A multi-model ensemble of decadal prediction experiments, performed in the framework of the EU-funded COMBINE (Comprehensive Modelling of the Earth System for Better Climate Prediction and Projection) Project following the 5th Coupled Model Intercomparison Project protocol is examined. The ensemble combines a variety of dynamical models, initialization and perturbation strategies, as well as data assimilation products employed to constrain the initial state of the system. Taking advantage of the multi-model approach, several aspects of decadal climate predictions are assessed, including predictive skill, impact of the initialization strategy and the level of uncertainty characterizing the predicted fluctuations of key climate variables. The present analysis adds to the growing evidence that the current generation of climate models adequately initialized have significant skill in predicting years ahead not only the anthropogenic warming but also part of the internal variability of the climate system. An important finding is that the multi-model ensemble mean does generally outperform the individual forecasts, a well-documented result for seasonal forecasting, supporting the need to extend the multi-model framework to real-time decadal predictions in order to maximize the predictive capabilities of currently available decadal forecast systems. The multi-model perspective did also allow a more robust assessment of the impact of the initialization strategy on the quality of decadal predictions, providing hints of an improved forecast skill under full-value (with respect to anomaly) initialization in the near-term range, over the Indo-Pacific equatorial region. Finally, the consistency across the different model predictions was assessed. Specifically, different systems reveal a general agreement in predicting the near-term evolution of surface temperatures, displaying positive correlations between different decadal hindcasts over most of the global domain.

A five-member ensemble of regional climate model (RCM) simulations for Europe, with a high resolution nest over Germany, is analysed in a two-part paper: Part I (the current paper) presents the performance of the models for the control period, and Part II presents results for near future climate changes. Two different RCMs, CLM and WRF, were used to dynamically downscale simulations with the ECHAM5 and CCCma3 global climate models (GCMs), as well as the ERA40-reanalysis for validation purposes. Three realisations of ECHAM5 and one with CCCma3 were downscaled with CLM, and additionally one realisation of ECHAM5 with WRF. An approach of double nesting was used, first to an approximately 50 km resolution for entire Europe and then to a domain of approximately 7 km covering Germany and its near surroundings. Comparisons of the fine nest simulations are made to earlier high resolution simulations for the region with the RCM REMO for two ECHAM5 realisations. Biases from the GCMs are generally carried over to the RCMs, which can then reduce or worsen the biases. The bias of the coarse nest is carried over to the fine nest but does not change in amplitude, i.e. the fine nest does not add additional mean bias to the simulations. The spatial pattern of the wet bias over central Europe is similar for all CLM simulations, and leads to a stronger bias in the fine nest simulations compared to that of WRF and REMO. The wet bias in the CLM model is found to be due to a too frequent drizzle, but for higher intensities the distributions are well simulated with both CLM and WRF at the 50 and 7 km resolutions. Also the spatial distributions are close to high resolution gridded observations. The REMO model has low biases in the domain averages over Germany and no drizzle problem, but has a shift in the mean precipitation patterns and a strong overestimation of higher intensities. The GCMs perform well in simulating the intensity distribution of precipitation at their own resolution, but the RCMs add value to the distributions when compared to observations at the fine nest resolution.

Atlantic tropical cyclone activity is known to oscillate between multi-annual periods of high and low activity. These changes have been linked to the Atlantic multidecadal oscillation (AMO), a mode of variability in Atlantic sea surface temperature which modifies the large-scale conditions of the tropical Atlantic. Cyclone activity is also modulated at higher frequencies by a series of other climate factors, with some of these influences appearing to be more consistent than others. Using the HURDAT2 database and a second set of tropical cyclone data corrected for possible missing storms in the earlier part of the record, we investigate, through Poisson regressions, the relationship between a series of climate variables and a series of metrics of seasonal Atlantic cyclone activity during both phases of the AMO. We find that, while some influences, such as El Nino Southern oscillation, remain present regardless of the AMO phase, other climate factors show an influence during only one of the two phases. During the negative phase, Sahel precipitation and the North Atlantic oscillation (NAO) are measured to play a role, while during the positive phase, the 11-year solar cycle and dust concentration over the Atlantic appear to be more important. Furthermore, we show that during the negative phase of the AMO, the NAO influences all our measures of tropical cyclone activity, and we go on to provide evidence that this is not simply due to changes in steering current, the mechanism by which the NAO is usually understood to impact Atlantic cyclone activity. Finally, we conclude by demonstrating that our results are robust to the sample size as well as to the choice of the statistical model.

Using a suite of lateral boundary conditions, we investigate the impact of domain size and boundary conditions on the Atlantic tropical cyclone and african easterly Wave activity simulated by a regional climate model. Irrespective of boundary conditions, simulations closest to observed climatology are obtained using a domain covering both the entire tropical Atlantic and northern African region. There is a clear degradation when the high-resolution model domain is diminished to cover only part of the African continent or only the tropical Atlantic. This is found to be the result of biases in the boundary data, which for the smaller domains, have a large impact on TC activity. In this series of simulations, the large-scale Atlantic atmospheric environment appears to be the primary control on simulated TC activity. Weaker wave activity is usually accompanied by a shift in cyclogenesis location, from the MDR to the subtropics. All ERA40-driven integrations manage to capture the observed interannual variability and to reproduce most of the upward trend in tropical cyclone activity observed during that period. When driven by low-resolution global climate model (GCM) integrations, the regional climate model captures interannual variability (albeit with lower correlation coefficients) only if tropical cyclones form in sufficient numbers in the main development region. However, all GCM-driven integrations fail to capture the upward trend in Atlantic tropical cyclone activity. In most integrations, variations in Atlantic tropical cyclone activity appear uncorrelated with variations in African easterly wave activity.

Using a statistical relationship between simulated sea surface temperature and Atlantic hurricane activity, we estimate the skill of a CMIP5 multi-model ensemble at predicting multi-annual level of Atlantic hurricane activity. The series of yearly-initialized hindcasts show positive skill compared to simpler forecasts such as persistence and climatology as well as non-initialized forecasts and return anomaly correlation coefficients of similar to 0.6 and similar to 0.8 for five and nine year forecasts, respectively. Some skill is shown to remain in the later years and making use of those later years to create a lagged-ensemble yields, for individual models, results that approach that obtained by the multi-model ensemble. Some of the skill is shown to come from persisting rather than predicting the climate shift that occur in 1994-1995. After accounting for that shift, the anomaly correlation coefficient for five-year forecasts is estimated to drop to 0.4, but remains statistically significant up to lead years 3-7. Most of the skill is shown to come from the ability of the forecast systems at capturing change in Atlantic sea surface temperature, although the failure of most systems at reproducing the observed slow down in warming over the tropics in recent years leads to an underestimation of hurricane activity in the later period.

We compare two 28-year simulations performed with two versions of the Global Environmental Multiscale model run in variable-resolution mode. The two versions differ only by small differences in their radiation scheme. The most significant modification introduced is a reduction in the ice effective radius, which is observed to increase absorption of upwelling infrared radiation and increase temperature in the upper troposphere. The resulting change in vertical lapse rate is then observed to drive a resolution-dependent response of convection, which in turn modifies the zonal circulation and induces significant changes in simulated Atlantic tropical cyclone activity. The resulting change in vertical lapse rate and its implication in the context of anthropogenic climate change are discussed.

Using the global environmental multiscale (GEM) model, we investigate the impact of increasing model resolution from 2A degrees to 0.3A degrees on Atlantic tropical cyclone activity. There is a clear improvement in the realism of Atlantic storms with increased resolution, in part, linked to a better representation of African easterly waves. The geographical distribution of a Genesis Potential Index, composed of large-scales fields known to impact cyclone formation, coincides closely in the model with areas of high cyclogenesis. The geographical distribution of this index also improves with resolution. We then compare two techniques for achieving local high resolution over the tropical Atlantic: a limited-area model driven at the boundaries by the global 2A degrees GEM simulation and a global variable resolution model (GVAR). The limited-area domain and high-resolution part of the GVAR model coincide geographically, allowing a direct comparison between these two downscaling options. These integrations are further compared with a set of limited-area simulations employing the same domain and resolution, but driven at the boundaries by reanalysis. The limited-area model driven by reanalysis produces the most realistic Atlantic tropical cyclone variability. The GVAR simulation is clearly more accurate than the limited-area version driven by GEM-Global. Degradation in the simulated interannual variability is partly linked to the models failure to accurately reproduce the impact of atmospheric teleconnections from the equatorial Pacific and Sahel on Atlantic cyclogenesis. Through the use of a smaller limited-area grid, driven by GEM-Global 2A degrees, we show that an accurate representation of African Easterly Waves is crucial for simulating Atlantic tropical cyclone variability.

The ability of four regional climate models to reproduce the present-day South American climate is examined with emphasis on La Plata Basin. Models were integrated for the period 1991-2000 with initial and lateral boundary conditions from ERA-40 Reanalysis. The ensemble sea level pressure, maximum and minimum temperatures and precipitation are evaluated in terms of seasonal means and extreme indices based on a percentile approach. Dispersion among the individual models and uncertainties when comparing the ensemble mean with different climatologies are also discussed. The ensemble mean is warmer than the observations in South Eastern South America (SESA), especially for minimum winter temperatures with errors increasing in magnitude towards the tails of the distributions. The ensemble mean reproduces the broad spatial pattern of precipitation, but overestimates the convective precipitation in the tropics and the orographic precipitation along the Andes and over the Brazilian Highlands, and underestimates the precipitation near the monsoon core region. The models overestimate the number of wet days and underestimate the daily intensity of rainfall for both seasons suggesting a premature triggering of convection. The skill of models to simulate the intensity of convective precipitation in summer in SESA and the variability associated with heavy precipitation events (the upper quartile daily precipitation) is far from satisfactory. Owing to the sparseness of the observing network, ensemble and observations uncertainties in seasonal means are comparable for some regions and seasons.

Four high resolution atmospheric general circulation models (GCMs) have been integrated with the standard forcings of the PRUDENCE experiment: IPCC-SRES A2 radiative forcing and Hadley Centre sea surface temperature and sea-ice extent. The response over Europe, calculated as the difference between the 2071-2100 and the 1961-1990 means is compared with the same diagnostic obtained with nine Regional Climate Models (RCM) all driven by the Hadley Centre atmospheric GCM. The seasonal mean response for 2m temperature and precipitation is investigated. For temperature, GCMs and RCMs behave similarly, except that GCMs exhibit a larger spread. However, during summer, the spread of the RCMs-in particular in terms of precipitation-is larger than that of the GCMs. This indicates that the European summer climate is strongly controlled by parameterized physics and/or high-resolution processes. The temperature response is larger than the systematic error. The situation is different for precipitation. The model bias is twice as large as the climate response. The confidence in PRUDENCE results comes from the fact that the models have a similar response to the IPCC-SRES A2 forcing, whereas their systematic errors are more spread. In addition, GCM precipitation response is slightly but significantly different from that of the RCMs.

The relative importance of regional processes inside the Arctic climate system and the large scale atmospheric circulation for Arctic interannual climate variability has been estimated with the help of a regional Arctic coupled ocean-ice-atmosphere model. The study focuses on sea ice and surface climate during the 1980s and 1990s. Simulations agree reasonably well with observations. Correlations between the winter North Atlantic Oscillation index and the summer Arctic sea ice thickness and summer sea ice extent are found. Spread of sea ice extent within an ensemble of model runs can be associated with a surface pressure gradient between the Nordic Seas and the Kara Sea. Trends in the sea ice thickness field are widely significant and can formally be attributed to large scale forcing outside the Arctic model domain. Concerning predictability, results indicate that the variability generated by the external forcing is more important in most regions than the internally generated variability. However, both are in the same order of magnitude. Local areas such as the Northern Greenland coast together with Fram Straits and parts of the Greenland Sea show a strong importance of internally generated variability, which is associated with wind direction variability due to interaction with atmospheric dynamics on the Greenland ice sheet. High predictability of sea ice extent is supported by north-easterly winds from the Arctic Ocean to Scandinavia.

This study presents an evaluation of the ability of 10 regional climate models (RCMs) participating in the COordinated Regional climate Downscaling Experiment-Africa to reproduce the present-day spatial distribution of annual cycles of precipitation over the South African region and its borders. As found in previous studies, annual mean precipitation is quasi-systematically overestimated by the RCMs over a large part of southern Africa south of about 20A degrees S and more strongly over South Africa. The spatial analysis of precipitation over the studied region shows that in most models the distribution of biases appears to be linked to orography. Wet biases are quasi-systematic in regions with higher elevation with inversely neutral to dry biases particularly in the coastal fringes. This spatial pattern of biases is particularly obvious during summer and specifically at the beginning of the rainy season (November and December) when the wet biases are found to be the strongest across all models. Applying a k-means algorithm, a classification of annual cycles is performed using observed precipitation data, and is compared with those derived from modeled data. It is found that the in-homogeneity of the spatial and temporal distribution of biases tends to impact the modeled seasonality of precipitation. Generally, the pattern of rainfall seasonality in the ensemble mean of the 10 RCMs tends to be shifted to the southwest. This spatial shift is mainly linked to a strong overestimation of convective precipitation at the beginning of the rainy season over the plateau inducing an early annual peak and to an underestimation of stratiform rainfall in winter and spring over southwestern South Africa.

A major source of uncertainty in regional climate model (RCM) simulations arises from the parameterisation of sub-grid scale convection. With increasing model resolution, approaching the so-called convection permitting scale, it is possible to switch off most of the convection parameterisations. A set of simulations using COSMO-CLM model has been carried out at different resolutions in order to investigate possible improvements and limitations resulting from increased horizontal resolution. For our analysis, 30 years were simulated in a triple nesting setup with 50, 7 and 2.8 km resolutions, with ERA40 reanalysis data at the lateral boundaries of the coarsest nest. The investigation area covers the state of Baden-Wurttemberg in southwestern Germany, which is a region known for abundant orographically induced convective precipitation. A very dense network of high temporal resolution rain gauges is used for evaluation of the model simulations. The purpose of this study is to examine the differences between the 7 and 2.8 km resolutions in the representation of precipitation at sub-daily timescales, and the atmospheric conditions leading to convection. Our results show that the highest resolution of RCM simulations significantly improves the representation of both hourly intensity distribution and diurnal cycle of precipitation. In addition, at convection permitting scale the atmospheric fields related to convective precipitation show a better agreement with each other. The results imply that higher spatial resolution partially improves the representation of the precipitation field, which must be the way forward for regional climate modelling.

EC-Earth, a new Earth system model based on the operational seasonal forecast system of the European Centre for Medium-Range Weather Forecasts (ECMWF), is presented. The performance of version 2.2 (V2.2) of the model is compared to observations, reanalysis data and other coupled atmosphere-ocean-sea ice models. The large-scale physical characteristics of the atmosphere, ocean and sea ice are well simulated. When compared to other coupled models with similar complexity, the model performs well in simulating tropospheric fields and dynamic variables, and performs less in simulating surface temperature and fluxes. The surface temperatures are too cold, with the exception of the Southern Ocean region and parts of the Northern Hemisphere extratropics. The main patterns of interannual climate variability are well represented. Experiments with enhanced CO2 concentrations show well-known responses of Arctic amplification, land-sea contrasts, tropospheric warming and stratospheric cooling. The global climate sensitivity of the current version of EC-Earth is slightly less than 1 K/(W m(-2)). An intensification of the hydrological cycle is found and strong regional changes in precipitation, affecting monsoon characteristics. The results show that a coupled model based on an operational seasonal prediction system can be used for climate studies, supporting emerging seamless prediction strategies.

The evolution in time of the thermal vertical stratification of the Baltic Sea in future climate is studied using a 3D ocean model. Comparing periods at the end of the twentieth and twenty first centuries we found a strong increase in stratification at the bottom of the mixed layer in the northern Baltic Sea. In order to understand the causes of this increase, a sensitivity analysis is performed. We found that the increased vertical stratification is explained by a major change in re-stratification during spring solely caused by the increase of the mean temperature. As in present climate winter temperatures in the Baltic are often below the temperature of maximum density, warming causes thermal convection. Re-stratification during the beginning of spring is then triggered by the spreading of freshwater. This process is believed to be important for the onset of the spring bloom. In future climate, temperatures are expected to be usually higher than the temperature of maximum density and thermally induced stratification will start without prior thermal convection. Thus, freshwater controlled re-stratification during spring is not an important process anymore. We employed a simple box model and used sensitivity experiments with the 3D ocean model to delineate the processes involved and to quantify the impact of changing freshwater supply on the thermal stratification in the Baltic Sea. It is suggested that these stratification changes may have an important impact on vertical nutrient fluxes and the intensity of the spring bloom in future climate of the Baltic Sea.

The impact of changes in volume, heat and freshwater fluxes through Arctic gateways on sea ice, circulation and fresh water and heat contents of the Arctic and North Atlantic Oceans is not fully understood. To explore the role played by each gateway, we use a regional sea-ice ocean general circulation model with a fixed atmospheric forcing. We run sensitivity simulations with combinations of Bering Strait (BS) and Canadian Arctic Archipelago (CAA) open and closed inspired by paleogeography of the Arctic. We show that fluxes through BS influence the Arctic, Atlantic and Nordic Seas while the impact of the CAA is more dominant in the Nordic Seas. In the experiments with BS closed, there is a change in the surface circulation of the Arctic with a weakening of the Beaufort Gyre by about thirty percent. As a consequence, the Siberian river discharge is spread offshore to the west, rather than being directly advected away by the Transpolar Drift. This results in a decrease of salinity in the upper 50 m across much of the central Arctic and East Siberian and Chukchi Seas. We also find an increase in stratification between the surface and subsurface layers after closure of BS. Moreover, closure of the BS results in an upward shift of the relatively warm waters lying between 50 and 120 m, as well as a reorganization of heat storage and transport. Consequently, more heat is kept in the upper layers of the Arctic Ocean, thus increasing the heat content in the upper 50 m and leading to a thinner sea ice cover. The CAA closing has a large impact on sea ice, temperature and salinity in the subarctic North Atlantic with opposite responses in the Greenland-Iceland-Norwegian Seas and Baffin Bay. It is also found that CAA being open or closed strongly controls the sea ice export through the Fram Strait. In all our experiments, the changes in temperature and salinity of the Barents and Kara Seas, and in fluxes through Barents Sea Opening are relatively small, suggesting that they are likely controlled by the atmospheric processes. Our results demonstrate the need to take into consideration the fluxes through the Arctic gateways when addressing the ocean and climate changes during deglaciations as well as for predictions of future climate.

Decadal prediction is one focus of the upcoming 5th IPCC Assessment report. To be able to interpret the results and to further improve the decadal predictions it is important to investigate the potential predictability in the participating climate models. This study analyzes the upper limit of climate predictability on decadal time scales and its dependency on sea ice albedo parameterization by performing two perfect ensemble experiments with the global coupled climate model EC-Earth. In the first experiment, the standard albedo formulation of EC-Earth is used, in the second experiment sea ice albedo is reduced. The potential prognostic predictability is analyzed for a set of oceanic and atmospheric parameters. The decadal predictability of the atmospheric circulation is small. The highest potential predictability was found in air temperature at 2 m height over the northern North Atlantic and the southern South Atlantic. Over land, only a few areas are significantly predictable. The predictability for continental size averages of air temperature is relatively good in all northern hemisphere regions. Sea ice thickness is highly predictable along the ice edges in the North Atlantic Arctic Sector. The meridional overturning circulation is highly predictable in both experiments and governs most of the decadal climate predictability in the northern hemisphere. The experiments using reduced sea ice albedo show some important differences like a generally higher predictability of atmospheric variables in the Arctic or higher predictability of air temperature in Europe. Furthermore, decadal variations are substantially smaller in the simulations with reduced ice albedo, which can be explained by reduced sea ice thickness in these simulations.

The ocean heat transport into the Arctic and the heat budget of the Barents Sea are analyzed in an ensemble of historical and future climate simulations performed with the global coupled climate model EC-Earth. The zonally integrated northward heat flux in the ocean at 70A degrees N is strongly enhanced and compensates for a reduction of its atmospheric counterpart in the twenty first century. Although an increase in the northward heat transport occurs through all of Fram Strait, Canadian Archipelago, Bering Strait and Barents Sea Opening, it is the latter which dominates the increase in ocean heat transport into the Arctic. Increased temperature of the northward transported Atlantic water masses are the main reason for the enhancement of the ocean heat transport. The natural variability in the heat transport into the Barents Sea is caused to the same extent by variations in temperature and volume transport. Large ocean heat transports lead to reduced ice and higher atmospheric temperature in the Barents Sea area and are related to the positive phase of the North Atlantic Oscillation. The net ocean heat transport into the Barents Sea grows until about year 2050. Thereafter, both heat and volume fluxes out of the Barents Sea through the section between Franz Josef Land and Novaya Zemlya are strongly enhanced and compensate for all further increase in the inflow through the Barents Sea Opening. Most of the heat transported by the ocean into the Barents Sea is passed to the atmosphere and contributes to warming of the atmosphere and Arctic temperature amplification. Latent and sensible heat fluxes are enhanced. Net surface long-wave and solar radiation are enhanced upward and downward, respectively and are almost compensating each other. We find that the changes in the surface heat fluxes are mainly caused by the vanishing sea ice in the twenty first century. The increasing ocean heat transport leads to enhanced bottom ice melt and to an extension of the area with bottom ice melt further northward. However, no indication for a substantial impact of the increased heat transport on ice melt in the Central Arctic is found. Most of the heat that is not passed to the atmosphere in the Barents Sea is stored in the Arctic intermediate layer of Atlantic water, which is increasingly pronounced in the twenty first century.