SRNWP Expert Team on upper air physics Work plan

Version 0.3, 5 September 2008

Expert team workplan

The expert team had its first meeting on 3 September 2008 at ECMWF. During this meeting the possible tasks and the way to execute them were discussed. In this section the outcome from these discussions are described, together with the work packages that are resulting. The work packages from the expert team consist of three main items:
- make available and accessible the information that is necessary or at least very helpful for the (mesoscale) physics development
- stimulate the exchange of information and the participation in already existing networks for comparison of model output with observations.
- (co-)organize workshops on issues that are deemed important or not giving enough attention in the already existing working groups (e.g. GEWEX, MAP, COPS etc).

Information on models and development methods

The expert team will exchange the methods with which the physics are monitored, tested and validated, and make them available at a website, preferably the SRNWP one. In addition we will build up a portfolio of well controlled, academic or academic-type baseline cases. These cases will have to be well documented in the literature and good datasets for comparison will have to be available to study the phenomena that are of interest. This will be done especially for the mesoscale models. We will also try to advise on a minimum set of model output parameters, which are necessary to enable a thorough comparison of the output of the mesoscale models that want to use the baseline cases for their model development.

This document is a first step in the exchange of information regarding the contents of all NWP-models that are currently used in the European consortia. It is therefore a good starting point and will be extended with the items that were discussed above.

Exchange of model data for comparison at special observation sites

A regular comparison with special observations is very helpful in finding model deficiencies. Already there are working groups that invite modellers to send in their model data for comparison with observations at specific sites. Examples of this are e.g. Cloudnet and the mast verification project in HIRLAM. Not all consortia are participating in these real time comparisons with observations. The expert team thinks that this is a missed opportunity and a very easy way for the different consortia to learn from each others strengths and weaknesses. We will make an overview of all the initiatives for observation comparison and stimulate all consortia to participate in these initiatives.

One of the phenomena that can be studied through the exchange of model data is the stable boundary layer. The processes under these conditions are so sensitive that it is very easy to do something wrong, with large errors in e.g. the wind speed or the depth of the mixing layer as a result. The exchange of model profiles of the lowest few 100 metres for observation sites like Sodankyla, Lindenberg. Cabauw and Valladolid is one of the ways to get a clear view of the seriousness of the problem and the opportunity to learn from other models if they are performing better in certain conditions.

This is also true for the clouds, which are studied in Cloudnet. Participating in this project can help to quickly understand the shortcomings of models with regard to clouds, convection and precipitation formation.

Other initiatives that can contribute to the development of the physics will be studied and the participation of the consortia will be stimulated whenever it is deemed positive for the physics development.

Workshops and working weeks.

One of the way to exchange information on the methods and performance of models is the organisation of and participation in workshops and working weeks. The expert team will actively stimulate the participation of all consortia in the already existing working groups and workshops. Further we will identify the areas that we feel are not getting enough attention. For these areas we will then organize workshops or working weeks, or we will try or get a session in an already organized workshop with a subject relatively close to the one we propose.

One example is the NetFAM workshop about moist processes that will be organized by SMHI next year. One of the subjects that the expert team finds important for further study and the exchange of model development results is the transition of shallow to deep convection. This area is very important for the right behaviour of deep convection (right intensity and timing of initiation) in mesoscale models. We will try to append a session on this issue to this particular workshop. This session could be build around the GEWEX GCSS daily cycle of deep convection case and take the form of a model intercomparison for NWP models. Until now the participating models in this case have mostly been academic CRMs.

The results from the interoperability work will highly increase the probability that model comparisons are possible. Software that enables the direct comparison of model output from the different consortia is one of the missing links that hampers the exchange of model data.

Other topics that may be interesting to cooperate on more closely or for which workshops or intercomparisons could be organized are microphysics schemes that go further than the Kessler formulations and physics dynamics coupling studies. In the near future the interaction of the cloud microphysics when prognostic aerosols will also be included in (mesoscale) models. This may lead to a revision of the warm rain parameterization.

Overview of physics developments area

The physics parameterizations in NWP models can be divided in five different groups. These are the surface exchange processes (the topic of another expert team), radiation, vertical diffusion and orography related turbulent fluxes, deep and shallow convection and condensation (microphysics). For all these processes there are many different parameterizations and all consortia seem to use different ones for the same process, so there are many alternative physics parameterizations in Europe.

In addition to the differences between the consortia, when taking the models into account that are used for similar resolutions, also many different resolutions are used in the limited area models. Nowadays, the resolutions that are used or studied for operational implementation range from around 20 km to 1.5 km. At the former resolution most physical processes need to be parameterized, at the high resolution of 1.5 km deep convection is more or less resolved (and conventional 1D parametrizations are certainly not appropriate). This means that the parameterizations that were used in the coarse resolution models cannot be used without adjustments at the very high resolution. The transition range from fully-parameterized to fully-resolved deep convection is often called the grey zone.

For the highest resolution models many new aspects become important. In addition to the grey zone for deep convection, a new grey zone will emerge for the shallow convection and largest boundary layer eddies, when the model resolution is close to or finer than the depth of the boundary layer. For the current mesoscale models these effects can be found already as the horizontal resolution is quite close to or in some cases finer than the boundary-layer depth. Other aspects that are important in the high-resolution mesoscale models and that were not so important in the regional models are 3D turbulence and 3D radiation.

All these different options that are used in the different consortia of course hamper the initiatives for cooperation on specific subjects. On the other hand the large variety of schemes has the advantage that we can learn a lot from each other, especially what the strong points and the weak points are of all the schemes that are used in Europe. Experience shows that it is often difficult to isolate the behaviour of individual schemes within 'real world' forecast models, as the interaction with other schemes can affect behaviour. Work on understanding the behaviour of different schemes therefore requires careful design, coordination and execution, often through more idealised intermediate studies. This will form a central part of the work of the SRNWP ET.

Parametrization Schemes Overview

Below the different options that are used in the consortia for the upper air physics will be discussed briefly.


For the radiation there are two basically different options in the European limited area models. On the one hand we have the very sophisticated radiation schemes that are not called every timestep, on the other hand we have the more simple schemes that are called every time step.


Many more different flavours are available for the treatment of turbulence in the limited area models. Turbulence is primarily treated within the so-called Planetary Boundary Layer (PBL), so schemes are often called PBL schemes, but turbulence may also be important aloft. Schemes are usually based on Reynolds (ensemble) averaging and the assumption of horizontal homogeneity is usually made, leading to 1D schemes, though both the averaging and homogeneity assumptions may not be what is required as very high (O(1 km)) horizontal resolution is approached.

There are local and non-local schemes, K-profile and TKE schemes and in the near future there will also be eddy-diffusion mass-flux schemes, where the turbulence and shallow convection are combined in one scheme, which then naturally gives you a non-local scheme. Turbulent transport is commonly treated as a diffusion process, though the formulation of eddy diffusivities and viscosities may be very non-linear and other, non-gradient or counter-gradient terms are often introduced, especially in unstable boundary layers.

In the local schemes the vertical turbulent transport is dependent on the local properties (primarily gradients) only. These schemes can be relatively simple and easy to tune like the Louis scheme, but also more sophisticated schemes like the CBR scheme (Cuxart, Bougeault and Redelsperger) are also local schemes, usually based on higher order closures of the Reynolds averaged equations and so requiring additional prognostic variables and more tuneable constants.

In the non-local schemes the vertical turbulent transports are not only a function of the local gradients, but also of larger scale characteristics like the boundary layer depth, the inversion strength and the average flux in the boundary layer. The non-local term in these schemes allows a slightly stable profile to develop in the upper part of the day-time boundary layer, something that is not possible in the local schemes but which is observed in LES studies.

A relatively new option for vertical diffusion is the combined eddy-diffusion mass-flux scheme. In this type of scheme the local turbulence is handled by an eddy diffusion scheme while the mixing caused by the strongest and largest eddies is handled by a mass flux scheme. This enables the observed boundary layer structure in LES models to develop in NWP-models also. An additional advantage of such a scheme is the handling of shallow convection by the boundary layer scheme, reducing the number of schemes necessary for vertical mixing and enabling the smooth transition from the boundary layer to the shallow clouds at the top of the boundary layer and slightly beyond.

Shallow and deep convection

A wide variety of schemes exist for convection, many inherited from larger-scale models. Though different approaches exist at very large scale, the majority of schemes used at mesoscale are based on the 1D massflux approach, in which the convective clouds are treated as a collection of near point plumes, with compensating subsidence between. Such schemes differ greatly in detail, but can often be summed up as having three compone nts:

  1. A (1D) 'cloud model' which models the vertical growth of massflux in the convective core, including some form of entrainment, detrainment, cloud condensation and (usually very simple) microphysics. Many schemes have a complementary 'convective downdraught' representation.
  2. A diagnosis or 'trigger function' which determines when the scheme should operate.
  3. A closure assumption to enable the cloud base mass flux to be detrmined (i.e. the boundary condition for the cloud model).

The detail of these schemes can lead to very different behaviour, but an even bigger problem is their essential 1D nature, which assumes horizontal homogeneity. For deep convection schemes this means, in practice, that they are appropriate for averaging length scales of order 100 km, and their use even at 20 km, which begins to match the spacing of individual deep convective clouds, is suspect and often requires tuning for specific applications. Below 10 km horizontal resolution, such 1D schemes are very difficult to use with any scientific rigour, though explicit convective clouds are not properly resolved until resolutions rather better than 1 km are used - thus 1-10 km is the so-called 'grey zone'. Designing schemes for this regime is very difficult, and there seems little doubt that a closer coupling to model dynamics is required as clouds become partially resolved - some attempts have been or are being developed (e.g. Gerard and Geleyn) though there is much less experience of such schemes.

In principle, shallow convection still needs to be parametrized even at resolution around 1 km, though in practice many models do not do so. Shallow convection schemes are often simplified versions of their deep brethren, though often using different closure and trigger functions as they are driven by boundary-layer fluxes and precipitation plays no role. Interaction between convection and boundary-layer schemes is often a problem, and approaches are under development which treat shallow cumulus more as a turbulence process (Grant). As already mentioned, the combined eddy-diffusion mass-flux scheme also has some flavour of this approach.


In lower resolution models, the primary role of microphysical parametrizations is to provide cloud properties for the radiation scheme; as resolution increases, the role of cloud microphysics in producing transport of water (in all phases) and hence buoyancy through latent heating becomes increasingly important (as well, of course, as the straightforward forecasting requirement to predict the amount and phase of precipitation).

Though research models may use more explicit 'bin' approaches to represent the spectrum of sizes (and shapes) of hydrometeors, this is prohibitively expensive for current operational models, and a 'bulk' formulation is invariably used, in which the hydrometeors are split into a number of classes, and the size spectrum of each parametrized through a small number of parameters. Since a spectral shape for each class is usually assumed, a number (n) of parameters defining the shape can be allowed to vary independently if that number of prognostic variables is carried by the model. Schemes are often thus described as having m classes and n moments, though m may differ amongst class members.

The need to represent a class prognostically is closely related to the mean (terminal) fall speed of particles in the class, V. Crudely, if ∆x/U is greater than ∆z/V, where ∆x is the horizontal grid spacing, U a typical mean wind speed and ∆z the typical distance the hydrometeor falls, then most of the hydrometeors fall out in the time to advect to the next grid box, so a very good quasi-steady state approximation may be made by neglecting advection by the wind. In practice, ∆z/V is greater than or about 5000 s for small ice and snow, so any model with horizontal resolution better than 10 km (taking U about 20 m s-1) should represent these prognostically. On the other hand, ∆z/V is usually less than about 1000 s for rain, so benefits from prognostic rain may only be evident once resolution is better than a few km.

These considerations mean that cloud water must be carried as a prognostic, though the high speed of the condensation/evaporation process means that alternative, more conservative prognostic variables may be used.

Many flavours of microphysics scheme exist, but typical configurations are prognostic cloud water and ice/snow at 10 km, and additional prognostic rain and graupel at so-called cloud resolving scales closer to 1 km. Single moment schemes are common, but dual (and even triple for graupel) moment schemes are often used in research models and some operational models.

In addition to hydrometeor representation, some consideration of sub-grid variability is needed. This may be confined to the liquid cloud condensation scheme, plus overlap assumptions in the radiation scheme, but at lower resolution, treatment of sub-grid variability can markedly change effective process rates (such as growth of snow by riming), and at higher resolution there are strong arguments that the sub-grid variability needs to be closely coupled to the turbulence scheme as it provides local buoyancy variability.

Un-resolved orography parameterization

Even at 1 km resolution, most land still presents un-resolved orography which can exert additional stress (and modify scalar transports, though this is often ignored). Even the basic concepts behind the parametrization of this differs widely between models (which often leads to confusion). Orography with wavelengths longer than a few km often produces vertically propagating gravity waves which may break, often at upper levels, to produce significant stress divergence. This process is called gravity wave drag (GWD).

Models with resolution a few km (even up to 10-15 km) are usually assumed to resolve this process explicitly and often run without GWD. However, other processes, largely associated with smaller-scale orography, occur, usually producing drag at lower levels. These include the impact of pressure forces and additional surface friction ('orographic roughness') and flow-blocking associated with stable flows. There may also be enhanced surface fluxes driven by stable drainage flows. These smaller scale processes are often treated (explicitly or implicitly) through modifications to the surface exchange process (though changes to e.g. surface roughness or stability functions) and may be covered by the surface processes ET but it needs to be remembered that these are really boundary-layer processes.

To add to confusion, all 'unresolved orography drag' is sometimes lumped together (either in the actual parametrization or in terminology) under the term 'GWD', though this should be discouraged.

What is currently used in reference versions of the models?

Regional models (O 10km) Radiation Vertical diffusion
ALADIN Ritter and Geleyn p-TKE
HIRLAM Savijarvi O1.5 TKE
COSMO Ritter and Geleyn O2 TKE
UNIFIED MODEL Edwards and Slingo Non-local 1st order (Lock)
Regional models (O 10km) Convection Microphysics
HIRLAM Kain-Fritsch Rasch-Kristjansson
COSMO Tiedke (1989) HYDCI (Doms and Schattler 2004; Reinhardt and Seifert 2006)
UNIFIED MODEL Modified Gregory and Rowntree with CAPE closure and CMT Wilson and Ballard
Regional models (O 10km) Gravity wave drag
UNIFIED MODEL Bulk GWD parametrization (Webster)

Mesoscale models (O 2km) Radiation Vertical diffusion
COSMO Ritter and Geleyn O2.5 TKE
UNIFIED MODEL Edwards and Slingo Non-local 1st order (Lock)
Mesoscale models (O 2km) Convection(?) Microphysics
AROME none Meso-NH
COSMO none HYDCI (Doms and Schattler 2004; Reinhardt and Seifert 2006)
UNIFIED MODEL 4 km : Gregory-Rowntree Shallow, Gregory-Rowntree deep with restricted massflux.
1 km : None. Considering GR shallow.
Enhanced Wilson and Ballard
Single moment bulk cloud water, rain, small ice, snow, graupel. Operational models currently run with combined ice/snow and no graupel but likely to change.

Current cooperation on NWP-physics developments within Europe

Most important items in physics developments in different consortia