Increasing the aerosol number in warm-phase clouds is thought to decrease the rain formation rate, whereas the physical processes taking place in mixed-phase clouds are more uncertain. Increasing number concentrations of soluble aerosols may reduce the riming efficiency and therefore also decrease precipitation. On the other hand, the glaciation of a cloud by heterogeneous freezing of cloud droplets may enhance the formation of graupel and snow. Using a numerical weather prediction model with coupled aerosol microphysics, it is found, in a statistical framework with 270 clean and polluted 2D simulations of mixed-phase precipitation over an Alpine transect, that the presence of the ice phase determines the magnitude and the sign of the effect of an increasing aerosol number concentration on orographic precipitation. Immersion/condensation freezing is the only ice-nucleating process considered here. It is shown that this indirect aerosol effect is much less pronounced in cold simulations compared to a warmer subset and that cloud glaciation tends to compensate the loss of rain in polluted situations. Comparing the clean and polluted cases, a reduction of rain by 52%, on average (std dev = 25%), over the transect in the polluted cases is found. For frozen precipitation a much broader range of differences is found (mean = +4%, std dev = 60%). Furthermore, this study shows that in comparison with the clean cases more precipitation spills over to the leeward side of the major ridge in the polluted cases (median = +14.6%).
Earliest human cultures detected that orography has a strong influence on precipitation and therefore plays an important role in defining local hydrology and climate. Orography is considered a major factor for modifying or amplifying precipitation (Smith 2006). A variety of different mechanisms that trigger orographic precipitation have been described (Roe 2005). Classically, orographic precipitation is associated with moist-adiabatic upslope ascent in a stably stratified environment. In this case, a moisture-laden air mass is mechanically forced above the lifting condensation level (LCL) where the atmosphere is supersaturated with respect to water so that water vapor condenses onto cloud condensation nuclei (CCN) and latent heat is released. The formation of an orographic cloud may be followed by precipitation. Precipitation in a warm-phase orographic cloud is initiated by collisional growth and collection of cloud droplets. Once the air mass is lifted above the freezing level, cloud droplets may eventually freeze. With the onset of the Wegener–Bergeron–Findeisen process, which describes the deposition of water vapor upon ice crystals at the expense of evaporating cloud droplets due to the differences in vapor pressure over water and ice in a supercooled cloud between 0° and −38°C, ice crystals grow to larger sizes. Finally, snowflakes can form due to aggregation of ice crystals. Graupel particles are generated when frozen drops or ice crystals collect liquid droplets that freeze on the ice surface.
Estimating the role of aerosols for cloud and precipitation formation is considered one of the key challenges in current climate science according to the latest Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) (Solomon et al. 2007). The role of microphysical processes for orographic clouds was reviewed by Khvorostyanov (1995). Aerosols can act as CCN and ice nuclei (IN) and thereby influence the microphysical characteristics of clouds (Pruppacher and Klett 1997). Increasing the number of potential CCN in a cloud with constant liquid water content (LWC) leads to more but smaller cloud droplets (Twomey et al. 1984; Peng et al. 2002; Lowenthal et al. 2002). Since smaller droplets have lower collision efficiencies than larger droplets (Pruppacher and Klett 1997), and collision/coalescence is a major process for the formation of rain in warm-phase clouds, precipitation is potentially suppressed due to a reduction of the precipitation efficiency. Observations from Borys et al. (2000, 2003) suggest that in mixed-phase clouds the snowfall rate tends to decrease with higher anthropogenic aerosol load owing to an inhibition of the accretion of cloud droplets by snowflakes (snow riming). As a consequence of this inhibition in combination with the slowdown of rain formation, a suppression of precipitation can generally be expected in mixed-phase clouds in such cases.
However, insoluble aerosol components serving preferably as heterogeneous IN such as dust, and perhaps black carbon, may cause an inverse effect and enhance precipitation formation via the ice phase [glaciation indirect aerosol effect (IAE) (Lohmann 2002; Lohmann and Diehl 2006)]. Their presence may lead to an enhanced freezing of cloud droplets, which would amplify the Wegener–Bergeron–Findeisen process and subsequently also aggregation and riming.
Heterogeneous ice formation occurs through deposition, condensation/immersion freezing, and contact freezing, depending on ambient temperature and ice supersaturation (Pruppacher and Klett 1997; Vali 1985). Most of the current models still apply simple diagnostic schemes to predict ice crystal number concentrations as a function of temperature (Cooper 1986; Fletcher 1962) or supersaturation with respect to ice (Meyers et al. 1992). Recently, several parameterizations have been developed to account for the interactions between different IN types and clouds (Khvorostyanov and Curry 2004; Phillips et al. 2008; Diehl and Wurzler 2004). These schemes include surface properties of the IN and therefore account for the importance of the chemical composition and the presence and efficiency of active sites on a particle (Eidhammer et al. 2009). Both the parameterizations of Phillips et al. (2008) and Diehl and Wurzler (2004) make the assumption of the singular hypothesis. The validity of the singular hypothesis is supported by various recent observational studies (DeMott 1990; Marcolli et al. 2007; Möhler et al. 2006). In the present study, a modified version of the scheme developed by Diehl and Wurzler (2004) is applied (Muhlbauer and Lohmann 2009).
Because of the aforementioned complex and competing effects, no conclusive quantification of the indirect aerosol effect on precipitation in mixed-phase clouds has been made so far (Denman et al. 2007). Statistical approaches based on annual precipitation data by Givati and Rosenfeld (2004), Rosenfeld and Givati (2006), Rosenfeld et al. (2007), and Jirak and Cotton (2006) suggest that the second indirect aerosol effect presumably leads to precipitation losses of 15%–25% downwind over topographic barriers in Colorado and Israel. Precipitation was suppressed on the windward slope, and an enhancement of precipitation was found on the lee side. This effect was shown for relatively polluted areas in which aerosol emissions increased over the last decades, whereas in more pristine surroundings no changes in the precipitation data were detected. On the other hand, Alpert et al. (2008) find contradicting results regarding changes of orographic precipitation over time in Israel, which reflects the large uncertainty of aerosol effects on precipitation. Modeling studies by Lynn et al. (2007) obtained a 30% decrease of precipitation if aerosol conditions were changed from maritime to continental. In a sensitivity study, they show that the relative humidity and the horizontal wind speed are important factors governing the cloud structure and precipitation amounts. Muhlbauer and Lohmann (2008) conducted idealized simulations and showed that the sensitivity of orographic precipitation to changes in the aerosol depends strongly on the orographic flow dynamics. Furthermore, Muhlbauer and Lohmann (2009) suggest that the aerosol indirect effect on orographic precipitation can change sign depending on the mixing state of the aerosols and their inherent freezing properties.
To improve our understanding and to quantify uncertainties regarding the indirect effect of aerosols on orographic precipitation we evaluate the microphysical processes in mixed-phase clouds for different thermodynamic conditions in a statistical sense. We conducted 270 pairs of 2D simulations with clean and polluted remote continental aerosol configurations typical for the Alpine region and Switzerland, using a numerical weather prediction model coupled to a sophisticated two-moment cloud and aerosol microphysics scheme. Permitting very large variability in the thermodynamic initial state and boundary conditions, the statistical perspective allows us to give a more reliable estimate of the potential range of changes in orographic precipitation due to an increase of the aerosol number concentration. The primary goal of this article is to explore the importance of the ice phase for the indirect aerosol effect on precipitation in the presence of black carbon and dust as efficient ice nuclei in the condensation/immersion freezing mode. Estimating the large uncertainties in the physical representation of the underlying processes, however, is beyond the scope of this article.
a. Numerical model
In this study we apply the nonhydrostatic weather prediction Consortium for Small-Scale Modeling (COSMO) model, version 3.19. The model was initially developed and applied by the German Weather Service, the Swiss National Meteorological Office (MeteoSwiss), and other members within COSMO (http://www.cosmo-model.org) (Doms and Schättler 2002; Steppeler et al. 2003).
COSMO integrates the fully compressible hydrothermodynamical equations in conservation form. A split-explicit third-order Runge–Kutta scheme is used for time integration (Wicker and Skamarock 2002; Förstner and Doms 2004). The spatial discretization is based on a fifth-order upstream advection scheme on an Arakawa C-grid with Lorenz vertical staggering. In the vertical, height-based Gal-Chen coordinates are applied (Gal-Chen and Sommerville 1975).
The second-order Bott scheme is used for the advection of the moisture variables and the aerosols (Bott 1989). Vertical turbulent diffusion is treated with a turbulence scheme using Smagorinsky-type stability functions (Herzog et al. 2002). Open boundary conditions are imposed at the lateral boundaries (Davies 1976). For the upper boundary a Rayleigh damping layer starting at 11 km of altitude is introduced to reduce gravity wave reflections from the top.
Given the semi-idealized two-dimensional setup of our simulations, the influence of the Coriolis force is neglected to avoid wind perpendicular to the flow direction. As the focus of the present study is on microphysical effects only, aerosol–radiation interactions and likely accompanying feedback mechanisms are not accounted for. This may be considered a major caveat of the framework applied here. However, there is evidence that microphysical effects on clouds dominate the radiative effects below a certain threshold of aerosol optical depth (Koren et al. 2008). Given the very remote continental regime of this study (see details of the model setup below), one can expect that the microphysics should dominate.
Furthermore, as the focus of this study is on stably stratified orographic flow, no parameterization for convection is employed. Soil processes are neglected owing to the short time scale of our simulations.
A detailed description of the model topography and initial conditions is given below in section 2b.
1) Aerosol microphysics
The model is coupled to the aerosol module M7 (Vignati et al. 2004). It is also used within the framework of the general circulation model ECHAM5-HAM (Stier et al. 2005). The modal concept of M7 is represented by a superposition of seven lognormal size distributions, such that
In Eq. (1), Ni denotes the aerosol number concentration in mode i; σg,i is the geometric standard deviation representing the width of the particle size distribution of each mode. The count median radius can be determined from the respective aerosol number and mass in each mode. Number and mass concentrations are advected as tracers in the model.
The module consists of four internally mixed modes, containing both soluble and insoluble compounds, and three insoluble modes including aerosol species characterized by low water solubility. The internally mixed modes can contain sulfate (SU), black carbon (BC) and particulate organic matter (POM), sea salt (SS), dust (DU), and aerosol water. In M7, the modal composition depends on aerosol microphysics and, thus, processes such as homogeneous nucleation of sulfuric acid (Vehkamäki et al. 2002), intra- and intermodal coagulation, coating of particles with sulfate, and water vapor uptake (Jacobson et al. 1996; Stier et al. 2005). For the sake of brevity, we refer to Vignati et al. (2004) and Stier et al. (2005) for further information on the aerosol microphysics in M7. An overview of the modal structure of M7 is given in Table 1.
2) Cloud microphysics
Cloud microphysical processes are treated with a two-moment bulk scheme from Seifert and Beheng (2006), which is coupled to the aerosol microphysics module as discussed in detail in Muhlbauer and Lohmann (2008, 2009). A total of five hydrometeor classes are accounted for: cloud droplets, rain, ice crystals, snow, and graupel. Following Heymsfield and Kajikawa (1987), ice crystals are assumed to be hexagonal plates, whereas snowflakes have the shape of mixed aggregates (Locatelli and Hobbs 1974). In addition, the properties for so-called lump graupel (Heymsfield and Kajikawa 1987) are applied. Generalized gamma distributions based on the corresponding shape parameters are applied for the size distributions of the five classes.
The simulated warm-phase processes are the nucleation of cloud droplets (source of cloud water), autoconversion of cloud droplets to rain drops, and accretion of rain by cloud droplets, as well as self-collection of cloud droplets and rain (sources for rain but sinks for cloud water), evaporation, and breakup of rain drops (sinks for rainwater). Microphysical processes including the ice phase are heterogeneous freezing in condensation/immersion mode, diffusional growth of ice crystals (sources for ice), aggregation, self-collection (sources for snow, sinks for ice), riming, conversion to graupel (sources for graupel; sinks for ice, cloud droplets, rain, and snow), melting, and sublimation (sinks for ice). Graupel shedding of liquid water and Hallett–Mossop ice multiplication are considered as well. The Wegener–Bergeron–Findeisen process is represented implicitly such that, in a lifting air parcel that is supersaturated with respect to ice but not saturated with respect to water, all condensate evaporates and ice crystals grow by vapor diffusion. For a description of the corresponding parameterizations of these processes, we refer to Seifert and Beheng (2006) and Muhlbauer and Lohmann (2008, 2009).
In the framework of this study, contact nucleation was neglected since all aerosols were assumed to be internally mixed and, hence, contributions from insoluble dust or soot particles were not taken into account. Observational findings from Weingartner et al. (2002) support this simplification in a climatological mean. Additionally, homogeneous nucleation is not considered, as temperatures below −38°C are required for this process. These are typically found in the upper troposphere (Pruppacher and Klett 1997) in a midlatitude atmosphere and do not play a role in low- or midlevel orographic clouds over the Alps.
Furthermore, Ansmann et al. (2008) argue based on lidar observations that in mixed-phase clouds the time required by deposition freezing to form ice crystals may be too short. Water saturation is reached in the updrafts already after several minutes. They indicate that, for deposition freezing to be efficient in nature at temperatures of about −10°C and relative humidities of 90%–95%, a complete absence of vertical motion is necessary. Thus, we also neglect deposition nucleation.
Condensation and immersion freezing are implicitly combined in our model as pointed out by Muhlbauer and Lohmann (2009). The scheme combines the singular and the stochastic hypothesis. Thereby it is assumed that heterogeneous freezing occurs inside supercooled cloud droplets. The probability of freezing depends on supercooling and droplet volume (Pruppacher and Klett 1997). The probability of freezing is enhanced for a fixed temperature by the immersed aerosols depending on its immersion nucleation efficiency. Reasonable arguments have been provided recently by Marcolli et al. (2007) and Vali (2008) that a pure stochastic approach may not be appropriate for condensation/immersion freezing. The parameterization used here is based on the freezing efficiencies of black carbon and mineral dust presented by Diehl and Wurzler (2004) and is described in great detail in Muhlbauer and Lohmann (2009). The major equations are given in the appendix.
b. Model setup
1) Computational domain
A two-dimensional setup along a meridional transect through the Alps with 50 vertical model levels is used for this study. The horizontal grid spacing is 2.2 km as used operationally by MeteoSwiss. The time step applied in all simulations is 20 s. Integrations over 12 h are performed.
Due to the availability of local aerosol size distribution measurements, the effect of aerosols on stratiform orographic precipitation is evaluated in the surroundings of the Jungfraujoch (JFJ), a high mountain ridge in central Switzerland that is oriented almost parallel to the Alps. The JFJ is located at 46.6°N, 8.0°E and its altitude corresponds to 3571 m with respect to the Mediterranean Sea. In the model the height at the JFJ is 3454 m due to filtering of the topography. Figure 1 shows the operational topography and the corresponding transect used for this study.
3) Initial conditions
To perform a statistical analysis, a series of pairs of model simulations was conducted. The initial state and the lateral boundary conditions of all simulations are given by an aerosol size distribution and a 0000 UTC sounding from the station of Payerne (LSMP 06610), a station northwest of the JFJ. The sounding provides vertical profiles for absolute temperature T, pressure p, relative humidity RH, and the wind components u (zonal) and υ (meridional). The vertical profiles are interpolated linearly from the original soundings to the model levels. For the horizontal wind speed, the absolute value of the wind vector is applied to avoid artificial shear layers in the 2D simulations. Nighttime soundings are used to minimize the possibility of small-scale (embedded) convection.
The soundings have been selected such that the average relative humidity in the lowest 3000 m is between 80% and 90% in order to yield orographic clouds. Hence, the temperature determines the specific humidity such that more moisture is available in warm simulations. The second criterion for the choice of the soundings was that temperatures and wind speeds correspond to climatological measurements at the JFJ. A comparison is provided in section 3. A set of 270 initial soundings from all seasons of the years 2002–06 were simulated separately with a clean and a polluted aerosol setup. Several simulations showed no orographic precipitation within the domain. Therefore, we only analyze simulations that produced more than 0.5 mm (12 h)−1 precipitation at a given gridpoint in the domain. The number of remaining clean and polluted simulations that are analyzed further is 166. Figure 2 shows the range of values for temperature T, specific humidity qv, horizontal wind speed U, and equivalent potential temperature Θe from all soundings used in the statistical analysis. At the surface, T and Θe range from −5° to +20°C and from 0° to 50°C, respectively. The specific humidity decreases rapidly with height. Surface values are between 2 and 15 g kg−1. The surface horizontal wind is set to zero. Near the highest mountain crests of the domain, the winds range from 0 to about 33 m s−1. Note that all soundings are stably stratified or close to neutral stratification.
For the evaluation of the indirect aerosol effect on precipitation, two aerosol setups are applied separately for each sounding. Figure 3 shows the corresponding number and mass distributions that are obtained from a typical winter and a summer climatology of Scanning Mobility Particle Sizer (SMPS) measurements at the Global Atmosphere Watch station on top of JFJ (Weingartner et al. 1999). The total number (mass) concentration is 350 cm−3 (0.5 μg m−3) in winter (clean case) and 790 cm−3 (2.2 μg m−3) in summer (polluted case). The geometric standard deviations are 1.59 μg m−3 for the nucleation mode, 2.13 μg m−3 for the Aitken mode, 1.61 μg m−3 for the accumulation mode, and 2.00 μg m−3 for the coarse mode. For the nucleation and coarse mode, the values have been adopted from Stier et al. (2005), whereas the standard deviations of Aitken and accumulation mode have been taken from the aforementioned measurements.
The attribution of the mass fractions derived from the total aerosol mass in each mode to black carbon, particulate organic matter, sulfate, and dust is based on aerosol mass spectrometry data as discussed in Cozic et al. (2007). For the aerosol components potentially acting as immersion nuclei, 5.5% (4.9%) of the total mass of the Aitken (accumulation) mode are attributed to black carbon in the clean case. In the polluted case, the fractional contributions of black carbon are 3.8% (3.5%) in Aitken (accumulation) mode. Mineral dust is only considered in the accumulation mode at initialization, the corresponding dust fractions of the total mass being 11% (8.1%) in the clean (polluted) case. A contribution from sea salt is neglected given the choice of remote continental Alpine aerosol conditions. The particle number and mass concentrations are assumed constant with height.
Aerosols with a radius greater than 35 nm (the larger part of the Aitken mode, complete accumulation, and coarse mode) can be activated to form cloud droplets in the model, depending on the vertical velocity. The activation scheme is based on Lin and Leaitch (1997) and further explained in Muhlbauer and Lohmann (2008).
The same two aerosol configurations have previously been used by Muhlbauer and Lohmann (2008). Note that aerosol removal processes such as dry deposition, sedimentation, and wet scavenging are not taken into account because of the relatively short duration of the simulations and the lack of appropriate emission inventories on the alpine scale.
In this section we analyze the indirect aerosol effect of the 2D simulations statistically with respect to mixed-phase cloud formation and orographic precipitation. We focus on the evaluation of the spillover factor, mean domain precipitation, and mixed-phase microphysical conversion rates.
a. Aerosol effects on precipitation patterns and mean domain precipitation
The reduction of the precipitation efficiency that is suggested to result from an increased aerosol number may lead to a shift in the precipitation pattern such that the windward precipitation is reduced while on the lee side an enhancement is found (Givati and Rosenfeld 2004; Rosenfeld and Givati 2006; Rosenfeld et al. 2007). Figure 4 depicts the pattern of the mean accumulated precipitation after 12 h of integration over all simulations. The largest amount of precipitation in the clean cases is found on the windward side of the major ridges R1, R2, and R3 (JFJ). In the mean for the polluted cases, the maximum total precipitation (sum of rain, snow, and graupel) is found at R2 because it is the primary orographic barrier (airflow from the left). The maximum is shifted by 9 km toward the top of R2 as compared to the clean case mean owing to the slowdown of rain formation. The rainfall maxima coincide with the maxima of total precipitation. Table 2 lists the maxima for all precipitation types within the ranges indicated for the three ridges (see bottom panel of Fig. 4).
As frozen precipitation may also form in higher cold clouds, it is more widespread and thus of similar magnitude at R2 and R3. R1 is a comparatively weak orographic barrier due to lower height, such that less precipitation forms within its range as compared to R2. Frozen precipitation (sum of snow and graupel) is highest at R2 in the clean cases and R3 in the polluted cases. In the lee of R2 and at R3, the frozen precipitation increases in the polluted cases. The locations of the frozen precipitation maxima do not shift. The increase is therefore caused by enhanced mixed-phase conversion rates in the polluted cases as discussed below and not by a shift of the maxima.
Rain dominates most simulations and therefore the mean because snowflakes or graupel particles, which may have formed above, melt as they fall below the freezing level. The temperatures in the initial soundings are above 0°C in most cases below 1000-m altitude, as shown in Fig. 2. At the JFJ, the temperatures of the initial soundings range from −20° to 5°C (mean = −6.2°C, std dev = 6.7°C). The measured climatological daily mean over the period 2002–06 corresponds to −6.5°C (std dev = 6.6°C) at this station. The daily mean wind speed within this time period is 7 m s−1 (std dev = 4 m s−1). The corresponding values in the initial soundings at the altitude of the JFJ range from almost 0 std dev = 6.7°C to 28 m s−1.
According to studies by Jiang (2003, 2006, 2007), Miglietta and Buzzi (2004), and Colle (2004) orographic clouds tend to form close to the top of a ridge in the case of narrow and steep slopes or situations of windward blocking resulting from a strong orographic barrier and weak horizontal winds in the lower troposphere. In comparison with a clean case, these clouds may be advected upstream in a polluted case because of the slowdown of precipitation formation, leading to a shift in the precipitation pattern. Such a shift is commonly evaluated by the spillover factor (SP) (Jiang and Smith 2003; Muhlbauer and Lohmann 2008). It corresponds to the ratio of lee precipitation to total precipitation such that
In Eq. (2), b represents the location (grid point) of the mountain peak and a denotes the beginning of the slope on the windward side of the corresponding mountain; Pi is the accumulated precipitation after 12 h, while n simply equals the lowest altitude on the leeward side ([Pi] = kilograms water equivalent). We calculated SP and conducted Wilcoxon rank-sum tests1 individually for each of the three major ridges in the computational domain (Table 3). A median increase of leeside precipitation in the polluted cases is found for the two primary ridges R1 and R2. The change at R1 is comparatively small because the ridge is not such a strong orographic barrier as R2 and R3. Given the relatively high precipitating clouds (see below) in comparison with the altitude of R1, a smaller change is expected as more precipitation can spill over R1 than, for example, over R2.
A decrease of lee side precipitation is obtained for R3. This corresponds to an increase in precipitation on the windward side of R3, which is explained by the enhanced frozen precipitation beyond R2. The differences in SP at R2 and R3 are significant, as indicated in Table 3.
The changes in spillover precipitation at R1 and R2 due to an increase in the aerosol number are primarily related to the effect of these aerosols on warm-phase microphysics and, hence, the direct formation of rain from cloud droplets. As Fig. 4 indicates, the decrease in polluted case frozen precipitation on the windward side of R2 is much smaller in magnitude than its enhancement beyond that ridge. This suggests that ice-phase microphysics could potentially weaken, compensate for, or even reverse the second indirect aerosol effect on rain formation.
If the lee side precipitation is not enhanced enough to compensate for the loss of precipitation on the windward side, the accumulated mean domain precipitation is decreased in the polluted case as compared to the clean case. Here, we analyze the relative differences in mean domain precipitation between clean and polluted cases [relative precipitation difference (RPD)]. It is interpreted as the relative change in the net precipitation budget. RPD does not give information about the nature of the precipitation event (e.g., short and heavy burst or slow and widespread drizzle). However, the mixed-phase conversion rates reach an equilibrium within the first 1–3 h (not shown). Thereafter, precipitation increases evenly in general.
For a single case, RPD is defined such that
is the clean case precipitation at grid point i and corresponds to the precipitation in the polluted case. The domain size n equals 350. Negative values of RPD represent an overall loss of precipitation due to an increase in aerosol number, whereas positive values indicate that more precipitation is formed on average in the polluted cases as compared to the clean cases. Figure 5 shows the empirical cumulative distribution functions for RPD of rain, frozen, and total precipitation for all 166 simulation pairs (see section 2b for details on the selection of the simulations). In the case of rain, nearly all simulations yield negative values. More than half of the simulations show a decrease in rainfall of more than 50% (RPD < −0.5). The mean corresponds to −0.52 (std dev = 0.25).
The processes that form snow or graupel are much more complex than the warm-phase microphysics and therefore, as argued previously, the frozen precipitation may be both increased (glaciation indirect aerosol effect) or decreased (reduced riming efficiency) in the polluted cases. Therefore, the distribution of RPD for frozen precipitation is much broader with a mean of +4.2% (std dev = 60%). About half of the simulations produced more frozen precipitation in the polluted cases than in the clean cases. About 10% of all cases show an increase of frozen precipitation of more than 100% (RPD > 1). Therefore, the effect on total precipitation is reduced in comparison with the effect on rainfall, such that an average reduction of 36% is found.
b. Impact of environmental conditions on aerosol effects
To investigate under what conditions a strong (large negative RPD) or a weak second indirect aerosol effect (small negative or positive RPD) is to be expected, we filtered the data further to exclude cases in which heavy precipitation in both clean and polluted simulation is produced. In these cases, RPD tends to be small mainly because the difference between the polluted and the clean simulation is much smaller than the absolute value of the precipitation in the clean case [see Eq. (3)]. Figure 6 displays RPD as a function of the clean case mean domain precipitation. It shows that the spread of RPD values is very small for cases with stronger mean domain precipitation and rather large for those with less precipitation in the domain. Thus, RPD asymptotically approaches zero with increasing precipitation rate, implying that clouds are less susceptible to changes in the aerosol number in case of strong precipitation formation.
Therefore, simulation pairs with more mean domain precipitation than those in the gray box in Fig. 6 are excluded. The threshold corresponds to 1.5 mm (12 h)−1 (85th percentile) and was chosen arbitrarily. Hence, 112 simulation pairs remain for further analysis.
Of the 112 analyzed cases, 25% have an RPD > 14.1%. They are hereafter referred to as the cases showing a weak second indirect aerosol effect (IAE). The 25% of cases with the strongest IAE have an RPD < −67.6%. Thus, we expect the strong effect to be related to rather warm simulations that primarily produce rain and exhibit only negligible contributions of ice-phase microphysics. On the other hand, enhanced cloud glaciation (ice crystal formation through condensation/immersion freezing) in cold soundings is suggested to counteract the second indirect effect and increase precipitation via aggregation and riming to form snow and graupel, presumably yielding a weak second IAE.
As shown in Fig. 7, the simulations featuring a strong IAE on orographic precipitation indeed produce more rain than the cases corresponding to a weaker effect. At ridge R2, the maximum rainfall for the cases with a strong (weak) IAE is 13 mm (12 h)−1 [0.5 mm (12 h)−1] in the respective clean case mean. The cases with a weak IAE show a different precipitation pattern with a maximum beyond R3. In the polluted situation, the cases with a strong IAE show the expected large decrease in precipitation [−12 mm (12 h)−1 at R2]. No average change in precipitation is found for the cases with a weak IAE. The latter exhibit frozen precipitation of 5 mm (12 h)−1 at R2, indicating that the cold phase is of much higher importance in these simulations compared to the subset with a strong IAE in which the maximum frozen precipitation is smaller than 0.5 mm (12 h)−1.
We use the distance from the freezing level (LFR: the atmospheric level above which temperatures are below 0°C) to the lifting condensation level from the initial soundings as an indicator for the importance of ice-phase processes in the simulated mixed-phase clouds. The lower the freezing level at constant LCL, and therefore the smaller the distance between LFR and LCL, the more important is the ice-phase contribution to hydrometeor formation. With an LFR higher than 5000 m, there would be no ice formation at all since there is hardly any moisture available above this altitude (see Fig. 2). Box plots for the distributions of (LFR − LCL) for both the subset with the weak as well as the strong IAE are displayed in Fig. 8a. The medians of the two individual distributions (0 m for a weak IAE, 2000 m for a strong IAE) differ significantly (p < 0.001), as determined with the previously applied rank-sum test. Another indicator for the potential strength of the ice phase is the distance from the LFR to the mixed-phase cloud top. The latter is also larger in the cases with a weak IAE since the LFR is lower than in the subset with a strong IAE for a similar cloud-top height (Figs. 10a,b and 11a,b).
Figure 8b shows that a strong IAE is related to larger differences in the warm-phase conversion rates (autoconversion + accretion) between the polluted and the clean cases. On average, the polluted case rates rw,poll are 3 μg m−3 s−1 smaller than the ones for the clean cases rw,clean, as rain drop formation is decelerated in case of a strong IAE. The differences between clean and polluted mixed-phase conversion ration rates (ri,clean, ri,poll) are about of the same order of magnitude for the weak IAE but remain near 0 for the cases with a strong IAE (Fig. 8c). This indicates that riming and aggregation rates can either increase or decrease in the polluted cases, depending on the thermodynamic conditions of the atmosphere. It also shows that warm-phase processes cause the major difference between the two subsets. These processes are much less important in the cases with a weak IAE.
Thus, the major difference between the two subsets is the vertical distribution of temperature (Fig. 9a). The temperatures within the lowest 7000 m of the atmosphere are 7°–16°C colder in the subset with a weak IAE than in the cases with a strong IAE. The latter is therefore also referred to as the warm subset, whereas the simulations with a weak IAE are referred to as the cold subset. The temperature differences are near −7°C in the areas of the dry downdrafts where strong evaporative cooling occurs in the warm subset. At an altitude of 4000 m the average temperature of the cold subset is about −20°C as compared to −8°C in the warm subset. Below 2500 m the temperatures in the cold subset are negative, whereas they are positive in the warm subset. Therefore, graupel and snow can melt below 2500 m in the warm subset and thus fall out as rain. Adiabatic warming on the lee side of the major mountain ridges explains the temperature increase in comparison with windward side conditions.
Another important aspect affecting the cloud formation and precipitation pattern in the two subsets is the vertical velocity. Depending on the location and strength of up- and downdraft regions, more or less cloud droplet nucleation or evaporation occurs. The vertical velocities are mainly determined by the incident horizontal wind and the steepness of a mountain slope. Even under stably stratified conditions, vertical velocities can be higher than 1 m s−1 in typical mountain waves (Lilly and Kennedy 1973).
In addition to the warmer temperatures, the subset with a strong IAE is also characterized by stronger horizontal winds (Fig. 9b). A maximum difference of 17.5 m s−1 is found in the downdraft region of the JFJ. As a result, these downdrafts are weaker by more than 1 m s−1 in the cold subset (Fig. 9c). Additionally, the weaker horizontal winds also slow down the updraft on the windward side of R3.
Due to a combination of lower temperatures and smaller vertical velocities in the updraft zones, one order of magnitude less cloud water is formed within 12 h in the cold subset (Figs. 10a,b). It can be assumed that the temperature effect is more important since there is a strong dependence of the liquid water content on temperature, for example, in frontal mixed-phase clouds below −10°C (Korolev et al. 2003). In absolute terms, an increase in cloud water of up to 0.5 g kg−1 is found in the polluted case mean of the warm subset in comparison with the corresponding clean case mean. Much smaller absolute differences between the polluted and clean are obtained in the cases with a weak IAE, O(0.01 g kg−1).
In contrast to the changes in cloud water, rainwater is neither reduced nor enhanced in the cold subset (Fig. 10c). The differences between the clean and the polluted mean are negligible. However, a reduction from 0.09 to 0.04 g kg−1 is obtained in the warm subset on the windward side of R2 (Fig. 10d). Consistently, R2 shows the strongest difference in accumulated precipitation between clean and polluted cases, as mentioned above. Thus, in the polluted cases of the subset with a strong IAE not only more cloud water is formed owing to warmer temperatures and stronger vertical winds, but also less is converted to rain via autoconversion and accretion.
This finding implies that, under conditions of warm and moist air advection from the south, (e.g., from the Mediterranean region) accompanied by strong winds, the largest effect of aerosols on precipitation can be expected. Such events are most frequent during the late spring and early summer months and are often related to transport of Saharan dust toward the Alps (Coen et al. 2004). Dust transported this far would most probably be internally mixed. In air warm enough for immersion nucleation to be inefficient, as in the warm subset, dust would be deactivated as an IN (Hoose et al. 2008). Dust would preferably act as CCN and, hence, contribute to a stronger IAE on precipitation rather than enhance glaciation.
In the cold subset, more cloud droplets freeze via condensation/immersion freezing as a consequence of the lower temperatures. Thus, the increase in cloud water due to slower autoconversion and accretion is partly compensated by glaciation and subsequent aggregation and consumption of cloud droplets by riming processes. Figure 11a shows that cloud ice forms more efficiently in the subset with a weak IAE.
In the warm subset, the average mixing ratio of cloud ice reaches a maximum of only 6 mg kg−1 in a small ice cloud detached from the major updraft zone of ridge R2. Note that this cloud is not formed in the clean cases (Fig. 11b). Below this cloud almost no ice forms along the mountain slopes in contrast to the simulations with a weak IAE.
Hence, the formation of frozen precipitation is much stronger in the cold than in the warm subset (Figs. 11c,d). On average, the atmospheric mixing ratio of graupel and snow reaches values of about 0.06 g kg−1 in the proximity of R2 and the JFJ in the cold subset. Consistent with the horizontal and the vertical extent of cloud ice, much smaller areas of the domain are subject to the formation of frozen precipitation in the cases with a strong IAE. The snowflakes and graupel particles generated here only reach the surface near the crest of R2, causing very small contributions of frozen precipitation. Most of the precipitation in the warm subset falls to the surface as rain.
The differences in cloud droplet number concentrations between the warm and the cold subset are consistent with those for the cloud water mixing ratio. Owing to weaker horizontal, and thus vertical, winds fewer cloud droplets nucleate in the cold subset (Fig. 12a). Additionally, more cloud droplets freeze in the cold subset due to stronger condensation/immersion freezing. Therefore, the cloud droplet number concentrations are generally smaller in the cold subset that in the warm subset (Figs. 12a,b). A maximum of 140 cm−3 is found in the polluted case mean of the subset with a strong IAE at the top of the JFJ (R3), where cloud droplet nucleation is strongest because of the highest updraft velocities. Another maximum is located above R2. In the subset with a weak IAE, the maximum cloud droplet number concentration is obtained above R1 (50 cm−3). Note that these average values are rather small given the remote continental aerosol concentrations in our simulations. In addition, the averaging was performed over all grid points and not just the cloudy part. The cloud droplet number concentration maximum over all 112 analyzed simulations is 1150 cm−3 in a polluted simulation and 140 cm−3 in a clean simulation. About 30% of all cloudy grid points in all polluted (clean) simulations have cloud droplet number concentrations above 100 cm−3 (20 cm−3).
Mixed-phase clouds form between ground level and a height of 5500 m in the cold subset (Fig. 12c). The temperatures are within a range of −8° to −24°C in and close to the main orographic clouds. The ice crystal number concentrations of the mixed-phase clouds are within 0.01 and 1 L−1 in the mean polluted case of the subset with a weak IAE and up to three orders of magnitude smaller in the warm subset (Figs. 12c,d). In this subset, temperatures near the surface are too warm. Ice crystals are only found above 2000 m (Fig. 12d).
An orographic wave cloud consisting only of ice is found at a distance of 400 km from the domain edge and above 4500 m, where the temperatures are near −38°C in the mean (Figs. 9a and 12c). This cloud precipitates neither in the clean cases nor in the polluted cases. There, ice crystal numbers are larger than 100 L−1 on average, which leads to very small ice crystals.
To illustrate the physical mechanism behind the formation of these clouds, we show the ice crystal number concentration after 12 h as a function of the corresponding temperature at each grid point for the polluted cases over all 112 analyzed simulations (Fig. 13). Despite the large spread in the data caused by differences in thermodynamics, two major peaks can be determined. The first peak at approximately −5°C is associated with Hallett–Mossop ice multiplication (Hallett and Mossop 1974), which is active between −3° and −8°C. The corresponding 95th percentile is 10 L−1. Note that this peak already disappears in the 90th percentile and is not visible in the median, indicating that Hallett–Mossop ice multiplication is not the major process responsible for large ice crystal numbers in our simulations. In the range from 0° to −20°C the median of the ice crystal number concentrations Nice increases approximately log-linearly with decreasing temperature from 10−3 to 1 L−1. Below −25°C Nice fluctuates around 0.1 L−1. Condensation/immersion freezing causes the second, much broader peak in the 95th percentile at about −30°C and colder. Temperatures in this range are very rare in our simulations at grid points where ice is nucleated. Hence, the high-level orographic wave cloud mentioned above is a consequence of strong ice nucleation of supercooled droplets (strong updrafts, adiabatic cooling, and temperatures below −32°C).
As shown by Eqs. (A1)–(A3), in our model the condensation/immersion freezing rate for ice water depends on the mean cloud droplet mass, the liquid water content, and the immersion nucleation rate JIFR. The latter increases linearly with the immersion nucleation efficiency of the immersed ice nuclei and the Lagrangian cooling rate. The frozen fractions of black carbon and dust as predicted from the ice nucleation scheme are displayed in Fig. 14 for a typical cloud droplet radius of 10 μm. Measurements between −20° and −40°C made in the Aerosols Interaction and Dynamics in the Atmosphere (AIDA) chamber at the Karlsruhe Institute of Technology in Germany by Field et al. (2006) show frozen fractions of dust from 0.1 to 1. For black carbon, frozen fractions from laboratory measurements do not exceed 1%, even at temperatures below −40°C (DeMott 1990; DeMott et al. 1999; Möhler et al. 2005). Consequently, the ice nucleation scheme applied here is within the range of measurements for dust with a tendency to overpredict the frozen fraction. In the case of black carbon the scheme yields too high frozen fractions at temperatures colder than −35°C but agrees with observations above this temperature (Eidhammer et al. 2009). The large ice crystal number concentrations in the orographic wave cloud are a direct consequence of freezing in the respective temperature range. Nevertheless, it has to be noted that ice crystal number concentrations of several hundred per liter are not unusual under the conditions presented here (e.g., Cooper and Vali 1981). Korolev et al. (2003) found particle concentrations of 2–5 cm−3 in glaciated frontal clouds. Additionally, an empirical parameterization by Verheggen et al. (2007) based on measured data from the JFJ yields ice crystal number concentrations of more than 1 cm−3 in mixed-phase orographic clouds with an aerosol particle number of 100 cm−3 for ice mass fractions greater than 10%. Thus, our ice crystal number concentrations are rather on the low side of measurements at the JFJ.
Owing to the considerable number of parameters influencing the freezing mechanism and, in particular, the subsequent formation processes for graupel and snow, large variability results for the effect of increasing aerosol numbers on mixed-phase cloud formation and precipitation. There is a tendency toward an increase in precipitation in the polluted case when the ice-phase processes become dominant and, thus, if the cloud is glaciated. On the other hand, clouds in which warm-phase processes dominate show a distinct suppression of rainfall.
It is important to note at this point that the microphysical processes in the simulated mixed-phase clouds strongly depend on the environmental conditions. Thus, the sign and magnitude of the aerosol effects on cloud and precipitation formation are not only determined by the amount of aerosol present but also depend to a large extent on the ambient temperature, wind speed, and humidity.
4. Conclusions and discussion
In the present study, we conducted pairs of simulations with a clean and a polluted case. The aerosol number and mass concentrations used to initialize the model represent a climatology of winter and summer aerosol conditions in the proximity of the Jungfraujoch (JFJ), a high ridge in the Swiss Alps. The 270 simulation pairs are characterized by large variability in the thermodynamical initial state. We show that on average the spillover factor (SP) is increased at the first two major ridges by 2.4%–14.6% in the polluted cases as a consequence of the decelerated and reduced hydrometeor formation.
Our study provides cumulative frequency distributions for the relative change in total domain precipitation (RPD) with an average decrease in the polluted case of 36% for total precipitation. Rain is reduced by 52% (std dev = 25%), frozen precipitation increases on average in the polluted cases by 4%. We found a remarkably broader distribution of RPD for frozen precipitation (std dev = 60%).
The cases with the strongest IAE are shown to have little contribution from cold-phase microphysical processes. The ice phase significantly influences the strength of the IAE on orographic precipitation in the colder simulations. There, more efficient riming and aggregation, triggered by an enhancement of the condensation/immersion freezing mode, yields slightly more frozen precipitation in the mean of the respective polluted cases. In some simulations this is found to compensate for or even exceed the superimposed loss of rain. We therefore suggest that the magnitude and the sign of the IAE on orographic precipitation depends on the strength of the ice phase and particularly the freezing of cloud droplets in the presence of ice nuclei such as mineral dust or black carbon particles. One may argue that the IAE on precipitation is subject to seasonal variability due to the dependence on temperature.
Limitations of the present study are 1) the semi-idealized framework and the restriction to a two-dimensional setup, which does not allow for flow around the mountain ridges. However, a 2D configuration enables a large number of simulations. Additionally, flow around the JFJ is improbable because of its geographical structure. In essence, it is a long ridge surrounded by higher mountains. With the current setup, we cannot quantify aerosol effects on 3D flow over the Alps. 2) The quantification of the IAE on orographic precipitation is limited to the uncertainty range caused by the thermodynamical initial conditions. It is also important to study the effect of different aerosol mixing states. A major focus of future work in this field should be the quantification of uncertainties due to differences in the parameterization of the underlying physical processes, for example, heterogeneous ice nucleation, which would have gone beyond the scope of this work. Given the overprediction of ice nucleation for black carbon aerosols in the scheme applied here, an adaptation of the scheme at temperatures below −35°C could be an option for future studies. 3) Finally, the current study focuses on immersion/condensation freezing as the only ice-nucleating process considered. In the future, the sensitivity of other freezing modes should be investigated as well.
In summary, we showed that the environmental conditions have a strong impact on the effect of aerosols on orographic precipitation. It is the first study to quantify this effect in a statistical sense by providing distributions of possible changes in the precipitation budget in Alpine terrain. Previous studies based on a single control simulation and another one with altered conditions were not able to give a conclusive statement with respect to the sign and the magnitude of this second indirect aerosol effect on mixed-phase precipitation.
Above all, we thank the reviewers for their valuable comments and suggestions. Kassiem Jacobs and Sylvaine Ferrachat deserve many thanks for technical support. We thank the Swiss National Supercomputing Centre (CSCS) for providing the large amount of computing resources within the ALPS-Climate project and MeteoSwiss for providing the soundings from Payerne and the station observations from the Jungfraujoch. In addition, the first author wishes to thank Cathy Hohenegger for sharing her outstanding knowledge and experience with the COSMO model.
Parameterization of Condensation/Immersion Freezing
The rates of change for the ice crystal number density Nice (m−3) and mass density Lice (kg m−3) are given by
Here Lc and xc are the mass density and the mean mass of cloud droplets, respectively; νc is the shape parameter of the gamma size distribution of cloud droplets (νc = 1) and ρw is the density of water. Parameter a equals 1 K−1; b is the surface-weighted average ice nucleation efficiency of the immersed aerosol components acting as ice nuclei. The immersion freezing rate in the modified parameterization is only calculated for negative Lagrangian cooling rates dT/dt and temperatures below T0 = 273.15 K. Note that the original formula of Diehl and Wurzler (2004) for JIFR was adapted using exp[a(T0 − T)] − 1 instead of simply exp[a(T0 − T)] to obey the obvious no-freezing boundary condition for T → 273.15 K (Khain et al. 2000).
Additionally, the maximum predicted crystal number concentration due to immersion freezing is limited by the available number concentration of potentially nucleating particles in condensation/immersion freezing mode as discussed in Eidhammer et al. (2009). Therefore, b is defined according to Muhlbauer and Lohmann (2009) such that
Bj is material specific and equals 2.91 × 10−3 m−3 for BC (soot) and 32.3 m−3 for DU (montmorillonite). The total surface area Sj of species j is given by the sum of the surface areas of components j in each internally mixed mode i, only taking the activated fraction into account (Aitken mode larger than 35 nm, accumulation, and coarse mode). Formally, the surface areas Sj and S are
In Eq. (A7) Ni denotes the prognostic aerosol number density of the ith lognormal aerosol mode with the geometric standard deviation is the count median radius and mj,i the prognostic mass density of aerosol component j in aerosol mode i.
Consequently, the ice nucleation efficiency changes with the chemical composition of an aerosol particle. The importance of taking the chemical composition into account for describing the ice-nucleating ability of an aerosol has been stressed by numerous studies (e.g., Sassen et al. 2003; DeMott et al. 2004; Cozic et al. 2007).
Following Gibbons (1985), the Wilcoxon rank-sum test is nonparametric (independent of the distribution of the underlying samples, a Gaussian distribution is not required) and two-sided, and is equivalent to a Mann–Whitney U test. It is therefore also equal to a classical analysis of the variance of ranks. The test examines whether two samples of the same size have equal values. The Wilcoxon rank-sum test is applied for each of the statistical analyses since our data is not normally distributed.