An Earth system model of intermediate complexity is developed for the study of climate dynamics on decadal and millennial time-scales. The main task is to identify the driving mechanisms and potential thresholds responsible for climate transitions. Emphasis is placed on the multiple states of the system and the interaction of the dominant pattern of atmospheric variability with the ocean and land surface. In contrast to conventional time-slice experiments, the present approach is not restricted to equilibrium transitions and is capable to utilise all available data for validation. Transient simulations for the past and future climate will examine feedback mechanism in the climate system. Pathways of orbital forcing into climate response will reveal the understanding of climate records. The model explicitely resolves the three-dimensional atmospheric and oceanic dynamics and is therefore conceptual different from statistical dynamical models.

At time-scales comparable to the glacial-interglacial cycles, the importance of each Earth system component has not been estimated. At this early stage of exploration for the coupled system, it is necessary to be able to carry out a large number of sensitivity experiments. It is expected that, compared to e.g. scenario calculations for the next century, different processes will dominate Earth system interactions at such long time scales. This has to be accounted for a suitable modelling framework in order to allow cooperation between expert groups of different focus. The Earth system model shall have a modular structure with portable Fortran code. The modular structure of the model enables the user to modify model configurations according to the application. New schemes and model components can be included with a minimum of technical expense which is particularly useful for paleoclimatic applications.

The atmospheric model has been coupled to the ocean model LSG [Maier-Reimer et al., 1993] designed especially for long-term climate studies (time step of one month). This model contains a parameterization for the bottom boundary layer [Beckmann and Döscher, 1997] which drastically improves the density-driven downslope flows [Lohmann, 1998] and is essential for the interpretation of paleoclimatic records [Lohmann and Schulz, 2000].

Thermodynamic sea ice model including a simple momentum balance for advection of sea ice.

In the LSG ocean model , the HAMOCC carbon cycle model [Maier-Reimer, 1993] is included. First experiments exist for the terrestrial biosphere model LPJ (Lund-Potsdam-Jena) [Sitch et al., 2000], which is forced with the PUMA-LSG climate. The LPJ model is a dynamic global vegetation model combining mechanistic treatments of terrestrial vegetation dynamics, carbon and water cycling.

The present version of the model uses a T21/L5 resolution for the atmosphere and 11 vertical levels with horizontal resolution of 5 degrees for the ocean. The model contains representations of sub-grid scale fluxes over land, ice, and open sea. The coupled atmosphere-ocean-sea ice model does not require flux adjustments.

• Langranian tracers are implemented in the atmosphere [Hardenberg, 2000; Bagliani et al., 2000].
• A sediment model is included in the marine carbon cycle [Heinze et al., 1999].

Limitation of the current model version is due to the missing feedbacks by cryosphere and vegetation. Further limitations are due to the simplified radiation codes used (chemistry and dust). The coarse resolution and the neglected non-linear terms in the oceanic momentum balance restrict studies to spatial scales of more than a thousand kilometers, the model cannot adequately resolve interannual climate variability in the tropical Pacific.

The required CPU time for the coupled atmosphere-ocean model is less than 8 min per year on the present Cray in Hamburg, i.e. approximate two orders of magnitude faster than complex GCMs. In the asynchronously coupled mode, where atmosphere and ocean model components take even parts in computing time, about 1000 years of model integration can be performed on one day.

The model components have been extensively tested in studies of paleoclimate [Winguth et al., 1999], future climate change scenarios for the next century [Lunkeit et al., 1998], and storm track variability [Frisius et al., 1998]. The model is flexible enough to turn various feedback processes off and on, to study the cause and relationships of the climate components. Simulations and sensitivity studies will focus on the following topics:

• Process of glaciation and deglaciation.
• Interaction of vegetation, atmospheric dynamics, and the oceanic circulation (e.g. vegetation-snow feedback, vegetation distribution).
• Diagnostic and prognostic calculations of atmospheric CO₂ (including carbon isotopes) on long time scales.
• Importance of northern versus southern hemispheric forcing and Atlantic-Pacific teleconnections.
• Dependence of climate variability on the background state.
• Simulation of the last glacial cycle.