• Collins, M., , S. F. B. Tett, , and C. Cooper, 2001: The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 17, 6181, doi:10.1007/s003820000094.

    • Search Google Scholar
    • Export Citation
  • Feistel, R., , G. Nausch, , and N. Wasmund, 2008: State and Evolution of the Baltic Sea, 1952–2005: A Detailed 50-Year Survey of Meteorology and Climate, Physics, Chemistry, Biology, and Marine Environment. J. Wiley and Sons, 712 pp.

  • Fonselius, S., , and J. Valderrama, 2003: One hundred years of hydrographic measurements in the Baltic Sea. J. Sea Res., 49, 229241, doi:10.1016/S1385-1101(03)00035-2.

    • Search Google Scholar
    • Export Citation
  • Gräwe, U., , R. Friedland, , and H. Burchard, 2013: The future of the western Baltic Sea: Two possible scenarios. Ocean Dyn., 63, 901921, doi:10.1007/s10236-013-0634-0.

    • Search Google Scholar
    • Export Citation
  • Gräwe, U., , M. Naumann, , V. Mohrholz, , and H. Burchard, 2015: Anatomizing one of the largest saltwater inflows into the Baltic Sea in December 2014. J. Geophys. Res. Oceans, 120, 76767697, doi:10.1002/2015JC011269.

    • Search Google Scholar
    • Export Citation
  • Grinsted, A., , J. C. Moore, , and S. Jevrejeva, 2004: Application of the cross wavelet transform and wavelet coherence to geophysical time series. Nonlinear Processes Geophys., 11, 561566, doi:10.5194/npg-11-561-2004.

    • Search Google Scholar
    • Export Citation
  • Gustafsson, B., 1997: Interaction between Baltic Sea and North Sea. Dtsch. Hydrogr. Z., 49, 165183, doi:10.1007/BF02764031.

  • Gustafsson, B., 2000: Time-dependent modeling of the Baltic entrance area. 2. Water and salt exchange of the Baltic Sea. Estuaries, 23, 253266, doi:10.2307/1352831.

    • Search Google Scholar
    • Export Citation
  • Hagen, E., , and R. Feistel, 2005: Climatic turning points and regime shifts in the Baltic Sea region: The Baltic winter index (WIBIX) 1659–2002. Boreal Environ. Res., 10, 211224.

    • Search Google Scholar
    • Export Citation
  • Hansson, D., , and E. Gustafsson, 2011: Salinity and hypoxia in the Baltic Sea since A.D. 1500. J. Geophys. Res., 116, C03027, doi:10.1029/2010JC006676.

    • Search Google Scholar
    • Export Citation
  • Hünicke, B., , E. Zorita, , and S. Haeseler, 2010: Baltic Holocene climate and regional sea-level change: A statistical analysis of observations, reconstructions and simulations within present and past analogues for future changes. Final Rep. of the DFG Research Unit SINCOS-2, 70 pp. [Available online at http://www.hzg.de/imperia/md/content/hzg/zentrale_einrichtungen/bibliothek/berichte/gkss_berichte_2010/gkss_2010_2.pdf.]

  • Kaspar, F., , T. Spangehl, , and U. Cubasch, 2007: Northern Hemisphere winter storm tracks of the Eemian interglacial and the last glacial inception. Climate Past, 3, 181192, doi:10.5194/cp-3-181-2007.

    • Search Google Scholar
    • Export Citation
  • Killworth, P. D., , D. Stainforth, , D. J. Webb, , and S. M. Paterson, 1991: The development of a free-surface Bryan–Cox–Semtner ocean model. J. Phys. Oceanogr., 21, 13331348, doi:10.1175/1520-0485(1991)021<1333:TDOAFS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kjellström, E., , J. Brandefelt, , J.-O. Näslund, , B. Smith, , G. Strandberg, , A. H. L. Voelker, , and B. Wohlfarth, 2010: Simulated climate conditions in Europe during the Marine Isotope Stage 3 stadial. Boreas, 39, 436456, doi:10.1111/j.1502-3885.2010.00143.x.

    • Search Google Scholar
    • Export Citation
  • Kjellström, E., , G. Nikulin, , U. Hansson, , G. Strandberg, , and A. Ullerstig, 2011: 21st century changes in the European climate: Uncertainties derived from an ensemble of regional climate model simulations. Tellus, 63A, 2440, doi:10.1111/j.1600-0870.2010.00475.x.

    • Search Google Scholar
    • Export Citation
  • Lass, H. U., , and W. Matthäus, 1996: On temporal wind variations forcing salt water inflows into the Baltic Sea. Tellus, 48A, 663671, doi:10.1034/j.1600-0870.1996.t01-4-00005.x.

    • Search Google Scholar
    • Export Citation
  • Lean, J., , J. Beer, , and R. Bradley, 1995: Reconstruction of solar irradiance since 1610: Implications for climate change. Geophys. Res. Lett., 22, 31953198, doi:10.1029/95GL03093.

    • Search Google Scholar
    • Export Citation
  • Legutke, S., , and R. Voss, 1999: The Hamburg Atmosphere–Ocean Coupled Circulation Model ECHO-G. DKRZ Tech. Rep. 18, 62 pp. [Available online at http://mms.dkrz.de/pdf/klimadaten/models/ReportNo.18.pdf.]

  • Matthäus, W., , and H. Franck, 1992: Characteristics of major Baltic inflows—A statistical analysis. Cont. Shelf Res., 12, 13751400, doi:10.1016/0278-4343(92)90060-W.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2005: Modeling the age of Baltic seawater masses: Quantification and steady state sensitivity experiments. J. Geophys. Res., 110, C02006, doi:10.1029/2004JC002607.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2006: Baltic Sea climate in the late twenty-first century: A dynamical downscaling approach using two global models and two emission scenarios. Climate Dyn., 27, 3968, doi:10.1007/s00382-006-0124-x.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2007: Modeling the pathways and ages of inflowing salt- and freshwater in the Baltic Sea. Estuarine Coastal Shelf Sci., 74, 717734.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , and F. Kauker, 2003a: Modeling decadal variability of the Baltic Sea: 2. Role of freshwater inflow and large-scale atmospheric circulation for salinity. J. Geophys. Res., 108, 3368, doi:10.1029/2003JC001799.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , and F. Kauker, 2003b: Sensitivity of the Baltic Sea salinity to the freshwater supply. Climate Res., 24, 231242, doi:10.3354/cr024231.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , R. Döscher, , and T. Faxen, 2003: A multiprocessor coupled ice–ocean model for the Baltic Sea: Application to salt inflow. J. Geophys. Res., 108, 3273, doi:10.1029/2000JC000521.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , A. Höglund, , R. Döscher, , H. Andersson, , U. Löptien, , and E. Kjellström, 2011: Quality assessment of atmospheric surface fields over the Baltic Sea from an ensemble of regional climate model simulations with respect to ocean dynamics. Oceanologia, 53, 193227, doi:10.5697/oc.53-1-TI.193.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , R. Hordoir, , H. Andersson, , C. Dieterich, , K. Eilola, , B. Gustafsson, , A. Höglund, , and S. Schimanke, 2012a: Modeling the combined impact of changing climate and changing socio-economic development on the Baltic Sea environment in an ensemble of transient simulations for 1961–2099. Climate Dyn., 39, 24212441, doi:10.1007/s00382-012-1339-7.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., and et al. , 2012b: Impact of climate change on ecological quality indicators and biogeochemical fluxes in the Baltic Sea? A multi-model ensemble study. AMBIO, 41, 558573, doi:10.1007/s13280-012-0320-3.

    • Search Google Scholar
    • Export Citation
  • Min, S.-K., , S. Legutke, , A. Hense, , and W.-T. Kwon, 2005a: Internal variability in a 1000-yr control simulation with the coupled climate model ECHO-G—I. Near-surface temperature, precipitation and mean sea level pressure. Tellus, 57A, 605621, doi:10.1111/j.1600-0870.2005.00133.x.

    • Search Google Scholar
    • Export Citation
  • Min, S.-K., , S. Legutke, , A. Hense, , and W.-T. Kwon, 2005b: Internal variability in a 1000-yr control simulation with the coupled climate model ECHO-G—II. El Nino Southern Oscillation and North Atlantic Oscillation. Tellus, 57A, 622640, doi:10.1111/j.1600-0870.2005.00132.x.

    • Search Google Scholar
    • Export Citation
  • Mohrholz, V., , M. Naumann, , G. Nausch, , S. Krüger, , and U. Gräwe, 2015: Fresh oxygen for the Baltic Sea—An exceptional saline inflow after a decade of stagnation. J. Mar. Syst., 148, 152166, doi:10.1016/j.jmarsys.2015.03.005.

    • Search Google Scholar
    • Export Citation
  • Nikulin, G., , E. Kjellström, , U. Hansson, , G. Strandberg, , and A. Ullerstig, 2011: Evaluation and future projections of temperature, precipitation and wind extremes over Europe in an ensemble of regional climate simulations. Tellus, 63A, 4155, doi:10.1111/j.1600-0870.2010.00466.x.

    • Search Google Scholar
    • Export Citation
  • Omstedt, A., , and L. Axell, 1998: Modeling the seasonal, interannual, and long-term variations of salinity and temperature in the Baltic proper. Tellus, 50A, 637652, doi:10.1034/j.1600-0870.1998.t01-4-00005.x.

    • Search Google Scholar
    • Export Citation
  • Omstedt, A., , and D. Hansson, 2006: The Baltic Sea ocean climate system memory and response to changes in the water and heat balance components. Cont. Shelf Res., 26, 236251, doi:10.1016/j.csr.2005.11.003.

    • Search Google Scholar
    • Export Citation
  • Rodhe, J., , and P. Winsor, 2002: On the influence of the freshwater supply on the Baltic Sea mean salinity. Tellus, 54A, 175186, doi:10.1034/j.1600-0870.2002.01307.x; Corrigendum, 55, 455–456, doi:10.1034/j.1600-0870.2003.00037.x.

    • Search Google Scholar
    • Export Citation
  • Samuelsson, P., and et al. , 2011: The Rossby Centre regional climate model RCA3: Model description and performance. Tellus, 63A, 423, doi:10.1111/j.1600-0870.2010.00478.x.

    • Search Google Scholar
    • Export Citation
  • Savchuk, O. P., , K. Eilola, , B. G. Gustafsson, , M. R. Medina, , and T. Ruoho-Airola, 2012: Long term reconstruction of nutrient loads to the Baltic Sea, 1850–2006. Baltic Nest Institute Tech. Rep. 6, 9 pp. [Available online at http://www.balticnest.org/download/18.76309165137811e731dbb5/1381789019579/TR+6+-++Long-term+reconstruction+of+nutrient+loads+to+the+Baltic+Sea+1850-2006_final.pdf.]

  • Schenk, F., , and E. Zorita, 2012: Reconstruction of high resolution atmospheric fields for northern Europe using analog-upscaling. Climate Past, 8, 16811703, doi:10.5194/cp-8-1681-2012.

    • Search Google Scholar
    • Export Citation
  • Schimanke, S., , H. E. M. Meier, , E. Kjellström, , G. Strandberg, , and R. Hordoir, 2012: The climate in the Baltic Sea region during the last millennium simulated with a regional climate model. Climate Past, 8, 14191433, doi:10.5194/cp-8-1419-2012.

    • Search Google Scholar
    • Export Citation
  • Schimanke, S., , C. Dieterich, , and H. E. M. Meier, 2014: An algorithm based on sea-level pressure fluctuations to identify major Baltic inflow events. Tellus, 66A, 23452, doi:10.3402/tellusa.v66.23452.

    • Search Google Scholar
    • Export Citation
  • Schinke, H., , and W. Matthäus, 1998: On the causes of major Baltic inflows—An analysis of long time series. Cont. Shelf Res., 18, 6797, doi:10.1016/S0278-4343(97)00071-X.

    • Search Google Scholar
    • Export Citation
  • Stevens, D. P., 1991: The open boundary condition in the United Kingdom Fine-Resolution Antarctic Model. J. Phys. Oceanogr., 21, 14941499, doi:10.1175/1520-0485(1991)021<1494:TOBCIT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stigebrandt, A., 1983: A model for the exchange of water and salt between the Baltic and the Skagerrak. J. Phys. Oceanogr., 13, 411427, doi:10.1175/1520-0485(1983)013<0411:AMFTEO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stigebrandt, A., , and B. G. Gustafsson, 2003: Response of the Baltic Sea to climate change—Theory and observations. J. Sea Res., 49, 243256, doi:10.1016/S1385-1101(03)00021-2.

    • Search Google Scholar
    • Export Citation
  • Stouffer, R. J., , S. Manabe, , and K. Y. Vinnikov, 1994: Model assessment of the role of natural variability in recent global warming. Nature, 367, 634636, doi:10.1038/367634a0.

    • Search Google Scholar
    • Export Citation
  • Strandberg, G., , J. Brandefelt, , E. Kjellström, , and B. Smith, 2011: High-resolution regional simulation of the last glacial maximum climate in Europe. Tellus, 63A, 107125, doi:10.1111/j.1600-0870.2010.00485.x.

    • Search Google Scholar
    • Export Citation
  • Torrence, C., , and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor. Soc., 79, 6178, doi:10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and et al. , 2005: The ERA-40 re-analysis. Quart. J. Roy. Meteor. Soc., 131, 29613012, doi:10.1256/qj.04.176.

  • Wagner, S., and et al. , 2007: Transient simulations, empirical reconstructions and forcing mechanism for the mid-Holocene hydrological climate in southern Patagonia. Climate Dyn., 29, 333355, doi:10.1007/s00382-007-0229-x.

    • Search Google Scholar
    • Export Citation
  • Winsor, P., , J. Rodhe, , and A. Omstedt, 2001: Baltic Sea ocean climate: An analysis of 100 yr of hydrographic data with focus on the freshwater budget. Climate Res., 18, 515, doi:10.3354/cr018005.

    • Search Google Scholar
    • Export Citation
  • Winsor, P., , J. Rodhe, , and A. Omsted, 2003: Erratum: Baltic Sea ocean climate: An analysis of 100 yr of hydrographical data with focus on the freshwater budget. Climate Res., 25, 183183.

    • Search Google Scholar
    • Export Citation
  • Zorita, E., , and A. Laine, 2000: Dependence of salinity and oxygen concentrations in the Baltic Sea on large-scale atmospheric circulation. Climate Res., 14, 2541, doi:10.3354/cr014025.

    • Search Google Scholar
    • Export Citation
  • Zorita, E., , J. F. Gonzalez-Rouco, , H. von Storch, , J. P. Montavez, , and F. Valero, 2005: Natural and anthropogenic modes of surface temperature variations in the last thousand years. Geophys. Res. Lett., 32, L08707, doi:10.1029/2004GL021563.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Topography of the Baltic Sea with the names of some of the locations used in this study.

  • View in gallery

    Annual mean salinity of the hindcast simulation at BY15 at a depth of 200 m (black line) and the strongest linear reductions in salinity (three yellow lines) for periods of 5, 15, and 25 yr. Observations based on Baltic Environmental Database (BED) and Svenskt Havsarkiv (SHARK) data are shown as red crosses, and corresponding maximum negative trends are shown by red lines.

  • View in gallery

    Deep water salinity at BY15 in the long climate simulation (black line). The red dashes indicate time slices with a reduction in salinity with a regression at least as big as in the hindcast simulation for 5-, 10-, 15-, and 20-yr periods of decrease.

  • View in gallery

    Probability distributions for regression coefficients over all segments for (a) 5- and (b) 16-yr trends. The dashed lines show fitted distributions using lognormal and normal distributions for 5- and 16-yr trends, respectively. The 5-yr distributions are shown for observations, the hindcast, and the climate simulations, whereas the bottom panel includes only the hindcast and the climate simulation. The observational time series is just too short to estimate robust distributions for trends during longer segments (see text).

  • View in gallery

    Time series of salinity (blue) and river discharge (green). Both series have been low-pass filtered with a cutoff frequency of 25 yr. To highlight the anticorrelation, river discharge time series was shifted to the right by 15 yr with respect to the other. The shift is according to the lag of the maximum cross correlation (cf. Table 2).

  • View in gallery

    The mean salinity of the Baltic Sea (red) in combination with the MLR based on the entire series (black) and the first 200 yr as training periods with the validation period thereafter (blue). Also included are results for the (top) 10- and (bottom) 25-yr low-pass-filtered analyses.

  • View in gallery

    Continuous wavelet power spectra following Torrence and Compo (1998), conducted for the time series of (a) Baltic Sea salinity, (b) river discharge, (c) precipitation, (d) temperature, (e) NAO index, (f) zonal wind, and (g) meridional wind. Black lines indicate significant power at the 95% level compared to red noise based on an AR(1) coefficient. The cone of influence is shown by the white areas. The right part of the figures includes the time-averaged variance and the significance level.

  • View in gallery

    Squared wavelet coherence following Grinsted et al. (2004) between the time series of mean salinity and (a) river discharge, (b) precipitation, (c) zonal wind, (d) meridional wind, (e) temperature, and (f) the NAO index. Colors represent the coherence, and arrows indicate the relative phase relationship between the series [i.e., pointing right (left) in phase (antiphase); and up, the parameter leads the mean salinity by 90° (equal to a quarter of a period)].

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 56 56 12
PDF Downloads 31 31 4

Decadal-to-Centennial Variability of Salinity in the Baltic Sea

View More View Less
  • 1 Swedish Meteorological and Hydrological Institute, Norrköping, and Bolin Centre for Climate Research, Stockholm University, Stockholm, Sweden
  • | 2 Leibniz Institute for Baltic Sea Research, Rostock, Germany, and Swedish Meteorological and Hydrological Institute, Norrköping, Sweden
© Get Permissions
Full access

Abstract

A transient multicentury simulation mimicking natural variability has been performed for the Baltic Sea. The simulation is used for investigations of long-term trends of salinity in the Baltic Sea with special focus on periods of salinity reduction. Periods with decreasing salinity over 10 yr are found to appear approximately once per century. Considering extended periods of salinity reduction, as observed from 1976 to 1992, such events are found to be quite exceptional. Based on the climate simulation, a return period of 200 yr is estimated. River discharge, net precipitation (precipitation minus evaporation), and zonal wind are identified as the most important drivers for salinity variations in the Baltic Sea. For multidecadal periods, almost two-thirds of the salinity variability can be explained by annual means of river discharge, precipitation, both wind components, temperature, and the North Atlantic Oscillation, when multilinear regression techniques are used. However, the evaluation of wavelet coherences among the time series highlights that this relationship is not constant in time. At least three periods exist, each spanning roughly 50 yr, where the coherence between salinity and runoff as the common main driver is rather weak. This indicates that the importance of river discharge might be limited for certain periods, and drivers such as zonal wind may become more important. Finally, the variability of the Baltic Sea salinity shows increased power on time scales of 100 yr and longer. Such periodicity has never been shown for Baltic Sea salinity, and the driving mechanism remains unclear.

Denotes Open Access content.

Corresponding author address: Semjon Schimanke, Swedish Meteorological and Hydrological Institute, SE-601 76 Norrköping, Sweden. E-mail: semjon.schimanke@smhi.se

Abstract

A transient multicentury simulation mimicking natural variability has been performed for the Baltic Sea. The simulation is used for investigations of long-term trends of salinity in the Baltic Sea with special focus on periods of salinity reduction. Periods with decreasing salinity over 10 yr are found to appear approximately once per century. Considering extended periods of salinity reduction, as observed from 1976 to 1992, such events are found to be quite exceptional. Based on the climate simulation, a return period of 200 yr is estimated. River discharge, net precipitation (precipitation minus evaporation), and zonal wind are identified as the most important drivers for salinity variations in the Baltic Sea. For multidecadal periods, almost two-thirds of the salinity variability can be explained by annual means of river discharge, precipitation, both wind components, temperature, and the North Atlantic Oscillation, when multilinear regression techniques are used. However, the evaluation of wavelet coherences among the time series highlights that this relationship is not constant in time. At least three periods exist, each spanning roughly 50 yr, where the coherence between salinity and runoff as the common main driver is rather weak. This indicates that the importance of river discharge might be limited for certain periods, and drivers such as zonal wind may become more important. Finally, the variability of the Baltic Sea salinity shows increased power on time scales of 100 yr and longer. Such periodicity has never been shown for Baltic Sea salinity, and the driving mechanism remains unclear.

Denotes Open Access content.

Corresponding author address: Semjon Schimanke, Swedish Meteorological and Hydrological Institute, SE-601 76 Norrköping, Sweden. E-mail: semjon.schimanke@smhi.se

1. Introduction

The Baltic Sea is one of the largest brackish sea areas of the world. The sensitive state of the Baltic Sea is sustained through a freshwater surplus by river discharge and net precipitation (precipitation minus evaporation) on one hand and by inflows of highly saline and oxygen-rich water from the North Sea on the other hand. Major Baltic inflows (MBIs), which are crucial for the renewal of the deep water below the permanent halocline, occur intermittently with a mean frequency of approximately one per year (Matthäus and Franck 1992). MBIs are driven by atmospheric variability on time scales of roughly 40 days, mainly as a result of changing wind directions from easterly to westerly (Schimanke et al. 2014). However, since 1976 the number of MBIs has been reduced considerably (Fonselius and Valderrama 2003; Feistel et al. 2008). During the past four decades about one event per decade was observed. After the small MBI in January 1983, actually a decade without any major inflow was observed (Lass and Matthäus 1996). According to climate reconstructions, similar stagnation periods (periods without or with reduced major inflows) like the one from 1983 to 1992 have very likely occurred before, for example, as during the 1920s/1930s (Meier and Kauker 2003a) or during earlier periods of the past 500 yr (Hansson and Gustafsson 2011), and each of these stagnation periods led to a reduction in oxygen concentration in the deep Baltic Sea (Fonselius and Valderrama 2003). Now, the phase with a reduced number of MBIs seems to have been completed. Four MBIs have been registered in the last 2 yr. One of the largest saltwater inflows ever observed occurred during December 2014 (Mohrholz et al. 2015; Gräwe et al. 2015) and was followed by small and moderate MBIs in March and November 2015 (http://www.io-warnemuende.de/news-details/items/autumn-gales-again-drive-salt-into-the-baltic-third-major-baltic-inflow-within-15-years.html), as well as a case during February 2016.

Depending on the amount of saltwater inflow and the freshwater supply, the deep water salinity of the Baltic proper varies between 11‰ and 14‰ on the decadal scale (Fonselius and Valderrama 2003). An ocean climate model simulation covering the period 1500–1995 suggests that the salinity has increased by 0.1‰ century−1 (Hansson and Gustafsson 2011) whereas other, shorter simulations and observations of the twentieth century showed no significant trend (Winsor et al. 2001; Meier and Kauker 2003a). Moreover, several periods of rapid salinity change occur over the modeled 500-yr time span. A mean estimate of these changes is of the order of 0.6‰ decade−1 (for the total water column) (Hansson and Gustafsson 2011).

Reasons for the occurrence of the latest stagnation period are still debated. Changes in wind speed and direction (Lass and Matthäus 1996) or the corresponding development of sea level pressure (SLP) anomalies (Schimanke et al. 2014) are likely contributors, as may be changes in river runoff and precipitation (e.g., Schinke and Matthäus 1998; Meier and Kauker 2003a). Moreover, it might be speculated whether the observed stagnation period is a consequence of the beginning impact of anthropogenic climate change.

On decadal time scales (with periods larger than 4 yr), Baltic Sea salinity variations of about 1‰ are linked inter alia to runoff variations. Analyzing almost 100 yr of observations, Winsor et al. (2001) found that variations in freshwater storage are closely correlated to accumulated changes in river runoff. From 1902 to 1998 the average freshwater inflow to the Baltic amounts to 16 115 m3 s−1 with additional contributions from river runoff (14 085 m3 s−1) and net precipitation over the Baltic Sea (2030 m3 s−1) (Meier and Kauker 2003a). This freshwater inflow results in a residence time of freshwater in the Baltic Sea of about 35 yr (Winsor et al. 2001; Meier and Kauker 2003a; Omstedt and Hansson 2006). Decadal variations in the two parts of the freshwater inflow are well correlated. Based on 4-yr annual mean values, the correlation between net precipitation over the Baltic Sea and river runoff amounts to r = 0.78 (Meier and Kauker 2003a).

However, decadal variations in Baltic Sea salinity are also sensitive to long-term changes in westerly winds. Two studies, based upon either observations (Zorita and Laine 2000) or model results (Meier and Kauker 2003a), suggest that stronger than normal westerlies cause lower than normal salinity in the upper and lower layers in all areas of the Baltic Sea. Zorita and Laine (2000) identified a North Atlantic Oscillation (NAO) like pattern that describes 61% and 41% of the salinity variability in the shallower and deeper layers, respectively. The dynamic explanation for this observation is that during periods with anomalous strong westerlies, the flow of high-salinity water through the Danish straits into the Baltic proper is hampered because of an additional barotropic pressure gradient caused by anomalous high sea levels in the Baltic Sea (Meier and Kauker 2003a). In addition, the large-scale precipitation over the region is correlated to the NAO as well (Zorita and Laine 2000). Meier and Kauker (2003a) concluded that from 1902 to 1998 about half of the decadal variability of salinity (on time scales longer than 4 yr) is explained by the variability in freshwater inflow. The other half is explained by the low-frequency variability of the zonal wind associated with the large-scale SLP over Scandinavia and by the high-frequency variability of saltwater inflows. Other potential drivers like sea ice cover and river regulation that have changed the seasonality in runoff since the 1970s seem to be unimportant for decadal salinity variations averaged for the entire Baltic Sea.

In several modeling studies, the sensitivity of Baltic Sea salinity to changes in external drivers like freshwater supply, wind speed, and sea level variations in Kattegat was investigated (e.g., Stigebrandt 1983; Gustafsson 1997, 2000; Meier and Kauker 2003b; Stigebrandt and Gustafsson 2003; Meier 2005, 2006; Omstedt and Hansson 2006). Meier (2006) found an e-folding time scale of the response of salinity in the Baltic to changes in the atmospheric and hydrological forcing of about 20 yr. If the time scale of a forcing anomaly is larger than the response time scale, the Baltic Sea system will drift into a new state with a significantly changed salinity, and with only slightly altered stability and deep water ventilation (Meier 2005, 2006).

The goal of this study is to understand the contribution of different driving factors for the decadal-to-multidecadal variability of salinity in the Baltic Sea. Moreover, we will investigate to what extent long-lasting periods of salinity decreases are a common feature of the Baltic Sea. However, since continuous measurement series of salinity exist only since 1892 (Fonselius and Valderrama 2003), they are not sufficiently long for such investigations. Therefore, a transient climate simulation of 850 yr (longer than any other reconstruction performed so far) has been carried out with a regional climate model of the Baltic Sea. So far, significantly enhanced spectral power for certain frequencies of the Baltic Sea salinity variations has not been documented at all.

Our analysis focuses on the role of variations in river discharge and precipitation, changes in wind speed and direction, fluctuations in temperature, and shifts in large-scale pressure patterns (e.g., NAO). Therefore, the length of the simulation will allow identification of mechanisms acting on decadal-to-multidecadal time scales. Moreover, how likely it is that long periods of salinity decreases occur under natural climate variability, and whether the observed period of salinity decreases (1976–93) might be related to the initiation of climate change, will be discussed.

The paper is organized as follows. In section 2, the models and the experimental setup are described. Section 3 investigates how exceptional long-lasting periods with a reduction in salinity are, whereas in section 4, control mechanisms for the low-frequency variability of salinity of the Baltic Sea are examined. The results are summarized and discussed in section 5.

2. Model description and experimental setup

a. Model description

1) The Rossby Centre Atmosphere model

The regional atmospheric model used in this study is the Rossby Centre Atmosphere model, version 3 (RCA3; Samuelsson et al. 2011). In the present setup it operates on a rotated latitude–longitude grid with a horizontal resolution of 0.44° (~50 km), 24 vertical levels, and a time step of 30 min. The model domain spans from northern Africa beyond the northern tip of Scandinavia, covering the entirety of Europe except for the easternmost parts. Lake surfaces are simulated with a lake model, whereas the sea surface temperatures (SSTs) are taken from the driving general circulation model (GCM) ECHO-G. ECHO-G has been used and evaluated for many climate studies (e.g., Zorita et al. 2005; Kaspar et al. 2007). The reader is referred to Legutke and Voss (1999) for all details of ECHO-G.

RCA3 has been used and verified for many applications ranging from paleo- (Kjellström et al. 2010; Strandberg et al. 2011) to future climate simulations (Kjellström et al. 2011; Nikulin et al. 2011). The model is in good agreement with observations if it is forced with 40-yr ECMWF Re-Analysis (ERA-40; Uppala et al. 2005) boundary conditions. For a more detailed model description and validation, we refer to Samuelsson et al. (2011).

2) The Rossby Centre Ocean model

We use the Rossby Centre Ocean model (RCO), which is a regional model covering the entire Baltic Sea (Fig. 1). RCO is a Bryan–Cox–Semtner primitive equation circulation model with a free surface (Killworth et al. 1991). It has a horizontal resolution of 2 n mi (about 3.7 km) and consists of 83 vertical layers with a thickness of 3 m, giving a maximum depth of almost 250 m. The baroclinic and barotropic time steps amount to 150 and 15 s, respectively. The open boundary (following Stevens 1991) is located in the northern Kattegat between Denmark and Sweden. The sea surface heights (SSHs) at the open boundary are calculated from SLP gradients over the North Sea taken from the RCA3 simulation (Meier et al. 2012a). In the case of inflowing water, temperature and salinity are nudged toward climatological profiles.

Fig. 1.
Fig. 1.

Topography of the Baltic Sea with the names of some of the locations used in this study.

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

RCO is forced with 10-m wind, 2-m air temperature, 2-m specific humidity, precipitation, total cloudiness, and SLP fields from RCA3. The river runoff is calculated with a statistical model based on RCA3 data following Meier et al. (2012a).

For a detailed model description and validation, the reader is referred to Meier et al. (2003) and Meier (2007). In the latter studies, observed atmospheric surface fields were used. In Schimanke et al. (2012), results were presented and analyzed when the RCO model was forced by RCA3 model data driven by ECHO-G at the lateral boundaries.

b. Experimental setup

For this study a transient climate simulation with RCO spanning 850 yr has been carried out. It is driven with atmospheric data dynamically downscaled with RCA3 from an ECHO-G simulation mimicking natural climate variability. It should be noted that the variability of the NAO generated by ECHO-G is similar to the variability in observations and the correlations of the NAO with precipitation and the 2-m air temperature are adequate (Min et al. 2005a). Moreover, the relationship between large-scale circulation and small-scale physics such as temperature, precipitation, and wind in the model chain has been investigated in depth in Schimanke et al. (2012). For instance, it was shown that the relation between the NAO and temperature is in good agreement in the model chain when compared with proxy data and observations.

The ECHO-G simulation covers several millennia, starting at 7000 BP (Wagner et al. 2007; Hünicke et al. 2010). Therefore, ECHO-G is forced with variations in orbital parameters, solar irradiance, and greenhouse gases between 7000 BP and AD 1998.

Solar variability is prescribed following the reconstruction of Lean et al. (1995). Its amplitude between solar maxima and minima such as the Maunder Minimum can be considered to be moderate in the ongoing discussion. Moreover, it should be mentioned that volcanic eruptions have not been considered as forcing. The main reason is the large uncertainty in the early part of the simulation. Finally, aerosols have been constant throughout the simulation. Since anthropogenic aerosol loads have a cooling effect, the recent warming is overestimated. Hence in this study, for the downscaling with RCA3, only data from AD 950 until AD 1800 are considered to avoid the effects of the overestimated warming at the end of the simulation. More information regarding the forcing data can be found in Wagner et al. (2007) and Schimanke et al. (2012).

In the next step, the downscaled atmospheric data are used to force RCO. Note that the driving RCA3 data need to be corrected for systematic biases. For air temperature and wind speed, spatially varying, climatological monthly mean bias adjustments (additive for temperature and a factor for wind speed) are applied. Hence, the variability within the RCA3 simulation remained unchanged. The biases are mainly caused by the prescribed SSTs for the Baltic Sea, because the fields had to be interpolated from the driving GCM that did not resolve the Baltic Sea. More details can be found in Schimanke et al. (2012).

In general, the experimental setup follows Schimanke et al. (2012). The major difference is that the ocean model was used for the entire 850-yr period (AD 950–1800) in a transient simulation instead of a series of time-slice experiments, as was done by Schimanke et al. (2012). The first 100 yr are neglected and form the spinup period, providing a 750-yr dataset for investigations. We will call this setup the climate simulation throughout the manuscript.

In addition, we use an RCO hindcast simulation for 1850–2006, which is driven by reconstructed historical forcing to examine the model performance compared to observations. In this simulation, RCO was driven by daily, high-resolution atmospheric fields that are homogeneous and physically consistent by making use of both long European historical station data since 1850 and atmospheric surface fields from a regional atmosphere model over northern Europe for the period 1958–2006, driven by reanalysis data at the lateral boundaries (Schenk and Zorita 2012). The reconstruction of monthly nutrient loads from rivers and point sources and of atmospheric nitrogen deposition for 1850–2006 is based on available historical data (Savchuk et al. 2012). A full description of this simulation can be found in Meier et al. (2012b).

3. Salinity trends

The longest observed stagnation period occurred in the 1980s and 1990s. In general, the stagnation period refers to the years 1983–92, when no MBI was registered. However, a general decrease of deep water salinity at BY15 (a monitoring station located in the middle of the eastern Gotland basin; see Fig. 1) had already started by 1976, giving a 16-yr period with a reduced number of inflows (Lass and Matthäus 1996), and hence a loss of salt in the Baltic Sea (Fig. 2). In this section, we will investigate how far such a long-lasting and strong reduction in salinity is exceptional in an undisturbed climate. Therefore, we follow the approach of the probability of occurrence of trends greater than in the observations, as performed in earlier studies (e.g., Stouffer et al. 1994; Collins et al. 2001; Min et al. 2005a). Here, maximum trends found in observations or reconstructions are compared with the likelihood of such trends in free-model simulations.

Fig. 2.
Fig. 2.

Annual mean salinity of the hindcast simulation at BY15 at a depth of 200 m (black line) and the strongest linear reductions in salinity (three yellow lines) for periods of 5, 15, and 25 yr. Observations based on Baltic Environmental Database (BED) and Svenskt Havsarkiv (SHARK) data are shown as red crosses, and corresponding maximum negative trends are shown by red lines.

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

First, we show briefly that RCO is capable of resembling observed characteristics, as has been shown in many previous studies (e.g., Meier and Kauker 2003a). Here, we use annual means only since we are focusing on long time scales. Using reconstructed historical data for the atmosphere and river discharge, the deep water salinity in the Gotland Deep develops similarly to the observations (Fig. 2). MBIs are reasonably well represented and the stagnation period (1983–92) agrees with the observations. Also, in terms of characteristic numbers, RCO performs close to the observations. The mean deep water salinity in the Gotland Deep for the hindcast simulation is 11.8‰ compared to 12.3‰ in the observations. The standard deviation of the hindcast simulation (0.37‰) is also in the range of the observations (0.55‰). Even better results with a standard deviation of 0.48‰ were obtained when the ocean model was driven by regionalized reanalysis atmospheric surface fields available for the period 1960–2008 (Meier et al. 2011). Moreover, the long climate simulation shows similar characteristics. With a mean deep water salinity of 12.8‰, it has a standard deviation of 0.52‰, which is also close to the observations. It should be noted that other salinity measurements also agree well with observations such as, for instance, the vertical profile and the mean salinity of the Baltic Sea, which has a long-term mean of 7.36‰ in RCO (Meier and Kauker 2003a).

Figure 2 illustrates the modeled deep water salinity at BY15 and highlights the strongest linear trends over 5-, 15-, and 25-yr periods in yellow. The corresponding regression coefficients as well as all regression coefficients for periods between 5 and 25 yr can be found in Table 1. For instance, the strongest decrease in salinity over 15 yr occurred from 1977 to 1992, with a regression coefficient of −0.094‰ yr−1. The modeled regression coefficients are in good agreement with the regression coefficients computed from observations. Agreements exist for the timing and the trend, though observed trends are somewhat stronger for all considered segments (Fig. 2). The observed regression coefficient for the period 1977–92 amounts to 0.116‰ yr−1. However, despite a slight underestimation, we can conclude that trends modeled with RCO agree reasonably well with observations and can be used for the trend analysis.

Table 1.

Largest trends and the number of occurrences in the climate simulation. According to the length of the salinity reduction period (first column) and the corresponding largest regression coefficient of the hindcast simulation (second column), we computed the number of occurrences in the climate simulation (third column). The last column gives the relative occurrence rate, i.e., the number of occurrences divided by the number of overlapping segments of the climate simulation (i.e., 745 for 5-yr segments, 744 for 6-yr segments, etc.).

Table 1.

It should be pointed out that a period with a reduction in salinity is not identical to a stagnation period in its classical sense. Usually, stagnation periods are considered to be periods without any MBI. However, the salinity might decrease although MBIs occur. For instance, within the period 1976–92, which is characterized by decreasing salinity, four MBIs were registered in 1977, 1979, 1982, and 1983 (Feistel et al. 2008). Nonetheless, the salinity has decreased, since the original salt content was very high after the strong MBIs in 1975 and 1976. Moreover, it should be noted that even smaller saltwater inflows, which are not considered to be MBIs, have an impact on the reduction rate of salinity in the Baltic Sea.

After establishing that the RCO behaves similar to the observations, we investigate how usual or unusual the observed reduction period (1977–92) and the stagnation period (1983–92) are in the long climate simulation. Therefore, we check the number of occurrences of periods with negative salinity trends of the same magnitude, or larger, as in the hindcast simulation. For instance, the strongest decrease over a 5-yr period in the hindcast simulation is −0.130‰ yr−1 (Table 1). After computing the trends for all 745 overlapping 5-yr periods of the climate simulation, we find that such a strong decrease occurred 37 times, or with a relative occurrence rate of 0.05 (37 occurrences divided by 745 segments of 5 yr). Obviously, such a trend is not exceptional and the occurrences are widely spread over the simulation (Fig. 3). We then repeat this analysis for 6-yr trends on up to 25 yr (Table 1). Trends with the length of roughly 10 yr, such as those found in the hindcast simulation, are still rather common in the long climate simulation, having a relative occurrence rate of 0.012 (nine occurrences in total). However, when the periods with decreasing trends become longer, the number of events in the climate simulation decreases quickly and the large trends of the hindcast simulation with lengths of 16 and 17 yr do not occur in the climate simulation at all. On the other hand, considering even longer periods (20 yr and more), we find a reasonable number of such trends in the climate simulation with relative occurrence rates larger than 1%.

Fig. 3.
Fig. 3.

Deep water salinity at BY15 in the long climate simulation (black line). The red dashes indicate time slices with a reduction in salinity with a regression at least as big as in the hindcast simulation for 5-, 10-, 15-, and 20-yr periods of decrease.

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

To investigate the trends in more detail, we examine their probability distributions and fit proper functions to them. Figure 4a shows the distribution of regression coefficients over all 5-yr segments, for instance as with the 745 segments for the climate simulation. The distributions and their fitted functions agree well between the simulations and observations regarding the general shape. The highest probability can be found in simulations and observations for slightly negative regression coefficients with values close to −0.05‰ yr−1. These values correspond to the general prevailing freshening of the Baltic Sea related to the positive freshwater balance during periods when strong inflows are absent. However, there might be smaller inflows within the 5-yr segments, reducing the freshening. The upper end of the distributions corresponds to MBIs with a rapid increase of deep water salinity at BY15. It is obvious that positive values can have much higher regression coefficients than negative trends. However, here our main interest is in the lower end of the distributions, which illustrate strong reductions in salinity or stagnation periods.

Fig. 4.
Fig. 4.

Probability distributions for regression coefficients over all segments for (a) 5- and (b) 16-yr trends. The dashed lines show fitted distributions using lognormal and normal distributions for 5- and 16-yr trends, respectively. The 5-yr distributions are shown for observations, the hindcast, and the climate simulations, whereas the bottom panel includes only the hindcast and the climate simulation. The observational time series is just too short to estimate robust distributions for trends during longer segments (see text).

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

Based on the fitted distributions, it is possible to estimate return periods for certain trends or trends related to return periods, respectively. For instance, the largest negative regression coefficient over a 5-yr segment modeled in the climate simulation is −0.174‰ yr−1. This trend has a return frequency of 525 yr based on the fitted distribution. The trend corresponding to a 100-yr period is −0.150‰ yr−1. Based on observations, the maximum decrease of −0.196‰ yr−1 was identified around 1930 (see Fig. 2). Based on the fit of observed trends, such a decrease has an estimated return period of 260 yr. Moreover, a trend with a return period of 100 yr is expected to be −0.174‰ yr−1. However, the blue bars in Fig. 4a do not represent a smooth distribution, indicating that the sample size is rather small for a good estimation. Continuous measurements start only in 1950, just before the all-time high of 1951. Therefore, trends seem to be biased toward negative values in the observations since the general trend since 1951 is negative. Thus, even the fitted distribution of observed trends is likely distorted toward negative trends.

For 16-yr trends, the distribution among the trends based on observations is not meaningful (not shown). Because of the short time series, the sample size is too small and the distribution is strongly affected by the decreasing trend after the large inflow in 1951. However, based on the simulations, it is clear that the distributions follow normal distributions, peaking around zero quite clearly (Fig. 4b). The fitted distributions of both the hindcast and climate simulations agree remarkably well. This is an indication that the variabilities in the reconstructed forcing and in the regionalization of the GCM results create the same pattern of behavior for salinity when using the Baltic Sea model.

In general, the distribution of the 16-yr trends peaks around zero. This result is also expected, since salinity changes in the Baltic Sea over longer periods are expected to be small. Here, periods of salinity decrease and inflow periods cancel each other out on average. However, by examining the lower tail of the fitted distribution, we see that a salinity reduction such as for the period of 1976–92 is very unusual. The modeled regression coefficient of the hindcast simulation for that event (−0.091‰ yr−1; see Table 1) has a return period of 199 yr in the fitted distribution of the climate simulation (Fig. 4b). Much stronger events become very unlikely. Events with a return period of 1000 yr are estimated to have a trend of −0.113 ‰ yr−1, corresponding to a reduction of 1.808‰ within 16 yr.

The distribution of 25-yr trends is also a normal distribution with zero as the most likely regression coefficient (not shown). Estimated salinity reductions over 25 yr are not much larger than for the estimated 16 yr of reduction, since the regression coefficients become smaller for longer periods. For instance, the 1000-yr return period has a regression coefficient of −0.082‰ yr−1, or roughly 2‰ in 25 yr.

The conclusion we draw from the trend analysis is that a 10-yr period with a reduction in salinity such as has been observed from 1983 to 1992 is a regularly recurring event. We find a total of nine instances in the climate simulation (Table 1). Hence, such an event can be assumed to happen approximately once per century in an undisturbed climate. This conclusion is also supported by observations, although the observed time series are too short for such an analysis. For instance, the somewhat more than 100 yr of observations include one strong stagnation period. However, for the period of extended reduction in salinity (1976–92), we cannot find similar periods in the climate simulation. This might be due to some special circumstances. Besides the extraordinary high values during the 1950s, the deep water salinity concentration was very high at the beginning of the reduction period in 1976. Moreover, at the end of the stagnation, the lowest concentrations of the entire time series were reached. Whereas it is natural that a long-lasting stagnation period leads to very low salinity, strong trends can only be reached if the stagnation period starts from very high values, as was the case in 1976. Hence, this might have been a very special situation, being so extraordinary that it did not happen during the climate simulation. However, on the other hand, it cannot be ruled out that the extended period with a reduction in salinity is related to changing climate and therefore the changing weather regimes and changes in the river discharge.

4. Understanding multidecadal salinity changes in the Baltic Sea

a. Regression analyses

In this section, we investigate which parameters steer the long-term variability of salinity concentrations in the Baltic Sea. We consider annual means of river discharge, precipitation, temperature, both wind components, and the NAO index as driving forces. Precipitation, temperature, and wind are considered only over the Baltic Sea. In contrast to the previous section, we focus here on the mean salinity of the Baltic Sea. However, the deep water salinity in the eastern Gotland basin is closely related to the mean salinity of the Baltic Sea. For instance, the correlations among these quantities in the RCO model are 0.79, 0.87, and 0.92 for annual means and 10- and 25-yr low-pass-filtered time series, respectively. Whereas there is no time lag for the annual values, the maximum correlation for low-pass-filtered series can be found with a time lag of 4 yr for 25-yr low-pass-filtered time series. Here, the Baltic Sea mean salinity is lagging behind the deep water salinity at BY15.

As a starting point, we compute the correlation for all driving parameters with the mean salinity, considering a possible delay in the response (Table 2). The correlations for annual means are rather small, and range from −0.08 to −0.26. The correlation between salinity and runoff is largest. Moreover, all correlations obtain their maximum with a positive lag, thus indicating that the parameters lead the variability of the mean salinity. In general, an immediate response of the Baltic Sea cannot be expected, and the delay is connected to the long residence time of freshwater. All parameters except the NAO index lead the salinity by 2 or 3 yr. The NAO index has a lead time of 11 yr. However, the correlation with the NAO is so small that, in combination with the long lead time, a significant correlation cannot be established for annual means.

Table 2.

Correlations of single parameters with the averaged Baltic Sea salinity. Correlation coefficients are computed based on annual means and low-pass-filtered time series with cutoff frequencies of 10 and 25 yr, respectively. Values in parentheses indicate the lag in years at which the maximum correlation is found, and positive lags indicate that the parameter leads.

Table 2.

The anticorrelation of all parameters agrees well with physical interpretations established in previous studies. Stronger freshwater supply (runoff and precipitation) leads to a reduction of mean salinity for various reasons, such as dilution and hampering of saltwater intrusion (e.g., Meier and Kauker 2003b). Stronger westerly winds keep the Baltic Sea filled and reduce the possibility for inflows from the North Sea.

Moreover, increased wind speed (both components) enhances the vertical mixing, which dilutes the inflowing saltwater plumes (e.g., Meier 2005; Gräwe et al. 2013). As saltwater inflows penetrate as density-driven currents, the interleaving of inflowing saltwater into the Baltic proper deep water might be shifted toward smaller depths or might even be hampered completely. The NAO index represents a large-scale feature, which has a broad impact on the European climate. It might affect the Baltic salinity through its correlation with precipitation, wind speed, and other parameters influencing the Baltic Sea. Temperature is linked to sea ice cover and evaporation. Here, Meier and Kauker (2003a) have shown that severe ice conditions can hamper the inflow activity, although the effect seems small compared to other drivers.

To investigate long-term fluctuations in the Baltic Sea mean salinity, we now smooth all of the time series by applying a digital Butterworth low-pass filter with cutoff frequencies of 10 and 25 yr, respectively. Table 2 indicates that for this reason the correlation increases for all parameters. Note that the first and the last 10 yr of the smoothed series are neglected to exclude artificial behavior at the edges of the series. River discharge keeps the strongest influence on salinity with correlations of 0.62 and 0.70 for 10- and 25-yr smoothed time series, respectively. The anticorrelation of river discharge and mean salinity is clearly visible by plotting both 25-yr smoothed time series together (Fig. 5). On the other hand, out of the parameters considered, the correlation is lowest for the meridional wind component, with values of 0.27 and 0.34 for 10- and 25-year smoothed time series, respectively.

Fig. 5.
Fig. 5.

Time series of salinity (blue) and river discharge (green). Both series have been low-pass filtered with a cutoff frequency of 25 yr. To highlight the anticorrelation, river discharge time series was shifted to the right by 15 yr with respect to the other. The shift is according to the lag of the maximum cross correlation (cf. Table 2).

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

Now, we perform a multiple-linear regression (MLR) analysis to investigate the combined effects of the driving parameters to perceive the total influence on long time scales. We conduct the MLR analysis by splitting the time series into a calibration and a validation period, as well as for the entire time series. When split, the first 200 yr are used as a training period for the MLR model, which is then applied for the adjacent validation period. Table 3 gives an overview of the results where the number of parameters has been increased successively. Note that the lags identified in the correlation analysis have been considered for the MLR.

Table 3.

MLR of the forcing parameters with the Baltic Sea mean salinity. Correlation coefficients and the explained variances in parentheses are given. The number of considered parameters increases successively throughout the table, always including parameters up to the given line (e.g., values in the line zonal wind are based on an MLR including runoff, precipitation, and zonal wind). However, in the two columns to the right runoff and precipitation are not considered. In general, the columns differentiate MLRs based on the 200-yr training period and the entire time series, as well as low-pass-filtered series with a cutoff frequency of 10 and 25 yr. For the MLR based on the training period, correlation coefficients are given for the validation period. Note that the lags found for the maximum correlation given in Table 2 are taken into account for all MLRs.

Table 3.

A large degree of the salinity variability is based on the amount of river discharge. However, every additionally considered parameter increases the correlation, and by this, the explained variance in the validation period of the 10-yr low-pass-filtered time series. Considering the entire climate simulation and all driving parameters, the correlation becomes 0.72 for the 10-yr smoothed series or, in other words, they explain 51% of the Baltic Sea salinity variance. Figure 6a illustrates that the regressions explain multidecadal variability quite well. However, there is higher variability in the 10–25-yr band of the regressions that is not present for the salinity. Moreover, it is interesting to note that the regressions based on the training period and the entire time series behave quite similarly, highlighting that the MLR model is not valid only for the training period.

Fig. 6.
Fig. 6.

The mean salinity of the Baltic Sea (red) in combination with the MLR based on the entire series (black) and the first 200 yr as training periods with the validation period thereafter (blue). Also included are results for the (top) 10- and (bottom) 25-yr low-pass-filtered analyses.

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

Using the complete series filtered with a cutoff frequency of 25 yr, the correlation increases to 0.8 when all driving parameters are considered in the MLR (Table 3). However, the MLR based on the training period does not show an improvement when the meridional wind component is added. Hence, this parameter can be neglected for long periods. Nevertheless, the chosen parameters explain almost two-thirds of the Baltic Sea mean salinity variance. That is the case for the MLR based on the training period only, as well as based on the entire time series. In general, the regressions fit the salinity variability very well (Fig. 6b). However, they have problems in representing some of the maxima and minima of the salinity.

Most of the parameters used for the MLR are strongly correlated among each other as with, for instance, zonal wind and precipitation, or precipitation and river discharge (Meier and Kauker 2003a; Schimanke et al. 2012). Because of this coherence among the parameters, the addition of further parameters does not necessarily enhance the correlations significantly. However, apart from the low-frequency variability of the meridional wind, every parameter improves the MLR model.

Performing the MLR without river discharge and precipitation limits the correlation significantly (right columns in Table 3). In this case, maximum correlations of 0.55 and 0.65 are reached for 10- and 25-yr low-pass-filtered series, respectively, when considering the whole simulation. These correlations are lower than considering only river discharge, which highlights once more the leading role of the freshwater supply as a driver of salinity variations.

In general, a perfect match between the driving parameters and the Baltic Sea mean salinity cannot be expected. The main reason for that is that parts of the Baltic Sea salinity change—namely major Baltic inflows—are controlled by atmospheric variability over a period of roughly 40 days (Lass and Matthäus 1996; Schimanke et al. 2014). This cannot be measured in our analysis, which is based on annual means. Meier and Kauker (2003a) showed in sensitivity studies that some decadal variability is left even after high-pass filtering the forcing data, which should remove decadal variability.

b. Wavelet analyses

The relationships between the parameters and the Baltic Sea mean salinity do not necessarily need to be steady in time. For instance, Meier and Kauker (2003a) showed that the freshwater input lagged the wind forcing by 7 yr during the 1920s stagnation period, but preceded it during the 1980s by 4 yr. Therefore, we apply a continuous wavelet analysis, following Torrence and Compo (1998), to analyze the variability in the time-frequency domain. Afterward, we explore interactions between the parameters and the mean salinity using wavelet coherencies (Grinsted et al. 2004). Note that we use the annual mean time series to perform the wavelet analyses.

The variability of the mean salinity shows only very little periodicity for time scales shorter than roughly 30 yr (Fig. 7a). However, the wavelet analysis indicates that there is significantly enhanced power on longer frequencies and especially for periods of 100 yr or longer. The time-averaged variance (Fig. 7a, right) is clearly above the 95% significance level for centennial periods. The significance is estimated by comparing with a background Fourier power spectrum based on the autoregressive model with lag-1 autocorrelation (Torrence and Compo 1998). River runoff and the zonal wind component have significant power on the same frequencies (Figs. 7b,f), supporting the finding of enhanced power on time scales of 100 yr and longer. However, one needs to be careful, since large parts of the long-term variability might be affected by the artificial behavior close to the edge of the resolved frequency domain, as indicated by the cone of influence. Therefore, even longer simulations would be needed for a careful examination of centennial periods. On the other hand, these findings are supported by the study of Hagen and Feistel (2005), who found significant power on time scales between 85 and 130 yr in the Baltic winter index that they designed.

Fig. 7.
Fig. 7.

Continuous wavelet power spectra following Torrence and Compo (1998), conducted for the time series of (a) Baltic Sea salinity, (b) river discharge, (c) precipitation, (d) temperature, (e) NAO index, (f) zonal wind, and (g) meridional wind. Black lines indicate significant power at the 95% level compared to red noise based on an AR(1) coefficient. The cone of influence is shown by the white areas. The right part of the figures includes the time-averaged variance and the significance level.

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

The wavelet analysis of the river discharge reveals features similar to those for the salinity, which is in line with the strong correlation described in the previous section. More or less every region in the time-frequency space where the salinity shows enhanced variability has a counterpart in the wavelet of the river discharge, as seen, for example, around the year 1250 and the frequency band between 10 and 16 yr (Figs. 7a,b). Moreover, the river discharge has additional power on time scales of up to a decade. These time scales are most likely damped in the Baltic Sea salinity as a result of the long response time. Precipitation over the Baltic Sea shows very similar features as well (Fig. 7c), supporting the strong relation to river discharge (e.g., Meier and Kauker 2003a).

Temperature and zonal wind show sporadic variability over all frequencies (Figs. 7d,f), which seems to be at least partly in agreement with the salinity. On the other hand, the annual NAO index and the meridional wind only have variability up to at most 20 yr (Figs. 7e,g). At least for the NAO, this is a typical feature. For instance, Min et al. (2005b) found that the NAO index of a control simulation of the driving ECHO-G GCM, as well as the estimations based on observations, have a white noise spectrum.

The continuous wavelet analysis highlights only variability for one parameter. The relation among two can be investigated by computing wavelet coherencies (Grinsted et al. 2004), as shown in Fig. 8. This basically confirms our results from the MLR and earlier studies, showing that the river discharge has the strongest impact on the Baltic Sea salinity. Figure 8a shows that the river discharge is strongly coherent with the mean salinity for all periods longer than 3–4 yr. The figure also highlights that the time lag between the river discharge and the salinity as indicated by the arrows is mostly 90° (equal to a quarter of a period), regardless of the length of the period; in other words, there is a lag of 5 yr for a period of 20 yr. It should also be noted that the remaining driving parameters show a lag of 90°, which indicates that all driving parameters lead the mean salinity. The results agree with the findings of the cross-correlation analysis, which showed longer lags for longer periods, though the lags are larger than 90° (cf. Table 2).

Fig. 8.
Fig. 8.

Squared wavelet coherence following Grinsted et al. (2004) between the time series of mean salinity and (a) river discharge, (b) precipitation, (c) zonal wind, (d) meridional wind, (e) temperature, and (f) the NAO index. Colors represent the coherence, and arrows indicate the relative phase relationship between the series [i.e., pointing right (left) in phase (antiphase); and up, the parameter leads the mean salinity by 90° (equal to a quarter of a period)].

Citation: Journal of Climate 29, 20; 10.1175/JCLI-D-15-0443.1

Besides the overall good coherence of Baltic Sea salinity and river discharge, gaps exist where the coherence is small, especially for decadal-to-multidecadal periods (Fig. 8a). For periods of 10–32 yr, we see only a weak coherence around the years 1170, 1310, and 1580. Each of these gaps last for roughly 50 yr. For the latter two periods, we find an increased coherence with the zonal wind component (Fig. 8c), although for the first period there is only little coherence with any of the driving parameters considered (Fig. 8). This indicates that there are periods where the river discharge does not play the dominant role for the long-term variability of salinity, but that other drivers seem to be in charge.

Another reason for the reduced coherence between salinity and river discharge during the above-mentioned periods might be the strong increase in salinity. Both the deep water salinity at BY15 (Fig. 3), as well as the mean salinity of the Baltic Sea (Figs. 5 and 6), show pronounced minima with strong increases afterward at the time of the low coherence (years 1170, 1310, and 1580). Obviously, the periods are characterized by more saltwater intrusions from the North Sea, which cannot be explained by an underlying change in river discharge, but are more likely related to changes in the zonal wind.

It is also worth mentioning that the precipitation and the zonal wind over the Baltic Sea are generally coherent with the salinity over many periods (Figs. 8b,c). However, the coherence is not that large, as compared to the coherence between the salinity and river discharge. Temperature also shows consistent coherency patterns at the centennial time scale, whereas the coherence is relatively small for shorter periods (Fig. 8e). Finally, the coherence found between the meridional wind or the NAO and the mean salinity is very restricted on all time scales (Figs. 8d,f).

5. Conclusions and discussion

For the first time, a transient 850-yr-long simulation has been performed for the Baltic Sea. It is driven by forcing data mimicking natural climate variability, providing the opportunity to investigate very rare events and long-term variability. Here, we focus our investigations on long-lasting periods with a reduction in salinity ranging from 5 to 25 yr. In addition, we examine the possible driving parameters for the long-term variability of the mean salinity in the Baltic Sea.

The longest observed stagnation period happened from 1983 to 1992, whereas the reduction in deep water salinity had already started in 1976 (Lass and Matthäus 1996). By investigating the probability of trends greater than in a hindcast simulation, we find that periods of strong salinity decrease over 10 yr, appear approximately once per century. Hence, these periods are rare but they are obviously a natural phenomenon. On the other hand, considering the extended period (1976–92) of salinity reduction, we cannot find a similar reduction within our climate simulation. This might point to an anthropogenic contribution to the event. However, it seems more likely that such events are natural but have long return periods. The fitted probability distribution indicates that such events have a return frequency of almost 200 yr. Here, we would like to remind the reader that we have investigated trends, and not a lack of inflows themselves. Strong trends need several conditions to be fulfilled, which might make these patterns even rarer. For instance, to create strong trends, it is not sufficient to only have a lack of inflows. In addition, salinity concentrations need to be very high at the beginning of the cycle to create large reductions. Moreover, the lack of such events can be related to underestimated long-term variability in the forcing data (Schimanke et al. 2012). Within this context, we should remind the reader that the forcing of the hindcast simulation is based on atmospheric surface data reconstructed from historical observations, whereas the forcing for the climate simulation is based on downscaled GCM data. Although the underlying ECHO-G GCM is supposed to generate large-scale variability similar to observations [e.g., for NAO, see Min et al. (2005a)], some uncertainties might be related to the forcing data.

The investigation of driving parameters shows that river discharge, precipitation, and the zonal wind play the most important role in the decadal variability of the Baltic Sea salinity. This is basically a confirmation of earlier studies like those of Zorita and Laine (2000), Winsor et al. (2001, 2003), and Meier and Kauker (2003a). With the help of models, the sensitivity of the salinity in the Baltic Sea has been investigated thoroughly (e.g., Stigebrandt 1983; Omstedt and Axell 1998; Gustafsson 2000; Rodhe and Winsor 2002; Meier and Kauker 2003b; Stigebrandt and Gustafsson 2003; Meier 2005; Omstedt and Hansson 2006). The underlying mechanisms are well understood and are based on dilution, restriction of saltwater intrusions, changes in vertical mixing, and more (e.g., Meier and Kauker 2003a; Gräwe et al. 2013). In contrast to earlier studies, in this study we have focused on longer time scales, which was not possible before, because of the restricted observational time series and shorter simulations. For multidecadal periods, almost two-thirds of the Baltic Sea salinity variability can be explained by annual means of river discharge, precipitation, both wind components, temperature, and the NAO when MLR techniques are used. The remaining variability is most likely related to fluctuations of the atmospheric forcing on shorter periods (e.g., fluctuations that basically cause saltwater inflows). Short-term variations of freshwater input can be considered to be less important (Meier and Kauker 2003a). To depict the variability of saltwater inflows, which can be forced over periods of roughly 40 days, daily fields of SLP might be addressed (Schimanke et al. 2014). However, here we tested only the impact of both wind components during the winter at annual time scale. Winter means might be more important than annual means, since winter is the main season for saltwater intrusion (Matthäus and Franck 1992). However, in our approach, winter means have a lower correlation with the Baltic Sea salinity than do annual means, and do not improve the results of the MLR.

The calculation of wavelet coherencies between the driving parameters and Baltic Sea salinity illustrates that the river discharge generally has the strongest relation to the salinity of the Baltic Sea. However, the relationship is not prevailing throughout the simulation. At least three gaps exist, each spanning roughly 50 yr, where the coherence is rather small. This indicates that the importance of river discharge might be limited for certain periods, and that drivers such as the zonal wind then become more important. Moreover, the periods with reduced linkage to river discharge are characterized by strong increases in salinity, indicating that high-frequency variability triggering inflows is an important driver during these periods.

A very interesting outcome of the wavelet analysis is the high power of the Baltic Sea mean salinity variability on low frequencies, in other words, on those of 100 yr and more. For the Baltic Sea salinity, such periodicity has never been observed, but it might be related to the findings of Hagen and Feistel (2005). They found power at similar frequencies in their Baltic winter index. Unfortunately, even our dataset spanning 750 yr is too short for profound examinations of periods lasting 100 yr and more. Hence, the analysis has to be extended to understand this low-frequency variability. For instance, variations with similar periods should be investigated through the underlying GCM in order to identify the source of the variability.

Acknowledgments

The research presented in this study is part of the Baltic Earth Programme (Earth System Science for the Baltic Sea Region; online at http://www.baltic.earth), and was funded by the European Community’s Seventh Framework Programme (FP/2007-2013) under Grant Agreement 217246 made with BONUS, the joint Baltic Sea Research and Development Programme, and from the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning (FORMAS, Reference 2008-1885) within the project BONUS INFLOW (Holocene saline water inflow changes into the Baltic Sea, ecosystem responses and future scenarios), as well as from the Swedish Research Council (VR) within the “Reconstruction and projecting Baltic Sea climate variability 1850–2100” project (Grant 2012-2117). Additional funding by FORMAS within the “Impact of accelerated future global mean sea level rise on the phosphorus cycle in the Baltic Sea” project (Grant 214-2009-577) and from Stockholm University’s Strategic Marine Environmental Research Funds [Baltic Ecosystem Adaptive Management (BEAM)] is gratefully acknowledged. We are grateful to Eduardo Zorita for providing the boundary data from the ECHO-G model.

REFERENCES

  • Collins, M., , S. F. B. Tett, , and C. Cooper, 2001: The internal climate variability of HadCM3, a version of the Hadley Centre coupled model without flux adjustments. Climate Dyn., 17, 6181, doi:10.1007/s003820000094.

    • Search Google Scholar
    • Export Citation
  • Feistel, R., , G. Nausch, , and N. Wasmund, 2008: State and Evolution of the Baltic Sea, 1952–2005: A Detailed 50-Year Survey of Meteorology and Climate, Physics, Chemistry, Biology, and Marine Environment. J. Wiley and Sons, 712 pp.

  • Fonselius, S., , and J. Valderrama, 2003: One hundred years of hydrographic measurements in the Baltic Sea. J. Sea Res., 49, 229241, doi:10.1016/S1385-1101(03)00035-2.

    • Search Google Scholar
    • Export Citation
  • Gräwe, U., , R. Friedland, , and H. Burchard, 2013: The future of the western Baltic Sea: Two possible scenarios. Ocean Dyn., 63, 901921, doi:10.1007/s10236-013-0634-0.

    • Search Google Scholar
    • Export Citation
  • Gräwe, U., , M. Naumann, , V. Mohrholz, , and H. Burchard, 2015: Anatomizing one of the largest saltwater inflows into the Baltic Sea in December 2014. J. Geophys. Res. Oceans, 120, 76767697, doi:10.1002/2015JC011269.

    • Search Google Scholar
    • Export Citation
  • Grinsted, A., , J. C. Moore, , and S. Jevrejeva, 2004: Application of the cross wavelet transform and wavelet coherence to geophysical time series. Nonlinear Processes Geophys., 11, 561566, doi:10.5194/npg-11-561-2004.

    • Search Google Scholar
    • Export Citation
  • Gustafsson, B., 1997: Interaction between Baltic Sea and North Sea. Dtsch. Hydrogr. Z., 49, 165183, doi:10.1007/BF02764031.

  • Gustafsson, B., 2000: Time-dependent modeling of the Baltic entrance area. 2. Water and salt exchange of the Baltic Sea. Estuaries, 23, 253266, doi:10.2307/1352831.

    • Search Google Scholar
    • Export Citation
  • Hagen, E., , and R. Feistel, 2005: Climatic turning points and regime shifts in the Baltic Sea region: The Baltic winter index (WIBIX) 1659–2002. Boreal Environ. Res., 10, 211224.

    • Search Google Scholar
    • Export Citation
  • Hansson, D., , and E. Gustafsson, 2011: Salinity and hypoxia in the Baltic Sea since A.D. 1500. J. Geophys. Res., 116, C03027, doi:10.1029/2010JC006676.

    • Search Google Scholar
    • Export Citation
  • Hünicke, B., , E. Zorita, , and S. Haeseler, 2010: Baltic Holocene climate and regional sea-level change: A statistical analysis of observations, reconstructions and simulations within present and past analogues for future changes. Final Rep. of the DFG Research Unit SINCOS-2, 70 pp. [Available online at http://www.hzg.de/imperia/md/content/hzg/zentrale_einrichtungen/bibliothek/berichte/gkss_berichte_2010/gkss_2010_2.pdf.]

  • Kaspar, F., , T. Spangehl, , and U. Cubasch, 2007: Northern Hemisphere winter storm tracks of the Eemian interglacial and the last glacial inception. Climate Past, 3, 181192, doi:10.5194/cp-3-181-2007.

    • Search Google Scholar
    • Export Citation
  • Killworth, P. D., , D. Stainforth, , D. J. Webb, , and S. M. Paterson, 1991: The development of a free-surface Bryan–Cox–Semtner ocean model. J. Phys. Oceanogr., 21, 13331348, doi:10.1175/1520-0485(1991)021<1333:TDOAFS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kjellström, E., , J. Brandefelt, , J.-O. Näslund, , B. Smith, , G. Strandberg, , A. H. L. Voelker, , and B. Wohlfarth, 2010: Simulated climate conditions in Europe during the Marine Isotope Stage 3 stadial. Boreas, 39, 436456, doi:10.1111/j.1502-3885.2010.00143.x.

    • Search Google Scholar
    • Export Citation
  • Kjellström, E., , G. Nikulin, , U. Hansson, , G. Strandberg, , and A. Ullerstig, 2011: 21st century changes in the European climate: Uncertainties derived from an ensemble of regional climate model simulations. Tellus, 63A, 2440, doi:10.1111/j.1600-0870.2010.00475.x.

    • Search Google Scholar
    • Export Citation
  • Lass, H. U., , and W. Matthäus, 1996: On temporal wind variations forcing salt water inflows into the Baltic Sea. Tellus, 48A, 663671, doi:10.1034/j.1600-0870.1996.t01-4-00005.x.

    • Search Google Scholar
    • Export Citation
  • Lean, J., , J. Beer, , and R. Bradley, 1995: Reconstruction of solar irradiance since 1610: Implications for climate change. Geophys. Res. Lett., 22, 31953198, doi:10.1029/95GL03093.

    • Search Google Scholar
    • Export Citation
  • Legutke, S., , and R. Voss, 1999: The Hamburg Atmosphere–Ocean Coupled Circulation Model ECHO-G. DKRZ Tech. Rep. 18, 62 pp. [Available online at http://mms.dkrz.de/pdf/klimadaten/models/ReportNo.18.pdf.]

  • Matthäus, W., , and H. Franck, 1992: Characteristics of major Baltic inflows—A statistical analysis. Cont. Shelf Res., 12, 13751400, doi:10.1016/0278-4343(92)90060-W.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2005: Modeling the age of Baltic seawater masses: Quantification and steady state sensitivity experiments. J. Geophys. Res., 110, C02006, doi:10.1029/2004JC002607.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2006: Baltic Sea climate in the late twenty-first century: A dynamical downscaling approach using two global models and two emission scenarios. Climate Dyn., 27, 3968, doi:10.1007/s00382-006-0124-x.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., 2007: Modeling the pathways and ages of inflowing salt- and freshwater in the Baltic Sea. Estuarine Coastal Shelf Sci., 74, 717734.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , and F. Kauker, 2003a: Modeling decadal variability of the Baltic Sea: 2. Role of freshwater inflow and large-scale atmospheric circulation for salinity. J. Geophys. Res., 108, 3368, doi:10.1029/2003JC001799.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , and F. Kauker, 2003b: Sensitivity of the Baltic Sea salinity to the freshwater supply. Climate Res., 24, 231242, doi:10.3354/cr024231.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , R. Döscher, , and T. Faxen, 2003: A multiprocessor coupled ice–ocean model for the Baltic Sea: Application to salt inflow. J. Geophys. Res., 108, 3273, doi:10.1029/2000JC000521.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , A. Höglund, , R. Döscher, , H. Andersson, , U. Löptien, , and E. Kjellström, 2011: Quality assessment of atmospheric surface fields over the Baltic Sea from an ensemble of regional climate model simulations with respect to ocean dynamics. Oceanologia, 53, 193227, doi:10.5697/oc.53-1-TI.193.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., , R. Hordoir, , H. Andersson, , C. Dieterich, , K. Eilola, , B. Gustafsson, , A. Höglund, , and S. Schimanke, 2012a: Modeling the combined impact of changing climate and changing socio-economic development on the Baltic Sea environment in an ensemble of transient simulations for 1961–2099. Climate Dyn., 39, 24212441, doi:10.1007/s00382-012-1339-7.

    • Search Google Scholar
    • Export Citation
  • Meier, H. E. M., and et al. , 2012b: Impact of climate change on ecological quality indicators and biogeochemical fluxes in the Baltic Sea? A multi-model ensemble study. AMBIO, 41, 558573, doi:10.1007/s13280-012-0320-3.

    • Search Google Scholar
    • Export Citation
  • Min, S.-K., , S. Legutke, , A. Hense, , and W.-T. Kwon, 2005a: Internal variability in a 1000-yr control simulation with the coupled climate model ECHO-G—I. Near-surface temperature, precipitation and mean sea level pressure. Tellus, 57A, 605621, doi:10.1111/j.1600-0870.2005.00133.x.

    • Search Google Scholar
    • Export Citation
  • Min, S.-K., , S. Legutke, , A. Hense, , and W.-T. Kwon, 2005b: Internal variability in a 1000-yr control simulation with the coupled climate model ECHO-G—II. El Nino Southern Oscillation and North Atlantic Oscillation. Tellus, 57A, 622640, doi:10.1111/j.1600-0870.2005.00132.x.

    • Search Google Scholar
    • Export Citation
  • Mohrholz, V., , M. Naumann, , G. Nausch, , S. Krüger, , and U. Gräwe, 2015: Fresh oxygen for the Baltic Sea—An exceptional saline inflow after a decade of stagnation. J. Mar. Syst., 148, 152166, doi:10.1016/j.jmarsys.2015.03.005.

    • Search Google Scholar
    • Export Citation
  • Nikulin, G., , E. Kjellström, , U. Hansson, , G. Strandberg, , and A. Ullerstig, 2011: Evaluation and future projections of temperature, precipitation and wind extremes over Europe in an ensemble of regional climate simulations. Tellus, 63A, 4155, doi:10.1111/j.1600-0870.2010.00466.x.

    • Search Google Scholar
    • Export Citation
  • Omstedt, A., , and L. Axell, 1998: Modeling the seasonal, interannual, and long-term variations of salinity and temperature in the Baltic proper. Tellus, 50A, 637652, doi:10.1034/j.1600-0870.1998.t01-4-00005.x.

    • Search Google Scholar
    • Export Citation
  • Omstedt, A., , and D. Hansson, 2006: The Baltic Sea ocean climate system memory and response to changes in the water and heat balance components. Cont. Shelf Res., 26, 236251, doi:10.1016/j.csr.2005.11.003.

    • Search Google Scholar
    • Export Citation
  • Rodhe, J., , and P. Winsor, 2002: On the influence of the freshwater supply on the Baltic Sea mean salinity. Tellus, 54A, 175186, doi:10.1034/j.1600-0870.2002.01307.x; Corrigendum, 55, 455–456, doi:10.1034/j.1600-0870.2003.00037.x.

    • Search Google Scholar
    • Export Citation
  • Samuelsson, P., and et al. , 2011: The Rossby Centre regional climate model RCA3: Model description and performance. Tellus, 63A, 423, doi:10.1111/j.1600-0870.2010.00478.x.

    • Search Google Scholar
    • Export Citation
  • Savchuk, O. P., , K. Eilola, , B. G. Gustafsson, , M. R. Medina, , and T. Ruoho-Airola, 2012: Long term reconstruction of nutrient loads to the Baltic Sea, 1850–2006. Baltic Nest Institute Tech. Rep. 6, 9 pp. [Available online at http://www.balticnest.org/download/18.76309165137811e731dbb5/1381789019579/TR+6+-++Long-term+reconstruction+of+nutrient+loads+to+the+Baltic+Sea+1850-2006_final.pdf.]

  • Schenk, F., , and E. Zorita, 2012: Reconstruction of high resolution atmospheric fields for northern Europe using analog-upscaling. Climate Past, 8, 16811703, doi:10.5194/cp-8-1681-2012.

    • Search Google Scholar
    • Export Citation
  • Schimanke, S., , H. E. M. Meier, , E. Kjellström, , G. Strandberg, , and R. Hordoir, 2012: The climate in the Baltic Sea region during the last millennium simulated with a regional climate model. Climate Past, 8, 14191433, doi:10.5194/cp-8-1419-2012.

    • Search Google Scholar
    • Export Citation
  • Schimanke, S., , C. Dieterich, , and H. E. M. Meier, 2014: An algorithm based on sea-level pressure fluctuations to identify major Baltic inflow events. Tellus, 66A, 23452, doi:10.3402/tellusa.v66.23452.

    • Search Google Scholar
    • Export Citation
  • Schinke, H., , and W. Matthäus, 1998: On the causes of major Baltic inflows—An analysis of long time series. Cont. Shelf Res., 18, 6797, doi:10.1016/S0278-4343(97)00071-X.

    • Search Google Scholar
    • Export Citation
  • Stevens, D. P., 1991: The open boundary condition in the United Kingdom Fine-Resolution Antarctic Model. J. Phys. Oceanogr., 21, 14941499, doi:10.1175/1520-0485(1991)021<1494:TOBCIT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stigebrandt, A., 1983: A model for the exchange of water and salt between the Baltic and the Skagerrak. J. Phys. Oceanogr., 13, 411427, doi:10.1175/1520-0485(1983)013<0411:AMFTEO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Stigebrandt, A., , and B. G. Gustafsson, 2003: Response of the Baltic Sea to climate change—Theory and observations. J. Sea Res., 49, 243256, doi:10.1016/S1385-1101(03)00021-2.

    • Search Google Scholar
    • Export Citation
  • Stouffer, R. J., , S. Manabe, , and K. Y. Vinnikov, 1994: Model assessment of the role of natural variability in recent global warming. Nature, 367, 634636, doi:10.1038/367634a0.

    • Search Google Scholar
    • Export Citation
  • Strandberg, G., , J. Brandefelt, , E. Kjellström, , and B. Smith, 2011: High-resolution regional simulation of the last glacial maximum climate in Europe. Tellus, 63A, 107125, doi:10.1111/j.1600-0870.2010.00485.x.

    • Search Google Scholar
    • Export Citation
  • Torrence, C., , and G. P. Compo, 1998: A practical guide to wavelet analysis. Bull. Amer. Meteor. Soc., 79, 6178, doi:10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and et al. , 2005: The ERA-40 re-analysis. Quart. J. Roy. Meteor. Soc., 131, 29613012, doi:10.1256/qj.04.176.

  • Wagner, S., and et al. , 2007: Transient simulations, empirical reconstructions and forcing mechanism for the mid-Holocene hydrological climate in southern Patagonia. Climate Dyn., 29, 333355, doi:10.1007/s00382-007-0229-x.

    • Search Google Scholar
    • Export Citation
  • Winsor, P., , J. Rodhe, , and A. Omstedt, 2001: Baltic Sea ocean climate: An analysis of 100 yr of hydrographic data with focus on the freshwater budget. Climate Res., 18, 515, doi:10.3354/cr018005.

    • Search Google Scholar
    • Export Citation
  • Winsor, P., , J. Rodhe, , and A. Omsted, 2003: Erratum: Baltic Sea ocean climate: An analysis of 100 yr of hydrographical data with focus on the freshwater budget. Climate Res., 25, 183183.

    • Search Google Scholar
    • Export Citation
  • Zorita, E., , and A. Laine, 2000: Dependence of salinity and oxygen concentrations in the Baltic Sea on large-scale atmospheric circulation. Climate Res., 14, 2541, doi:10.3354/cr014025.

    • Search Google Scholar
    • Export Citation
  • Zorita, E., , J. F. Gonzalez-Rouco, , H. von Storch, , J. P. Montavez, , and F. Valero, 2005: Natural and anthropogenic modes of surface temperature variations in the last thousand years. Geophys. Res. Lett., 32, L08707, doi:10.1029/2004GL021563.

    • Search Google Scholar
    • Export Citation
Save