Forecasting Models

Intermediate level forecasting models:

Despite strong efforts in the design, in the choice of the numerical methods and in the implementation and optimization of the software system, MEMO/DYMOS is still not suitable for forecast purposes because of the tremendous computation time required. Even with the fastest computers currently available it takes at least several hours to simulate a 24-hour-period. Even if this shortcoming seems to be temporary in view of rapidly growing computer power, there has been a recent need for forecast purposes from environmental agencies. In order to furnish these agencies with the possibility to run a photochemical dispersion model in a forecast mode two additional models have been included in the ECOSIM modelling system. These models are slim-line versions of the advanced MEMO/DYMOS system and are able to forecast ozone under restricted circumstances in an appropriate time, usually within two hours for a 24-hours-period. Nevertheless, the MEMO/DYMOS system is in principle able to operate in a forecast mode as soon as the prerequisite computer power is available.

The Ozone Model REGOZON

The principle design of the model REGOZON is similar to the MEMO/DYMOS system. A wind field model is coupled with a dispersion model and a chemical module: REGOZON includes a meteorological model which is much more simple than MEMO. The "time-variable meteorology" comes mainly from the internal energy budget computation. If there are any information about changing geostrophical winds or cloud cover they can be incorporated. Initial meteorological input includes:

• vertical soundings (potential temperature profile)
• geostrophic wind
• cloud cover
• humidity
• surface water temperature.

For the computation of meteorological values and the dispersion, an enhanced version of the Eulerian grid model REWIMET is used. It is based on the conservation laws for impulse, mass, energy and passive constituents comparable to MEMO/DYMOS but with a higher level of approximations. A hydrostatically stratified atmosphere is assumed, which is dry and incompressible. The model equations are expressed in three vertical layers. The first (surface layer) follows the ground level and has a fixed vertical thickness of 50 m above ground. It is turbulently mixed and its physical behaviour is strongly coupled with the surface characteristics.

Emissions from traffic, from households and from industrial sources with low emission heights are introduced into the surface layer. The second layer (mixed layer) reaches from the upper level of the surface layer to the upper level of the atmospheric boundary layer, up to the mixing height. This layer is also turbulently mixed and shows the characteristic diurnal variation of the thickness of the atmospheric boundary layer. Emissions from higher emission sources, for example high stacks from power stations, enter the mixed layer. The third layer (temporary layer) is located above the mixed layer. It is assumed to be free of turbulence. Since the atmospheric boundary layer can expand to the suprascale inversion, it is possible for the temporary layer to disappear. It will be recreated when the atmospheric boundary layer sinks. No substances are emitted in this layer but it transports the suprascale background concentrations of ozone and ozone precursor substances above the atmospheric boundary layer. For the computation of the temperature regime a surface energy budget routine has been added.

The computation of the dispersion is carried out immediately after the determination of the meteorological values. The transport model uses the same vertical structure. The dispersion equation is solved in two dimensions but with an allowed vertical exchange according to the determined stability. The chemical changes form a source or sink term in the dispersion equation. The photochemical scheme of CBM IV is applied; the chemical module is nearly identical with the DYMOS module. As characteristic time steps for the solution of the chemical system are much smaller, the chemical computations have been decoupled from the determination of transport. Usually, chemistry is not computed every transport time step.

Dry deposition velocities and biogenic emissions are computed as a function of land use. The forecast system REGOZON is applicable under the following restrictions:

• relatively flat terrain without strong orographic features and land/sea breeze effects
• fair weather conditions with moderate wind velocity
• stable, barotropic synoptic situations.

Because of the strong constraints and the rough vertical model resolution, the computational time for the REGOZON model is small. A comparison between this relatively simple model and a nonhydrostatic 3d model with 35 vertical layers shows a similar ozone production for the selected episodes which conform with the above restrictions. The extent of computation time required by the advanced MEMO/DYMOS system is approximately an order of magnitude higher. The REGOZON forecast model for a 50x50 horizontal grid needs approximately 2 hours for a 24-hour-period on a workstation (SUN E3000).

The Multilayer Dispersion Model MUSE

Whereas REGOZON is a combined mesoscale meteorological and dispersion model, MUSE is a dispersion model only, requiring the meteorological quantities usually computed by MEMO. On one hand it enhances the computation time in comparison to REGOZON, on the other hand it could be applied to a lot more situations because of the lower constraints on the meteorological parts. For example, whereas REGOZON seems to be unable to describe the transport processes related to sea breeze, MUSE is able to perform this task. The model equations solved by MUSE are those of the fully 3D version of the dispersion model DYMOS. However, the multilayer model MUSE differs from the ECOSIM main dispersion model in the following points:

• Instead of discretizing the model domain into 10-30 non-equidistantly distributed vertical layers, three layers are being used in the vertical direction. This modelling concept is very similar to the REGOZON model structure. The depth of the lower layer (which practically corresponds to the surface layer) is prescribed and kept constant in time. The depth of the middle layer is permitted to vary following the diurnal variation of the mixing height. The latter is described by Deardorff's prognostic equation which was shown to lead to realistic results when the variation of the mixing height with time is strongly influenced by surface heating. For periods of stability, an algebraic parameterization based on the friction velocity and the Monin-Obukhov length is applied. The limit of the upper layer coincides with the fixed domain top. The assumption of a time dependent depth of the middle and upper layer leads to the need of an entrainment term.

• The description of vertical transport due to turbulent diffusion at the lower limit of the middle layer is based on the turbulent kinetic energy defined at the interface of the lower and middle layers. To avoid unrealistic vertical diffusion rates associated with the growth of the mixing height, one-sided concentration gradients are calculated at this interface. Turbulent diffusion in the upper limit of the middle layer is neglected.

• Advective transport is described using the scheme designed by Smolarkiewicz. The average wind speed in the middle and upper layers is calculated by integrating the fluxes of the corresponding layers of the fully 3D wind field model.

• The differential equation system in MUSE is solved with a backward difference (BDF) solution procedure (by applying the Gauss-Seidel iteration scheme). Because of the nature of this semi-implicit algorithm, vertical diffusive transport and chemical transformation of pollutants have to be treated separately.

• Aspects related to the formation of photochemical oxidants could be analyzed with the multilayer model MUSE in conjunction with the chemical reaction mechanism KOREM (a modified version of the Bottenheim-Strausz mechanism). This is a more compressed chemical reaction scheme.

  With the above simplifications, the MUSE code becomes approximately 5 times faster than the full 3D version. For example, MUSE with chemistry needs about 2 hours for a 24-hour-period (IBM RISC 6000). At the same time, the memory requirements are reduced by 90 %. The model configuration MEMO/MUSE can be used for forecast purposes in more complex terrain than REGOZON under defined conditions.