Implementation of aerosol-cloud interactions in the regional atmosphere-aerosol model COSMO-MUSCAT and evaluation using satellite data

The regional atmospheric model Consortium for Small Scale Modeling (COSMO) coupled to the MultiScale Chemistry Aerosol Transport model (MUSCAT), is extended in this work to represent aerosol-cloud interactions. Previously, only one-way interactions (scavenging of aerosol and in-cloud chemistry) and aerosol-radiation interactions were included in this model. The new version allows for a microphysical aerosol effect on clouds. For this, we use the optional two-moment cloud microphysical scheme in COSMO and the online-computed aerosol information for cloud condensation nuclei (CCN) concen5 trations, replacing the constant CCN concentration profile. In the radiation scheme, we implement a droplet-size-dependent cloud optical depth, allowing now for aerosol-cloud-radiation interactions. In order to evaluate the model with satellite data, the Cloud Feedback Model Inter-comparison Project Observational Simulator Package (COSP) has been implemented. A case study has been carried out to understand the effects of the modifications, in which the modified modeling system was applied over the European domain with a horizontal resolution of 0.25◦× 0.25◦. It is found that the online coupled aerosol introduces 10 significant changes for some cloud microphysical properties. The cloud effective radius shows an increase of 2 to 10 μm, and the cloud droplet number concentration is reduced by 10 to 50 cm−3. For both quantities, the new model version shows a better agreement with the satellite data. The microphysics modifications have a smaller effect on other parameters such as optical depth, cloud water content, and cloud fraction.


Introduction
The quantification of aerosol-cloud interactions in models continues to be a challenge (IPCC, 2013).Estimates of effective radiative forcing and assessments of the radiative effects due to aerosol-cloud interactions to a large extent rely on numerical modeling.A large effort has been made to represent such effects in general circulation models (GCMs) (Penner et al., 2006;Quaas et al., 2009;Ghan et al., 2016).However, GCMs do not resolve the processes relevant for cloud dynamics well.Improving the understanding of processes of aerosol-cloud interactions thus largely relies on simulations with cloud-resolving models and large-eddy simulations (LESs) (Ackerman et al., 2000(Ackerman et al., , 2004;;Xue and Feingold, 2006;Sandu et al., 2008;Seifert et al., 2015;Berner et al., 2013).However, LESs often focus on case studies and use idealized boundary conditions as well as an idealized representation of the aerosol.This leads to uncertainties specifically because, when analyzing cloud systems or cloud regimes, rather than individual clouds, aerosolcloud-precipitation interaction processes often are buffered (Stevens and Feingold, 2009).Regional climate modeling is a powerful tool to overcome these limitations of smalldomain idealized LESs, and much higher resolutions are possible than for GCMs.Compared to an LES that only simulates individual cloud systems, regional climate models are able to simulate the feedbacks between clouds and aspects of the large-scale circulation and its variability reasonably well.Even though regional models do not describe part of the large-scale feedbacks, they may be considered a good optimal compromise (Bangert et al., 2011;Van den Heever and Cotton, 2007;Chapman et al., 2009;Forkel et al., 2015;Yang et al., 2012).A still often applied cloud microphysics parameterization in numerical weather prediction is a bulk one-moment scheme (Kessler, 1969;Lin et al., 1983) which uses the specific masses for different hydrometeor species as prognostic variables.However, it cannot treat aerosol-cloud interactions because it calculates only one moment of the size distribution and does not carry information about size or number concentration of cloud droplets.In contrast, bin microphysical schemes numerically resolve the size spectrum and are thus able to predict the spatiotemporal behavior of a number of size categories for each hydrometeor type explicitly (Khain et al., 2000;Simmel et al., 2015).However, this approach is numerically very expensive especially when applied for regional atmospheric models.As a compromise between these two approaches, two-moment microphysical schemes can predict the number concentration of the liquid and ice hydrometeors in addition to mass variables (Cotton et al., 1986;Meyers et al., 1997;Seifert and Beheng, 2006).Furthermore, numerous studies have shown that using twomoment schemes is a promising avenue in future operational forecast models (Reisner et al., 1998;Tao et al., 2003;Seifert and Beheng, 2006) and is also computationally efficient.
At present, several weather prediction and global models apply with two-moment cloud microphysical schemes.For example, the Weather Research and Forecasting (WRF) model is available with different types of two-moment microphysical schemes (Thompson et al., 2008;Morrison and Gettelman, 2008;Lim and Hong, 2010).Morrison et al. (2009) showed that using a two-moment scheme in the WRF model produced more trailing stratiform precipitation in an idealized two-dimensional squall case, which is more consistent with observations.In another study, Li et al. (2008) investigated the effect of aerosol on cloud microphysical processes with a two-moment microphysical scheme in the WRF model.Also, Lim and Hong (2010) have included a prognostic equation for cloud water and cloud condensation nuclei number concentration (C ccn ), which could reduce the uncertainty in investigating the aerosol effect on cloud properties and the precipitation process in the WRF model.Seifert et al. (2012) and Weverberg et al. (2014) compared the operational one-moment and two-moment microphysical schemes in the Consortium for Small-Scale Modeling atmospheric model (COSMO).Further, some groups previously implemented aerosol-cloud interactions in COSMO, albeit with a different aerosol scheme (Bangert et al., 2011;Zubler et al., 2011;Possner et al., 2015) and very few are coupled to the radiation scheme.Seifert et al. (2012) compared the operational one-moment microphysics scheme to a two-moment scheme.They found that aerosol perturbation has a significant effect on radiation and near-surface temperature, rather than the resulting surface precipitation.
In this paper, we discuss the improved cloud microphysics parameterization in the COSMO model (Doms and Schättler, 1999) via the online-coupled aerosol model, Multi-Scale Chemistry Aerosol Transport (MUSCAT; Wolke et al., 2004Wolke et al., , 2012)).The two-moment cloud microphysical scheme in the COSMO model (Seifert and Beheng, 2006) uses fixed profiles of C ccn .Rather than this simplification, here we use C ccn predicted on the basis of the simulated aerosol from the MUSCAT module.This will enable the COSMO model to have temporally and spatially varying C ccn values at each grid point, which are fully consistent with the cloud and precipitation fields, as well as with dynamics (e.g., scavenging is taken into account, as is vertical transport) to represent aerosol-cloud interactions.In two further steps, (i) the radiation scheme is slightly revised to consider the cloud droplet size information (so far considered constant even when applying the two-moment cloud microphysical scheme) and (ii) a diagnostic tool, the Cloud Feedback Model Intercomparison Project Observational Simulator Package (Bodas-Salcedo et al., 2011, 2008;Nam and Quaas, 2012) is implemented, which allows for a consistent evaluation using satellite observations.The paper is organized as follows: Sect. 2 gives a brief introduction to the coupled model system of COSMO-MUSCAT, data, and methodology.The comparison between the improved two-moment cloud microphysical parameterization with the available two-moment scheme making use of the COSP satellite simulator is discussed in Sect.3. Finally, concluding remarks are found in Sect. 4.

The COSMO-MUSCAT model and revised cloud activation
The non-hydrostatic three-dimensional model, COSMO, which was originally developed for limited-area operational numerical weather predictions (NWPs), is used in this study (Doms and Schättler, 1999;Steppeler et al., 2003).This model has been used operationally in convection-permitting configurations since 2007 by the German Weather Service (Deutscher Wetterdienst, DWD; Baldauf et al., 2011).In this study, we have used COSMO version 5.0, which is initialized and forced by reanalyzed data provided by the global meteorological model GME (Global Model of the Earth) of DWD, which is a hydrostatic weather prediction model (Majewski et al., 2002).GME operates on an icosahedralhexagonal grid having a horizontal resolution of approximately 40 km and vertical resolution of 40 layers up to 10 hPa.The COSMO model is initialized with the interpolated GME initial state and nested within GME with hourly updates of lateral boundary values.In this study, the COSMO model has been configured in a non-convection-permitting mode with a uniform horizontal grid resolution of 0.25 • (≈ 28 km).The two-moment cloud microphysics scheme in the COSMO model (Seifert and Beheng, 2006) distinguishes between five hydrometeor classes, namely cloud droplets, rain, ice crystals, snow, and graupel.Processes in the warm (liquid) phase considered by this scheme include the nucleation of cloud droplets, autoconversion of cloud droplets to form rain, accretion, and self-collection of rain droplets.
The formulations have been derived by Seifert and Beheng (2001) from the theoretical formulation of Beheng and Doms (1986).However, the radiation scheme does not yet make use of the additional information about cloud particle sizes provided by the two-moment microphysics.It uses the Ritter and Geleyn (1992) parameterization for the cloud optical properties in radiation scheme.According to Ritter and Geleyn (1992), the cloud optical properties are approximated by the relation between specific liquid water content q c and cloud effective radius r e .Thus, cloud optical depth τ c is expressed as where dz is layer thickness, and c 1 and c 2 are constants.Similarly, the effective radius r e is related to specific cloud water content and is approximated as where c 3 and c 4 are constants (Ritter and Geleyn, 1992).In order to take the two-moment microphysics into account in the radiation scheme, the cloud optical properties have to be modified.The cloud effective radius r e is derived by dividing the third and second moments of the size distribution (Martin et al., 1994;Morrison and Gettelman, 2008) which, after rearranging, yield where µ is the spectral shape parameter (here, µ = 2), is the gamma distribution function, and λ is the slope parameter, which is given by where ρ is the density of the air, ρ w = 1000 kg m −3 is the bulk density of liquid water, N d is the droplet number concentration, and q c is the specific water content.The corresponding cloud optical depth is given by where dz is the layer thickness, ρ w is the bulk density of liquid water, and the droplet size spectrum is considered vertically constant in the grid layer.
The online-coupled model system of COSMO-MUSCAT (Wolke et al., 2004(Wolke et al., , 2012;;Renner and Wolke, 2010) is used for prognostic cloud condensation nuclei in the cloud microphysics parameterization in the COSMO model.The chemistry-aerosol transport model MUSCAT treats atmospheric transport as well as chemical reactions with the Regional Atmospheric Chemistry Mechanism (RACM; Stockwell et al., 1997).In MUSCAT, all meteorological fields are given with respect to the uniform horizontal meteorological grid from the online-coupled COSMO2M (COSMO with two-moment scheme) model, whereas the aerosol information is fed back to the COSMO2M model from MUS-CAT.In the previous setting, the interactions only considered the radiative effects of aerosols (scattering and absorption of solar radiation), as well as the scavenging of aerosol and in-cloud aerosol chemistry.A diagram illustrating the COSMO-MUSCAT modeling setup is shown in Fig. 1.In the COSMO model with the two-moment approach, the nucleation of cloud droplets has been treated explicitly and the aerosol activation parameterization is based on empirical activation spectra, which is in the form of the following power law: where S is supersaturation, C ccn = 1.26 × 10 9 m −3 , and k = 0.308 for continental conditions or C ccn = 1.0 × 10 8 m −3 and k = 0.462 for maritime conditions (Khain et al., 2001).Seifert et al. (2012) investigated the influence of substantially perturbing C ccn from 100 to 3200 cm −3 (see above for a brief discussion of this paper).Accordingly, the grid-scale explicit nucleation rate is calculated from the time derivative of the activation relation (Seifert and Beheng, 2006): and S < S max , 0 else (7) The above parameterization scheme uses constant C ccn , and S max varies with atmospheric conditions (in maritime conditions, C ccn assumes that at S max = 1.1 %, all C ccn are already activated).In the above equation, nucleation explicitly depends on grid-scale supersaturation in combination with saturation adjustment assumed in the cloud scheme, which has limitations (Seifert and Beheng, 2006).As an initial step, a coupled model simulation is carried out by setting S max = 2.0 %, the condition for intermediate aerosols in the COSMO model.In a second step, we have used simulated sulfate (SO 4 ) aerosol mass concentration information from the MUSCAT model.The emission inventory in the MUS-CAT model is provided by the TNO (Netherlands Organisation for Applied Scientific Research) for the Air Quality Model Evaluation International Initiative (AQMEII) project (Pouliot et al., 2012).C ccn is derived using the following empirical relation (Boucher and Lohmann, 1995): where mSO 4 is the sulfate aerosol mass concentration in µg m −3 .The constant C ccn in Eq. ( 7) is replaced by the spatially and temporally varying C ccn values, derived from Eq. ( 8), using the sulfate aerosol mass concentration from the MUSCAT module.This empirical relationship which links sulfate aerosol mass concentration to C ccn is subject to substantial uncertainty.Representing sulfate aerosol as a surrogate for all aerosols is probably too simplistic to capture the complexity of the whole activation process.Future work will introduce a more complex aerosol-cloud coupling, taking into account also other aerosol compounds.The Seifert and Beheng (2006) cloud microphysical scheme considers both phases.Also, mixed-phase clouds are affected by the revision of the CCN parameterization, e.g., via the Bergeron-Findeisen process.Nevertheless, the current analysis focuses on the liquid phase only; investigations on mixed and icephase clouds are left for future research.

Model evaluation method
Satellite retrievals have been used to evaluate the performance of the numerous GCMs and NWP models (e.g., Quaas et al., 2004Quaas et al., , 2009;;Zhang et al., 2005;Brunke et al., 2010;Cherian et al., 2012;Nam et al., 2014).However, a meaningful evaluation of modeling with satellite observations is challenging because of the difference in the model variables and the satellite retrievals.Mühlbauer et al. (2014Mühlbauer et al. ( , 2015)).The output diagnostics include a variety of cloud properties, which facilitate consistent model-to-observation comparisons as well as consistent intermodel comparisons.
In the next section, we evaluate the model results (derived using MODIS and the ISCCP satellite simulators) with MODIS level-2 and ISCCP satellite observations, in terms of cloud optical and microphysical properties (cloud optical depth, effective radius, liquid water path, and cloud fraction).The MODIS satellite simulator uses profiles of particle size for liquid and ice and corresponding optical depths within each layer of the subcolumn as a function of model levels.Using the cloud overlap assumption, 0 or 1 cloudiness in each subcolumn is created in each level.The diagnostics are then integrated over the cloudy subcolumns to obtain in-cloud average cloud optical depth and liquid water path.In turn, cloud effective radius is sampled at the cloud tops, which is not a vertical integral.Further, the ISCCP simulator aggregates pixel-scale cloud retrievals (fraction of the subcolumn with τ ≥ 0.3) to estimate cloud fraction (for more details, see Pincus et al., 2012).
The precision of weather forecasts for longer times is inherently limited by the non-linear nature of the problem.As forecast progresses, the uncertainty in weather prediction also increases.In turn, the earliest forecast time steps are still substantially affected by the initialization.Hence, we have considered the third day of the simulation for evaluating the model with satellite observations.The synoptic condition which is discussed in the next section.To compare with model simulations, different swath data sets of MODIS level-2 on 17 February 2007 (daytime overpass only) are combined and gridded to the model domain and model outputs are averaged between 08:00 and 14:00 UTC (corresponding ap-proximately to the MODIS Terra overpass time over the domain).Also, MODIS level-2 products and model simulations are screened for liquid-phase clouds because in COSMO-MUSCAT only cloud microphysics for liquid clouds were modified.Additionally, MODIS cloud optical depth and effective radius are applied with threshold values of 5 and 2 µm (Sourdeval et al., 2016;Z. Zhang et al., 2012).Since the analysis is carried out for winter, satellite retrievals can be affected by snow cover on the ground.However, the MODIS retrieval (Platnick et al., 2001) uses a combination of absorbing spectral channels for which the snow/ice albedo is relatively small which makes it suitable for retrieving cloud properties over snow.Furthermore, the COSPdiagnosed model clouds are compared to ISCCP daily cloud products.For that, modeled ISCCP cloud products are regridded from 28 to 280 km resolution (ISCCP satellite resolution) using a grid interpolation method and averaged daily.Besides these satellite observations, Clouds and the Earth's Radiant Energy System (CERES; Loeb et al., 2012) satellite observations are also used for model evaluation, which are daily products (Kato and Loeb, 2003).One should keep in mind that the satellite products, just like models, are prone to biases.Nonetheless, the spatial correlation of the cloud structures is well represented (Noble and Hudson, 2015;Min et al., 2012).

Numerical simulations
To isolate and analyze the effects of the model modifications, we have performed three different model simulations with the same interpolated GME initial conditions for the time period of 10 days (15-24 February 2007).They are (a) a standard COSMO two-moment simulation with fixed C ccn (3.0 × 10 8 m −3 ) (COSMO2M), (b) a COSMO twomoment simulation with radiation coupled to cloud microphysics (COSMO2MR) which uses Eqs.(3)-( 5) in the radiation scheme (here, also C ccn is kept fixed as in COSMO2M), and (c) a coupled COSMO-MUSCAT simulation, i.e., using interactive rather than prescribed C ccn and treating radiation in the same way as COSMO2MR (COSMO-MUSCAT).In most of the discussion, we have used simulations (a) and (c).In all three model versions (COSMO2M, COSMO2MR, and COSMO-MUSCAT), we make use of the COSP diagnostics for the MODIS and ISCCP satellite simulators.

Synoptic situation
The simulation starts on 15 February and ends on 24 February 2007.At the beginning of the simulation (00:00 UTC), the meteorological condition is dominated by a low-pressure system over the North Atlantic and a high-pressure system over land.The 2 m temperature shows a temperature gradient with a warm ocean and a cool continent, mostly in the northeastern part of the domain.The winds are mostly strong southwesterly over the Atlantic and northerly and northwesterly in the southern region (Fig. S1 in the Supplement).Since the case study has been conducted for 17 February, the model-derived key meteorological parameters at 12:00 UTC are illustrated in Fig. 2. On 17 February, the low-pressure system has moved to the French Atlantic coast and a cyclonic circulation has set up over the region.Furthermore, a strong high pressure is seen over northeastern Europe.The 2 m temperature shows that prominent winter synoptic conditions still exist in the northern part with a warm oceanic region (Atlantic) and cold northeastern part.The southern region has a maximum temperature of 20 • C, whereas the northeastern continental region experiences a minimum temperature of −20 • C. The cyclonic circulation drives the air mass from the oceanic region and results in the formation of clouds along the frontal systems.In addition, the high pressure in the eastern part of the domain results in a cloudfree region due to subsidence.However, most of the domain is cloud covered, with cloud fraction close to 100 %.Furthermore, rainfall around 100 mm (accumulated precipitation over 96 h) on 17 February is observed along with the cyclonic circulation and the southeastern part of Europe, south of the low-pressure system.The modeled convective cloud bases are located between 500 and 4000 m over the domain.

Evaluation with satellite data
The model-derived cloud fraction is averaged daily (0:00 to 24:00 UTC) to illustrate the comparison between the model (COSP) and ISCCP satellite retrievals (Fig. 3).The observed cloud fraction shows more cloud-free regions compared to the model simulations.Nevertheless, the modelderived cloud fraction is in broad agreement with ISCCP satellite retrievals, allowing now for a more detailed analysis of the cloud microphysical properties with a fine resolution which is the focus of this study.Furthermore, a comparison of radiative fluxes with CERES (Clouds and the Earth's Radiant Energy System) satellite products is discussed in Sect.3.3.
Figure 4 shows the comparison between MODIS-retrieved (Fig. 4a-c) and COSMO-MUSCAT-simulated (Fig. 4d-f) cloud optical depth, cloud droplet effective radius, and cloud liquid water path.From the figure, it can be noticed that the simulated cloud optical depth exhibits a spatial pattern similar to the observations, albeit with higher magnitude in the MODIS level-2 retrievals (Fig. 4a and d).In the satellite retrievals, it varies between 5 and 100 and in the model between 5 and 45, with maximum values observed over similar geographical regions.However, the satellite-derived cloud optical depth and liquid water path are larger in comparison to model (COSMO-MUSCAT) outputs.The model-derived cloud effective radius exhibits both a similar spatial pattern and magnitude compared to that of the MODIS satellite retrievals (Fig. 4b and e).The modeled cloud droplet effec- tive radius varies between 3 and 16 µm, whereas it is in the range between 2 and 30 µm in the satellite retrievals.Similar to cloud optical depth, liquid water path also exhibits comparable spatial patterns for both model and observations.However, the modeled liquid water path varies between 0.025 and 0.425 kg m −2 , and in the satellite observation it varies between 0.  optical properties.The above cloud optical properties are calculated using Eqs.( 3) and ( 5) in the models, although their correlations are valid only for that particular model layer/levels.In the following, the statistical distribution of the satellite and the model cloud microphysical properties are compared and evaluated in terms of probability density functions (PDFs).Figure 5 represents the probability density function of the spatiotemporal distribution of cloud optical depth, effective radius, and liquid water path, defined as the normalized count of occurrence per bin width of cloud optical property.The cloud optical depth PDF shows that thin clouds (cloud optical depth < 10) in all model versions occur substantially more frequently than in the satellite retrievals, and thick clouds (cloud optical depth > 30) occur less frequently.The modeled cloud effective radius PDFs are constrained to 3 and 16 µm, whereas the satellite retrievals show a range of 4 to 30 µm.A shift of the PDF is found in the COSMO-MUSCAT-derived PDFs, which indicate the increased droplet size for the interactive C ccn .For liquid water path, modeled PDFs overestimate the clouds with low liquid water path and underestimate clouds with high water paths.The differences in PDFs largely follow what is found for the cloud optical depth, but model deficiencies compared to the satellite retrievals are substantially larger.The analysis also illustrates an increase in cloud optical PDF from the COSMO-MUSCAT simulation.Certainly, the drop and preponderance of modeled cloud optical properties can be influenced by model tuning, an approach which, however, has not been performed yet for the COSMO-MUSCAT model version.
The outcome of the cloud microphysics modifications can be analyzed by considering the difference between the two simulations (COSMO-MUSCAT and COSMO2M), which is shown in Fig. 4g-i.The version considering the interactive aerosol number concentration (COSMO-MUSCAT) exhibits an increase in the cloud droplet effective radius by a range of 1-4 µm throughout the domain with an overall increase of 9.5 %, although a slight reduction can be noticed in a few areas.For cloud optical depth and liquid water path, both generally show increases despite the reduction in a few areas.The revised parameterization in the coupled model has modified the spatial distribution of the cloud optical depth in the range of ±15 and the liquid water exhibits a variation in the range of ±0.12 kg m −2 .However, the domain-averaged cloud optical depth and liquid water path have been increased by 4.1 and 14.2 %, which is also observed in the PDF analysis.
The cloud droplet number concentration N d can be used as a diagnostic for aerosol cloud interactions.From satellite observations, it can be expressed in terms of cloud optical depth τ c and effective radius r e (Quaas et al., 2006), which are given by where α = 1.37 × 10 −5 m −0.5 .Likewise, the model-derived N d is also estimated using Eq. ( 9), which uses COSP-derived (MODIS simulator) cloud optical depth and effective radius.Figure 6a-c show the comparison between modeled (COSMO2M and COSMO-MUSCAT) and observed N d .On 17 February 2007, the domain-averaged N d values are 153, 120, and 378 cm −3 for COSMO2M, COSMO-MUSCAT, and MODIS, which indicates an underestimation of modelderived values (Fig. 6a-c) compared to MODIS (Y.Zhang et al., 2012;Storelvmo et al., 2009).The intermodel comparison (COSMO2M and COSMO-MUSCAT) reveals that COSMO-MUSCAT-derived N d shows a decrease of 21.5 %.
Figure 6d shows the spatial distribution of sulfate aerosol number concentration (aerosol number concentration proxy) below the convective cloud base (representative of aerosols in the model and it is also averaged for 8-14 h on 17 February 2007), where high number concentrations are simulated over southeastern Europe.In contrast, N d values are smaller over the same region.This is because the models with Boucher and Lohmann (1995) parameterization saturate N d over high aerosol or polluted regions (Penner et al., 2001), and the high pressure in this region results in trapping aerosol in the boundary layer.From the above analysis, it can be inferred that COSMO-MUSCAT can be used as a tool for regional aerosol-cloud interaction estimates.The interactive aerosol coupling results show an increase in cloud droplet effective radius by 9.5 % and a reduction in N d by 21.5 %.

Impact on radiative balance
In addition, we have also implemented aerosol-cloudradiation interactions in the COSMO model by revising the radiation scheme to make use of a droplet-size-dependent cloud optical depth  simulations, which are also averaged daily (Fig. 8).From the figure, the regions with reduced cloud optical depth and liquid water path are correlated with increased net shortwave flux at the surface and decreased net longwave flux at the TOA.To illustrate the combined effect of the revised radiation scheme and interactive aerosols, COSMO-MUSCATderived radiative fluxes are compared to CERES satellite observations (Kato and Loeb, 2003;Loeb et al., 2012).For comparison, we have considered computed CERES fluxes (derived based on state and composition of the atmosphere, surface, and the incoming solar radiation) with spatial resolution 1 • × 1 • , and care must be taken while interpreting the results.Also, during winter, the uncertainty in CERES flux observations is slightly higher (Guo et al., 2007).The spatial pattern and the magnitude of model-simulated fluxes are comparable to satellite observations in which the surface net downward shortwave flux ranges between 20 and

Conclusions
This paper discusses the modification of the Seifert and Beheng (2006) two-moment scheme in the COSMO model.This has been done with aerosol information from the onlinecoupled MUSCAT model, which allows for a microphysical aerosol effect on clouds.It has been achieved by replacing the constant cloud condensation nuclei profile in the COSMO two-moment scheme with gridded aerosol information derived from the online-coupled MUSCAT model, using the Boucher and Lohmann (1995) parameterization, which takes sulfate aerosol as a proxy for all aerosols.In addition, the radiation scheme is revised to a droplet-size-dependent cloud optical depth, allowing now for aerosol-cloud-radiation interactions.In order to facilitate an evaluation using satellite retrievals, the COSP satellite simulator has been incorporated into the modeling system which runs online in the model.The model results are evaluated with satellite observations from the ISCCP, MODIS, and CERES projects and instruments, respectively.Since the cloud microphysics modification has been done for cloud droplet nucleation, the anal-ysis is restricted to the liquid part of clouds in the model, and MODIS level-2 cloud products are screened for liquidphase cloud products.Although the two-moment cloud microphysics and radiation scheme in the COSMO model has been modified, the model was not re-tuned to get reasonable 2 m temperature or precipitation.
A case study has been carried out to compare the model output with observations.The incorporated COSP satellite simulator serves as a link between model and satellite comparisons.Despite the resolution, the COSP-derived ISCCP cloud fraction shows a similar spatial pattern and magnitude.Further, MODIS level-2 cloud optical products such as cloud optical depth, effective radius, and liquid water path are compared.The COSMO-MUSCAT-derived cloud optical properties show a similar spatial distribution compared to the MODIS observation.In COSMO-MUSCAT, the cloud optical depth has been increased by 4.1 %, cloud droplet effective has been increased by 9.5 %, and liquid water path has been increased by 14.2 % in comparison to COSMO2M.In turn, the cloud droplet number concentration estimated from the COSMO-MUSCAT model shows a reduction of 21.5 % compared to the COSMO2M model.Furthermore, considerable changes in the radiation budget have been found.This analysis indicates that the coupled model (COSMO-MUSCAT) with interactive aerosol treatment results in an increase in cloud droplet size and a reduction in cloud droplet number concentration by activation and growth of droplets, which illustrates implicit aerosol-cloud interactions.Also, the cloud properties in COSMO-MUSCAT agree reasonably well with observations, so that it can be used for regional aerosol-cloud interaction studies.As a next step, further improvement in the two-moment scheme will be carried out through the use of the newly included aerosol model M7 (Vignati et al., 2004) framework in the MUSCAT model, which is able to provide aerosol number concentration information to the COSMO two-moment scheme by replacing Boucher and Lohmann (1995) parameterization.This can result in more precise cloud droplet activation parameterization, involving different aerosol species such as C ccn , and thus improving the microphysical aerosol effect on clouds (Lohmann et al., 2007).Also, the role of aerosols on ice nucleation will be addressed.
Code and data availability.The COSMO-MUSCAT(5.0)model is freely available under public license policy.The source code, external parameters, and documentation can be obtained through Ralf Wolke (wolke@tropos.de).

Figure 2 .
Figure 2. Model synoptic conditions for 17 February 2007 at 12:00 UTC: (a) surface pressure in contours and 2 m temperature ( • C) as color shading, (b) 500 mb wind vectors and total cloud area fraction as color shading.
25 and 1.0 kg m −2 .The white regions (missing val-ues) in satellite retrievals can be explained by the very strict quality filtering of the MODIS cloud products.The domainaveraged cloud optical depth, effective radius, and liquid water path are 23.34,11.30 µm, and 0.175 kg m −2 , whereas the COSMO-MUSCAT-derived values are 7.60, 9.93 µm, and 0.056 kg m −2 , which illustrates an underestimation of all simulated quantities compared to the satellite-derived cloud

Figure 7 .
Figure 7.Comparison and difference between shortwave and longwave radiation fluxes at the surface and top of the atmosphere, and the difference between the two simulations (COSMO2MR radiation coupled minus COSMO2M).

Figure 8 .
Figure 8. Daily averaged cloud optical depth and liquid water path differences between COSMO2MR and COSMO2M on 17 February 2007.

Figure 9 .
Figure 9.Comparison between shortwave and longwave fluxes at the surface and top of the atmosphere with CERES satellite fluxes and correlation between satellite (CERES) and models (COSMO2MR and COSMO2M).
(Bodas-Salcedo et al., 2011) et al., 2012), Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO;Chepfer et al., 2010), and the CloudSat cloud radar(Marchand et al., 2009).To produce similar output to satellite data, COSP requires the grid mean vertical profile of temperature, humidity, hydrometer mixing ratios, cloud optical thickness and emissivity, the surface temperature, and emissivity from the model.It produces output comparable with satellite data in three steps.First, it addresses the mismatch between model and satellite pixel resolution by generating subcolumns using model information about subgrid-scale variability, e.g., from the assumption on vertical overlap of fractional cloudiness.Second, vertical profiles of individual subcolumns are passed to each instrument simulator.Lastly, the COSP statistic module gathers the output from all instrument simulators(Bodas-Salcedo et al., 2011).COSP is implemented online in the COSMO model with hourly output.While using COSP facilitates a more consistent comparison between model output and satellite data, differences between the model simulation and the satellite can, for example, still arise due to displacements in simulated storm tracks.COSP has previously been used with COSMO by Rossow and Schiffer, 1999))al., 2011)has been developed.The COSP satellite simulator produces model diagnostics which are fully consistent with satellite products such as the International Satellite Cloud Climatology Project (ISCCP;Rossow and Schiffer, 1999), Moderate Resolution Imaging Spectroradiometer (MODIS;