A workshop with invited Norwegian experts was held at Triaden Hotel outside Oslo in March 1997. At the meeting The Board of the Research Programme for Changes in Climate and the Ozone Layer (The Programme Board) invited relevant institutions to propose a co-ordinated project, with the purpose of quantifying regional climate change in Northern Europe and bordering sea areas, given a global change as predicted by global climate models.


With some amendments of the first research proposal, the project with the acronym RegClim with six participating institutions was initiated in the autumn of 1997. An extension was applied for in 1998, which enters into force mid-1999. In their first endorsement of the project, The Programme Board only supported a first phase, and requested a mid-term updated project re-application for a second phase. The present project description belongs to the application for a second phase application. The updating of the project description has been co-ordinated by the project leader (Trond Iversen) in co-operation with the project leader group and the principal investigators. Its content was endorsed at the all-staff meeting at Thorbjørnrud, Jevnaker, June 2, and after subsequent discussions by e-mail.


1.1       On the basis for regionalization of climate scenarios


Climate scenarios of time-horizon up to a thousand years have to take into account changes in atmospheric circulation, changes in oceanic circulation and temperatures (including sea-ice), and to some extent the land-based ice-mass. The general geology can be taken as constant, except for the effects of volcanic activity through emissions of radiatively active contaminants.

In RegClim we define a climate scenario as an estimated change in the statistical distribution of key parameters characterizing the weather, given a systematic change in radiative forcing. We do not address changes in radiative energy from space. We assume a long-term radiative equilibrium between the earth and the universe, but that systematic changes in the absorption-efficiency of solar radiation by the globe (changes in planetary albedo), and how the terrestrial radiation to the universe is brought about (changes in the greenhouse effect), may occur.


In routine weather prediction effects of long-term changes in radiative forcing is irrelevant. The time-horizon for weather prediction is determined by the rate by which typical observation errors grow. On average there is little weather predictability left after about two weeks from a given observation time, except for occasional seasonal predictability linked to anomalies in sea-surface temperature (SST) in certain regions (e.g. ENSO). Beyond the limit of weather predictability, any prediction skill will not be better than if we selected predictions by random from a climatological, statistical distribution of the weather elements for the date and place in question.


Climate predictability up to a thousand years is mainly determined by our knowledge of processes in the atmosphere and the oceans. In order to perform climate scenarios in practice it is common today to formulate mathematical expressions for these processes as accurately as knowledge and available computer resources permit, and code these for approximate solution by super-computers. Such codes are termed numerical climate models. The output from these models is statistical distributions of weather elements, given of the climate system. Whilst the quality of weather forecasts to a large extent is determined by the accuracy of initial conditions and of process descriptions in the atmosphere alone, the quality of climate scenarios fully depend on the accuracy of multi-compartment process modelling and the boundary conditions.


Climate modelling requires a global scope. Numerical models with boundary conditions describing the external forcing global coverage are termed global climate models (GCM). A GCM which is not interactively coupled to other compartments than the atmosphere, is abbreviated AGCM. If interactively coupled to the oceans and the sea-ice we use the acronym AOGCM, and these are the state-of-the-art climate models. Regional climate scenarios are estimates of distributions of weather-elements in a limited area of the earth given external forcings. The reason for addressing a region only, is the possibility to use better spatial resolution to represent the involved compartments. In this way geophysical processes and heterogeneities of the earth’s physical properties is better resolved than in a model having a global coverage. The information of processes taking place outside, or having scales larger than covered by the region in question, have to be provided to the regional climate model in a proper way. Given that the regional climate model is linked to a global climate description of satisfactory quality, dynamical downscaling (regional climate scenario) should be possible in principle. We decide to use the acronym ARCM for an uncoupled atmospheric regional climate model, and AORCM for a regional model that calculates some sea-surface conditions (not deep ocean). We use the acronym AGCMS for a global atmospheric climate model with a variable horizontal resolution (S = stretched grid). 


Most of the geophysical processes that determine the earth’s climate are non-linear. An important consequence is that impacts of processes not explicitly resolved in a model on those that are resolved must be parameterized. Such parameterizations probably cause the major shortcomings of today’s climate scenarios. A study of Boyle (1993) shows that little improve­ment is seen in the wintertime zonal mean diagnostics when the model-resolution in an AGCM increases beyond T42 (ca. 2.8 degrees or 300 km). This is a consequence of the fact that important atmospheric motion systems, which are responsible for transport of quasi-conservative quantities, are well resolved by T42, whilst physical processes at smaller scales are only poorly known. According to Table 6.5 of IPCC (1996), regionally and seasonally averaged present-day surface air temperatures and precipitation amounts were not simulated convincingly better with an ARCM, than if taken directly from the coarse global data. The regional detail appears more realistic in the regional models, however.


The spatial scale of the oceanic motion systems is much smaller than in the atmosphere. The model resolution of the oceanic components of the AOGCMs were, at the time of writing the IPCC(1996)-report, much too coarse to resolve these systems. Since then several AOGCMs have considerably improved their oceanic calculations most probably by increasing the model resolution (NCAR, UKMO Hadley Centre). See e.g. Carson (1999) for details. This leaves some hope that considerable new knowledge can be gained by using ORCM for the North Atlantic Ocean and the Nordic Seas.


Satisfactory regional climate modelling will require improvements of process-descriptions in the global climate models, as well as a better incorporation of regional scale radiative forcing and atmosphere-ocean interactions. Thus there is a clear need to include studies in RegClim that focus the role of processes taking place in the Nordic Seas, as well as the impacts of heterogene­ous radiative forcing (ozone and aerosols).

As an alternative approach to running ARCMs for regional climate scenarios, empirical methods are also used in RegClim. According to IPCC (1996), when optimally calibrated, empirical downscaling models have been quite successful in reproducing different statistics of local surface climatology. In complex physiographic settings, climate change scenario projections of local temperature and precipitation generated using empirical methods are significantly different from those directly interpolated from the driving GCMs. Because of the significant­ orographic influences on the regional climates in Norway, the approach of empirical downscaling is probably an appropriate method of establishing local and regional climate scenarios. A potentially weak point is the assumption that empirical relations are unchanged when the global climate changes. Also the analysis of extreme events may be hampered by low statistical significance.


Even with moderate expectations, RegClim is an important project for Norway. Regional climate modelling in our area has been practically absent before RegClim. The geographical position of Norway relative to the Arctic and a climatically sensitive sea area, necessitates studies of crucial processes in the region. Furthermore, Norway has a long history in numerical modelling of atmospheric and oceanic processes, starting out from the pioneering era of numerical weather prediction in the late 1940s. Both dynamic models and chemical transport models have a long history in the education of meteorologists and oceanographers at Norwegian Universities, and DNMI have more than 30 years of experience in operational use of weather prediction models. Hence there is ample Norwegian competence, which is now allocated to climate studies in RegClim.


1.2   RegClim in relation to international research

The ultimate goal of research in human-induced climatic change is to obtain climate projections which are sufficiently accurate to fully assess potential impacts on the natural environment and on human activities. Progress towards this goal depends on determining the likely global magnitude and rate of human-induced climate change, as well as the regional and local expressions of this. The use of sophisticated, “state-of-the-art” climate models is very important in order to achieve this progress. The resolution of present physical climate models does, however, not allow for impact evaluations on the level of detail frequently necessary. 


There are four main centres running AOGCMs for global climate scenarios. The Hadley Centre (UKMO) in Bracknel, U.K.; Max-Planck Institute (MPI) in Hamburg, Germany; The Geophysical Fluid Dynamics Laboratory (GFDL) in Princeton, USA, and the National Center for Atmospheric Research (NCAR) in Boulder, USA. The models from the first two centres give better scores when used to reproduce the present-day global climate, e.g. the World Climate Research Programme’s Atmospheric Model Inter- comparison Project (AMIP)-runs. Research taking place in these and other climate centres as well as in other relevant groups in the world (e.g. in France and Canada), to a large extent focus on improving and refining the climate model tools in order to contribute to the progress towards the goal. This research comprises studies to increase the understanding of key processes and their quantitative description in the models, as well as developing tools to remedy the shortcomings of the global models.


Processes have been subject for research all the time since geophysics became a quantitative and theoretical science, and they still are. The development of alternative tools are in many ways necessitated by the limits set by the presently available computer power for the models. Examples of alternative tools are dynamical and empirical downscaling as described above.


The confidence in global climate models are increasing as more processes are better accounted for, and significant improvements have been obtained during the period 1990-1995 (IPCC, 1990; 1992; 1996). Several factors are listed in e.g. IPCC (1996). One example of a parameterization that up to recently was necessary to avoid unphysical climate drift, is the flux-correction procedure. Using coupled models to simulate the climate under present forcing have yielded large deviations (”climate drift”) from the present climate. To dispense with this problem without really solving it, artificial exchange of heat between oceans and the atmosphere have been added (“flux correction”) and kept constant when simulating future scenarios. Such adjustments compensate for model errors without really knowing what physical processes are causing their need. Flux adjustments were often large at high latitudes (Carson, 1999) and this contributed to a lack of confidence in future climate projections for our region. Quite recently, however, a considerable breakthrough has been made, since several models have been run without flux correction with considerable success, first by NCAR. Not much has been published about these runs yet, but the most documented and tested experiments are probably those at The Hadley Centre (see e.g. Carson, 1999). It seems that the old problem was mainly due to lack of horizontal resolution in the ocean model. The new Hadley Centre run uses 1x1 degree resolution. In the Nordic Seas, there still is some climate drift. To even further increase the resolution in regional simulations of these sea areas therefore is natural to do in RegClim.


Coupled models are now tested in so-called CMIP1-runs

(http://www-pcmdi.llnl.gov/covey/cmip/cmiphome.html), where external forcing terms (CO2, solar luminosity, etc.) are kept constant


Key scientific questions in RegClim

The following list of urgent scientific problems is addressed. They are all related to the determination of the rate and magnitude of climate change and regional patterns.

1.    Coupling of scales between global climate models and regional and smaller scale models.

2.    Simulations with higher resolution climate models.

3.    Factors controlling the distributions of clouds and their radiative characteristics.

4.    The distribution and time evolution of ozone and aerosols and their radiative characteristics.

5.    The response of marine systems to change in climate forcing and their positive and negative feedbacks.

6.    The coupling between the atmosphere and ocean, and ocean circulations.

7.    Ocean-surface processes and their link to atmospheric processes.


All these points are among those listed in IPCC (1996) as the primary scientific problems for research. RegClim addresses the problems listed above with emphasis on processes particularly important in our region. Dynamical and empirical downscaling cover pts.1 and 2. Applications for similar regions as ours, have been done for some time at the Danish Climate Center (Danish Meteorological Institute) and more recently at the Rossby Center in Sweden under their project Sweclim. This kind of activity has increased considerably since IPCC (1996). Our main contacts are the MPI and the other Nordic groups, but also The Hadley Centre is an important link. RegClim is regularly participating in the Workhops on dynamical downscaling organised as a part of the EU-project MERCURE.


The process-related studies (pts. 3-7) proceedes simultanously. Some of these further developments and extensions of the earlier research projects. The Max-Planck-Institut in Hamburg and the UKMO Hadley Centre for Climate Prediction and Research centre will be central contacts also for the process-related research, since they are in the very forefront in several of the relevant areas. The formal contact person at MPI is the Director, Dr. Lennart Bengtsson. But other groups are also important, in particular MeteoFrance (Dr. M. Déqué), NCAR (Dr. Philip Rasch), since RegClim uses their AGCMs for process studies.


The scope of the proposed research is accordingly within the scope of international research covered by The World Climate Research Programme (WCRP), sponsored by The World Meteorological Organization (WMO), the international Council of Scientific Unions (ICSU) and the Intergovernmental Oceanographic Commission (IOC) of UNESCO. It is related to several components of the WCRP, in particular: Climate Variability and Predictability Programme (CLIVAR), but also World Ocean Circulation Experiments (WOCE) and Arctic Climate System Study (ACSyS). Connections are also found to some research components under the International Geosphere-Biosphere Programme (IGBP), in particular under the International Global Atmospheric Chemistry Project (IGAC).


1.3     Problem descriptions and methodologies

Our region is characterized by complex topography, and the presence of complicated coastlines and sea-ice borders separating surfaces with contrasting thermal properties. Present climate models resolve these features only crudely. Thus the need for regionalization of climate change estimates is pronounced.


In dynamical downscaling, the first approach would be to run an ARCM with boundary conditions and the largest scales taken from one of the best AOGCM-models. ARCMs are linked to AOGCM output or other global climate data through one-way influence nesting, for which there is no feedback to AOGCM-results from the ARCM.  In such runs one will use the same radiative forcings as in the AOGCM, as well as the oceanic conditions estimated by these. Thus, the ARCM will produce a regional, dynamically downscaled interpretation of the global, coarse-resolution climate simulations.


It is quite certain that even today’s best AOGCM-estimate, driving large scale and boundary conditions for ARCMs, is subject to systematic errors, e.g. with respect to major storm tracks. In our region storm-tracks in climate models tend to be too zonal, with an overestimate of cyclones entering central parts of Europe to the south of Scandinavia (e.g. Stendel and Roecner, 1998). One cannot expect downscaling to remedy this shortcoming, since it is probably in part due to remote errors. For this reason, results from downscaling only give rise to moderate expectations without improving the process-descriptions. Some improvements are possible in dynamical downscaling if surface processes are allowed to develop meso-scale structures in regional balance with the atmosphere. In Norway this applies to a coupling of the ARCM to a slab ocean, in which the surface temperature and ice-cover is calculated. Furthermore, RegClim aims at reducing the uncertainties in the estimates in our region, by considering key processes in addition to downscaling. This includes regional-scale radiative forcing, the role of air-sea interaction, and the quality of certain aspects of clouds and their radiative interaction.


Incorporation of influences from heterogeneous forcings, not included in the GCM used to drive the largest scales, necessitates incorporation of the influences of the new forcings on the global scales as well as on the regional scales. In many cases important information about the climate system sensitivity to a new forcing pattern might be made available using an AGCM only. In IPCC (1996) the largest contribution to the relative uncertainty of the global and annual anthropogenic radiative forcing is stated to be the indirect effect of tropospheric aerosols, and considerable uncertainties are also connected with tropospheric ozone and the direct effect of tropospheric aerosols. Furthermore, the strong regional variations of their forcing over our region makes studies of these important. Full climatological response scenarios can presently not be made in RegClim due to insufficient computer power. Estimates of first estimates of climate sensitivities from runs with an AGCM are under development, however. This situation may change during the running of the project.


The important connections between our region’s climate and the conditions of the adjacent sea areas, necessitate a special study of the role of the Nordic Sea and parts of the Arctic Ocean.


The SST and ice-cover in these waters are vitally connected with the relatively warm climate in our region. The prolongation of the North Atlantic Drift Current into the Nordic Seas (known as the Norwegian Atlantic Current, NAC) is the main signature of these conditions. The major flux of warm surface water into the Nordic Seas takes place between Iceland and Scotland. The driving mechanism behind NAC may partly be wind stress of predominant direction, and partly thermohaline forcing. Measure­ments taken at the weather ship M (Østerhus and Gammelsrød, 1999), show signs that the thermohaline circulation may have weakened considerably during the latest 10-15 years. RegClim aims to better understand these observations, and to investigate if they can be taken as a first sign of an abrupt, regional climate change.


In Iversen et al. (1997) a so-called quasi-coupling procedure was proposed. As described in the extension application in 1998, this procedure is now replaced by a full coupling between an AGCMS and an oceanic regional climate model (ORCM) including a dynamic-thermodynamic ice model. It is imperative that the boundaries of the ORCM is put sufficiently far from the area of main interest. If computer power permits during the project, even an OGCMS (with variable grid-resolution) will be tested for the oceanic compartment.


In order to evaluate regional climate simulations by numerical RCMs, it is necessary to have available a good set of data describing the present-day regional climate of our region. Norway has long series of climate measurements from surface stations and radio-soundings. Statistics from such stations will be used for comparisons with model statistics. Measurements from surrounding countries and the adjacent sea areas will also be made use of. The same data will be used to establish relationships for empirical downscaling. As far as possible, RegClim plans to make use of data from satellite (ATOVS) for evaluation of the vertical distribution of water-vapor, which is the most efficient “greenhouse-gas” and is responsible for important feedback-processes.


The use of paleoclimatic and proxy-data is beyond the scope of RegClim presently. However, such data are important sources of information on natural climate variations. The modelling community in Norway should be able to address also this very interesting field of climate research. If the recently started NORPAST-project is successful in systemising available data, GCM simulations of climatic scenarios in periods more than 100 years ago, should definitely be addressed at a later stage. Presently RegClim focus on validation with today’s climate, for which data are of better quality and the knowledge of external and internal processes influencing the radiation budgets are considerably more certain.


1.3.1            Atmosphere dynamics

Modelling the dynamics of the atmosphere is an unavoidable issue for climate simulations. Traditional climate parameters are the air-temperature and the precipitation amounts felt by vegetation, but modern climatology also includes statistics for a larger selection of geophysical parameters. When applying models to simulate climate change up to a few hundred years, it is necessary to include other geophysical systems, in particular the oceans and the sea ice. Since oceans are generally more inert than the atmosphere, one can make pure atmospheric simulations over a few years with a prescribed ocean state, or with an active ocean mixed layer and a prescribed deep ocean. Such experiments are called time-slice experiments.


Climate scenario simulations necessarily have a global scope. The atmospheric global circulation model (AGCM) run as a part of a coupled climate model (AOGCM) must cover the whole global troposphere and major parts of the stratosphere. With present-day computer resources, this precludes the possibility to use a horizontal resolution that is sufficient for many important impact studies. To obtain the necessary detail, one can use some statistical technique (Empirical Downscaling), or interpret the global results with a dynamic model with a fine-resolution. The latter can be made with a fine-scale AGCM or an ARCM run in time-slice mode. Obviously, such models need to resolve complex physiographic features. Furthermore, large-scale atmospheric circulations or heterogeneities in sea-surface temperature provide internal forcing of smaller-scaled features that are responsible for extreme weather events. Typical examples are ageostrophic frontogenesis, shallow frontal cyclones embedded in the jet-streams, and development of polar lows.


An ARCM with a completely prescribed ocean or an active oceanic mixed layer (slab ocean), is a tool to achieve better climate statistics pertaining to a specific region, provided a qualified knowledge of the large-scale systems. This includes ground-level extreme weather events, such as high wind-speeds, high precipitation intensities, long dry periods etc. Continuous ARCM-simulations over several years are now normal (e.g. Christensen et al., 1998; Machenhauer, 1999). This allows estimate of seasonal and inter-annual variation in the chosen region. In European regions examples of such simulations have been given by e.g. Machenhauer, et al., (1994), Jones et al., (1995 and 1997), and Christensen et al., (1997 and 1998). IPCC (1996) summarizes the main development of nested ARCM since the beginning of the 1990s. Experience gained so far only gives rise to a moderate expectation to downscaling.


A potential problem addressed by Jones et al., (1995 and 1997) was the equilibrium between the ARCM dynamics and that provided by the AGCM on scales resolved by the latter. A problem may arise when the ARCM is large enough to develop its own climate also for the scales that are determined by the open boundaries. Kida et al. (1991) proposed to control this problem by relaxing the larger scales in the ARCM towards those resolved in the AGCM. In RegClim this problem has not been detected, and the ARCM is free to develop any dynamics it is capable of describing.


A forcing pattern for atmospheric circulations, e.g. regionally heterogeneous heating of the atmosphere, may have an atmospheric response on all scales. This is one basic property of the chaotic nature of the atmospheric motions. In particular there is a typical up-scale development of small-scale structures (e.g. Palmer, 1996). In an ARCM nested into an AOGCM, there will be differences in forcing between the two models, even if the model physics are the same (this is the reason why we do downscaling in the first place). The ARCM is then only able to capture adequately the responses inside its own domain, whilst in reality responses can be global. As long as local responses dominate, ARCM-simulations can be better than the AOGCM. Nevertheless, studies of effects of specific regional forcing patterns, such as caused by aerosols or major ocean anomalies, ought to be investigated in global models. According to recent hypotheses proposed e.g. by Corti et al. (1999) the effect of a slight change in the forcing pattern of the climate system, the response can be dramatic if the pattern have large projections onto dynamic instability patterns of the climate system. The repsonse will then not bear much resemblance of the forcing pattern itself, but of the patterns connected with flow regimes receiving a larger state population after the forcing is changed. This emphasize the global nature of the climate response to a regional forcing pattern. 


A major uncertainty in climate models is the parameterization of clouds in general and sub-grid scale clouds in particular. Convection and cloud-topped boundary layers are typical examples of this. RegClim makes an effort to improve the description of shallow convection and stratocumulus, which is believed to be particularly important at high latitudes. Shallow convective clouds are ubiquitous and cover at any time huge areas on the globe. Furthermore, they interact strongly with aerosols, due to their typically low altitude and high fractional cloud co­ver. Non-precipitating low- and mid-level clouds are “oxidation chambers” for SO2 to sulphate, and their albedo are also efficiently influenced by even moderate changes in concentrations of cloud condensation nuclei and aerosols (Hobbs, 1993).


1.3.2    Ocean dynamics

There is ample evidence that the currently relatively temperate regional climate in Norway and in North western Europe at large is linked to the circulation in the Nordic Seas and the Arctic Ocean, particularly the existence of the Norwegian Atlantic Current (NAC). Hence, in order to understand how the regional climate changes in response to the global-mean climate change, as for instance stated by IPCC (1996), it is necessary to understand how the circulation in the Nordic Seas in general, and the NAC in particular, responds to such a change. In such an endeavor the coupling to the atmosphere is essential. Presently available coupled atmospheric-oceanic global climate models (AOGCM) have a too coarse grid resolution to resolve many of the important climate relevant processes taking place in the Nordic Seas and the Arctic Ocean.


To be able to predict regional changes in the climate in response to a global-mean warming, one of the foremost requirements is to understand the existence and variability of the NAC. A key question addressed through the present proposed research is; what is the driving force behind the NAC? Is it the prevailing wind systems of the region or is it the thermohaline circulation? A third possibility is that this is a signature of a chaotic coupled system so that these questions are meaningless.


The NAC is a relatively warm and saline northward flowing current, essentially following the shelf break in the eastern part of the Nordic Seas. It enters the Nordic Seas across the Iceland-Scotland Ridge, and thus brings warm and saline North Atlantic water masses close to the western coast of Norway. It also limits the ice to form in the Norwegian Sea during winter. Any changes in the behavior of this current may therefore potentially have a significant impact on the Norwegian climate. Hence, to make adequate scenarios for future changes in this flow, it is of vital importance to know the sensitivity of NAC to changes in forcing. Several driving mechanisms have been suggested. Deep and bottom water formation takes intermittently place in the cyclonic gyres of the Greenland and Iceland Seas. Warm and saline surface water is expected to be drawn into these gyres, cool off and sink during winter, and to exit as dense deep water. If this process is the dominant forcing, a reduction in the deep and bottom water formation as e.g. suggested by Schlosser et al (1991) and Meincke et al. (1992), would have a dramatic effect on the NAC. The findings of Schlosser and Meincke is further corroborated by Gammelsrød and Østerhus (1999) who report findings suggesting a reduced overflow across the Greenland-Shetland ridge to the North Atlantic. They also report that the temperature and salinity in the Atlantic water have decreased in the recent years, a decrease akin to that observed during the “Great Salinity Anomaly” in the late seventies as described by Dickson et al. (1988).

Some climate models have estimated the effects of increased greenhouse gases on the strength of the thermohaline circulation in the North Atlantic Ocean. The most recent is made by the HadCM3-model at the UKMO Hadley Centre, a model that does not need explicit flux corrections. After quadrupling the CO2 content in the atmosphere, the North Atlantic Deep Water (NADW) formation is reduced with 30-40 % of its value under pre-industrial conditions. Earlier the MPI ECHAM4/OPYC3 coupled model system was able to stop the NADW after quadrupling, whilst a doubling only lead to a 50% decrease and subsequent recovery (U. Cubasch, 1999, pers. comm.). However, to what extent this decrease in NADW is linked to considerable changes in storm-tracks or sea-ice cover, is left as an open question so far.


The view that the deep and bottom water formation in the Greenland Sea is the dominant driving force for the NAC, was challenged by Mauritzen (1994). She suggested that the dense water is formed east of the gyres within the NAC itself as it flows northward along the shelf slope west of Norway. Thus, she suggests that the NAC is continuously cooled off, and by the time it subducts in the Fram Strait it is dense enough to sink to the bottom of the North Atlantic and thereby replenish the dense bottom water. She also argues that the necessary water mass conversion needed to convert the Atlantic water to deep water without changing its density, takes place in the Arctic Ocean. Her theory may explain why the observed overflow of dense water is a continuous phenomenon.


Another obvious suggestion is that the circulation in the Nordic Seas, and thereby the NAC, is purely wind-driven and that it is upheld by the prevailing local atmospheric circulation of the Nordic Sea region. The latter features, in a mean sense, a low pressure centre located in the central parts of the Nordic Seas (the Icelandic low), connected with a major storm-track in the Nordic Sea. This in turn drives a circulation in the Nordic Seas with warm and saline water entering across the Iceland-Scotland Ridge and with cold and fresh water leaving through the Denmark Strait between Greenland and Iceland. Also the so-called JEBAR (Joint Effect of Baroclinicity And Relief) mechanism (Huthnance, 1984) may be important. This mechanism relies on the fact that generally there is a large scale north-south density gradient in the ocean upheld by the heat loss to the atmosphere in the polar areas and the heat input in the tropics. This combined with an east-west bottom topography gradient gives a northward flowing current along the eastern slopes and a southward flowing current at the western slopes. Thus, the NAC could be a manifestation of the northern branch of the almost continuously northward flow found along the edge of the European continent (Hackett and Røed, 1996).


An important feature of the Nordic Seas is the presence of an ice cover. With the present climate conditions the NAC limits ice formation to the western part of the Nordic Seas during winter. This is important since a reduced or enhanced ice cover significantly changes the heat flux exchange between the atmosphere and the ocean, and thus the thermal contrast between the ocean and the adjacent continents. A slight change in the ice condition in the Nordic Seas may therefore have a momentous impact on the regional climate. The role of the freshwater drainage to the Nordic Seas and Arctic Oceans is also connected to this. Under a global warming scenario it is conceivable that the precipitation pattern over the Northern Hemisphere will increase, which may have a profound effect on the drainage. An increase in the drainage will cause larger influx of fresh water to the Nordic Seas and the Arctic Ocean and hence potentially change both the circulation and ice extent in the area.


To come with grips with these questions RegClim uses contemporary coupled ice-ocean numerical models. As outlined in Heburn and Johnson (1995) (and references therein) numerical modelling of the circulation of the Nordic Seas is not new. However, none of the earlier studies addresses the key scientific question posed here, and very few addresses the issues relating to a coupled ice-ocean model of the Nordic Seas and Arctic Ocean.


To study variability on interannual time-scales and longer, coupled atmosphere-ocean models are required. Especially in the Labrador and the Nordic Seas, higher resolution than in most present AOGCMs is needed to resolve the relevant processes and geometry. When the resolution increases, physics on shorter space and time scale will be resolved in the models. Correct formulation of these processes therefore needs a better understanding of the air-sea interaction processes on short time and spatial scale. Consequently, it is of vital importance that the coupling between atmosphere and ocean models is carried out in connection with a detailed study of the air-sea interaction processes. A focus on the processes in the Nordic- and Barents Sea will be made, including the role of a shifting sea-ice cover. This will give a better understanding of the physics at the air-sea boundary, and help to develop more sophisticated parameterizations to be used in high-resolution regional climate models.


The total downward flux of momentum from the atmosphere to ocean current is produced either directly through traction or indirectly through wave breaking. The latter is brought about due to the fact that waves can only carry a small part of the momentum. Thus the waves play an important role in transferring momentum across the air-sea boundary. The waves act as an intermediate-stage in the flux of atmospheric momentum to the ocean.


Traditionally, the fluxes at the air-sea interface are parameterized using a simple description like the Charnock’s relation (Charnock 1955), implying that the sea-surface roughness can be calculated from the local wind. Heat and moisture fluxes are calculated with similar relations, although the physical mechanisms for the transfer are different. Also these fluxes depend on the surface shape and its roughness.

From several field studies it has been demonstrated that the sea surface roughness is not only dependent on the local wind at a given moment, but may also strongly depend on the distribution of the wave spectrum. For young wind-sea, the wave energy is concentrated in the high frequency part of the spectrum. Such short waves contribute effectively to the momentum transfer, whereas the air flows smoothly over longer swells (Doneland et.al. 1997). If the wind is blowing steady, the waves will become longer and the surface roughness decreases.

Lionelli et. al. (1998) studied the effect of coupling on the development of an idealized mid-latitude baroclinic instability. They found that the increased surface roughness at the early stages of the process, lead to a significantly larger heat flux from the ocean to the atmosphere, however, with only minor effects on the development of the atmospheric low. They argued that the reason for this is that the life cycle and intensity of a mid-latitude low is strongly dominated by the baroclinic instability of the basic flow.

On this background we suggest that the changes in surface roughness due to wave age dependent stress may significantly alter life cycle of instabilities which has the heat flux from the ocean as its main driving force.


1.3.3            Regional radiative climate forcing

Ozone and aerosol particles are known to yield radiative forcing which exhibit significant regional gradients (e.g. IPCC, 1996 and references therein). Such differential radiative forcing can potentially constitute an important feedback on climate. The recent climate development simulation from Max-Planck-Institut für Meterorologie in Hamburg (MPI), which for the first time include effects of sulphate, soot and ozone in the radiation budget, confirms considerable effects in our region (Bengtsson and Machenhauer, 1998, pers. comm.; results not published yet). The gradients in the radiative forcing are a result of the relatively short atmospheric residence times of ozone and particles. Sulphate and carbonaceous particles have lifetimes of a few days, while ozone lifetimes vary between a few days in the planetary boundary layer to a few weeks in the upper troposphere. Their distributions reflect the regions where production is most efficient. Since ozone and sulphate precursors are strongly affected by the emission of pollutants, their concentrations in the lower troposphere are strongly enhanced over North America, Europe and Southeast Asia. Sulphuric aerosols also affect climate through their interaction with clouds, which in turn also influence the radiative balance. This indirect effect of the aerosols also exhibits large spatial inhomogenities.


Ozone (O3) is formed in the atmosphere by photochemical processes taking place in the sunlit atmosphere. The ozone distribution and changes in the free troposphere are strongly affected by photochemical production and loss that are determined by precursors like nitrogen oxides (NOx), non methane hydrocarbons (NMHC) and carbon monoxide (CO). Water vapour is also a key gas in the tropospheric chemistry as it is an important precursor for OH..


OH is a key compound in the tropospheric oxidation process, and changes in its distribution will affect a large number of compounds. Chemical schemes that are developed to simulate oxidation processes such as the regional scale formation of ozone and sulphate particles in the troposphere, need to yield a realistic representation of the distribution and changes in concentrations of OH, O3, and hydrogen peroxide (H2O2).


Ozone is an efficient greenhouse gas. It affects the radiation through the absorption of longwave thermal radiation in the 9.6 and 14 µm bands, as well as through the absorption of short-wave solar radiation in the UV-B region (wavelengths shorter than 310 nm) and partly in the visible region. The impact on radiative forcing is particularly large for ozone changes in the lower stratosphere and in the upper troposphere (Wang et al., 1980; Lacis et al., 1990).


The changes in radiative forcing due to ozone changes from pre-industrial time are largest at mid northern latitudes where the impact on ozone from emission of pollutants is largest (Forster et al., 1996; Berntsen, T. et al., 1997). Regional ozone changes in areas where there have been a strong increase in the emissions of pollutants over the last couple of decades (e.g. Southeast Asia), have caused an increase in the radiative forcing in parts of the region of up to 0.5 Wm-2 (Berntsen, T. et al., 1996; 1997). In Europe there has been up to a 0.4 Wm-2 increase in the radiative forcing.



It is now well established that anthropogenic aerosols, which stem mainly from combustion of fossil fuels and biomass burning, can have a significant effect on climate. The links between aerosols and climate are among the more uncertain components of the understanding of possible anthropogenic climate change (IPCC, 1996; Penner et al., 1994). The direct aerosol effect includes all changes in the radiative emission and extinction properties of the particles themselves. The indirect aerosol effect is caused by a possible increase of cloud condensation nuclei (CCN): The Twomey effect (Twomey, 1977) is an increase of cloud albedo due to smaller and more numerous droplets; the Albrecht effect (Albrecht, 1989) is an increase of global cloudiness caused by less efficient precipitation release. Discussions on aerosols and climate in the early 1990s mainly addressed the light-scattering and cloud-modifying properties of anthropogenic sulphate particles  (Albrecht, 1989; Wigley, 1989; Charlson et al., 1990, 1991, 1992; Isaksen et al., 1992). Due to the low surface albedo in areas of high abundance of anthropogenic sulphate, early estimates of sulphate radiative forcing nearly counteracted that of greenhouse gases. According to the estimates in IPCC(1996), the direct (radiative) effect yields a negative perturbation of the global energy balance at the top of the atmosphere of about 0.2 -0.8 Wm-2, as compared to a positive 2.5 Wm-2 perturbation due to longlived greenhouse gases. The indirect effect of the aerosols is very uncertain, with global estimates ranging between 0 and -1.5 Wm-2 (IPCC, 1996). Both effects exhibit large spatial inhomogenities, which can give rise to differential radiative forcing. The numbers for global radiative forcing are therefore misleading.


Aerosol particles are grouped into primary particles, which are directly emitted into the atmosphere, and secondary particles being produced in the atmosphere after emission of a gaseous precursor.


Primary background particles embrace sea-salt and continental crustal dust particles. These particles constitute the natural nuclei onto which gaseous matter are condensed by molecular diffusion, and other particles are coagulated by Brownian diffusion (e.g. Chuang and Penner, 1995; Iversen et al., 1998). Sensitivity tests by Kirkevåg et al. (1998) indicate that the direct effect of anthropogenic sulphate and fossil fuel BC is only moderately sensitive to the number density of the background aerosol. The contribution to CCN concentrations is considerably more sensitive, however.


Sulphate particles are formed in the atmosphere from SO2 by gas phase reactions initiated by reaction with OH, or in cloud droplets via reactions involving O3 and H2O2. The sulphate formation is therefore strongly coupled to the gas phase ozone chemistry, and the frequency of occurrence of clouds is decisive for the sulphate formation. Sulphate can contribute both to the direct and indirect effect. Calculations of radiative forcing due to sulphate aerosols have frequently been based on the distribution calculated by Langner and Rodhe (1991). Global distributions of sulphate using atmospheric global circulation models (AGCM) are now becoming numerous (Taylor and Penner, 1994; Boucher and Anderson, 1995; Dastoor and Pudykiewicz, 1996; Feichter et al., 1996; Pham et al., 1996; Kasibhatla et al., 1997; Barth et al., 1999; Rasch et al., 1999; Kiehl et al. 1999. Only very few models included feedback to the model dynamics (Taylor and Penner, 1994; MPI, 1998 (Bengtsson and Machenhauer, pers. comm.). Mitchell et al. (1995) simulated the direct sulphate-aerosol effect in the Hadley-centre coupled AOGCM, simply by changing the surface albedo according to Charlson et al. (1991). Interestingly, this produced results with much better agreement with the observed development of global temperatures in the 20th century.


Carbonaceous aerosols consist of black carbon (BC) and organic carbon (OC). BC is of the part responsible for absorption of solar radiation and is chemically very inert. In external mixtures BC is hydrophobic, but may transform to hydrophilic by internal mixing with hygroscopic material, e.g. by condensation of sulphuric acid gas or by coagulation with hygroscopic particles. BC is produced as small nucleation mode particles during incomplete combustion (Seinfeld and Pandis, 1998). In combustion engines awkwardly shaped BC-agglomerates (fractal particles) are also formed (e.g. Ström et al., 1992). OC embraces all organic particles in the atmosphere except BC. Both primary and secondary OC are important in the atmosphere. Soot is a mixture of BC and primary OC (Penner and Novakov, 1996). Both BC and OC are produced during combustion of fossil fuel and biomass (Penner et al., 1993; Cooke and Wilson, 1996; Liousse et al., 1996). There are also natural sources for OC from vegetation (Liousse et al., 1996). Carbonaceous particles may contribute both to direct and indirect effects, but externally mixed BC is hydrophobic. Studies of OC are very few in the context of climate change. Liousse et al. (1996) published an emission survey together with updated BC-emission data. These data predict a general OC/BC ratio of 6 for biomass burning and about 3 for fossil fuel combustion (all units in carbon-mass). Results from the TARFOX measurement campaign, show that OC are at least as important as sulphate both with respect to aerosol mass and optical depth (Hegg et al., 1997; Novakov et al., 1997).

Radiative effects of aerosols

The radiative forcing due to the direct effect has also been estimated off-line of climate models. Large regional changes in the radiative forcing have been calculated due to the direct effect of sulphate particles. E.g. Kiehl and Rodhe (1995) calculated a maximum in the forcing in July at -11 Wm-2 over central Europe and -7.2 Wm-2 over Eastern China. Although there are uncertainties in these numbers, they indicate that the sulphate plus greenhouse gas forcing since pre-industrial times has been negative, probably to a substantial degree, in these regions.


The role of anthropogenic BC on climate was discussed during the study of Arctic haze in the 1980s (Rosen et al., 1981; Clarke et al., 1984; Hansen and Rosen, 1984; Valero et al., 1984; Wendling et al., 1985; Sheridan, 1989). Due to the high surface albedo in the Arctic the can be considerable. Attempts at estimating the joint direct effects of sulphate and BC have included assumptions that the BC-column is a fraction of the sulphate-column. This yields a slightly positive forcing over high-albedo surfaces, whilst negative forcing remains elsewhere (Iversen and Tarrasón, 1995; Haywood and Shine, 1995 and 1997; Haywood et al., 1997; Myhre et al., 1998). Absorption is also important if layers of BC-containing aerosols are situated in or above clouds (Haywood and Shine, 1995). Thus, our region appears to be exposed to a considerable contrast in radiative forcing that is potentially important for the global circulation. Early attempts at estimating such effects were made by Blanchet (1989), but with artificial assumptions for the aerosol distribution.


Estimates based on BC-concentrations calculated by transport-models from grid emissions are given by Schult et al., (1997) and Haywood and Ramaswamy (1998) assuming an external mixture of sulphate and BC. Still there exists no global estimate of the combined radiative forcing of sulphate and BC, which includes a realistic description of the aerosol mixing. Iversen et al.. (1998) and Kirkevåg et al. (1999) include examples of such calculations based on concentration calculations by Seland and Iversen (1999). The transformation from external to internal mixing was determined by a parameterization of coagulation, and the main assumptions for the primary background particles were taken from d’Almeida et al. (1991).


The indirect effect of sulphate has so far been included in climate simulations in even more parameterized fashions. Jones et al. (1994) and Boucher and Lohmann (1995) both estimate cloud droplet concentrations as single-valued functions of sulphate mass using empirical data. Using the sulphate distribution by Langner and Rodhe (1991), the Twomey-effect was estimated to be comparable to the direct effect on a global basis. The incertitude is highly emphasized in these papers. In particular the single-valued formulas used imply that effects are direct consequences of the empirical formulas. Schemes that are more closely linked to first principles are needed. E.g. super-saturations should also be a function of the aerosol properties. Chuang and Penner (1995) proposed a method to estimate CCN concentrations based on aerosol size-distributions estimated by assuming background aerosols of marine and continental origins. First attempts at this in a global model were presented by Chuang et al. (1997), and this approach is elaborated further in RegClim. In Lohman et al (1999) a step into more realism of the effects on clouds by aerosols has been taken.


OC consists of very many different organic molecules (see e.g. Seinfeld and Pandis, 1998), and the hygroscopic and optical properties of these particles are not given as readily as for sulphate and BC (Liousse et al., 1996; Penner et al., 1996). If the atmospheric abundance of OC is as large as recent investigations suggest (e.g. TARFOX), hygroscopic OC-components may contribute more efficiently to the indirect effect than sulphate. To a large extent, OC are primary particles that all may act as new CCN (Penner et al., 1992; Novakov and Penner, 1993), as opposed to sulphate whose major portion are produced in already existing droplets. Preliminary estimates of indirect forcing of carbonaceous particles by Penner et al. (1996) were as high as -2.4 to -4.4 Wm-2.


It is widely appreciated that marine aerosols have considerable influence on CCN and atmospheric optics. Crustal aerosols both absorb and scatter solar radiation, and they also include hygroscopic components important for the CCN concentrations. Studies have shown that crustal aerosols contribute to a heating of the atmosphere, in particular over source regions (Coakley and Cess, 1985; Tegen and Lacis, 1996).


The large uncertainty in estimating the global impact of the indirect effect of aerosols is due to difficulties associated with modelling the couplings between aerosols, clouds and radiation. Until recently, most GCMs treated clouds in a mostly diagnostic fashion with very few degrees of freedom (Manabe et al., 1965; Senior and Mitchell, 1993). In such formulations, any incorporation of the impact of aerosols on cloud microphysics would be extremely difficult to carry out. Over the last few years, GCMs have gradually introduced more elaborate cloud parameterization schemes, inter alia, carry cloud water as a prognostic variable, subject to sources, sinks and transport (Le Treut and Li, 1991; Sundqvist, 1993; Del Genio et al., 1996). In such a configuration, aerosol effects could tentatively be introduced by, e.g., modifying the rate of auto-conversion of cloud water to precipitation. The «third generation» cloud parameterization schemes treat cloud microphysics even more explicitly, e.g. by introducing prognostic precipitating water, ice and snow (Fowler et al., 1996), or by describing the precipitation process in more detail. This includes carrying information on cloud droplet concentrations in every grid point (Boucher et al., 1995, Rasch and Kristjánsson, 1997). In this suite of cloud parameterization schemes, the impact of aerosols can be treated in a more direct way, by modifying the variable giving cloud droplet concentrations.


With a possible increase of CCN concentrations, a narrowing of the cloud droplet spectrum is expected, with fewer large droplets and more numerous small droplets, and reduced precipitation release (Albrecht, 1989).


In addition to the effect on precipitation release and cloud coverage, a general reduction in droplet size increases the cloud albedo, even if cloud thickness and coverage were unchanged (Twomey, 1977). According to simple calculations by Hobbs (1993), moderately thick clouds with about 50% coverage will be most affected.


Due to the short lifetime of aerosols, the indirect effect will probably mostly affect low, and possibly, middle clouds. This typically applies to, e.g., stratocumulus clouds associated with shallow convection over sea, a cloud type that is very common over the North Atlantic in summer. The cooling effect of such clouds is very large, since their LW warming effect is negligible, due to their high cloud top temperatures, while their SW cooling effect is large, due to low surface albedo and relatively high sun. It is evident that this cooling effect can have a large impact on regional-scale radiative forcing.

Convective clouds

In addition to research on the microphysical effects of aerosols, we also envisage an effort towards improving the parameterizations of convection (mainly shallow convection) and cloud fraction. The treatment of both shallow convection and cloud fraction is very central to the issue of indirect aerosol effects. Furthermore low-level clouds are essential in obtaining a correct liquid-phase oxidation of SO2 to sulphate.


1.3.4            Empirical downscaling and observation data

Climatological Data.

Climatological data appropriate for evaluation of regional climate models for Northern Europe resides in different data archives at different institutions. Most of the traditional climate data are archived at the Norwegian Meteorological Institute (DNMI), but important data (e.g. oceanographic data) are also archived elsewhere. To secure easy access to the data, it is a necessity that needed information about the data is made available for all participants within the project. This will be secured by establishing a common survey of a). data available within the project b). information about the available data (metadata), c) where the data can be obtained, and d). who to contact. The survey can be found on the dedicated internet address:


which also is linked from the regular RegClim home-page.


Empirical Downscaling.

An alternative technique to dynamical internet-address downscaling in order to establish projected changes in regional and local climate from AOGCM simulations, is to use empirical downscaling. This is basically a two-step process:

(i)       Development of empirical relationships between observed local climate elements (e.g. temperature and precipitation) and observed large-scale atmospheric predictor fields (e.g. sea level pressure or geopotential height).

(ii) Application of these relationships on large-scale fields from AOGCM simulations to infer changes in local climate characteristics.


According to IPCC (Kattenberg et al., 1996, p. 339), when optimally calibrated, downscaling models have been quite successful in reproducing different statistics of local surface climatology. It is also stated that in complex physiographic settings, climate change scenarios of local temperature and precipitation generated using downscaling methods are significantly different from, and have a finer spatial structure than those directly interpolated from the driving AOGCMs. Precipitation is often the most poorly simulated parameter in AOGCM experiments (Hewitson, 1995). AOGCMs do, however, produce reasonable simulations of the large-scale circulation, and this fact can be explicitly exploited by downscaling.


For different regions in various parts of the world, relations between large-scale predictors and local predictands have been developed to generate regional/local climate change scenarios. Von Storch et al (1993) used downscaling of AOGCM experiments to deduce regional scale scenarios of Iberian rainfall in winter from the large-scale North Atlantic sea level pressure (SLP). The empirical model was based on canonical correlation analysis (CCA). The resulting empirical model was found to give a good approximation of the Iberian rainfall from 1900 to present just based on the observed North Atlantic mean SLP distribution. The empirical model was also used on the simulated change of the mean North Atlantic SLP field from projected 2xCO2 AOGCM experiments, to estimate the implications for Iberian rainfall.


Zorita et al (1995) developed two statistical approaches for linking large-scale atmospheric circulation pattern and daily local rainfall. These two empirical models were applied to AOGCM results, to simulate local precipitation under altered climate regimes in two regions in the US. The first method was based on «Classification And Regression Tree (CART)» analysis. The CART method classifies observed daily SLP fields into weather types that are most strongly associated with the presence/absence of rainfall at selected index stations. After applying this method to historical SLP observations, precipitation simulation associated with AOGCM SLP output were validated in terms of probability of occurrence and life-time of weather types. Daily rainfall time series generated from CART weather classes were then generated for daily SLP fields derived both from historical observations and from AOGCM simulations. The second rainfall generator was based on an analogue method, and was using information about the evolution of the SLP field from several previous days. In this method a pool of past observations for the circulation patterns closest to each AOGCM SLP output was considered. For a given SLP field simulated by an AOGCM experiment, this method selected from the daily historical observations the closest possible SLP field, and then used the observed precipitation on that day as the simulated precipitation.


Beersma et al. (1997) used a MPI AOGCM to study the effect of greenhouse warming on the North Atlantic wind climatology. The analysis was based on winter data from two high-resolution time-slice experiments: a control run and a 2xCO2 run. The storm track (500 hPa height variability), mean sea level pressure and average surface winds in the control experiment compared well with five years of analyses from ECMWF. The mean sea level pressure field however differed markedly from a pressure climatology for the period 1955-1993 prepared by DNMI and from ECMWF analysis for the period 1986-1990.


Kaas (1993) used regression analysis to establish relationships between observed monthly means of climate elements (temperature, precipitation) at Nordic weather stations and PCA (Principal Component Analysis) decomposition of the monthly mean large-scale flow. Predictors were 500 hPa heights and 500/1000 hPa thickness for the period 1961-87. The established relationships were applied on projected 2xCO2 runs from the Max-Planck-Institute (MPI) AOGCM, and temperature and precipitation scenarios were estimated for all four seasons of the year. Two Norwegian stations (Tromsø & Bergen) were included in the analysis.


Because of the significant orographic influences on the regional climates in Norway, the approach of empirical downscaling is an appropriate method for establishing local and regional climate scenarios here. The climate in Norway is strongly influenced by the large-scale atmospheric circulation over Northern Europe. Because of variation in the strength of the westerlies, the inter-annual variation of temperature and precipitation is large.


In Phase I of RegClim, empirical relationships are established between monthly SLP fields over the North Atlantic/Europe region, and temperature / precipitation in different sub-regions in Norway (Hanssen-Bauer, 1999). Correlation coefficients between observed and modelled regional series for most months range between 0.7 and 0.9. The models we used to investigate to which degree observed long-term trends and decadal scale variability can be explained variability in SLP alone. The analysis concludes that even though considerable parts of the regional temperature and precipitation variabilty is connected to SLP-variability, this should not be used as the only predictor.


The application of CCA, singular value decomposition (SVD), and multivariate regression (MVR) to establish empirical relationships between large-scale fields and local climate variables in selected Norwegian locations, has been discussed by Benestad (1998 b, c, d). The CCA SLP models in general demonstrate good skills for predicting Norwegian land-surface temperatures. The models have also been calibrated with different datasets of sea surface temperatures (SST), sea-ice, geopotential heights and 500hPa temperatures. The prediction skill varies with the seasons, with largest seasonal sensitivities for the SST models. The SST-models gave good results for winter temperatures.


1.3.5            Implications of Climate Change: hydrology

One important rational for downscaling global climate scenarios is to obtain data with spatial and temporal resolution that are sufficient for use in impact studies. RegClim does not presently aim at impact studies per se, however, some impacts are linked to feedbacks in the climate system in ways that defend their inclusion as parts of climate scenario developments. One such example is hydrological effects, such as flooding and draughts. The water substance occurs naturally in all three phases in the climate system, is an important carrier of latent heat, and is a key substance in both short-wave and long-wave radiative transfer.


Precipitation is itself one of the key climate parameters. In order to estimate short-term effects of changed precipitation on vital elements for the society, the precipitation amount is only one (although a key) input parameter amongst several. Precipitation and evaporation data are needed on reasonable catchment resolution along with data on geology and vegetation. The dynamically downscaled data produced in RegClim could be used for assessing hydrological effects in Norway; however, the hydrological module in the ARCM is relatively crude. Further improvement is needed before full hydrological impacts can be assessed, and our first approach is to use empirical downscaling to parameters that can be used directly in off-line water-balance models. At a later stage, as a part of a Nordic co-operation, it is hoped that hydrology will be developed as a fully integrated part of the ARCM with full feedback to the other parts of the climate system covered by the regional model.


For Norway the climate change impacts on water resources is of major importance, both concerning hydropower production, water supply, flooding, and dam safety. An extensive Nordic program on “Climate change and energy production” was carried out during 1991-96 (Sælthun et al., 1998), and a survey of effects of climate change on hydrological regimes in Europe was recently given by Arnell(1999). To adapt the RegClim scenarios for hydrological impact studies, the downscaled results will be used as input to conceptual water balance models (e.g. Bergström, 1992; Førland et al., 1996). Scenarios with sufficient spatial and temporal resolution will be worked out partly by empirical downscaling of large-scale fields from an AOGCM (ECHAM4/OPYC3), (see section 1.3.4 and Principal Task 3), and partly by empirical downscaling of the output from the ARCM-runs (HIRHAM) produced in RegClim (see section 1.4.1, and Principal Task 1). Results from dynamical downscaling of fields from ECHAM4/OPYC3 are so far presented in Bjørge and Haugen (1998), Haugen et al. (1999a,b). The first empirically downscaled scenarios of temperature and precipitation in various parts of Norway are presented in Benestad (1999a) and Hanssen-Bauer (2000).


1.4     Model tools


1.4.1        Atmospheric circulation models

Estimates of probable regional climate change given a best estimate of global change, requires an ARCM nested with one-way influence to output from a high-quality AOGCM. HIRHAM is a full climate version (Machenhauer, et al., 1994; Christensen et al., 1997) of the limited-area weather prediction model operated at The Norwegian Meteorological Institute (HIRLAM). It is a sophisticated ARCM, developed by the HIRLAM community (Gustafsson et al., 1993) for numerical weather prediction, adopted by MPI for regional climate simulations, and also used by the Danish Meteorological Institute (DMI) for regional climate simulations (Christensen et al., 1996).


We have decided to use the HIRHAM-model being run at MPI and at the Danish Meteorological Institute in RegClim. The latest version of the model was made available to the project from MPI during RegClim Phase I. The code has been implemented for massive parallel processing by Dag Bjørge, DNMI, and the full ECHAM4 physics implemented by him and Jan Erik Haugen, DNMI. This has been a success, since the model runs faster than anticipated when planning RegClim originally. Thus more ambitious plans can now be made for the dynamical downscaling pert of RegClim in Phase II.


HIRHAM is chosen because it is thoroughly tested and has the same physical parameteriza­tion as the AOGCM operated at MPI (ECHAM). In addition, there is expertise in running the HIRLAM model at DNMI for operational numerical weather prediction. According to model-intercomparisons from the Atmospheric Model Intercompar­ison Project (AMIP), the models at the UK Met.-Office (The Hadley Centre) and at MPI (ECHAM) are among the very best in the world (Slingo et al., 1995; D’Andrea et al., 1996). We will probably use results from ECHAM as input to RegClim’s HIRHAM-version focusing our region (Northern Europe and adjacent sea-areas including parts of the Arctic).


In order to build on already existing Norwegian competence on model tools for global climate simulations, we have presently the choice between two models. The ARPEGE-IFS model is operated for the Nordic Climate Project (NOCLIMP) in Geophysical Institute, University of Bergen. The model was originally developed by METEO FRANCE and ECMWF (Courtier et al., 1991), and extended to a climate version by Déqué et al. (1994). It is a spectral model with semi-lagrangian time integration. The model offers the possibility of two-way interaction nesting with varying horizontal resolution using stretched co-ordinates, as described by Courtier and Geleyn (1988). Application of this technique in regional climate simulation has been demonstrated in Déqué and Piedelievre (1995). Vertically, a terrain following hybrid coordinate is applied (Simmons and Burridge, 1981). Climatological boundary conditions and most of the parameterization schemes are described in Déqué et al. (1994). In later versions of the model an additional radiation scheme (Morcrette, 1989, 1991) and soil scheme (Noilhan and Planton, 1988) have become available in the parameterization library. Déqué and Piedelievre (1995) calculated 10 years simulations with a the ARPEGE-IFS model with stretched grid-coordinates yielding a T200 resolution over Europe and T21 over the antipode-region (South-Pacific). By comparing with results of regular T42 (2.8 degrees) and T106 models, they showed promising results for Europe. The winter-time large-scale dynamics are improved in the fine-mesh area, e.g. the representation of the Icelandic low is better than in the T106-model. On the other hand, the summer-circulation is worse than the T42-model. For area-mean, seasonally and yearly averaged temperatures the results are mixed. They conclude that the grid-stretching method is a valid alternative to other regionalization procedures being employed. It is an expensive method, however, the introduction of the semi-lagrangian integration scheme has limited the amounts of extra computer time involved with the variable resolution option.


The Department of Geophysics at the University of Oslo uses several versions of the Community Climate Model (presently: CCM3 with some extensions to be included in the upcoming CCM4) from the US National Center for Atmospheric Research (NCAR) (Hack et al., 1994; Kiehl et al., 1996). These are pure AGCMs, but can provide input data to an ARCM run with one-way influence nesting. The NCAR-CCM is used for research purposes and to develop improved parameterizations of cloud processes in co-operation with scientists at NCAR and others (Rasch and Kristjánsson, 1998; Kristjánsson et al, 1999). CCM3 is the fourth in a series of Community Climate Models from NCAR. It constitutes the atmospheric component of the coupled climate model NCAR CSM, which was the first to be run without flux adjustments. In order to make this coupling feasible, CCM3 has been considerably improved with respect to biases in the energy budgets compared to earlier model versions (Kiehl et al., 1996). The performance of CCM3 is described in several papers (Hack et al., 1998; Kiehl et al., 1998a and b; Rasch and Kristjansson, 1998). Its pre-decessor, the CCM2 has been documented by, e.g., Kiehl et al. (1994) and Hack et al. (1994). CCM3 is a spectral model, normally run at T42 resolution, but coarser and finer (T106) resolutions have been run (Philip Rasch, pers. comm.). There are normally 18 levels in the vertical, with a hybrid vertical coordinate (Simmons and Burridge, 1981). The time-integration is performed with a semi-implicit leapfrog scheme using a time-step of 20 min. The transport of moisture and trace species is done using a semi-lagrangian scheme, developed by Williamson and Rasch (1994). Physical parameteriza­tions include a non-local scheme for the vertical diffusion in the PBL (Holtslag and Boville, 1993), a deep convection parameteri­zation scheme of Zhang and McFarlane (1995) and a bulk exchange formulation for the exchange of heat, moisture and momentum between the atmosphere and ground. Information on the model, source codes and initial and boundary datasets for resolutions T5, T21, T31 and T42, is obtainable from internet: http://www.cgd.ucar.edu/cms/ccm3/.


We have decided to dedicate the ARPEGE-IFS to study of processes determining the conditions of The Nordic Seas and its feedback on the atmospheric state in our region. The possibility of using variable grid resolution makes this model well suited for studying air-sea-interactions over selected sea-regions.


For studies of first order effects of regional-scale radiative forcing and the indirect aerosol effects, we dedicate the NCAR CCM3/4 model. This is a favourable choice partly because a suitable prognostic cloud water scheme is included already (Rasch and Kristjáns­son, 1998). Extensive developments in this model, including an aerosol cycle model for sulphur and black carbon, parameterization of aerosol optical parameters and cloud condensation nuclei, as well as linking the cloud scheme interactively to the aerosol properties, are presently being made in RegClim. The radiation calculation is directly linked to the cloud parameteri­zation scheme. Already a T42 resolution will give much new insight, since very few experiments (if any) with time-resolved interaction between dynamics, heterogeneously distributed contaminant concentrations, and their influence on clouds and radiative forcing have been made so far.


1.4.2    Ocean models

It is required that a coupled ice-ocean modeling tool, suitable for describing and studying the circulation of the Nordic Seas and Arctic Ocean (ORCM), is established. Since it is not obvious which type of ocean model is best suited for regional climate studies, two existing ocean community models will be utilized. Both of these models are in use at Norwegian institutions. The primary model will be the Miami Isopycnic Coordinate Ocean Model (MICOM), which is presently in use at NERSC and DNMI. The secondary model, or backup model, is DNMI's version of the Princeton Ocean Model (POM), which is the operational ocean model at DNMI. As such it is used to provide 48 hours forecast of ocean variables twice a day (Engedahl, 1995). A POM version is also in use at IMR, (Ådlandsvik and Hansen, 1997) and is also coupled with chemical and biological model components (Skogen, 1993).


MICOM is a layered model using isopycnic coordinates in the vertical and was originally developed by Rainer Bleck and coworkers at the University of Miami, USA (Bleck et al., 1992). It has been used at NERSC as a modeling tool to study such diverse problems as the carbon cycle in the North Atlantic (Drange, 1994, 1996) and the stratified circulation over abrupt topography on the Norwegian continental shelf (Budgell et al., 1994). An important aspect of MICOM, which is particularly important for climate studies, is its ability to conserve water masses. This is linked to the isopycnic nature of the model. MICOM is now adapted for use in Arctic areas and coupled to a sea ice model. Several inherent parameterizations relevant for the Arctic Ocean are being tested in RegClim. In RegClim MICOM is used both as a global and a limited area model with variable resolution.


At DNMI MICOM is in use as a research tool, in particular to study the influence of mesoscale features on the variability of exchanges across the Greenland-Iceland-Scotland ga. This study is of  direct relevance to RegClim.


POM is, in contrast to MICOM, a level model, which uses a bottom-following sigma coordinate as its vertical coordinate.Iin its present implementation it is not well suited for long term simulations, as for instance required for climate studies. Thus DNMI's POM version will be further developed toward this goal, and results will be compared with MICOM results whenever possible.. The POM was originally developed by G. L. Mellor and A. F. Blumberg at Princeton University (Blumberg and Mellor, 1987). Through analysis of results obtained through the regional atmospheric modeling task PT1 of Phase I, it has become clear that to do a proper dynamic downscaling a dynamic ocean surface needs to be taken into account. At present there is an inconcistency in the way the bottom boundary in the ARCM (HIRHAM) is treated over land and over ocean surfaces. Over land the soil temperature and snow cover is free to develop its own climate, and may therefore become different from the AOGCM input. This is not the case over the ocean surface where the sea surface temperature (SST) and ice/snow cover is fixed to those values specified in the AOGCM. To remedy this inconsistency we propose to implement a simplified version of MICOM, in essence a 1½-layer model or slab ocean model coupled with an ice/snow model, and couple it to the present ARCM (HIRHAM). This will allow the HIRHAM to develop its own SST and ice/snow cover consistent with its own dynamics (see section 1.5.3 and section

As ocean wave model we will use the third generation wave model WAM (Wave Modelling), developed by an international group of scientists (The WAMDI Group, 1988). At the Norwegian Meteorological Institute versions of both models have been developed for MPP-computers (Massive Parallel Processing), and are now being used for operational weather- and wave forecasting.

1.4.3    Air-Sea Interaction and Coupling

The ARPEGE-IFS climate model (Courtier et al., 1991; Déqué et al., 1994) is already chosen as the atmospheric model under PT4. In RegClim the model is used with variable resolution (Courtier and Geleyn 1988, Deque and Piedelievre 1995, Deque et al., 1998). ARPEGE has been adapted to different Norwegian super-computer platforms (Cray J90, SGI-Cray Origin 2000) and several climate simulations with prescribed sea-surface temperatures (SST) have been performed (Kvamstø, pers. comm.). We choose to couple ARPEGE with the ocean model MICOM. We propose to do the coupling between ARPEGE and MICOM in the frame of the software package OASIS developed by Dr. Terray at National Centre for climate modelling and global change, Toulouse, France (CERFACS). OASIS is well recognised and today widely used at many climate centres (e.g ECMWF, Max Planck Institute and CERFACS).


As described and justified in section 1.5.2, we aim at coupling HIRHAM with a slab ocean model, to get a better description of regional details of the ocean surface which is compatible with the resolution used for the dynamical downscaling. For the coupling we will apply the same OASIS software as proposed used to couple ARPEGE and MICOM


1.4.4            Models for radiatively active contaminants

Three-Dimensional Chemistry Transport Models.

A global 3-D Chemistry Transport Model (CTM) developed at the University of Oslo (UiO) is used in RegClim. This model has been developed to study tropospheric ozone distribution and changes (Berntsen et al., 1996; Berntsen and Isaksen, 1997; Jaffe et al., 1997). The model uses GCM generated wind fields (GISS) for 1 model year. The model has been run for several years with the full diurnal chemistry, reproducing global distributions of ozone NOx and CO that are in agreement with observations of the latitudinal, longitudinal and seasonal distributions in the lower troposphere. There is however a tendency like in most CTMs to underestimate the concentrations in the upper troposphere, particularly the NOx distribution (Jaffe et al., 1997). The chemical scheme is described in Berntsen and Isaksen (1997). The diurnal variation of approximately 50 chemical compounds are calculated with time steps of 20 minutes. Species in the oxygen, nitrogen, hydrocarbon (including CH4 and CO) families are included.


A new version with higher resolution of the global 3-D CTM (Oslo CTM2) has been developed at the Department of Geophysics by Sundet (1997). This new version is the basic model for global off-line atmospheric chemistry calculations in RegClim for testing and developing new chemistry schemes. It uses the same chemical scheme as in the coarse resolution model. It uses wind velocity, subgrid processes, temperature, humidity and surface pressure data that have been extracted from the ECMWF model. The initial data from ECMWF are extracted on a resolution corresponding to T63, and the CTM can be run with either T21, T42 or T63 resolution, corresponding to horizontal resolutions of 5.6, 2.8 and 1.9 degrees respectively. The CTM is set up with the same vertical resolution as the (optional) 19 layers version of the ECMWF model.


In connection with aerosol-calculations, a limited-area CTM for major parts of the northern hemisphere (e.g. Iversen et al., 1998; Seland and Iversen, 1999). The model uses isentropic co-ordinates, a grid-resolution of 150 km, and uses input meteorological data, e.g. from ECMWF. The model has been used in several studies such as Arctic Haze (Iversen, 1989) and transatlantic sulphur transport (Tarrason and Iversen, 1992 and 1998).


Short wave radiative transfer models

At solar wavelengths a model using the discrete ordinate method (DISORT, Stamnes et al., 1988) will be used. The model is a multi stream model and can be used for various numbers of streams depending on the accuracy needed in the calculations. The spectral resolution can also be varied, with 1 nm as the highest resolution. Clouds are included in the model, allowing three cloud layers.


Global distributions of temperatures and humidity will be taken from ECMWF in the calculations of the radiative forcing. Data for cloud amount and optical depth, as well as albedo, will be taken from International Satellite Cloud Climatology Project (ISCCP). Variable horizontal resolution can be used with 2.5° in longitudinal and latitudinal direction as the highest resolution.


Long wave radiative transfer models

Two models for thermal infrared radiation will be used, namely a broad band model (BBM) and a Line-by-Line (LBL) model.

The (BBM) includes about 50 bands among all the trace gases of importance for modelling of the terrestrial infrared radiation (Stordal, 1988; Myhre and Stordal, 1997). For ozone the main band at 9.6 mm and the spectroscopically weaker band at 14 mm are included. Clouds are included in the thermal infrared scheme. Three cloud layers are used in the radiative transfer calculations.


The BBM is compared to several LBL models, and in general the BBM model was in good agreement with LBL models, in fact in most cases within the variations between the individual LBL models (Myhre and Stordal, 1995). The longwave radiative transfer model uses the same data set of temperature, humidity, and clouds as the shortwave radiative transfer model in the calculations of the radiative forcing.


The LBL model is developed by Edwards (1992), and is used to calculate optical depths. The water vapour continuum is from Clough et al. (1989). HITRAN 1992 and HITRAN 1996 absorption data are used in the calculations. A model using the discrete ordinate method (Stamnes et al., 1988) is used to calculate radiative fluxes from the optical depths calculated with the LBL model.


AGCM for regional forcing studies

The final aim so far in RegClim for the study of the effects of regionally distributed radiative forcing, is to use time-slice runs with a pure atmospheric global climate model. For this purpose we use the NCAR CCM (presently an extended version of CCM3).




1.5            Interpretations in relation to the
Climate System


1.5.1 Conceptualisation of key properties of the climate system: The non-linear perspective.

The question if man influences the course of climate by perturbing the earth’s radiation budget (radiative forcing), is difficult to answer for at least two basic reasons: the climate system is highly non-linear so that processes on all scales feed back on each other, and important compartments of the climate system are intrinsically unstable. Due to the non-linearity almost any simplification of the governing equations for the climate system, will soon influence parts that are not simplified during a climate scenario calculation. The intrinsic instability precludes predictability due to the critical dependence on some assumed start condition. This leads to unpredictable variability on very different time-scales (natural variability, chaos, “noise”), which may mask any possible signal caused by e.g. anthropogenic radiative forcing (or any other type of forcing for that matter). Since the equations unavoidably include simplifications,  (numerical truncation, sub-grid scale parameterisations), the instabilities will be realized even if the start-conditions were perfect, since the approximations will cause errors that very soon grow to become comparable to typical errors in start-conditions.


The combination of non-linearity and intrinsic instability thus leaves us with a signal-detection problem that will have to involve statistical terms, and “climate change” can never be “proved” but at best be estimated at some level of statistical confidence. In order to understand and interpret the changes and variability that are observed in long climate observation series or calculated by models, quite sophisticated statistical tools may have to be used. Furthermore, ensembles of scenarios need to be run in order to cover the expected range of natural variability over selected periods (e.g. Tett at al., 1999; Hulme et al., 1999; Delworth and Knutson, 2000).


Another consequence of the non-linearity combined with model imperfections is that the usefulness of climate scenario calculations will depend on one’s conceptual ideas of the climate system. Much discussion amongst scientists is caused by their expectations on how the climate system should respond to changes in radiative forcing. For example, methods have been developed to detect anthropogenic signals from “noise” based on expected patterns or “fingerprints” (Hasselmann, 1993) that are presumed to bear some resemblance with the forcing patterns. Others claim that due to the system’s non-linearity and the relatively small external forcing caused by man, the response should not be expected to deviate from the natural modes of variability even if the forcing does (“the non-linear perspective” Palmer, 1999; Corti et al., 1999)


An important consequence of non-linearity is that the degree of instability (i.e. the speed at which small differences grow) depends on the actual state of the system. It is well known that weather forecasts under certain conditions can be made up 10 days with the same accuracy as it can be made up to only three days under different conditions. In other words, the predictability varies over the system’s attractor. Similarly, the sensitivity of the system for small perturbations in external forcing (compared to the total natural forcing) also varies over the system’s attractor. A consequence of this is that there tend to be an un-even frequency distribution of system states, with a tendency of clustering around states with small sensitivity and high predictability. The climate system has a tendency to develop flow regimes, in which the system tends to stay frequently. There are several examples of such flow regimes (see e.g. Wallace and Gutzler, 1981), and the system’s state may quickly flip between these, sometimes predictably but often not. Important regimes that involve atmosphere and ocean are ENSO in the tropics, and NAO (or AO, Thompson and Wallace, 1998). In the oceans, paleo-climatic data indicate that the deep worldwide thermohaline circulation may switch between regimes. A consequence of the varying sensitivity for changes in external forcing is that very little will happen with states in the regimes, whilst for states in unpredictable transitions between regimes, the sensitivity is high and the influence may be that the system switches regimes in a different manner.


Hence the response of a small change (e.g. anthropogenic) in forcing will probably be a changed frequency distribution of states in flow regimes that are otherwise unaffected. This is Palmer’s (1999) non-linear perspective. Furthermore, since the linearised equations that describe growth of small initial differences (linear instability theory) involve non-normal Jacobian operators, finite-time growth rates can be much larger than predicted by normal modes (exponential growth) as well as involving up-scale developments. Thus, the sensitivity patterns for external forcing that may cause a selection of one specific flow regime, may not at all resemble the regime pattern itself, but be spatially much more confined and to quite different regions as information is propagated in space over optimal finite time-intervals. Hence “fingerprints” for impacts of small changes in external forcing should be sought as fields resembling differences between flow regimes that should be expected to gain in frequency and those that should be expected to lose frequency. In order to pre-calculate such “fingerprints”, sensitivity patterns can be estimated by calculating the adjoint to a tangent-linear model  (e.g. Corti and Palmer, 1997), or by calculating “forcing singular vectors”.


In RegClim we tend to adopt the “non-linear dynamical perspective” of Palmer (1999), and will constitute (at least until further notice) our main conceptualisation of climate variability and change.


One consequence of this view is that limited-area, regional climate modelling may be of limited value, since major shortcomings over a region due to errors in GCMs may well be caused remotely. Furthermore, one should expect such errors due to the tendency of smaller scales in the forcing sensitivity patterns than in the flow regimes themselves (Corti and Palmer, 1998). The view is therefore well compatible with RegClim, which uses global models to improve forcing calculations (Overall Aim II), even though regionalisation is the original purpose (Overall Aim I).


The view is also in agreement with the recent paper by Corti et al. (1999), which conclude that the recent Northern Hemisphere warming is more directly related to the thermal structures of natural atmospheric circulation regimes than to any anthropogenic forcing pattern itself. This lack of change in the regimes themselves cannot be used as an evidence of no anthropogenic effect on climate. The anthropogenic impact must be investigated by comparing the changes in the frequency distribution with “natural variability” of the frequency distribution.


1.5.2            Compartments of the climate system

The Atmosphere

There is a long tradition in meteorology to define flow regimes in accordance with significant weather and its predictability. Well known examples are blocking and the high and low zonal index circulation modes (the index cycle). A number of regimes for the Northern hemisphere were systematically grouped by Wallace and Gutzler (1981) using simple flow indices and spatial autocorrelations for teleconnection patterns. Examples are sectorally defined flow regimes such as the Pacific-North-Amerian (PNA) pattern that correlates with ENSO, and the North-Atlantic Oscillation (NAO). Hemispheric regimes, which were defined by clusters of states in a low-dimensional phase-space of leading EOFs, were selected by Molteni et al.(1990). (Empirical orthogonal functions, EOFs, were originally introduced into meteorology by Lorenz(1956)). Hemispheric flow regimes are, however, somewhat controversial since that they seem to lack physical basis, and that they occur mainly by coincidence in a truncated phase-space (e.g. Wallace, 1996). Nevertheless, quite recently there seems to be quite good evidence of annular modes extending deep into the stratosphere defined on the basis of EOFs. These are the Arctic and Antarctic Oscillations (Thompson and Wallace, 1998: Baldwin and Dunkerton, 1999). The Arctic Oscillation (AO) seems to encompass the NAO, in that a strong stratospheric winter vortex is connected with a high NAO-index and v.v. Evidence of northern hemispheric flow regimes are also found for GCMs (e.g. Robertson et al., 1998), but due to model imperfections one must expect modelled and observationally based flow-regimes to be different. Furthermore, EOFs may also be different, a problem that can be circumvented by employing common EOFs.


Observed global warming in the later part of the 20th century is far from spatially homogeneous, but shows a clear regional pattern known as the “Cold Ocean Warm Land”-pattern (COWL) (IPCC, 1996). Also the free troposphere does not seem to have warmed with the same rate as on average on the ground, a feature that has generated much discussion about the realism of anthropogenic climate change. Climate models does not seem to give the same vertical structure of the warming, a feature that may be due to model imperfections. Corti et al (1999) explained these patterns in terms of some flow-regimes becoming more frequent at the expense of others, but that the structures of the regimes are basically unchanged. The difference in thermal structures between these regimes determines the realisation of the climate system response in terms of warming or cooling, and this difference may have little resemblance with the forcing pattern causing the change.


In RegClim the focal interest is the European-North-Atlantic sector for which NAO is a dominant mode of variability. A natural way of interpreting model results with a view to understand possible climate change in this region, may be to compare the structure of modelled regional regimes with observed ones, to check if changes in modelled regime populations agree reasonably with observed changes, and finally how regime populations may change under future emission scenarios.


To identify causes of regime-structure errors, a powerful method would be to use the adjoint of the tangent-linear version of the climate model to find the sensitivity pattern for the difference between model and observation (to the extent that errors are small enough for linearity to be approximately valid). To estimate the forcing patterns that are prone to release changes in regime populations in our region, regionally projected forcing singular vectors calculated by using the above-mentioned adjoint could be estimated. The projections of calculated radiative forcing on leading, locally projected, forcing singular vectors will then give a good estimate on expected response in our region. Simultaneously, it will be seen to what extent a regional climate model will be controlled remotely, and thus limit the expected improvement that can be obtained by regionalisation.


Unfortunately, such tools are presently not available for full AGCMs. However, we have available a quasi-geostrophic model with simplified forcing of T21- and with three layers resolution (Marshall and Molteni, 1993; Molteni and Palmer, 1993) that have all these features, including forcing singular vectors (Barkmeijer, ECMWF, personal communication). The model is the same as used by Corti and Palmer (1998) for estimating forcing sensitivities. This can be used as a tool to investigate the above-mentioned important problems to a larger extent than done before.


Amosphere-Ocean Interactions

The role of the oceans in extra-tropical, climate variability has been studied and widely debated since the 1960s. Three main paradigms are possible: the atmosphere responds passively to the ocean generated SSTs, the ocean responds passively to the atmospheric forcing, or the atmosphere and ocean are both active in producing the variability. The first of these ideas has been widely used in the tropics where it is believed that the boundary forcing of the atmosphere plays a significant role. Because of the strong internal atmospheric instabilities, this idea is less applicable to the mid-latitudes.


After examining interannual and inter-decadal SST and Sea-Level Pressure (SLP) correlations, Bjerknes (1964) proposed that the passive ocean paradigm was valid for inter-annual time scales, but also that basin-wide coupled ocean-atmosphere processes were necessary for longer inter-decadal time scales. Bjerknes noted, for inter-annual time scales, that the westerly jet showed local negative correlations with SST anomalies, and hence hypothesised that the SST anomalies were the result of the atmospheric fluxes on the oceanic mixed-layer. This has been confirmed for the central and eastern North Atlantic in coupled model experiments (O'Brien and Chassignet, 1995), and in a recent study that used fluxes based on COADS observations from 1952-92 to force a simplified oceanic mixed-layer model (Frankignoul et al., 1996).


For inter-decadal time scales, Bjerknes found basin-wide positive correlations, and hypothesised that this was due to changes in the poleward heat transport of both the atmosphere and the North Atlantic Ocean.  Bjerknes (1964) looked for SST patterns associated with the NAO. Kushnir (1994) found different MSLP/SST anomaly patterns for time-periods longer than, or shorter than, a decade. The shorter time-scale pattern was similar to that found previously by Deser and Blackmon (1993), with the pressure pattern being similar to the W.Atlantic pattern of Wallace and Gutzler (1981). Warm SSTs went with a surface easterly anomaly and a high to the north. This was like the barotropic response found by Palmer and Sun (1985), and Lau and Nath (1990). However, it differed from the steady, linear, baroclinic extra-tropical response to low-level heating given by Hoskins and Karoly (1981), and found in a range of prescribed SST AGCM experiments by Kushnir and Held (1994). In contrast, the longer time-scale pattern of Kushnir (1994) had an MSLP low downstream of the warm SSTs just as in the idealised models. The different nature of the thermodynamic and dynamic balances in the atmosphere, the surface fluxes, and the roles of the transient motions in the two cases are presently not thoroughly understood.


The ocean

During the last four decades, we have experienced two extreme states of the climate in the North Atlantic sector of the northern hemisphere (Hurrell, 1995). This is strongly linked to the large-scale shift in atmospheric mass between the Azorean high and the Icelandic low, known as the North Atlantic Oscillation (NAO). Recently it has been demonstrated that NAO probably is linked to annular mode-shifts of the Arctic Oscillation (AO) (Thompson and Wallace, 1998). From record low values in the 1960s, the NAO index, based upon the difference in sea level pressure between the two poles, has shown a steep increasing trend towards the highest values ever recorded during the 1990s.


Only recently one has become fully aware of the large connections between the NAO and the oceanography of the Nordic Seas and the Arctic Ocean. There has been a suppression of the deep-water convection in the Greenland Sea (Dickson et al. 1996), which has resulted in a warming of the deep waters and a reversal of the deep ocean flow between the Greenland Sea and the Norwegian Sea (Østerhus and Gammelsrød 1999).  In the upper water, the front position between the water masses of polar and Atlantic origin has moved eastwards (Blindheim et al. 1999), and the Norwegian Atlantic Current (NWAC) seems to have become more narrow (Mork and Blindheim 2000) and warmer (Furevik 2000a).  An increased volume and heat transport to the Arctic Ocean may explain why the AW layer in the Arctic Ocean has become warmer and thicker (Carmack et al. 1997; Swift et al. 1997), and the cold halocline layer, which isolates the ice from the warm waters beneath, has become thinner  (Steele and Boyd 1998). Recent reports by Rothrock et al. (1999) and Johannessen et al. (1999) show that both the horizontal extent and the thickness of sea ice are decreasing rapidly.


A major question is whether these developments are natural fluctuations of AO that will oscillate back to the opposite phase, or if the anthropogenic forcing has forced the system into a state with low probability of swing-back. RegClim can certainly not give the final conclusion to this problem, but the plan is to contribute significantly to a future answer.

1.5.3 Tools and methods

Statistical methods and tools are needed to diagnose possible flow regimes in different parts of the climate system. During later years, a number of papers have been published applying different methods (SVD, CCA, MVR and EOF analysis combined with correlation or covariance maps) for finding coupled patterns in observed climate data and/or model results (e.g. Bretherton et al. 1992, Zorita et al. 1992, Santer et al. 1993, Busuioc et al.1999). EOF or, when comparing model results and observations, common EOF (Benestad 1999c), form an excellent basis for classification of flow regimes. In a low-dimensional phase-space defined by EOFs, regimes can by estimated by cluster analysis, by rotation of the EOFs (Huth, 1996), or by estimation of probability density functions (PDF) (Corti et al. 1999).  Estimated PDFs should be accompanied by tests for statistical significance in order to establish the realism of possible multiple regimes.  Complex EOFs (CEOFs) can be used for detecting propagating features as well as standing oscillations (Horel 1984).  Wavelet analysis has also been demonstrated to be a useful tool for climate signal detection (Lau and Weng 1995).


Estimates of forcing patterns that optimally trigger regime switches as defined by forcing singular vectors, as well as estimates of forcing sensitivity as defined by the adjoint of selected patterns, can now be approximated with a global T21L3 quasi-geostrophic atmospheric model. The tool can be run on workstations.


RegClim’s contribution

In RegClim, advanced statistical and dynamical tools will be used for the following two purposes:

·        improve the interpretation of results from RegClim’s own model simulations in view of the non-linear dynamical perspective (atmospheric models, ocean models, air-sea-interactions, radiative forcing due to anthropogenic aerosols and tropospheric ozone);

·        study the sensitivity and response of prevailing atmospheric flow regimes in the RegClim region w.r.t. changes in the radiative or oceanic forcing.