Chemistry-Transport Model System COSMO-MUSCAT

The modeling department of the TROPOS has developed the state-of-the-art multiscale model system COSMO-MUSCAT (Wolke et al., 2004; 2012). It is qualified for process studies as well as the operational forecast of pollutants in local and regional areas (Heinold et al., 2007; Renner and Wolke, 2010). The model system consists of two online-coupled codes. The operational forecast model COSMO is a non-hydrostatic and compressible meteorological model and solves the governing equations on the basis of a terrain-following grid (Schättler et al., 2008; Baldauf et al., 2011). Driven by the meteorological model, the chemistry transport model MUSCAT (Multi-Scale Chemistry Aerosol Transport) treats the atmospheric transport as well as chemical transformations for several gas phase species and particle populations (Knoth and Wolke, 1998a; Wolke et al. 2012). The transport processes include advection, turbulent diffusion, sedimentation, dry and wet deposition.

The physical and chemical processes that determine the distribution of air pollutants occur on a wide range of temporal and spatial scales. Multiscale models can provide finer resolution in certain key regions, e.g. urban areas. The paper focuses on some numerical aspects of modelling urban and regional scale interactions as well as on requirements on the used parameterizations in this context. Multiblock grid techniques (“two-way nesting”) and implicit-explicit (IMEX) time integration schemes are suitable for an efficient numerical treatment of such scale interactions. In the online-coupled model system COSMO-MUSCAT, both approaches are implemented for the chemistry-transport code.

The modelling system is used for several air quality applications (e.g., Stern et al., 2007; Hinneburg et al., 2009) and the investigation of the large-scale transport of Saharan dust, including its sources and sinks (e.g., Heinold et al., 2011a). The simulation results are evaluated by measurements and satellite data within the framework of the project SAMUM. In addition to parameterizing dust particle fluxes, the influence of dust by modifying solar and thermal radiative fluxes on temperature, wind fields, and cloud dynamics is estimated (Helmert et al., 2007; Heinold et al., 2011b). Furthermore, Saharan dust export to the North Atlantic is quantified, where nutrient input by dust is suspected to enhance marine productivity. The performance of the modeling system was evaluated in several model intercomparison studies, e. g. in the framework of AQMEII (Solazzo et al., 2012 a, b; Im et al. 2014 a, b).

Averaged PM10 concentration (left) and fraction of secondary PM (right) for October 2006 in Germany.


Treatment of Atmospheric Processes.

The model system COSMO-MUSCAT is qualified for the operational forecast of pollutants in regional areas and also for detailed studies of tropospheric processes. Gas phase processes, especially the formation of photooxidants as well as the transport and the transformation of particulate matter, can be investigated. The chemical reaction mechanisms are given in ASCII data files. All information required for the computation of the chemical term and the corresponding Jacobian is generated from this input file. Therefore, changes in the chemical mechanism can be performed in a simple and comprehensive way. Several gas phase mechanisms, e. g. RACM-MIM2 (Stockwell et al., 1997; Karl et al., 2006) with more than 90 species and over 200 reactions, are used successfully in 3D case studies. Time resolved anthropogenic emissions are treated in the model as point, area and line sources. The different time evolution of several emitting groups is taken into account for the emission intensity.  It is distinguished between several emitting groups. Biogenic emissions are parameterized in terms of land use type, temperature, and radiation. Dry and wet deposition processes are also included. Modeled dust emissions depend on surface wind friction velocities, surface roughness, soil particle size distribution, and soil moisture from COSMO [Heinold et al., 2011a, b].

For the description of the particle size distribution and aerosol dynamical processes the modal aerosol model M7 (Vignati et al., 2004) extended by the treatment of nitrate and ammonium is used. In this approach, the total particle population is aggregated from seven log-normal modes with different compositions. For simulation of particulate matter, the size distribution and the aerosol dynamical processes (condensation, coagulation, sedimentation, and deposition) are described using a modal technique. The mass fractions of all particles within one mode are assumed to be identical. Particle size distribution changes owing to various mechanisms, which are divided into external processes like particle transport by convection and diffusion, deposition, and sedimentation as well as internal processes like condensation and coagulation. Alternatively, a more simplified mass based approach (similar to EMEP) is available. A more detailed description of MUSCAT is given by Wolke et al. (2004, 2012).


Multiblock Grid Structure.

In MUSCAT a static grid nesting technique is implemented. The horizontal grid is subdivided into so-called “blocks”. Different resolutions can be used for individual subdomains in the multiblock approach. This allows fine resolution for the description of the dispersion in urban regions and around large point sources. This structure originates from dividing an equidistant horizontal grid (usually the meteorological grid) into rectangular blocks of different size. By means of doubling or halving the refinement level, each block can be coarsened or refined separately. This is done on condition that the refinements of neighboring blocks differ by one level at the most.  The vertical grid is the same as in the meteorological model. The spatial discretization is performed by a finite-volume scheme on a staggered grid. Such schemes are known to be mass conservative because of the direct discretization of the integral form of the conservation laws. For the approximation of the surface integrals, point values of the mixing ratio and its first derivative are needed on the cell surfaces. To approximate the mixing ratio at the surface we implemented both a first order upwind and a biased upwind third order procedure with additional limiting (Hundsdorfer et al., 1995). This scheme has to be applied to non-equidistant stencils which occur at the interface of blocks with different resolutions (Knoth and Wolke, 1998a).


IMEX Time Integration Scheme.

For MUSCAT a novel implicit-explicit (IMEX) time integration scheme was developed (Knoth and Wolke, 1998b; Wolke and Knoth, 2000) to combine in an efficient way the slow process of horizontal advection and the fast processes of vertical exchange and gas phase chemistry. Whereas the slow processes are integrated by explicit Runge-Kutta methods any suitable solver can be applied to the fast processes.  A change of the solution values as in conventional operator splitting is avoided in this approach. Within the implicit integration in the chemistry-transport code MUSCAT, the stiff chemistry and all vertical transport processes (turbulent diffusion, advection, deposition) are integrated in a coupled manner by the second order BDF method. We apply a modification of the code LSODE (Hindmarsh, 1983) with a special linear system solver and an adapted restart procedure (Knoth and Wolke, 1998a). The error control can lead to several implicit time steps per one explicit step. Furthermore, different implicit step sizes may be generated in different blocks. The size of the “large” explicit time step depends on the CFL number. Higher order accuracy and stability conditions for this class of IMEX schemes are investigated in Knoth and Wolke (1998b). These methods can also be applied to multiphase processes. Furthermore, multirate time integration techniques are implemented and tested in COSMO-MUSCAT. Due to the choice of different advection steps in different model regions a significant reduction of the computational costs can be reached especially in cases with few large point sources (Schlegel et al., 2012a, b).


Parallelization and Dynamic Load-Balancing in MUSCAT.

The code is parallelized by distributing the blocks (rectangular subsets of the grid) on the available processors using MPI for communication. A static partitioning may lead to load imbalances, since each block has its own time step size control defined by the implicit time integrator. Therefore, a dynamic load balancing has been developed, which periodically redistributes the blocks. The graph partitioning library ParMETIS (Karypis et al., 2003) is utilized to calculate an improved partitioning from the workload of the blocks and their adjacency.


Coupling of COSMO-MUSCAT.

In the past, the “concurrent” coupling scheme has been used in COSMO-MUSCAT, where both models operate concurrently on distinct sets of processors (Wolke et al., 2004)}. Since an adaptive time step control is applied in MUSCAT, the overall workload fluctuates during runtime, especially at scenarios with highly dynamical behavior of the simulated chemical processes. These fluctuations led to processor idle time at the synchronization points of the two models. Fluctuations of the overall MUSCAT workload caused processor idle time at the synchronization points of the two models.  To achieve a higher efficiency, an alternative coupling scheme has been implemented, which is based on the “sequential” approach (Lieber and Wolke, 2008). Benefits are an increased performance and a simplified model startup, since no processor partitioning (determination of processors for COSMO and MUSCAT) has to be defined a priori.  By making this crucial choice unnecessary, a potential source of inefficiency is removed. The bidirectional exchange of data fields between COSMO and MUSCAT is handled by an independent library (Lieber, 2005).



Baldauf, M., A. Seifert, J. Förstner, D. Majewski, M. Raschendorfer, T. Reinhardt, 2011, Operational convective-scale numerical weather prediction with the COSMO model: description and sensitivities. Monthly Weather Review, DOI: 10.1175/MWR-D-10-05013.1.

Heinold, B., J. Helmert, O. Hellmuth, R. Wolke, A. Ansmann, B. Marticorena, B. Laurent and I. Tegen, 2007, Regional Modeling of Saharan Dust Events using LM-MUSCAT: Model Description and Case Studies, J. Geophys. Res. 112, D11204, doi: 10.1029/2006JD007443.

Heinold, B., Tegen, I., Schepansli, K., Tesche, M., et al., 2011a, Regional modeling of Saharan dust and biomass-burning smoke. Tellus B, 63: 781–799. doi: 10.1111/j.1600-0889.2011.00570.x

Heinold, B., Tegen, I., Bauer, S. and Wendisch, M., 2011b, Regional modeling of Saharan dust and biomass-burning smoke. Tellus B, 63: 800–813. doi: 10.1111/j.1600-0889.2011.00574.x

Helmert, J., B. Heinold, O. Hellmuth, R. Wolke, A. Ansmann, B. Marticorena, B. Laurent and I. Tegen, 2007, Regional Modeling of Saharan Dust Events using LM-MUSCAT: Model Description and Case Studies, J. Geophys. Res. 112, D11204, doi: 10.1029/2006JD007443.

Hindmarsh, A.C., 1983, ODEPACK: A systematized collection of ODE solvers, in: R.S. Stepleman, Ed., Scientific Computing, pages 55–74.

Hinneburg D., Renner E., Wolke R., 2009, Formation of secondary inorganic aerosols by power plant emissions exhausted through cooling towers in Saxony. Env. Sci. Pollut Res. 16:25-35.

Hundsdorfer, W.,  B. Koren, M. van Loon and J.G. Verwer, 1995, A positive finite-difference advection scheme, J. Comput. Phys. 117, 35–46.

Im, U., et al., 2014a, Evaluation of operational online-coupled regional air quality models over Europe and North America in the context of AQMEII phase 2. Part I: Ozone, Atmos. Env., 2014, in press, doi:10.1016/j.atmosenv.2014.09.042.

Im, U., et al., 2014b, Evaluation of operational online-coupled regional air quality models over Europe and North America in the context of AQMEII phase 2. Part II: Particulate matter, Atmos. Env., 2014, in press, doi:10.1016/j.atmosenv.2014.08.072.

Karl M., Dorn H.-P., Holland F., Koppmann R., Poppe D., Rupp L., Schaub A., Wahner A., 2006, Product study of the reaction of OH radicals with isoprene in the atmosphere simulation chamber SAPHIR, J. Atmos. Chem., 55 (2), 167-187.

Karypis, G.,  K. Schloegel, and V. Kumar, 2003, ParMETIS: Parallel graph partitioning and sparse matrix ordering library (Version 3.1), University of Minnesota.

Knoth, O. and R. Wolke, 1998a, An explicit-implicit numerical approach for atmospheric chemistry-transport modelling, Atmos. Env. 32, 1785-1797.

Knoth, O. and R. Wolke, 1998b, Implicit-explicit Runge-Kutta methods for computing atmospheric reactive flows. Appl. Numer. Math. 28, 327–341, 1998.

M. Lieber, 2005, Die Optimierung der Kopplung von Simulationsmodellen mit unterschiedlichen Gitterstrukturen auf Parallelrechnern, Diplomarbeit, Hochschule für Technik und Wirtschaft Dresden.

Lieber M, Wolke R., 2008, Optimizing the coupling in parallel air quality model systems. Environ. Modell. Softw., 23:235–243.

Renner E., Wolke R., 2010, Modeling the formation and atmospheric transport of secondary inorganic aerosols with special attention to regions with high ammonia emissions. Atmos. Env., 44, 1904-1912.

Schättler, U., Doms, G., Schraff, C., 2008, A Description of the Nonhydrostatic Regional COSMO-Model. Deutscher Wetterdienst, Offenbach.

Schlegel, M., O. Knoth, M. Arnold, and R. Wolke, 2012a, Implementation of splitting methods for air pollution modeling, Geosci. Model Dev., 5, 1395-1405.

Schlegel, M., O. Knoth, M. Arnold, and R. Wolke,2012b, Numerical solution of multiscale problems in atmospheric modeling, Appl. Numer. Math., 62(10), 1531-1543, doi:10.1016/j.apnum.2012.06.023.

Solazzo, E., et al., 2012a, Operational model evaluation for particulate matter in Europe and North America in the context of AQMEII. Atmos. Env. 53, 75-92.

Solazzo, E., et al., 2012b, Ensemble modelling of surface level ozone in Europe and North America in the context of AQMEI. Atmos. Env. 53, 60-74.

Stern, R., Builtjes, P., Schaap, M., Timmermans, R., Vautard, R., Hodzic, A., Memmesheimer, M., Feldmann, H., Renner, E., Wolke, R., Kerschbaumer, A., 2008, A model intercomparison study focussing on episodes with elevated PM10 concentrations. Atmos. Env. 42, 4567-4588.

Stockwell, R.W., F. Kirchner, M. Kuhn and S. Seefeld, 1997, A new mechanism for regional atmospheric chemistry modeling, J. Geophys. Res. 102, 25,847–25,879.

Vignati, E., J. Wilson and P. Stier, 2004, M7: An efficient size-resolved aerosol microphysics module for large-scale aerosol transport models, J. Geophys. Res. 109, D22202, doi: 10.1029/2003JD004485.

Wolke, R. and O. Knoth, 2000, Implicit-explicit Runge-Kutta methods applied to atmospheric chemistry-transport modelling, Env. Mod. & Software 15, 711–719.

Wolke, R.,  O. Knoth, O. Hellmuth, W. Schröder and E. Renner, 2004, The parallel model system LM-MUSCAT for chemistry-transport simulations: Coupling scheme, parallelization and application, in: G.R. Joubert, W.E. Nagel, F.J. Peters, and W.V. Walter, Eds., Parallel Computing: Software Technology, Algorithms, Architectures, and Applications, Elsevier, Amsterdam, The Netherlands, 363-370.

Wolke, R., W. Schroeder, R. Schroedner,  E. Renner, 2012,  Influence of grid resolution and meteorological forcing on simulated European air quality: A sensitivity study with the modeling system COSMO-MUSCAT. Atmos. Env., 53, 110-130.


Dr. Ralf Wolke, Dr. Oswald Knoth, Bernd Heinold

Leibniz Institute for Tropospheric Research
Permoserstraße 15
04318 Leipzig, Germany
Tel: +49 (0) 341-235-2860
Fax: +49 (0) 341 235-2139