The Baltic Sea Model Intercomparison Project (BMIP) – a platform for model development, evaluation, and uncertainty assessment

Abstract

While advanced computational capabilities have enabled the development of complex ocean general circulation models (OGCMs) for marginal seas, systematic comparisons of regional ocean models and their setups are still rare. The Baltic Sea Model Intercomparison Project (BMIP), introduced herein, was therefore established as a platform for the scientific analysis and systematic comparison of Baltic Sea models. The inclusion of a physically consistent regional reanalysis data set for the period 1961–2018 allows for standardized meteorological forcing and river runoff. Protocols to harmonize model outputs and analyses are provided as well. An analysis of six simulations performed with four regional OGCMs differing in their resolution, grid coordinates, and numerical methods was carried out to explore intermodel differences despite harmonized forcing. Uncertainties in the modeled surface temperatures were shown to be larger at extreme than at moderate temperatures. In addition, a roughly linear increase in the temperature spread with increasing water depth was determined and indicated larger uncertainties in the near-bottom layer. On the seasonal scale, the model spread was larger in summer than in winter, likely due to differences in the models' thermocline dynamics. In winter, stronger air–sea heat fluxes and vigorous convective and wind mixing reduced the intermodel spread. Uncertainties were likewise reduced near the coasts, where the impact of meteorological forcing was stronger. The uncertainties were highest in the Bothnian Sea and Bothnian Bay, attributable to the differences between the models in the seasonal cycles of sea ice triggered by the ice–albedo feedback. However, despite the large spreads in the mean climatologies, high interannual correlations between the sea surface temperatures (SSTs) of all models and data derived from a satellite product were determined. The exceptions were the Bothnian Sea and Bothnian Bay, where the correlation dropped significantly, likely related to the effect of sea ice on air–sea heat exchange. The spread of water salinity across the models is generally larger compared to water temperature, which is most obvious in the long-term time series of deepwater salinity. The inflow dynamics of saline water from the North Sea is covered well by most models, but the magnitude, as inferred from salinity, differs as much as the simulated mean salinity of deepwater. Marine heat waves (MHWs), coastal upwelling, and stratification were also assessed. In all models, MHWs were more frequent in shallow areas and in regions with seasonal ice cover. An increase in the frequency (regionally varying between ∼50 % and 250 %) and duration (50 %–150 %) of MHWs during the last 3 decades in all models was found as well. The uncertainties were highest in the Bothnian Bay, likely due to the different trends in sea ice presence. All but one of the analyzed models overestimated upwelling frequencies along the Swedish coast, the Gulf of Finland, and around Gotland, while they underestimated upwelling in the Gulf of Riga. The onset and seasonal cycle of thermal stratification likewise differed among the models. Compared to observation-based estimates, in all models the thermocline in early spring was too deep, whereas a good match was obtained in June when the thermocline intensifies.