The high quality of the atmospheric surface fields is important because physical parameters have a large impact on the Baltic Sea ecosystem. Some functional dependences are even non-linear and include thresholds. For instance, water masses appropriate for the reproduction

of cod should have salinities and oxygen concentrations larger than 11 PSU and 2 ml/l respectively (e.g. MacKenzie et al. 2007). The cod eggs sink until they reach waters of salinity of ca 11 PSU. If oxygen levels at the corresponding depth of neutral buoyancy (typically the depth of the permanent PD0325901 datasheet halocline in the south-western Baltic proper) are less than 2 ml/l, the eggs will not survive. Consequently, it is important to simulate both salinity and oxygen concentrations realistically. Horizontal and vertical salinity variations in the Baltic Sea are large owing to the freshwater supply from the land. Sea surface salinities range from more than 20 PSU in the northern Kattegat to less than 2 PSU in the northern Bothnian Bay.

As the Baltic Sea catchment area is four times larger than the Baltic Sea surface area, basically the difference between precipitation and evaporation over Panobinostat manufacturer land controls the salinity gradients in the Baltic Sea. Thus, relatively small biases of simulated precipitation or evaporation over land in the RCM can have large impacts on salinity if the errors in precipitation and evaporation are not roughly equal, thus isothipendyl compensating each other.

As a consequence, any shortcomings of the simulated water cycle may significantly affect the model results of cod reproduction in particular and the Baltic Sea ecosystem in general. Both coupled physical-biogeochemical and food web modelling requires high-quality atmospheric and hydrological forcing fields (Figure 1). Forcing biases could affect biodiversity and food web functioning, and in the worst case they might result in the complete loss of species and finally in a breakdown of the simulated food web. As the marine ecosystem depends not only on mean hydrographical conditions but also on extremes, the variability of extreme variables needs to be simulated correctly by the RCM in addition to the mean states. Hence, the presented effort (as outlined in Figure 1) is a further development of earlier investigations based upon the delta approach, which assumes that the high-frequency variability of the atmospheric and hydrological forcing does not change (Meier 2006, Meier et al. 2011). Thus, it is important to select the applied GCMs carefully and to quantify uncertainties. Biases of sea level pressure (SLP), air temperature, and precipitation over Europe’s land area in GCM driven RCM simulations have been studied by Kjellström et al. (2011). They used the Rossby Centre Atmosphere model version 3 (RCA3; see Samuelsson et al.