep waters in the central Baltic Sea. Moreover, the 1976inflow is stratified in intermediate waters.

Fig. 9. Oxygen time series from the central Eastern Gotland Sea. Observationsare shown in the upper panel and simulated oxygen in the lower panel.Shaded areas indicate hydrogen sulfide counted as negative oxygenequivalents.


We have summarized the models strengths for theBornholm Sea station in a Taylor diagram shown in Fig. 7.The ratio of the standard deviations (standard deviations ofthe simulated data are divided by corresponding standarddeviation of the observations), the correlation of two data sets(observations and simulations), and their centered RMSdifference are marked by one point in a 2D diagram (Taylor,2001). To assess the correspondence between model andobservation we have interpolated the measurements onto amonthly time scale. From all data the seasonal cycle isremoved. For nutrients in the surface water we only considerwinter values.

The variability of simulated temperature and salinityappears to agree well with observations, although the bottomtemperature shows a higher variability. This is due to themodelslightly overestimating the temperature of inflowing water inthe summer compared to observations. The reason might bethat the summer sea surface temperature is overestimated by 1to 2 K. The model achieves a good performance for surfacetemperature and bottom oxygen, while the correlation of thesurface salinity isweak. ATaylor diagramfrom theEGS is shownin Fig. 8. Again a goodperformance for temperature, salinityandoxygen is achieved. However, the simulated variability isunderestimated compared to the observations.

As an example for the oxygen dynamics we show a timeseries from the EGS in Fig. 9. Since the general tendencyagrees, consequences of the missed inflow in 1976 becomeobvious. The stagnation period starts in the model already inthemid 1970. Another divergence from the observations is theweaker vertical gradient in the simulated oxygen distribution.

3.2. Biogeochemical variables

In Fig. 10 we show winter nutrient concentrations atdifferent stations in the surface water. In particular we

Fig. 8. Comparison of simulated data to observed data from the EasternGotland Sea. The parameters are temperature (T), salinity (S), dissolvedinorganic nitrogen (DIN) and dissolved inorganic phosphorus (DIP). s and bindicate surface and bottom values, respectively. Observations are repre-sented by 1 at the x axis. (This figure is in color in the online version.)

consider dissolved inorganic nitrogen (DIN: ammonium andnitrate) and phosphate. Fig. 10 gives examples for centralregions, the Bornholm Sea and the Gotland Sea. In general thesimulated nutrient concentrations follow the eutrophicationtrend with a considerable increase in the 1970. However,compared to observations, the simulated increase is lesspronounced. This is reflected in the lesser variabilityrepresented in Figs. 7 and 8, while the correlation is quitereasonable. DIN and phosphate concentrations in the Born-holm Sea and phosphate concentration in the Gotland Searoughly agree with the observations after the eutrophicationbut show too high values before. DIN in the Gotland Seaagrees during the pre-eutrophication period and shows too-low values after the eutrophication.

We have to note that the model does not start from initialconditions for the biogeochemical fields. As outlined inSection 2.3, biogeochemical variables have been allowed toadjust to external forcing and physical conditions due toseveral spin-up simulations. Consequently, shown concentra-tions at the beginning of the simulation are a quasi-equilibrium of model dynamics and forcing.

Ingeneral themodel system is able to reproduce interannualand decadal variability of the physical and biogeochemicalvariables. The latter tend to show less variability than thephysical variables compared to the observations. On a decadaltime scale, the model reproduces stagnation periods and

Fig. 10. Winter (DJF) concentrations of nutrients in surface water for different central regions. solid line: observed data; dashed line: simulated data. a) DINBornholm Sea, b) phosphate Bornholm Sea, c) DIN Eastern Gotland Sea, d) phosphate Eastern Gotland Sea.

Fig. 11. Nitrogen fixation rate averaged over the entire study period (upperpanel) andhorizontally integratednitrogenfixation (lowerpanel). (Thisfigure isin color in the online version.)


periods with frequent saltwater inflows. However, singleevents, in particular in the deep water of the central BalticSea, are under- or overestimated by the model.

The observed increase of winter nutrient concentrationsduring the eutrophication period is underestimated by themodel; nevertheless the tendency of a concentration increaseis reproduced. There are several possible explanations for thediscrepancies. We cannot be sure that all important, so-calledfirst-order processes are included in the model. Furthermore,parameterizations for the unresolved processes may not besuitable. We made series of sensitivity studies for processparameterizations, which did not significantly improve theunderestimated nutrient increase. Another problem is theuncertainty of the forcing data, especially for nutrient loads. Itconcerns riverine sources, diffusive loads, and atmosphericdeposition as well. Accessible data have their origin mostly inthe HELCOM documents (e.g. HELCOM, 2003). These data areaverages for countries over several years. Information onregional, annual or even seasonal scale are not available. Fromour point of view, the quality and availability of the data needto be improved if we are to develop well-grounded ecosystemmodels. An acceptable and independent alternative arecatchment models, which have started to become available(Mörth et al., 2007).

Fig. 12. Annual nitrogen fixation in the control experiment (solid line) andthe high load experiment, wherein, nutrient loads were held at 1987 levels(dashed line).


4. Nitrogen fixation

Nitrogen fixation is a substantial internal source fornitrogen in the Baltic Sea. Reported fixation rates range up to318 mmol m− 2 year− 1 , which corresponds to 1000 kt year− 1

(e.g. Rolff et al., 2007). This almost is in the order of riverinenitrogen loads and atmospheric deposition. In this section wewill discuss nitrogen fixation estimates from our ecosystemmodel.

The average nitrogen fixation pattern is shown in Fig. 11.We found the highest rates in the western part of the Gulf ofFinland (GoF). Another region with elevated fixation rates isaround the Island Gotland and at theMiddle Bank.Wasmundet al. (2001) reviewed literature and found fixation rateestimates from 1 to 263 mmol−2 year−1. From a carbondioxide budget approach, Schneider et al. (2003) estimated afixation rate of 318mmolm−2 year−1, which amounts to about1000 kt nitrogen per year. Referring to observation-basedestimates, our simulated rates are in a reasonable range. Thelower panel of Fig. 11 shows the time evolution of nitrogenfixation. From the beginning of the simulation until 1990,nitrogen fixation shows a downward trend from about 400 ktyear−1 to about 200 kt year−1. Between 1988 and 1992nitrogen fixation rapidly increased to about 700 kt year−1. Ahint at increased nitrogen fixation in the 1990 is given inKahru et al. (2007).

The slight decreasing trend in the 1970 and 1980 can beexplained with increasing nutrient loads. Nutrient loads tothe Baltic Sea deliver excess nitrogen (Fig. 4) to the system,which reduces nitrogen fixation. The connection betweenreduced nutrient loads and increased nitrogen fixation andvice versa has been demonstrated in Neumann et al. (2002)and Neumann and Schernewski (2005). At the same timenitrogenfixation levels rapidly increased (starting about 1990)nutrient loads decreased. To determine whether decreasednutrient loading was the mechanism behind increasednitrogen fixation, we reran the model starting in 1987 withriverine, atmospheric, and point and diffusive source nutrientloads constant at 1987 levels. This was termed the “high loadexperiment”. The annual nitrogen fixation in the high loadexperiment was approximately 200 kt less than in the controlexperiment. However, nitrogen fixation has more thandoubled between 1988 and 1992 (Fig. 12). Consequently, thenitrogen fixation increase cannot be explained by decreasingnutrient loads alone; hence, internal dynamics or meteorolo-gical forcing must have contributed to the shift in nitrogenfixation. From the simulation data we did not find any hintsof internal dynamics; neither the pool of phosphorus storedin sediments nor the pools of phosphorus and nitrogen inthe water column changed significantly. Therefore, we con-clude that it is most likely that meteorological forcing hascaused the increased nitrogen fixation. To highlight thecause–effect chain from the meteorological forcing tonitrogen fixation we applied an EOF (empirical orthogonalfunction) analysis to different data sets. All data used forthe EOF analysis have been de-trended, weighted by area,and standardized by standard deviation. EOFs and principalcomponents have been normalized that each EOF variesbetween −1 and +1.

In Fig. 13a the first EOF of nitrogen fixation is shown. Thecorresponding principal components (lower panel) has the

same time tendency as the nitrogen fixation in Fig.11. This factyields the conclusion that the EOF pattern shows theinterdecadal variability. The most pronounced variabilitycan be seen in the western part of the Baltic proper. TheGoF with the highest fixation rate (Fig. 11) obviously showsanother interdecadal variability and, hence, is controlled byother processes.

Excess phosphate is considered to play an important rolefor nitrogen fixation (e.g. Laamanen and Kuosa, 2005; Kahruet al., 2007) because the lack of oxidized nitrogen provides acompetitive advantage to nitrogen fixing cyanobacteria.Usually excess phosphate is defined as the amount ofphosphate which is in excess compared to dissolved inorganicnitrogen with respect to the Redfield ratio eP ¼ DIP− DIN

16

� �. In

the simulated data, we found (especially for the wintermonths January, February, and March) a signal in the surfaceexcess phosphate corresponding to the interdecadal nitrogenfixation variability. The first EOF mode of the winter surfaceexcess phosphate is shown in Fig. 13b. Pattern and timeevolution are similar to the patterns of nitrogen fixation.Enrichment of the surface water with nutrients (especiallyexcess phosphate) in winter is due to upwelling, convection,and vertical mixing. We diagnosed the vertical flux ofnutrients through a horizontal cross section in 15 m depth.Fig. 13c shows the first EOF pattern of the vertical excessphosphate flux in winter. Again, the principal componentschange drastically at the beginning of the 1990. A positive signin the pattern refers to an upward flux. The pattern suggeststhat the upward flux has increased in the western part anddecreased in the eastern part. However, the horizontallyintegrated flux increases as well.

Westerly and southwesterly winds favor upwelling at thewestern coast and vertical mixing. Since westerly windsprevail in winter over the Baltic Sea, we hypothesized thatthe wind speed has changed. We analyzed the wind stress,which is the momentum entering the ocean, again with anEOF analysis. The first pattern (Fig. 13d), together with theprincipal components, also indicates a drastic changebetween 1988 and 1990, that is, the wind stress increased



during this time. The North Atlantic oscillation (NAO) is to alarge extent dominating the climate variability in northernEurope (e.g. Hurrell et al., 2003). In Fig. 13d the NAO winterindex is shown as a red line. Clearly visible is the correlationbetween NAO and wind stress.

We can summarize the cause–effect chain as follows: Anintensified winter NAO enhances the transport of cyclonestowards the Baltic Sea area and, hence, wind speed intensifies.Increased wind stress promotes upwelling and deep mixing,which result in an enlarged upward transport of excessphosphate. This excess phosphate may partly remain in thesurface layer or be caught in the intermediate winter water.During summer it becomes available for nitrogen fixingcyanobacteria due to, for example, mixing or upwellingevents. This relationship was first described by Janssen et al.(2004), albeit on a shorter time scale. However, from winterNAO to nitrogen fixation in summer several processes areinvolved, which modify available excess phosphate andcyanobacteria blooms. These are inter alia the concrete windfields, vertical distribution of excess phosphate and meteor-ological conditions in summer. Consequently, high NAO indexis a precondition for a strong cyanobacteria bloom, but thestrength of the bloom may be modified by other processestoo.

5. Conclusions and summary

In this study we have presented an ecosystem model ofthe Baltic Sea that was able to reproduce the eutrophicationperiod of 1960 to 2000. An important step to reach this goalwas the inclusion of sedimentary processes in the biogeo-chemical model of Neumann et al. (2002). A weakness of themodel system is that the observed strong and rapid increaseof winter nutrient concentrations in the central basins in the1970 is underestimated by the model.

Availability of reliable data for nutrient loads is poor.Compilations from HELCOM (e.g. HELCOM, 2003) are onlyroughly resolved in time and space. A preferable option iscatchment models, which provide water and nutrient fluxesto the Baltic Sea. In this study we have used this approach forthe River Oder.

Hindcast simulations are test cases for models whichshould evolve a predictive capability. On the other hand,hindcast simulations can provide insight into the cause–effectchains of changes and variability of a systems state. This isimportant especially for marine systems which cannot besampled sufficiently.

The simulated Baltic Sea ecosystem shows a strongdependence on forcing conditions. Atmospheric forcingtogether with sea surface elevation at the open boundarycontrol, in particular, how the inflow of saline water is

Fig. 13. Most significant EOF and principal components of various oceanographic anpattern and the lower the respective principal components (bars). The black line is amode of nitrogen fixation (explained variance: 0.34). b) First EOF mode of winter emode of vertical excess phosphate flux in winter (explained variance: 0.29). Positiv(explained variance: 0.84). Red (dotted in B/W version) line shows the winter NAinterpretation of the references to color in this figure legend, the reader is referred

represented by models. Inflow events into the central basindepend on recent forcing but also on previous inflows. Anoverestimated inflow may prevent a consecutive inflow.

A prominent characteristic of the simulated ecosystem isthe increased nitrogen fixation in about 1990. It occurs at thesame time that regime shifts were detected in the Baltic andNorth Sea. Alheit et al. (2005) report about synchronousregime shifts in the North and Baltic Sea visible in almost alltrophic levels of the ecosystem and relate the shifts toincreased air and sea surface temperatures in winter as thedriving forces caused by the NOA.

However, we found for nitrogen fixation that increasedwind speed is the main driver of the shift. This highlightsthat not only temperature affects the ecosystem, in fact allexternal forcing has to be considered as potential driversfor ecosystem shifts. Here we want to note that theinterdecadal variability of nitrogen fixation in the GoF isdifferent from that in the Baltic proper, although it showsthe highest fixation rate. Winter mixing and upwelling isnot that important for cyanobacteria bloom dynamics inthis area.

Changing nutrient loads influence nitrogen fixation aswell (Neumann and Schernewski, 2005). The slow decreasefrom 1960 to 1988 may be related to increasing nutrientloads. Nevertheless, in the presented simulation changingnutrient loads can explain the nitrogen fixation increase onlypartly. Although after 1988 the nutrient loads havedecreased, they do not sufficiently explain the nitrogenfixation increase.

Altogether, interdecadal variability of nitrogen fixation canbe attributed to changed nutrient loading and changes inmeteorological forcing. Especially the wintertime wind speedhas an important influence on the nitrogen fixation. Theincrease that started in about 1990, which is consistent withother observed regime shifts, can related to the NAO and,hence, to intensified wind speed in winter. The Baltic Seaecosystem is sensitive to the processes in the catchment(nutrients, runoff) as well as to changes in climate forcing.

Acknowledgements

The work was partly supported by BMBF within theframework of Project IKZM-Oder (03F0403A) and by DFG-grant NE G17/3-1. Supercomputing power was provided byHLRN (Norddeutscher Verbund für Hoch-und Höchstleis-tungsrechnen).We thank themodeling group of the Baltic SeaResearch Institute for providing support for the circulationmodel and Horst Behrendt for providing us with nutrient loaddata for River Oder from his catchment model. We would liketo thank two anonymous referees who helped to improve themanuscript substantially.

d meteorological data sets. The upper of each pair of panels shows the EOFrunning mean. All quantities are normalized and dimensionless. a) First EOFxcess phosphate in the surface water (explained variance: 0.23). c) First EOFe sign denotes an upward flux. d) First EOF mode of the winter wind stressO index. Every higher EOF explains no more than 0.1 of the variance. (Forto the web version of this article.)


References

Alheit, J., Möllmann, C., Dutz, J., Kornilovs, G., Loewe, P., Morholz, V.,Wasmund, N., 2005. Synchronous ecologocal regime shifts in the centralBaltic and the North Sea in the late 1980s. ICES Journal of Marine Sciences62, 1205–1215.

Behrendt, H., Dannowski, H. (Eds.), 2005. Nutrients and Heavy Metals in theOdra River System. Weißenseeverlag Berlin.

Conley, D.J., Humborg, C., Rahm, L., Savchuk, O.P., Wulff, F., 2002. Hypoxia inthe Baltic Sea and basin-scale changes in phosphorus biogeochemistry.Environmental Science and Technology 36 (24), 5315–5320.

Department of Systems Ecology, Stockholm University, 2006. Baltic Environ-mental Database. URL http://data.ecology.su.se/Models/bed.htm.

Graf, G., Rosenberg, R., 1997. Bioresuspension and biodeposition: a review.Journal of Marine Systems 11, 269–278.

HELCOM, 1996. Baltic Sea Environment Proceedings No. 64 B. Baltic MarineEnvironment Protection Commission, Third Periodic Assessment of thestate of the marine Environment of the Baltic Sea 1989–1993.

HELCOM, 2003. Executive summary of the fourth Baltic Sea pollution loadcompilation, plc-4. Baltic Marine Environment Protection Commission.HELCOM, 2005. Airborne nitrogen loads to the Baltic Sea. Baltic MarineEnvironment Protection Commission.

Hurrell, J., Kushnir, Y., Visbeck, M., Ottersen, G., 2003. The North Atlanticoscillation: climate significance and environmental impact. Vol. 134 ofGeophysical Monograph Series. AGU, Ch. 1: An Overview of the NorthAtlantic Oscillation, pp. 1–35.

Janssen, F., Neumann, T., Schmidt, M., 2004. Inter-annual variability incyanobacteria blooms in the Baltic Sea controlled by wintertimehydrographic conditions. Marine Ecology Progress Series 275, 59–68.

Janssen, F., Schrum, C., Backhaus, J., 1999. A climatological dataset oftemperature and salinity for the North Sea and the Baltic Sea. GermanJournal of Hydrography 9, 1–245.

Kahru, M., Savchuk, O., Elmgren, R., 2007. Satellite measurements ofcyanobacterial bloom frequency in the Balic Sea: Interannual and spatialvariability. Marine Ecology Progress Series 343, 15–23.

Kauker, F., Meier, H.E.M., 2003. Modeling decadal variability of the Baltic Sea: 1.Reconstructing atmospheric surface data for the period 1902–1998. Journalof Geophysical Research 108 (C8), 3267.

Laamanen, M., Kuosa, H., 2005. Annual variability of biomass and heterocystof the N2-fixing cyanobacterium Aphanizomenon flos-aquae in the BalticSea with reference to Anabaena spp. and Nodularia spumigena. BorealEnvironmental Research 10 (1), 19–30.

Lääne, A., Pitkänen, H., Arheimer, B., Behrendt, H., Jarosinski, W., Lucane, S.,Pachel, K., Räike, A., Shekhovtsov, A., Swendsen, L.M., Valatka, S., 2002.Evaluation of the implementation of the 1988 ministerial declarationregarding nutrient load reductions in the Baltic Sea catchment area. Tech.Rep. 524, Finish Environment Institute.

Laima, M.J.C., Matthiesen, H., Christiansen, C., Lund-Hansen, L.C., Emeis, K.C.,2001. Dynamics of P, Fe and Mn along a depth gradient in the SW BalticSea. Boreal Environment Research 6, 317–333.

Larsson, U., Elmgren, R., Wulff, F., 1985. Eutrophication and the Baltic Sea:causes and consequences. Ambio 14, 9–14.

Meier, H.E.M., 2006. Baltic Sea climate in the late twenty-first century: adynamical downscaling approach using two global models and twoemission scenarios. Climate Dynamics 27 (1), 39–68.

Meier, H.E.M., Kauker, F., 2003. Modeling decadal variability of the Baltic Sea:2. role of freshwater inflow and large-scale atmospheric circulation forsalinity. Journal of Geophysical Research 108 (C11), 3368.

Meysman, F.J.R., Middelburg, J.J., Heip, C.H.R., 2006. Bioturbation: a fresh lookat Darwin's last idea. TRENDS in Ecology and Evolution 21 (12), 688–695.

Mörth, C.-M., Humborg, C., Eriksson, H., Danielsson, Å., Medina, M.R., Löfgren,S., Swaney, D.P., Rahm, L., 2007. Modeling riverine nutrient transport tothe Baltic Sea: a large-scale approach. Ambio 36 (2–3), 124–133.

Nehring, D., Matthäus, W., 1991. Current trends in hydrographic and chemicalparameters and eutrophication in the Baltic. Int. Rev. Gesamten Hydro-biologie 76, 276–316.

Neumann, T., 2000. Towards a 3d-ecosystem model of the Baltic Sea. Journalof Marine Systems 25 (3–4), 405–419.

Neumann, T., Fennel, W., Kremp, C., 2002. Experimental simulations with anecosystemmodel of the Baltic Sea: a nutrient load reduction experiment.Global Biogeochemical Cycles 16 (3). doi:10.1029/2001GB001450.

Neumann, T., Schernewski, G., 2005. An ecological model evaluation of twonutrient abatement strategies for the Baltic Sea. Journal of MarineSystems 56/1–2, 195–206.

Omstedt, A., Elken, J., Lehmann, A., Piechura, J., 2004. Knowledge of the BalticSea physics gained during the BALTEX and related programmes. Progressin Oceanography 63, 1–28.

Pacanowski, R.C., Griffies, S.M., 2000. MOM 3.0 manual. Tech. rep., GeophysicalFluid Dynamics Laboratory.

Redfield, A.C., Ketchum, B.H., Richards, B.H., 1963. The influence of organismson the composotion of sea water. In: Hill, L. (Ed.), The Sea, Vol. 2.Interscience, New York, pp. 26–77.

Rolff, C., Almesjö, L., Elmgren, R., 2007. Nitrogen fixation and abundance ofthe diazotrophic cyanobacterium Aphanizomenon sp. in the Baltic proper.Marine Ecology Progress Series 332, 107–118.

Sanden, P., Rahm, L.,1993.Nutrient trends in theBaltic Sea. Environmetrics 4 (1),75–103.

Schernewski, G., Neumann, T., 2005. The trophic state of the Baltic Sea acentury ago: a model simulation study. Journal of Marine Systems 53,109–124.

Schneider, B., Nausch, G., Nagel, K., Wasmund, N., 2003. The surface water co2budget for the Baltic proper: a new way to determine nitrogen fixation.Journal of Marine Systems 42, 53–64.

Taylor, K.E., 2001. Summarizing multiple aspects of models performance in asingle diagram. Journal Geophysical Residence 106 (D7), 7183–7192.

Uppala, S., Kallberg, P., Simmons, A., Andrae, U., da Costa Bechtold, V., Fiorino,M., Gibson, J., Haseler, J., Hernandez, A., Kelly, G., Li, X., Onogi, K.,Saarinen, S., Sokka, N., Allan, R., Andersson, E., Arpe, K., Balmaseda, M.,Beljaars, A., van de Berg, L., Bidlot, J., Bormann, N., Caires, S., Chevallier, F.,Dethof, A., Dragosavac, M., Fisher, M., Fuentes, M., Hagemann, S., Holm, E.,Hoskins, B., Isaksen, L., Janssen, P., Jenne, R., McNally, A., Mahfouf, J.-F.,Morcrette, J.-J., Rayner, N., Saunders, R., Simon, P., Sterl, A., Trenberth, K.,Untch, A., Vasiljevic, D., Viterbo, P., Woollen, J., 2005. The era-40 re-analysis. Quarterly Journal of the Royal of Meteorological Society 131,2961–3012.

Voss, M., Emeis, K., Hille, S., Neumann, T., Dippner, J., 2005. Nitrogen cycle ofthe Baltic Sea from an isotopic perspective. Global Biogeochemical Cycles19. doi:10.1029/2004GB002338.

Wasmund, N., Voss, M., Lochte, K., 2001. Evidence of nitrogen fixation by non-heterocystous cyanobacteria in the Baltic Sea and re-calculation of abudget of nitrogen fixation. Marine Ecological Progress Series 214, 1–14.

Wulff, F., Savchuk, O., Sokolov, A., Humborg, C., Mörth, C.M., 2007. Manage-ment options for effects on a marine ecosystem: assessing the future ofthe Baltic. Ambio 36 (2–3), 243–249.

http://data.ecology.su.se/Models/bed.htm

http://dx.doi.org/10.1029/2001GB001450

http://dx.doi.org/10.1029/2004GB002338

Documents

Eutrophication in the Baltic Sea and shifts in nitrogen fixation analyzed with a 3D ecosystem model