This document may be downloaded as a pdf file

The full title of the project is: "Project On Tendency Evaluations using New Techniques to Improve Atmospheric Long-term Simulations".

The short title (acronym) is: POTENTIALS

There are three main objectives within the project:

1. Use of high quality ECMWF Re-Analyses (ERA) data to construct space-time maps of the tendency or forcing errors in four GCMs. The maps of forcing errors should be more usefull as a guide when improving the GCMs than the more traditional long term mean systematic errors, i.e. differences between observed climatology and multi-year model simulations.
2. Reduce the models forcing errors by making changes in the physical parameterization schemes and by parameterizing the forcing errors empirically. Reduction of the forcing errors of a general circulation model can be a powerful diagnostic tool for improving current GCMs, e.g. Klinker and Sardeshmukh (1992).
3. Investigate the benefits of a reduction of model forcing errors with special attention to the following two applications:

• Simulation of regional climate, particularly over Europe. Since global climate models are used to drive regional limited area high resolution climate models, errors in the former are directly transferred to the latter. Also variable resolution global models show similar systematic errors over their high resolution regional area of interest. These induced systematic errors are the main reason for large biases over Europe in temperature and precipitation simulated by the regional models. For this application, guided by the forcing error maps, changes which as far as possible eliminate the systematic errors will be made in physical parameterization schemes of GCMs.
• Seasonal prediction. Reducing the long term systematic errors is very important for dynamical prediction of seasonal climate. This is because most of the long term predictive skill has its origin in certain situations of anomalous forcing at the lower boundary of the atmosphere, e.g. by sea surface temperature (SST) anomalies related to El Niño/Southern Oscillation (ENSO). The prediction of the (remote) atmospheric response to such lower boundary forcing is critically dependent on errors in the three dimensional structure of the atmosphere.

Summarizing, the goals of the project are to improve the simulation of regional aspects of climate in climate models and to improve the capability for producing seasonal forecast.

The work will to some extend have an iterative nature and will include five main tasks

1. Construction of tendency errors (section 2.2.1),
2. Re-tuning/improvement of physical parameterization (section 2.2.2)
3. Climate simulations with re-tuned models (section 2.2.3)
4. Empirical parameterization of forcing errors (section 2.2.4)
5. Simulations including empirical parameterization of forcing (section 2.2.5)

Section 2.2.6 provides a breakdown of the work into a number of subtasks.

As mentioned above four GCMs (ARPEGE, ECHAM, PE and QG) of different complexity will be used and compared in the different tasks. An overview of these models can be found in section 2.2.7.

An observed tendency vector () of the prognostic variable F (observed state vector) appearing in a given atmospheric model can formally be split into two terms:

 where is a tendency as it is modelled, given the instantaneous observed atmospheric state and the state of the lower boundary of the atmosphere, e.g. Sea Surface Temperature (SST), temperature and humidity of the soil including possible snow cover etc. The residual vector R is the difference between the two tendencies - a number which generally is much smaller than .

Assuming no errors in observations (analyses) and exact calculation of observed tendencies, the size and time/space structure of R reflects all model deficiencies. In addition to its dependency on the given model, it is obvious that the so calculated R will depend also on the quality and homogeneity of the analyses used. Furthermore, if there are little or no data in a given geographical region, R will merely reflect differences between the model used for data-assimilation and the actual model rather than model errors.

By calculating initial model tendencies for a large sample of observed, initialised atmospheric states it is possible to infer a temporal average value of R which will balance the average of calculated tendencies (Klinker and Sardeshmukh, 1992). This may be called the direct residual method. The residual tendency method will in the present project be applied for the PE model to obtain seasonal average values of R, which will be compared to similar long term averages of R calculated via the nudging method (see below). This comparizon may ease detection of possible problems discussed below related to vertical and horizontal interpolation/truncation of the analyses to the model grid.

A direct way of calculating instantaneous values of R is to

1. take the raw analyses,
2. interpolate the analyses to the horizontal and vertical resolution of the actual model to be used,
3. calculate observed tendencies using a high order time differencing scheme (possibly local Fourier decomposition),
4. calculate the model tendency at the analysis times by taking a single time step with the model based on the interpolated analysis of the model variables (2), and finally
5. subtract from

to obtain R.

There will, however, be problems when R is calculated this way, particularly in models based on primitive equations allowing for gravity waves and in models including moist processes. This is because the interpolation (item 2) can introduce dynamical imbalances between the mass and velocity fields and also moisture spin up problems are introduced.

It will be tested how much these problems can be reduced by the use of a nudging technique recently tested by Jeuken et al. (1996) for a T21 version of ECHAM4. During nudging item 4 is changed as follows.

Instead of model tendencies calculated from model variables obtained directly from the interpolated re-analyses, tendencies are extracted from a continuous model integration in which the model variables, by a Newtonian relaxation, are nudged toward the analysed values, interpolated in time as well. Thus, in the continuous integration the interpolated analysed values are gradually assimilated into the model. The impact of relaxing variables dominating in Rossby modes (vorticity and surface pressure) strongly to the analysed values and variables dominating in gravity modes (divergence and temperature) less strongly, as suggested by Jeuken et al. (1996), will be investigated. In that way less gravity waves are excited and those which are will be damped gradually by dispersion and dissipation mechanisms in the model (geostrophic adjustment like in dynamical initialisation). Focusing at problems with moisture/precipitation spin up, similar experimentation will be made with the strength of the relaxation of humidity, including no relaxation at all. The nudging technique will due to its simplicity be used as the basic tool to obtain the forcing errors R, and it will be tested for all four GCMs. The nudging will generally be applied in spectral space. This allows the use of different relaxation times for different spatial scales.

A disadvantage of this standard nudging technique is, however, that the state variables used to compute model tendencies and thereby the forcing errors are not exactly equal to the interpolated analysed values which should be the most realistic ones. In order to try to alleviate this, an alternative approach building on diabatic non-linear normal mode initialisation and 100% relaxation will be tested. Instead of using "observed" precipitation (see e.g. Kasahara and Mizzi, 1986), it is possible, with historical data as dealt with here, to use "observed", i.e. analysed, temperature tendencies. By this technique one may expect to get realistic divergence fields and thereby vertical velocity fields for the T42 resolution. This is a first pre-requisite for an elimination of the spin up problem. A second necessary condition is that the moisture fields are realistic. As the analysed fields are most likely not, they will not be used. Instead it will be investigated if the model itself can establish realistic moisture fields when that field and the other prognostic variables involved in the hydrological cycle are predicted by the model (no relaxation as mentioned above) whereas the remaining prognostic variables are determined completely by the diabatic initialised re-analyses, smoothly interpolated in time. If that is not the case other possibilities will be investigated. Anyway, the non-hydrological prognostic variables will be determined completely by the diabatic initialised re-analyses, i.e. by a 100% relaxation to these analyses. Thus with this approach it is avoid that model influences decrease the realism of the non-hydrological variables. The test of diabatic NNMI and 100% relaxation will only be carried out for ECHAM4.

Another method for computing the model forcing error is the use of the adjoint model and a variational approach. The method consists in performing a short model integration, starting from an "observed" model state F, and obtaining a forecast of the next available "observation" of F. At this point, with the use of the adjoint method, a value of R can be estimated so as to minimise the difference between the model forecast and the final observed value of F. This allows to build a time series of R that is more model consistent. The method - which will only be tested for the QG model where the adjoint is already available - allows for a direct comparison with the nudging technique, as well as for an estimation of the time constants that must be used in the Newtonian relaxation. The adjoint method will also be applied when searching for a parameterization of the QG model residual - see section 2.3.2.

In addition to the caveats discussed above, artificial observed tendencies can be introduced due to errors and/or in-homogeneities in the analyses. These may arise from errors and changes of the assimilating atmospheric model and the analysis scheme used. For these reasons the re-analyses prepared by the ECMWF (ERA-data) will be used, since these data in this respect are homogeneous. The ERA-data are available from 1979 and onwards, every 6 hour at resolution T106 and with 31 hybrid levels in the vertical.

One must be aware, however, that some in-homogeneities in R will exist even in the ERA data. These are introduced via the raw observations entering the ERA data-assimilation system. The observations are not homogeneous since new conventional instruments and in particular new types of satellite data have been introduced over time. As an example of in-homogeneity problems on short time scale, it is noted that the number and kind of observations is different at the different analysis hours (00, 06, 12, 18 UTC).

The identified forcing errors will be used as a guide to improve and tune the physical parameterization in the 4 GCMs listed in section 2.2.7. It will, as described in section 2.2.3, be tested in long runs if the modified parameterization has lead to improved simulation of climate. The work will have an iterative nature involving model corrections, identification of forcing errors, simulation of long term mean climate, new model corrections, and so on. In this way a diagnosis and correction of the errors in forcing which cause the systematic errors in the flow patterns of the model simulations will be attempted.

The work on improvement of the physical parameterization will involve testing of individual parameterization schemes as well as entire packages.

There are reasons to suspect the forcing in the tropics to be wrong and thereby being a main reason for the systematic errors in the flow, e.g. that over Europe seen in ECHAM4 and ARPEGE. This suspicion is based on recent sensitivity experiments, i.e. Chase et al. (1996) where the leaf area index (LAI) was changed in the NCAR CCM2 resulting in anomalous flow very similar to the systematic errors in the ECHAM4 model. The LAI was mainly changed over the tropics and in the southern hemisphere and apparently it was the forcing differences in the tropics which caused flow differences in the Northern Hemisphere. These experiments suggests that modification/correction of tropical forcing, e.g. from release of latent heat, may be very important in reducing the systematic errors in the ECHAM4 and ARPEGE. Inspections of the three dimensional structure of the forcing errors combined with sensitivity experiments as described above will be used to find out which schemes needs improvements.

In addition to the changes (made within the present project) in the parameterization schemes, with special weight to tropical heating, also the changes developed and tested in the MERCURE project (see section 5) for regional climate models (RCMs) will be introduced and tested.

The horizontal diffusion in spectral GCMs takes place in spectral space and is usually formulated as a simple damping of the amplitudes of the different expansion coefficients. The actual choice of damping of a given wave component is, however, to some extend arbitrary and mainly introduced to cure in the best possible way the so-called spectral blocking. In the present project the spectral information in R will be used to tune and optimise the horizontal diffusion in climate models in a way which is strongly bound to physical reality. This is easily done since the amplitude of each wave component of R in principle is equal to the error in the models original spectral damping. Of course the resulting damping will vary with time, and to obtain a new set of damping coefficients for use in the model, averaging over a reasonably long period and different seasons is needed.

In addition to horizontal diffusion, the study of Klinker and Sardeshmukh (1992) will be re-considered in the context of a model which has not been used as data-assimilation model. In practice R calculated by the nudging technique is used to study and eventually reduce possible errors in the parameterization of the momentum equations, e.g. gravity wave drag.

From a theoretical point of view a direct way of minimising the forcing error at different time scales is to make use of the variational tuning approach. This will, however, only be done for the QG model, of which the adjoint formulation is already available. The use of this simple model also permits to simulate model errors in a perfectly controlled environment, by, for instance reducing the resolution or modifying deformation radii. It is intended to try such experiments, and validate error-parameterizations over a large set of data.

The effect of the choice of different optimisations on the climatology and variability of the model will be checked - see section 2.2.3.

State of the art atmospheric climate models

To be able to produce sufficiently reliable simulations with RCMs in the future it is essential that the systematic errors in the flow and pressure pattern of GCMs are reduced, which most likely must involve improvements in the parameterization schemes that influence the forcing in the tropics - as noted above. It is thus a main objective of the project to improve the simulation of regional climate in GCMs, possibly progressing iteratively towards this goal as mentioned in section 2.2.2.

Note that the emphasis will not only be on regional climate over e.g. Europe since it is important to avoid improvement of the model in one respect at the expense of others. For this reasons a series of standard statistics will be analysed after each simulation. These include global mean budget statistics, zonal mean fields and transports as well as global maps of seasonal averages and temporal variability of atmospheric variables, outgoing long wave radiation, etc.

To identify how much the systematic errors are reduced in the re-tuned and parameterized models ARPEGE and ECHAM4, a number (depending on the needed degree of iteration) of simulations will be performed. The length if these will depend on which tuning or parameterization is being tested. Some of these simulations may be in "perpetual mode" and some with full annual cycle including long AMIP II simulations, i.e. 17 year simulations with "observed" SSTs and sea-ice distributions. For ECHAM4, a final AMIP II simulation will be handed over to the MERCURE project (see section 5). Then the MERCURE project will validate the improvements of their RCMs by performing climate simulations forced with boundary conditions from this AMIP II.

The PE and QG-models are much simpler than ARPEGE and ECHAM4 but they do include simple physical parameterization packages which will be objectively optimised in the same way as for ARPEGE and ECHAM4. For these models it is possible to perform very long runs (2000-3000 years) at relatively little computer cost. Based on such simulations the ability of the models to reproduce the inter-annual and inter-decadal variability will be tested as part of the optimisation procedure.

The objective of this part of the work is to test to what extent the model errors R can be parameterized empirically as function of the prognostic model variables and of the lower boundary condition, and in this way be fed back to the model as (small) correction terms to the prognostic equations. In this way a mostly dynamical but also somehow statistical model will be build which should have smaller systematic errors than is seen without the parameterization of R. Such a model should be very well suited for extended range forecasts that are considered to be only marginally dependent on the initial state of the atmosphere, but are very dependent on the local interaction with the lower boundary of the atmosphere and the atmospheric dispersion properties.

The most simple way of parameterizing its tendency errors R is to use the long term ensemble of R for a given season as a constant correction or forcing. Such constant flux correction will be tested and intercompared for the ARPEGE, PE and QG models (see section 2.2.7).

Both linear and non-linear methods of empirical parameterization of the temporal anomalies R´=R- of R will be considered. To the extent the relationships between R’ and the actual observed flow are linear it is obvious to consider a multiple linear regression based on techniques like canonical correlation analysis or singular value decomposition. The outcome will be a parameterization of the type R´= AF, where A is a matrix with dimensions determined by the dimension of R and of the instantaneous state vector F. It will be considered to use only severely spectrally truncated fields (e.g. T10), of R´ and F in this parameterization, since preliminary tests have indicated that this is beneficial in order to minimise noise problems. Linear empirical parameterization will be tested in models ARPEGE and QG.

It is obvious that the relationship between R´ and F can not be strictly linear and that it is needed to consider non-linear techniques of parameterization as well. The analogue method is a simple and straightforward but for this application probably powerful approach which will be tested for ARPEGE and QG (see section 2.2.7). In the project, weighted analogues of the instantaneous large-scale flow will be sought for within the longest available data bases where an average residual can be calculated by approximating the first-order moment of the conditional distribution P(R/X). Here, X generally is a truncated atmospheric-oceanic state vector containing for instance the current leading EOF coefficients, a limited number of spectral expansion coefficients or empirical normal modes. Second-order moments may also be considered by looking at the covariance structure of this conditional distribution. A GCM with analogue type empirical parameterization of R´ may, however, for some operational applications be connected with practical file-handling and storage problems. Therefore, although not straightforward, it has been decided to test the application of classical non-linear techniques of estimation of transfer functions such as radial basis functions or neural networks. The application of these techniques, although not straightforward, will be tested only for the QG model.

The dependency of R to lower boundary forcing which is of fundamental importance for seasonal forecasting will be included in the parameterization for ARPEGE considering both linear and non-linear techniques. A successive error reduction may be used by estimating first the fraction (R¹ ) which is linearly explained simply by correlating R (column-wise in the vertical) with e.g. oceanic temperature anomaly. Non-linear relationships should then be used to parameterize the remaining part (R’-R¹).

Furthermore, the sensitivity to lower boundary anomalies will be investigated in the PE model by estimating the residual from different data sub-samples in periods with large and opposite SST anomalies.

The predictors in the statistical parameterization may be of local or global scale. For this reason a more general investigation of optimal selection - including scale - of predictors for parameterizing the forcing will be done for ARPEGE and the QG model.

It is very important that the empirical parameterization is numerically stable when it is incorporated as a correction in the GCMs. For this reason the different types of parameterization with the options described above need to be developed under this constraint. Stability testing of the models will currently be made as the parameterization is developed. It is noted that there are strong conservative forces in the models which will counteract large excursions from the mean atmospheric flow. This means that a simple way of obtaining stability is to force R towards zero or a constant value when the flow becomes critically abnormal. Certain conservative constraints as mass and energy conservation will also be considered.

The effect of incorporating the empirically parameterized forcing error in the models will be tested for models ARPEGE, PE and QG. First long climate simulations will be performed to check that the model systematic errors have a magnitude between the "nudging mode" and the "free mode". Then seasonal forecasts will be performed to evaluate how much the forecast scores have been improved with respect to forecasts in "free mode".

Climate simulations in "empirically forced mode" and the "free mode" will be compared. The length of these simulations varies from model to model. For ARPEGE they will cover relatively few years while they will be much longer (100-1000 years) for the computationally very cheap PE and QG models. The results of the long simulations will be compared with the results obtained from long runs with optimised parameterization (section 2.2.3).

Seasonal forecasts which, according to present day knowledge, only are marginally dependent on the initial state of the atmosphere should benefit from the introduction of the empirical parameterization described in section 2.2.4. This is because it is anticipated very important to have as small systematic errors as possible in order to obtain a realistic response to lower boundary forcing.

Two different approaches can be used to improve the quality of dynamical seasonal predictions:

1. a posteriori or after-the-fact correction where the systematic error is subtracted from the forecast after the forecast integration is completed, and
2. within-forecast correction, where the effect of errors in the governing model equations are currently eliminated during the forecast integration.

It is the second approach which will be followed in the present project since the first is already being tested as part of the PROVOST project on prediction of seasonal climate (see section 5). There is an ongoing debate which of the two approaches will lead to the best results.

A series of experiments which in their experimental design are equal to those in the PROVOST project (see section 5) will be carried out in the project. In these, ensembles - each including 9 members - of seasonal predictions (hindcasts) will be made, with and without the empirically (within forecast) corrected ARPEGE forced with observed SSTs, i.e. it is assumed that the state of the ocean is perfectly predicted in these experiments.

It is a basic hope - to be tested in the project - that even a relatively simple model as the QG model can perform successfully in seasonal forecasting when a proper parameterization of R is introduced. This relies on the fact that only large-scale features of the atmosphere can be expected to exhibit any significant skill, and variations of boundary forcing itself is expected to be important. Hence, the atmospheric model a priori needs not to represent many degrees of freedom, but a proper parameterization of boundary forcing and unresolved-scale forcing.

Once a parameterization of the residual is achieved, there are still some (presumably independent of the flow) errors left, which can be modelled as a stochastic forcing whose distribution (and not its actual value) may vary with the flow. By generating random values pulled from this distribution, one builds a stochastic model which can be used to produce ensemble forecasts. One particularly interesting question is whether the spread of forecasts is larger than the spread obtained by perturbing initial conditions (Toth and Kalnay, 1993; Molteni et al. 1996), in which case the problem of ensemble prediction should be tackled by first using model perturbations instead of initial condition perturbations. It will also be investigated whether there is any relation between spread and skill within this context. These interesting questions will be addressed only within the framework of the simplest model (QG).

This sub-section provides a break down into tasks of the work described above. There are 5 main tasks and underlying sub-tasks.

1. Coding and testing the nudging technique in all models. Testing in short periods.
2. Coding and testing 100% relaxation of diabatically initialised ERA for ECHAM4 and compare with 1a and observed precipitation. Decide if this method or 1a) should be used.
3. Applying adjoint technique for the QG model and compare with 1a.
4. Coding and testing residual tendency technique in the PE model for each season and for selected periods of anomalous SSTs, and compare with 1a.

1. Compare forcing/flux errors as well as long term mean errors in a full new version of ARPEGE with the original version. The development of the new version is not part of the project.
2. Optimising damping mechanisms in ARPEGE, ECHAM4 and the QG model.
3. Using repeated forcing error estimations to improve physical parameterization in ECHAM4 and ARPEGE with main emphasis on tropical heating.
4. Testing/coding of improved parameterization obtained in co-operation with the project MERCURE in combination with improvements from task 2b and 2c.
5. Optimising the simplified parameterization in PE and QG, possibly using the variational approach.

1. Relatively short development runs will be carried out for all models to test basic elements as stability and check for coding errors.
2. A number of 17 year AMIP II simulations with ECHAM4 to be compared to climatology and to a previous ECHAM4 AMIP simulation. The final improved AMIP II will be given to the MERCURE project.
3. Climate simulations with ARPEGE.
4. Very long runs with PE and QG for investigation of natural ultra low frequency variability.

1. Climate simulations with constant flux correction identified from the nudging technique and the direct residual tendency (ARPEGE, PE and QG).
2. Development of and climate simulations with linear and non-linear parameterization of the (non constant) forcing error, i.e. dynamical flux correction.

1. Seasonal hindcast experiments with ARPEGE in dynamical flux correction mode to be compared with results obtained in the existing PROVOST project.
2. Test the potentials of ensemble forecasts with model noise in the QG model.
3. Sensitivity experiments with constant flux correction identified during specific periods of anomalous SST distributions (PE model).

ARPEGE:

The ARPEGE/IFS model is a state of the art atmospheric model developed jointly by ECMWF and Météo France/CNRM. It is used for climate research studies at several laboratories in France and elsewhere in Europe, among them DMI and CNRM. A detailed description of the model and its performance is given in Déqué et al. (1994). For the project it is planned to use the climate version 2 of the model in a configuration with 31 (identical to those in the ERA data-assimilation system) sigma/pressure hybrid vertical levels mainly distributed in the troposphere and the lower stratosphere. The horizontal truncation is at total wave-number 42 (T42).

The ECHAM4 is also a state of the art climate model developed at MPI in Hamburg, Germany. The model is used for climate studies at several universities and research institutions in Europe. The model is described in detail in Roeckner et al. (1996) and will in the present project be used by MPI and to some extend by DMI, in its standard configuration, i.e. T42 horizontal resolution and 19 sigma/pressure hybrid vertical levels.

This model is a primitive equation model as ARPEGE and ECHAM4 but it has only 5 vertical levels. The model is based on a dynamical core from GFDL (Held and Suarez, 1994) and includes two different parameterisation packages: a simple relaxation of temperature and wind towards prescribed equilibrium fields, and a set of physically based schemes for radiation, large-scale condensation, convection and surface fluxes which all are much simpler than them in ARPEGE and ECHAM4. This PE-model is developed by and will be used mainly by CINECA.

This is a quasi-geostrophic 3 level, T21 spectral model (Marshall and Molteni, 1993). The basic model, which is run at many places in Europe, includes a Newtonian relaxation of temperature (driving), linear drag of wind at 800 hPa and horizontal diffusion of vorticity and temperature. The version to be used in the project includes a simple physical parameterization of basic processes. It will mainly be used by LMD.

The milestones are identified by the end of the duration of the selected groups of tasks from section 2.2.6 shown in the table below

The last months of the project are mainly devoted to writing of scientific papers and preparation of the final project report. A project status report will be given to the Commission at the end of the first 12 month period.

Another milestone is a planned extended working meeting/workshop for all the participants to take place approximately at year 1.25, with a duration of 7-10 days.

Refering to the list of tasks and sub-tasks in section 2.2.6 the following scheme identifies the role of each partner:

The following diagram illustrates the planned schedule for the individual sub-tasks listed in section 2.2.6. The darkness of the shading indicates the level of activity on a given task, with dark shading corresponding to the highest activity.

It is obvious that there is a considerable degree of interdependence between the tasks that are carried out iteratively (tasks 1, 2 &3, and to some extend tasks 4 and 5). However, the outcome of each task is still a result in itself. The major goals - reduction of systematic errors on the regional scale and improved seasonal prediction capability - are relatively independent. MPI will take the lead and be responsible for fulfilment of the first of these goals while DMI and CNRM will be the main responsible partners for the second. The responsibility at the level of individual tasks follows the list provided in section 3.

The deliverables of the project will be new versions of the 4 GCMs which are improved in the sense that the forcing errors and the cancellation of errors in opposite direction errors are reduced simultaneously . The improvements will be obtained by applying

• re-tuning,
• updated or new parameterization, or
• dynamical flux correction

to the models where the last item only is relevant for seasonal prediction. Furthermore the project will deliver:

• Validations of the reduction of systematic errors in simulations of the general circulation with the improved models and estimates of the improved capabilities of the GCMs to simulate regional climate. Specifically, data from a final and hopefully improved AMIP II simulation will be delivered to the regional climate modelling community (see MERCURE, section 5) for application and validation over Europe.
• Estimates of the importance of reduction of forcing errors in relation to seasonal prediction relative to correction of the forecasts themselves as in the PROVOST project (see section 5).

Results obtained in the project will to the greatest possible extend be published in international peer-refereed journals. Participants in the project will be encouraged to participate in international scientific conferences, e.g. the General Assembly of the European Geophysical Society.

An internet home page will be set up which describes the project and its current status.

Seasonal prediction of climate is one of the main themes in the present project. There is, however, no overlap with the existing PROVOST project (Contract no. ENV4-CT95-0109), since the method of (dynamical) flux correction undertaken in POTENTIALS is not part of PROVOST. Actually, POTENTIALS constitutes a scientifically and practically very interesting complement to the methods being tested in PROVOST for improving seasonal hindcasts, i.e. predictions with prescribed SSTs and sea ice.

As improvement of the simulations of seasonal climate in GCMs is part of POTENTIALS it has important relationships with the MERCURE (Modelling European Regional Climate: Understanding and Reducing Errors, proposal number PL970255) project. It has been found in previous projects on regional climate simulations RACCS (contract no.’s EV5V-CT92-0126 and EV5V-CT94-0505) that errors in regional climate simulations to a very high degree reflect errors in the GCMs which provide boundary conditions for the regional climate models. Since MERCURE deals with improvement of regional climate simulations in high resolution regional climate models the POTENTIALS project can provide valuable information to the MERCURE project. But the information also goes in the opposite direction since the work in MERCURE hopefully will result in improved parameterization algorithms, in particular land surface and soil parameterizations, which can be used also in the GCMs in POTENTIALS.

Chase T. N., R. A. Pielke, T. G. F. Kittel, R. Nemani and S. W. Running, 1996: Sensitivity of a general circulation model to global changes in leaf area index. J. Geophys. Res., 101, 7393-7408.

Déqué, M., Dreveton, C., Braun A., and Cariolle, D., 1994: The ARPEGE/IFS atmosphere model: a contribution to the French community climate modelling. Clim. Dyn. 10, 249-266

Held, I. M. and M. J. Suarez, 1994: A proposal for the intercomparison of the dynamical cores of atmospheric general circulation models. Bull. Am. Met. Soc., 75, 1825-1830.

Jeuken, A.B.M., P. C. Siegmund, L. C. Heijboer, J. Feichter and L. Bengtsson, 1996: On the potential of assimilating meteorological analyses in a global climate model for the purpose of model validation. J. Geophys. Res., 101, D12, 16939-16950.

Kasahara, A. and A. P. Mizzi, 1996: Use of Precipitation Data for Diabatic Initialisation to Improve the Tropical Analysis of Divergence and Moisture. Meteorol. Atmos. Phys., 60, 143-156.

Klinker, E., and P. D. Sardeshmukh, 1992: The diagnosis of mechanical dissipation in the atmosphere from large-scale balance requirements. J. Atmos. Sci., 49, 608-627.

Marshall, J. and F. Molteni, 1993: Towards a dynamical understanding of planetary-scale flow regimes. J. Atmos. Sci., 50, 1792-1818.

Molteni, F., 1996: On the dynamics of planetary flow-regimes. Part I: The role of high-frequency transients. J. Atmos. Sci., 53, 1950-1971.

Molteni, F., 1996: On the dynamics of planetary flow-regimes. Part II: Results from a hierarchy of orographically-forced models. J. Atmos. Sci., 53, 1972-1992.

Molteni, F. and T.N. Palmer, 1993: Predictability and finite-time instability of the northern winter circulation. Quart. J. R. Meteorol. Soc., 119, 269-298.

Roeckner, E., K. Arpe, L. Bengtsson, M. Christoph, M. Claussen, L. Dumenil, M. Esch, M. Giorgetta, U. Schlese and U. Schulzweida, 1996: The atmospheric general circulation model ECHAM-4: Model description and simulation of present-day climate. MPI Report No. 218.

Toth, Z., E. Kalnay, 1993: Ensemble forecasting at NMC: the generation of perturbations. Bull. Amer. Met. Soc., 74, 2317-2330.

There are five participating institutions in the project:

Danish Meteorological Institute (DMI)

Lyngbyvej 100

DK-2100 Copenhagen Ø

Denmark

Responsible scientist: Dr. Eigil Kaas

Phone, fax and E-mail: +45 39 15 74 24 +45 39 15 74 60 ek@dmi.min.dk

Météo-France CNRM/GMGEC/EAC

42 Avenue Coriolis

31057 Toulouse Cedex 01

France

Responsible scientist: Michel Déqué

Phone, fax and E-mail: +33 561 07 93 82 +33 561 07 96 10 deque@meteo.fr

Max Planck Institute for Meteorology

Bundesstrasse 55,

D-20146 Hamburg,

Germany

Responsible scientist: Dr. Bennert Machenhauer

Phone, Fax and E-mail: +494041173357 +494041173366 machenhauer@dkrz.de

Laboratoire de Météorologie Dynamique du CNRS

École Normale Supérieure

24, Rue Lhomond

75231 PARIS CEDEX 05

FRANCE

Responsible scientist: Dr. Robert Vautard

Phone, Fax and E-mail: +33 1 44 27 73 53 +33 1 44 27 62 72 vautard@lmd.ens.fr

CINECA - Consorzio interuniversitario per la gestione del centro di calcolo elettronico dell’Italia Nord-Orientale,

Via Magnanelli 6/3

40033 Casalecchio de Reno (Bologna)

ITALY

Responsible scientist: Dr. Franco Molteni

Phone, Fax and E-mail: +39 51 6171591 +39 51 6592581 molteni@cineca.it

An important element in the project is seasonal prediction. For this reason Tim Palmer (ECMWF) who is co-ordinating the EU-supported project PROVOST, see section 5 will be an external expert in the present project. This will ensure that full interaction of ideas and results can take place between the present project and the ongoing activities on seasonal prediction at ECMWF and elsewhere.

Also Richard Jones from the HADLEY Centre will be external expert in the project since he is co-ordinator on the project on regional climate, MERCURE, see section 5. This will ensure a continuous contact between the two projects which is important since good simulation of regional climate depends critically on a good simulation of the coarse mesh (GCM) simulation near the region of interest.

Both external experts will participate in the project meetings.