Simulation of a supercell storm in clean and dirty atmosphere using weather research and forecast model with spectral bin microphysics

Authors


Abstract

[1] The development of supercell storms was simulated using a 2-km-resolution weather research and forecast (WRF) model with spectral (bin) microphysics (WRF-SBM) and a recent version of the Thompson bulk-parameterization scheme. The simulations were performed in clean, semipolluted, and dirty air under two values of relative humidity, conditionally referred to as low and high humidity. Both SBM and the Thompson scheme simulated the development of supercell storm with storm splitting. Both SBM and the Thompson scheme demonstrated that an increase in relative humidity by ∼10% invigorates convection and increases precipitation by factor of 2, i.e., to much larger extent than can be achieved by variations of the aerosol concentration. At the same time the storms simulated by the schemes are quite different. The maximum updrafts in the Thompson scheme are about 65 m/s, and the left-moving storm prevails. The SBM predicts 35 m/s maximum updrafts, and the right-moving storm prevails in the SBM simulations. While the bulk scheme predicts decrease in precipitation in clean air at both low and high humidity, the SBM indicates decrease precipitation in polluted air under low humidity and increase in precipitation under high humidity. The SBM scheme shows a substantial effect of aerosols on spatial distribution of precipitation, especially in the low-humidity case. The sensitivity of the Thompson scheme to aerosols turns out to be much less than that of SBM. The difference in the results (vertical velocities, microphysical cloud structure, and precipitation) obtained by different schemes is much larger than the changes caused by variation of the aerosol concentration within each scheme. However, the average amount of precipitation in the Thompson scheme in each simulation was about twice that of the corresponding SBM simulation. The possible reasons for such difference are discussed. A scheme for classifying aerosol effects on precipitation from clouds and cloud systems is also discussed.

1. Introduction

[2] Observations and numerical studies indicate that atmospheric aerosols affect cloud microphysical structure. An increase in the concentration of submicron aerosol particles (AP) serving as cloud condensation nuclei (CCN) increases the concentration and decreases the size of droplets [e.g., Twomey, 1974; Albrecht, 1989; Rosenfeld and Lensky, 1998; Ramanathan et al., 2001; Andreae et al., 2004]. Numerical models with an accurate spectral-bin microphysics approach make it possible to reproduce the observed drop–aerosol concentration dependencies [e.g., Segal and Khain, 2006; Kuba and Fujiyoshi, 2006; Magaritz et al., 2007, 2008].

[3] The general interest in the possible impacts of aerosols on precipitation can be attributed to the continuous production of anthropogenic AP that could lead to long-lasting trends in precipitation in different regions. At the same time, the effect of aerosols on precipitation still remains a challenging problem (see detailed overview by Levin and Cotton [2009]). Aerosols represent one factor among many factors affecting precipitation and in many cases the aerosol factor it is not the strongest one. For instance the increase in relative humidity (RH) by only 10% can increase precipitation from deep convective clouds 2–3 times [e.g., Fan et al., 2007b].

[4] In several studies [e.g., Khain et al., 2003, 2004, 2005, 2008b; Kaufman et al., 2005; Lynn et al., 2005a, 2005b, 2007; Lynn and Khain, 2007; Matsui et al., 2006; Seifert and Beheng, 2006; Tao et al., 2007; Lee et al., 2008a, 2008b] it was shown that aerosol effects on precipitation amount depend on the environment conditions and on cloud type (e.g., stratocumulus clouds, single isolated cumulus clouds, severe storms, squall lines, etc.). Lynn et al. [2005a, 2005b] and Lee et al. [2008a, 2008b] showed that aerosols tend to redistribute precipitation between clouds of different type, decreasing precipitation from small clouds and increasing precipitation in the zone of deep convection (e.g., squall line).

[5] The reason that the precipitation response to aerosols depends on the type of clouds is related to the fact that surface precipitation is the difference between the generation of condensate by drop condensation and ice deposition and its loss by sublimation and evaporation. The higher the aerosol concentration (i.e., the smaller droplet size), the more likely droplets will ascend to a high levels and freeze (first of all by collisions with ice particles). This leads to two related consequences: to an increase in the generation of condensate mass (larger convective heating) and to an increase in the condensate loss by evaporation and ice sublimation (note that this comment is related largely to the situations with high freezing level). The sign of the surface precipitation response to aerosols depends on a small difference between the increase in the generation and loss of condensate. In case of small clouds, an increase in the condensate loss by evaporation dominates over the increase in the generation, i.e., aerosols tend to suppress precipitation. In deep tropical clouds, the situation may be just opposite [Khain et al., 2004, 2005, 2008b]. This idea will be expanded upon in section 4.

[6] Since precipitation at the surface represents often a small difference between two large values: the generation and the loss of the hydrometeor mass, an accurate calculation of both the items of the mass budget is required to properly calculate the precipitation amount. To reveal aerosol effects on precipitation (which is supposed to be from several percent to a few tens percent of the mean precipitation) imposes much heavier demands on the precision of the calculations of the components of the heat and moisture budget in numerical models than needed just for the calculation of precipitation. These demands are related to both microphysical schemes and to dimensionality of the models.

[7] Note that most detailed high-resolution investigations of precipitation response to aerosols have been performed using 2-D models [e.g., Khain et al., 2005, 2008b; Fan et al., 2007a, 2007b]. Results of Tompkins [2000] and Li et al. [2008] indicate that for highly two-dimensionally organized circulations, such as squall lines, a 2D model can successfully reproduce observations. But, it is highly preferable to use a 3D cloud model for random or clustered convection, especially in low-wind environments. Phillips et al. [2007] compared 2-D and 3-D results for deep convective clouds over the ocean. They found that the updrafts in the 3-D simulations were faster than those in the 2-D for the given domain that they occupied. Two-dimensional model hardly can reproduce many features of a essentially 3-D phenomenon such as a supercell storm. For instance, Schlesinger [1984] found “a much weaker main updraft in two dimensions [with] pronounced downshear tilt of the two-dimensional storm core versus an erect three-dimensional storm core, and a dry secondary updraft downshear of the main cloudy updraft in two dimensions with no analog in three dimensions.” Phillips and Donner [2006] extended the previous results of Schlesinger [1984] to the scale of cumulus ensemble for several diverse cases of deep convection.

[8] Three-dimensional simulations of mesoscale cloud systems (including supercell storms) have been performed using 3-D models with bulk parameterization of cloud microphysical processes [e.g., Lee et al., 2008a, 2008b; van den Heever et al., 2006; van den Heever and Cotton, 2007]. Spectral (bin) Microphysics (SBM) has been used in only a few 3-D simulation studies [Lynn et al., 2005b; Lynn and Khain, 2007]. In the latter, 3-D mesoscale model simulations of a Florida squall line showed that SBM produced more realistic cloud structure and better agreement of precipitation amounts with observations than obtained with various bulk-parameterization schemes. These simulations were produced using the Mesoscale Modeling System Version 5 (MM5) [Dudhia, 1993] as the hosting model. The SBM was recently included in the Weather Research Forecast Model [Skamarock et al., 2005], and used to simulate the effect of aerosols on orographic clouds in the Pacific Northwest [Lynn et al., 2007]. The Thompson bulk parameterization scheme [Thompson et al., 2006] gave better results than other bulk parameterization schemes previously available in MM5 that were used to simulate the general characteristics of a Florida squall line under maritime conditions [Lynn and Khain, 2007]. This scheme has the flexibility to possibly simulate the effects of aerosols on precipitation, by changing the initial drop concentration.

[9] This paper investigates the potential effects of aerosols in combination with other environment factors on dynamics and microphysics (including precipitation) of supercell storms. Supercell storms are among the most spectacular and severe of all storms. They are usually accompanied by falling of big hail and intense lightning. Supercell thunderstorms are a special breed as they continuously regenerate their updrafts in a preferred direction and thus propagate at various angles to the averaged wind in the cloud-bearing layer [Marwitz, 1972; Davies-Jones, 1984; Rotunno and Klemp, 1985; Weisman and Klemp, 1984; Rasmussen and Straka, 1998]. Supercell storms are beta-mesoscale phenomena with characteristic spatial scales 10–100 km. The minimum diameter of updraft in the middle atmosphere exceeds 4–6 km. Closer to the ground, drafts tend to have a larger diameter and lower speeds than do drafts higher in the cloud. Updraft speeds exceeding 20 m per second are common in the upper parts of large storms. Airplanes flying through large storms at altitudes of about 10,000 m have measured updrafts exceeding 30 m per second. The strongest updrafts occur in organized storms that are many tens of kilometers in diameter, and lines or zones of such storms can extend for hundreds of kilometers. In such storms maximum updrafts can exceed 40 m/s leading to formation of big hail stones of several centimeter size (Encyclopedia Britannica).

[10] In the present study we perform two sets of simulations using SBM with a sounding conducive to supercell development; comparatively low and high values of air humidity were used, typical of the Great Plains and Gulf Coast, respectively. The same set of simulations was repeated with the Thompson scheme, and the results are compared against those of the SBM.

2. Model and Experimental Design

2.1. Spectral Bin Microphysical Scheme

[11] The spectral bin microphysics (SBM) scheme used is similar in many aspects to that described by Khain et al. [2004], Lynn et al. [2005a], and Lynn and Khain [2007]. The model microphysics is based on solving kinetic equations system for size distribution functions for water drops, ice crystals (plate, columnar and branch types), aggregates, graupel and hail/frozen drops, as well as atmospheric AP. Each size distribution is described using 33 doubling mass bins, allowing the simulation of graupel and hail with sizes up to ∼1 cm in diameter. The model is specially designed to take into account the effects of AP on the cloud microphysics, dynamics, and precipitation.

[12] The initial (at t = 0) CCN size distribution was calculated using the empirical dependence N = NoSk, using the procedure described by Khain et al. [2000]. In the formula N is the concentration of activated AP (nucleated droplets) at supersaturation S (in %) is with respect to water. No and k are constants chosen to be typical of maritime and continental aerosols (see below). At t > 0, the prognostic equation for the size distribution of nonactivated aerosol particles (AP) is solved in the following way: using the value of S calculated at each time step, the critical AP radius is calculated according to the Kohler theory. The APs with radii exceeding the critical value are activated and new droplets are nucleated. For AP with radius below 0.03 μm the droplet size was calculated using the Kohler theory. For largest APs, droplet size was calculated by multiplying the dry aerosol radius by factor ranging from 3 to 8. The magnitude of the factor decreases with the increase in size of dry AP [Segal et al., 2007].

[13] The primary nucleation of each type of ice crystals follows Takahashi et al. [1991]. Ice nuclei activation is described using an empirical expression suggested by Meyers et al. [1992] and applying a semi-Lagrangian approach [Khain et al., 2000] allowing the utilization of the proposed diagnostic formulas in a time-dependent framework. Secondary ice generation is described according to Hallett and Mossop [1974]. Amount of collisions between graupel and hail, on one hand, and water droplets with diameter exceeding 24 microns, on the other hand, was calculated. According to the measurements, 250 collisions produced one ice splinter which was assigned to plate type of crystals with density of 0.9 g cm−3 with the size corresponding to the first mass bin. The rate of drop freezing follows the observations of immersion nuclei by Vali [1975, 1994], and homogeneous freezing according to Pruppacher [1995]. The diffusion growth/evaporation of droplets and the deposition/sublimation of ice particles are calculated using analytical solutions for supersaturation with respect to water and ice [Khain et al., 2008b]. An efficient and accurate method of solving the stochastic kinetic equation for collisions [Bott, 1998] was extended to a system of stochastic kinetic equations calculating water-ice and ice-ice collisions. The model uses height-dependent drop-drop and drop-graupel collision kernels following Khain et al. [2001] and Pinsky et al. [2001]. Ice-ice collection rates are assumed to be temperature-dependent [Pruppacher and Klett, 1997].

[14] New improvements in the SBM include a new remapping scheme applied for diffusion growth/evaporation which significantly reduces artificial numerical droplet spectrum broadening [Khain et al., 2008b]. Some changes were introduced as regards to the transformation of graupel to hail by riming. Simple evaluations show that graupel particles significantly increase their mass by collection of supercooled droplets during one collision time step if liquid water content exceeds 3 g m−3. This means that after a few time steps the density of graupel particles should be close to that of pure ice (0.9 g cm−3). Correspondingly, it was assumed that collisions of graupel with radii exceeding 500 μm and water droplet leads to hail formation if liquid water content exceeded 3 g m−3.

2.2. Thompson Bulk Parameterization Scheme

[15] In this paper, we compare WRF-SBM to the latest version of the single-moment bulk microphysical scheme previously referred to as “Reisner2” or “Reisner/Thompson” [Thompson et al., 2004], but here called the “Thompson scheme” [Thompson et al., 2006]. The version was from WRF Model Version 3.0.1 (5 August 2008). The scheme has been quite upgraded and is now designed to mimic processes currently only described using SBM type schemes. For instance, the probability of certain volume of drops freezing at specified temperatures is precomputed and stored in a lookup table; larger raindrops freeze into graupel whereas the smaller cloud droplets freeze into cloud ice. The fraction of ice mass with particle diameters greater than 125 μm is immediately transferred to the snow category. The new scheme also utilizes a look-up table with 100 size bins of rain and snow. In the case of rain collecting snow (and its inverse), the scheme scans the snow size bins in the table and if the mean mass of the water drop exceeds the mass of the snow particle, it is assumed that the two particles join as one thus freezing the drop into graupel. If, on the other hand, the water drops mass is less than the snow particle mass, the snow simply accretes the water drop, thus increasing the snow mass and decreasing the rain mass.

[16] The cloud droplet concentration is assumed constant in space and time, but the gamma size distribution for cloud water is directly affected by the value chosen for the cloud concentration. A clean, maritime CCN distribution is simulated with a very high value of the gamma shape parameter, thus increasing the droplet effective radius when compared to a continental CCN distribution in which the gamma shape parameter is small and resulting effective radius is small. This assumption directly affects process rates for autoconversion, accretion, riming, and evaporation. Other WRF bulk microphysics schemes (e.g., the Lin, WSM5, or WSM6 schemes) have no direct possibility to describe aerosol effects because they use traditional one-moment bulk-parameterization schemes where the rate of microphysical processes depend on mass contents only, and do not contain information on drop concentration.

2.3. Design of Simulations

[17] We used WRFV2.0 with 2-km horizontal resolution to simulate an idealized supercell storm under environmental conditions typically occurring in the Great Plains during supercell storm development. The computational area was 252 km × 252 km, and the maximum time step was 10 s. Automatic choice of time step was used to keep the numerical stability. The number of vertical sigma levels was set equal to 41. The top height of the model computational area is 20 km.

[18] Note that the model resolution is an important parameter that may affect the model results. The simulation of single clouds, especially small ones requires a grid resolution of a few hundred meters and even higher (see results of sensitivity simulations from Khain et al. [2004]). Usually two-dimensional simulations are used at grid resolutions of a few hundred meters. In three-dimensional simulations of deep convective clouds, typical resolutions 1 km to 3 km are used, since these simulations are highly computer resource intensive (see Table 1). Such simulations do not resolve small clouds and typically underestimate their effects to precipitation. Still, these simulations produce realistic vertical updrafts in large clouds, which is the most important condition for calculation of supersaturation and, consequently, droplet concentration. Note that clouds with diameters of 10 km are typically found in squall lines, supercell storms, and clouds in tropical cyclone eyewalls. In the present study we use 2 km resolution, which is within the range of grid resolutions used to simulate such phenomena. We realize, however, that the utilization of higher resolution would be desirable when computer resources are appropriate for such endeavors.

Table 1. Horizontal Resolution Used in Cloud-Resolved Models Using Bulk Parameterization and SBM Description of Microphysical Processesa
AuthorsPhenomenon SimulatedMicrophysical SchemeModel Geometry, TitleModel Resolution, kmMain Results
  • a

    The main results of simulations are presented as well. Additional information about the models is presented in the study by Tao et al. [2007]. Model title abbreviations: GCE, the Goddard Cloud ensemble model; HUJI, the Hebrew University cloud model; SAM, the System for Atmospheric Modeling; TAO, the model developed at Tel Aviv University, Israel; MM5, the Mesoscale Modeling System, version 5; RAMS, the Regional Atmospheric Meteorological System; and WRF, the Weather Research Forecasting model.

  • b

    H. Noppel et al., Simulations of a hail storm and the impact of CCN using an advanced 2-moment cloud microphysical scheme, submitted to Atmospheric Research, 2009.

Takahashi [1976]Single mixed phase cloudSBMaxisym0.2Hail simulation by hydrometeor recycling
Kogan [1991]Single warm rain cumulus cloudSBM3-D0.25Accurate simulation of a small warm cloud in 3-D framework
Khvorostyanov et al. [1989]Single mixed phase cumulus cloudSBM2-D0.5Investigation of cloud seeding effects
Hall [1980]Single mixed phase cumulus cloudSBM2-D0.2Explicit treatment of warm and ice microphysics
Khain et al. [1993]Breeze-induced cloud ensembleSBM2-D, (earlier HUCM)3.0Investigation of effect of sea-land temperature difference and wind speed on precipitation
Khain and Sednev [1996], Khain et al. [1999]Breeze-induced cloud ensembleSBM2-D HUCM0.3Formulation of HUCM, simulation of breeze-related precipitation, effects of aerosols on microphysics of clouds in the Eastern Mediterranean
Reisin et al. [1996]Single mixed phase cumulus cloudSBMAxisym, TAO0.15Investigation of role of ice, Simulation of cloud seeding effects on precipitation
Yin et al. [2000]Single mixed phase cumulus cloudSBM2-D, TAO0.3Simulation of cloud seeding effects on precipitation
Farley and Orville [1986], Farley [1987]Hail stormsOne-moment bulk scheme with improved description of hail formation2-D0.2Simulation of hail and test of hail suppression hypotheses
Seifert and Beheng [2006]Mixed phase deep cloudsTwo-moment bulk scheme <1.0The role of air humidity in determining the structure of storms and in precipitation response is stressed.
Tao et al. [2003]Squall linebulk2D, GCE1.0Effects of microphysics and radiation
Wang [2005]Deep tropical convectionTwo-moment bulk scheme3D2.0Convective invigoration and precipitation increase with the increase in AP loading
Lynn et al. [2005a, 2005b]Rain event over FloridaSBM, bulk schemes3-D, MM53.0SBM dramatically improves the prediction of rain distribution and precipitation rate
Tao et al. [2007]Squall linesSBM, bulk schemes2-D, GCE1.0SBM realistically reproduces the structure of squall lines. Air humidity plays an important role on precipitation response to aerosols
Lynn et al. [2007]Orographic cloudsSBM2-D, WRF1.0Aerosols affect the amount and spatial distribution of precipitation
Teller and Levin [2006]Single mixed phase cloudsSBM2-D, TAU0.3Simulation of clouds with the low freezing level
Khain et al. [2005]Maritime and continental cloudsSBM2-D, HUCM0.25–0.35Dependence of precipitation response to aerosols on environmental conditions
Khain et al. [2008b]Maritime, continental clouds, pyrocloudsSBM2-D, HUCM0.35Reproduction of in situ measured droplet size distributions
Li et al. [2009a, 2009b]Squall lineSBM, bulk schemes2-D, GCE0.5–1.0SBM describes the observed structure of a squall line much better than the bulk scheme used.
Fan et al. [2007a, 2007b]Single mixed-phase cloudsSBMGCE0.5The role of humidity and other parameters on precipitation response to aerosols
Fan et al. [2009]Arctic cloudsSBM3-D, SAM0.1Reproduction of size distributions of ice hydrometeors measured in situ
Iguchi et al. [2008]Frontal cloud systemsSBM, bulk schemes3-D Weather forecast model7.0SBM predicts cloud structure better than bulk schemes and allows calculation of effective radius measured from satellites
Lee et al. [2008a, 2008b]Cloud ensembleTwo-moment bulk scheme3-D, WRF3.0Aerosols invigorate convection and precipitation by intensification of secondary clouds
van den Heever et al. [2006]Rain events, hail stormsAdvanced (SBM mimic) bulk scheme3-D RAMS1.5Aerosols increase CWC and W in clouds
Noppel et al. [2009]bHail stormTwo moment bulk scheme3-D COSMO1.0Simulation of hail at the surface. Implementation of a special class for big hailstones
Ntelekos et al. [2009]Cloud system over the USATwo-moment bulk scheme3-D, WRF3.0Zones of rain enhancement in the Northeastern U.S. caused by urban aerosols are found
Li et al. [2009a, 2009b]Maritime convectionTwo-moment bulk scheme3-D, WRF2.0Nonmonotonic dependence of precipitation from deep clouds on aerosols
Zhang et al. [2007]Tropical cyclonesAdvanced (SBM mimic) bulk scheme3-D RAMS2.0Sensitivity of TC intensity on aerosols
Khain et al. [2009]Tropical cyclonesSBM3-D, WRF3.0Sensitivity of TC intensity on aerosols

[19] The waves (and other fluctuations) reaching the upper troposphere have been damped by Rayleigh damping to prevent their reflection from the upper boundary and their effect on the solution. Periodic boundary conditions were applied. The storm has been triggered by a temperature pulse of 3°C, decreased exponentially both horizontally and vertical with distance from the triggering location. The model was run with a standard set up, including subgrid-scale (TKE) 1.5 turbulence closure. Such scheme is widely used in advanced simulations of cloud related phenomena (see Tao et al. [2007] for details).

[20] The SBM simulations were carried out for clean (maritime) (No = 100 cm−3 and k = 0.308), semipolluted (No = 500 and k = 0.462), and polluted (continental) conditions (No = 1500 cm−3 and k = 0.462). Two sets of simulations were conducted to test the effect of humidity on supercell development using a WRF idealized sounding as a basis, which is typical of the Great Plains during supercell storm development. Soundings used are presented in Figure 1 and Table 2. In the first set, the humidity was generally between 10–15% lower than in the second. The humidity in the first set is closer to that of the Great Plains, while the second is more typical of summertime weather in Houston. We refer to these sounding as “low-humidity” and “high-humidity” soundings.

Figure 1.

Sounding data for simulation with (left) low humidity and (right) high humidity.

Table 2. Soundings Used in Low- and High-Humidity Simulationsa
HeightTemperatureHigh HumidityLow Humidityu-windv-wind
  • a

    The u-wind and v-wind are the x and y components, respectively, of wind speed.

250.0298.0195.0085.00−7.88−3.58
750.0293.9195.0085.00−6.94−0.89
1250.0290.2195.0095.00−5.171.33
1750.0286.7295.0093.00−2.762.84
2250.0283.1395.0080.000.013.47
2750.0279.6390.0080.002.873.49
3250.0276.2385.0075.005.733.49
3750.0272.7485.0075.008.583.49
4250.0269.3585.0073.0011.443.49
4750.0265.9685.0063.0014.303.49
5250.0262.5680.0060.0017.153.49
5750.0259.2780.0060.0020.013.49
6250.0255.8880.0060.0022.873.49
6750.0252.4780.0055.0025.733.49
7250.0249.0980.0059.6527.153.49
7750.0245.7870.0056.5027.153.49
8250.0242.3970.0052.0427.153.49
8750.0239.0070.0050.2527.153.49
9250.0235.1970.0051.5527.153.49
9750.0232.0070.0047.4027.153.49
10250.0228.7940.0036.6527.153.49
10750.0225.3140.0040.1727.153.49
11250.0221.9140.0034.5627.153.49
11750.0219.1340.0026.7827.153.49
12250.0218.7140.0034.7827.153.49
12750.0218.4340.0033.2727.153.49
13250.0218.1240.0031.9627.153.49
13750.0218.1440.0029.4727.153.49
14250.0218.1440.0027.2627.153.49
14750.0218.1240.0025.2527.153.49
15250.0218.0040.0023.7127.153.49
15750.0217.7740.0022.5527.153.49
16250.0217.4340.0021.7327.153.49
16750.0216.9940.0021.2027.153.49
17250.0216.4640.0020.9527.153.49
17750.0215.8340.0020.9627.153.49
18250.0215.1040.0021.2427.153.49
18750.0214.2840.0021.7927.153.49
19250.0213.3640.0022.6427.153.49
19750.0212.3640.0023.8227.153.49
20250.0211.2640.0025.3927.153.49
20750.0210.0840.0027.4327.153.49
21250.0208.8240.0030.0227.153.49
21750.0207.4740.0033.3227.153.49
22250.0206.0340.0037.5027.153.49
22750.0204.5240.0042.8227.153.49

[21] Simulations with the bulk parameterization have been performed at droplet concentrations Nd = 50 cm−3, Nd = 250 cm−3, and Nd = 500 cm−3, representing clean, semipolluted, and polluted conditions. Since about 30 to 50% of aerosol particles are transformed to droplets by nucleation [Ramanathan et al., 2001], the droplet concentrations used in the bulk parameterization approximately correspond to aerosol concentrations used in the SBM simulations. The thermodynamical conditions were similar to those used in the SBM simulations.

[22] Initially, each set of SBM simulations were run for almost 3 h, before one or more terminated owing to numerical instability, with 5-min output. After seeing the results from these simulations, the damping coefficient was reset from 0.01 to 0.015, and the set of simulations under high humidity were extended to 4 h (but at 15 min output intervals to conserve hard disk space). The Thompson scheme simulations were each run with a damping coefficient of 0.01.

3. Results of Simulations

3.1. Simulations Using the Spectral (Bin) Microphysics

[23] Figure 2 shows maximum vertical velocities in the SBM simulations with low and high humidity. One can see that the maximum vertical velocity ranges from 25 to 40 m/s. These values seem to be reasonable values for the Convective Available Potential Energy (5483 J/kg)) (maximum vertical speed is usually evaluated as Wmax = equation image). Development of the first cells takes place during the first 40 min. At this stage the maxima of vertical velocities is reached in cases of high humidity. In simulations with the same humidity, the maximum updrafts take place under high aerosol concentrations which indicates higher latent heat release in polluted conditions. At t > 80 min the difference in the maximum vertical velocities in different simulations decreases. The latter can be attributed to the process of recirculation, when large drops penetrate the cloud in vicinity of cloud base even in the polluted case.

Figure 2.

Maximum vertical velocities (left) in the SBM simulations and (right) in the simulations with the bulk-parameterization scheme. “Mar” refers to a maritime concentration of aerosols (i.e., clean air), “Int” refers to an intermediate concentration of aerosols (i.e., semipolluted air), and “Con” refers to a continental concentration of aerosols (i.e., polluted air).

[24] Figure 3 shows the spatial distribution of accumulated rain in the SBM simulations with low humidity at 165 min. One can see a dramatic impact of aerosols on the spatial distribution of surface precipitation. In most of the simulations, the storm formed two branches that propagated relative to each other with an angle between the main storm branches of about 30–40 degrees, and the right branch of the supercells produced more clouds and more precipitation in clean and semipolluted cases. The simulation in clean air produced more precipitation initially than in the semipolluted and polluted air, but the simulation in the semipolluted air produced more precipitation than in the clean air on the downwind side of both the left and right branches. The spatial extent of precipitation in the clean case in the west-east direction was considerably less than in the other simulations. The difference in structure of surface precipitation can be attributed to different vertical distribution of condensate. Figure 4a shows vertical profiles of horizontally averaged liquid water content (LWC) (the sum of cloud water and rainwater contents) at t = 25 min (Figure 4a, left) and 60 min for SBM (Figure 4a, top) and the Thompson scheme (Figure 4a, bottom). At t = 25 min LWC is close to cloud water content (small cloud droplets), while at t = 60 min the contribution of rainwater content (raindrops) to LWC is dominating, which determines large values of LWC near the surface. As one can see in Figures 4a and 4b in polluted case, water drops ascend to higher levels and produce more ice aloft where the westerly wind is especially strong (see Table 2). Accordingly, a significant mass of condensate is transported downwind at upper levels, which determines the surface precipitation elongation in the eastern direction. Falling from higher levels, condensate evaporates significantly, determining the decrease in total precipitation in the polluted case. Even though the overall precipitation was less, condensate that reaches higher levels was able to advect further downwind and then fall as precipitation. In all, the total amounts of precipitation (averaged over the entire area) were 2.4 mm, 2.1 mm and 2.0 mm in the clean, in semipolluted, and polluted air.

Figure 3.

SBM simulated precipitation amounts at 165 min under relatively low humidity.

Figure 4a.

Vertical profiles of horizontally averaged LWC at (left) t = 25 min and (right) t = 60 min for (top) SBM and (bottom) the Thompson scheme. The levels of location of the LWC maxima in simulations with low humidity and high (H) and intermediate (I) and low (L) aerosol concentrations. Abbreviations are as in Figure 2.

Figure 4b.

Vertical profiles of horizontally averaged total ice content at (left) t = 25 min and (right) t = 60 min for (top) SBM and (bottom) the Thompson scheme. The levels of location of the total ice content maxima in simulations with low humidity and high (H) and intermediate (I) and low (L) aerosol concentrations. Abbreviations are as in Figure 2.

[25] An increase in humidity leads to a dramatic change in precipitation amount and distribution. These changes are much more substantial than those following from changes in the aerosol concentration (Figure 5, left). Under high humidity, the total precipitation at 4 h in the clean, semipolluted, and polluted air mass simulations was 4.5 mm, 5.0 mm and 4.7 mm, respectively, i.e., more than twice as large as in the case of lower humidity. Note that under high humidity, maximum precipitation is no longer in clean air. Accumulated rain in semipolluted and polluted air turns out to be higher than in clean air. Thus, we see an opposite response of precipitation to aerosols in case of low and high humidity. This effect was reported earlier for singe convective clouds [Khain et al., 2005, 2008b] and squall lines [Tao et al., 2007].

Figure 5.

(left) SBM and (right) Thompson simulated precipitation amounts under conditions of relatively low and high humidity. Abbreviations are as in Figure 2.

[26] There were also significant spatial differences in precipitation between the high-humidity SBM simulations at 240 min (Figure 6). Both the semipolluted and polluted cases produced from both branches a larger area of lighter precipitation (5 to 15 mm), extending further down wind than the clean air simulation. It is interesting that only the right branch of precipitation developed in the polluted air in the high-humidity case. There was no such feature in case of low humidity. One of the reasons for this effect can be the following: when there was high humidity the cloud base was located lower and significant warm precipitation formed at lower levels, where the westerly wind was not so strong.

Figure 6.

SBM simulated precipitation amounts at 240 min under relatively high humidity.

[27] Figure 7 (top) shows the maximum amounts of cloud and rain mass produced in the SBM simulations in low- and high-humidity air. In both low and high humidity, the simulations in the dirty air initially produced more cloud water mass than in the semipolluted or clean air, with as much as 60% more cloud water mass produced. Initial maximum rain amounts in the clean air, however, were nearly double those in the polluted air, and more than in the semipolluted air.

Figure 7.

Simulated maximum (left) cloud water and (right) rainwater contents under relatively low and high humidity from (top) SBM and (bottom) Thompson. Abbreviations are as in Figure 2.

[28] However, by 160 min the simulated maximum rain amounts in the semipolluted and polluted air approached more closely (low humidity) or even exceeded (high humidity) those in the clean air. We propose two reasons of such effect: (1) the recirculation of downdraft air, containing cloud droplets, into the updrafts of new cloud cells and (2) the decreasing aerosol concentration over time within the storm caused by nucleation scavenging. Both mechanisms lead to acceleration of raindrop formation in initially polluted air and corresponding comparative decrease in CWC and increase in RWC in the polluted case.

[29] Figure 8 shows the precipitation rate (calculated for 5 min time increments) in the SBM simulations in low and high humidity. The results show that the simulations in clean air initially produced a higher precipitation rate than those in semipolluted or polluted air for both low and high humidity due to warm rain. After about 110 min under low humidity, though, the semipolluted SBM simulation produced a higher precipitation rate than the clean air simulation, while the polluted simulation produced a similar precipitation rate to the simulation in clean air. When the humidity was high, the differences in initial precipitation rates were even more extended. After 100 min, though, the precipitation rate trended downward in the clean air, but upward in the semipolluted and polluted air to meet or exceed the clean-air curve. The results indicate that aerosols tend to intensify secondary convective cells especially in the case of high humidity. Figure 8 shows that in case of high humidity a quasi-stationary state has not been achieved toward 160 min. The continuation of this set of simulations to 4 h indicated a successive increase in precipitation in the semipolluted and polluted air.

Figure 8.

Simulated maximum rain rates under relatively (left) low and (right) high humidity from SBM (top) and Thompson (bottom). Abbreviations are as in Figure 2.

[30] Figure 9 shows the time dependencies of integral (spatially averaged and vertically) contents of cloud water and rainwater in the SBM simulations. One can see that storms developing in the semipolluted and polluted air have more average mass of liquid cloud water, especially under conditions of high humidity. The simulations in clean and low-humidity air produced more rain mass than those in semipolluted or polluted air, but in the more humid air the simulations in semipolluted and polluted air eventually produced more rain mass than that in the clean air. Comparing the average quantities with the mean number concentrations (not shown) indicates that the simulations in polluted air are producing relatively smaller sized droplets than those in clean air.

Figure 9.

Simulated domain averaged (left) cloud water and (right) rainwater contents under relatively low and high humidity from (top) SBM and (bottom) Thompson. Abbreviations are as in Figure 2.

[31] The increase in the AP concentration led to decrease in amount of ice crystals (Figure 11, left), likely because of faster riming of the crystals to graupel in zones of high supercooled water content. At the same time the mass of snow, graupel increases with the increase in the AP concentration (Figure 10). Supposedly the most important result is the increase in the hail mass content with the increase in the AP concentration (Figure 11, right) at times exceeding about 1 h. Hail particles grow by riming falling within the cloud with high supercooled water content. These particles (graupel and hail) were responsible for increasing the rain rate as compared to that calculated in the clean air by 165 min. In the 4-h simulation, these same particles eventually grew large enough to lead to higher amounts of accumulated rain in the semipolluted air mass simulation than in the clean air simulation.

Figure 10.

Simulated domain averaged (left) snow and (right) graupel contents under relatively low and high humidity from (top) SBM and (bottom) Thompson. Abbreviations are as in Figure 2.

Figure 11.

Simulated domain averaged (left) ice crystal and (right) hail content under relatively low and high humidity from SBM. Abbreviations are as in Figure 2.

3.2. Comparison With the Results Obtained Using the Thompson Scheme

[32] Figure 2 (right) shows the maximum vertical velocities in the simulations using the Thompson scheme. A comparison with the vertical velocity in Figure 2 (left) indicates significant differences between the SBM and bulk-parameterization simulations. The Thompson scheme had maximum vertical velocities ranging from 45 m/s to 65 m/s. These values seem to be unrealistically high for such values of CAPE. The overestimation of vertical velocities and production of too strong convective precipitation (with the underestimation of the stratiform clouds area) is a typical feature of many bulk-parameterization schemes [see Lynn et al., 2005b, 2007; Tao et al., 2007; Li et al., 2009a, 2009b]. This effect can be attributed to some characteristic properties of the schemes. The accurate calculation of supersaturation requires utilization of very fine algorithms, which take into account the decrease in supersaturation during one time step due to drop and ice growth [e.g., Khain and Sednev, 1996; Khain et al., 2000, 2004, 2008b]. Besides, the supersaturation does not fall to zero by the condensational growth at each time step (as it is typically assumed in bulk-parameterization schemes). The result is that the bulk schemes tend to overestimate the latent heat release by condensation and deposition. The overestimation of evaporative cooling by bulk schemes was investigated by Li et al. [2009a, 2009b]. As a result, the vertical velocities both in updrafts and downdrafts turned out to be overestimated. The second possible reason is the treatment to drop freezing. It is known that large drops have a higher probability to freeze because of higher probability to contain immersion ice nuclei within them. It means that freezing should first eliminate large drops keeping small droplets unfrozen. However, bulk schemes assuming the Marshall-Palmer distribution of raindrops restore the tail of large drops at each time step which leads to new (artificial) freezing of large drops and extra latent heat of fusion. The third possible reason is that bulk schemes tend to overestimate the residential time of large hydrometeors within clouds that may lead to artificially strong riming accompanied by latent heat release of fusion. Actually many bulk schemes tend to release the CAPE during time period shorter than in the SBM scheme.

[33] Deng and Stauffer [2006] found that using 4-km grid resolution can delay the onset of precipitation, leading to unrealistically high vertical velocities. Although we did not think that this was of concern using 2-km grid resolution, we produced a supplemental simulation with 1-km resolution with the Thompson scheme. The results obtained were quite similar to those obtained with the resolution of 2 km. The maximum value of vertical velocity increased from 65 m/s to 67 m/s indicating a quite low sensitivity of the results to the resolution within this range of the model resolutions.

[34] Figure 4a (bottom left) shows that at t = 25 min the Thompson scheme produces significant amount of rainwater at the 3-km level independently of cloud droplet concentration and assumed humidity. This result is in contrary with the SBM simulations which predict a substantial elevation of the maximum water content (by several kilometers!) with the increase in the CCN concentration. Since the vertical updrafts are overestimated by the bulk-parameterization scheme, the production of warm rain below 3.5 km probably indicates an overestimation of the autoconversion rate and rapid formation of warm rain in the bulk-parameterization scheme. It is known that in the bulk-parameterization schemes precipitation dramatically depends on threshold parameters used for the description of autoconversion and other physical processes. The very rapid appearance of precipitation at the surface is further supported by Figure 12.

Figure 12.

Precipitation amounts at 165 min under relatively low humidity simulated by the Thompson scheme.

[35] Probably because of high updraft speed, the Thompson scheme generates a significant amount of ice in the upper troposphere at and above 12 km at t = 60 min (Figure 4b, bottom right). The maximum ice content in the bulk-parameterization scheme is located at height of 5 km independent of aerosol concentration or on air humidity. This result dramatically differs from that obtained using the SBM, which indicates a significant increase in the height of the maximum ice content with the increase in the CCN concentration and humidity (instability). Besides, the amount of total ice in the Thompson scheme was more than twice less that that in the SBM simulations. Figures 4a and 4b indicate a very weak impact of aerosols on vertical profiles of horizontally averaged LWC and total ice content in the Thompson scheme. Figure 5 (right) depicts the accumulated rainfall obtained in the simulations with the Thompson scheme under low and high humidity for different values of cloud droplet concentration, representing clean, semipolluted, and polluted air masses. In both cases the increase in the air humidity by ∼10% resulted in a doubling of the accumulated precipitation. The increase in the CCN concentration within a wide range from clean maritime to highly polluted air led in the bulk-parameterization scheme to a small (by ∼10 percent) decrease in accumulated precipitation both in the cases of low and high humidity. This result is in contrary to the SBM results showing larger accumulated rain in the polluted cases for the high-humidity atmosphere. Last, the accumulated precipitation in the Thompson scheme is twice as high as in the SBM. (G. Thompson pointed out that there was an error in calculation of in description of sink term for graupel and source term for rain below freezing level; this error was corrected in March 2009 (G. Thompson, personal communication, March 2009).

[36] Figure 7 shows the time dependence of maximum cloud and rain mass in relatively low and high humidity air, for each of the different cloud drop concentrations. The Thompson scheme produced in general about half the amount of maximum cloud mass as the SBM. This result appears contradictory because one could expect higher cloud water content under huge vertical updrafts generated by the bulk scheme. Yet, a comparison of rainwater content indicates that the bulk scheme produced higher rainwater content than the SBM. Thus, the lower amount of CWC and ice content in the bulk scheme on one hand, and twice as high accumulated rain and RWC as well as high precipitation rate (see Figure 8) compared with the SBM likely indicate an overestimation of raindrop formation rate by autoconversion (droplet-droplet collisions) in the bulk scheme.

[37] Figure 9 presents the domain averaged cloud and rainwater contents under relatively low and high humidity from SBM (Figure 9, top) and Thompson (Figure 9, bottom). One can see that the domain averaged rainwater contents in the Thompson scheme is on the same order of magnitude or lower than those obtained by the SBM. Taking into account that the bulk scheme produces higher maxima of RWC, we conclude that the bulk scheme produces stronger showers but over a smaller area as compared to the SBM. Thus conclusion is further supported by Figures 12 and 13 depicting the fields of accumulation surface precipitation at 165 min and 240 min, respectively.

Figure 13.

Precipitation amounts simulated using the Thompson scheme at 240 min under relatively high humidity.

[38] A comparison of the fields of accumulated rain simulated by the SBM (Figures 3 and 6) with those in Figures 12 and 13 shows that (1) each of the simulations produced two branches, like the SBM, but (2) in the SBM simulations the right-hand branch is dominating rather than the left-hand branch in the Thompson simulations. We attribute this difference to the differences in the vertical microphysical structure of clouds. In the SBM most precipitation forms at higher levels and ice (cold) precipitation contributes substantially in the polluted case. It seems that in the bulk scheme, warm rain formed at low levels dominates in all cases. The lengths of the right-hand branches are longer than those of the left-hand branches in all simulations. The detailed analysis of the mechanism leading to different distribution of precipitation between the branches in the SBM and the bulk-parameterization scheme is beyond the scopes of the work.

[39] Note here that in contrast to the SBM results, the spatial distribution of precipitation turns out to be insensitive to the droplet concentration in the bulk-parameterization scheme because of the reasons mentioned above.

[40] Figure 10 (bottom) shows the Thompson scheme's sensitivity of average snow and graupel mass to initial drop concentration. The Thompson scheme simulated similar snow sensitivity to initial drop concentration as the SBM. However, the Thompson scheme shows only a small sensitivity of graupel amounts to initial drop concentration. Moreover, the bulk scheme predicts the maximum graupel mass content in the clean air which is in contrary to the SBM indicating the increase in graupel and hail mass in polluted cases.

[41] Comparison of Figures 14 and 15 shows that the SBM produces more ice above the 5-km level. Note that the SBM generally produced substantially more ice in the area surrounding the convective cores and in the stratiform region. The tendency of the bulk-parameterization schemes to underestimate the area of stratiform ice precipitation was noted in many studies [e.g., Lynn et al., 2005b; Li et al., 2009a, 2009b]. Conversely, bulk schemes appear to overestimate the intensity of convective rain.

Figure 14.

SBM simulated maximum total ice content above 5 km height (time = 165 min).

Figure 15.

Same as Figure 14 but for the Thompson scheme.

[42] It is interesting that in polluted air the zone of cloud ice shifts in the positive y direction (“northward”) (Figure 14). It can be attributed to the fact that in the polluted case cloud ice located at higher levels is transported by the southern wind component northward. No such effect is seen in the bulk-parameterization scheme (Figure 15). The location of cloud ice in the vertical direction is not sensitive to aerosols in the bulk parameterization scheme.

4. Summary and Discussion: Classification of Aerosol Effects

[43] The development of a supercell storm in the atmosphere with high (more than 4 km) freezing level was simulated using a 2-km resolution WRF model with both spectral (bin) microphysics and the updated Thompson bulk-parameterization scheme. To investigate the effects of aerosols on storm structure and precipitation, the simulations were performed under clean (maritime aerosol concentrations), semipolluted (intermediate continental aerosol concentration), and dirty (high aerosol, continental concentration) conditions. To show that aerosol effects on precipitation depend on environmental conditions and to compare aerosol effects with impact of other factors the simulations were carried out under moderate (more typical of the Great Plains) and higher relative humidity more typical of the southern Gulf coast. The difference between the relative humidity in these simulations was about 10%.

[44] Similar to some other simulations of deep convection [e.g., Khain et al., 2008b; Fan et al., 2007b; van den Heever et al., 2006], both the SBM and the Thompson scheme showed that the effect of aerosols on deep convection is weaker than that of relative humidity: the increase in relative humidity by 10% leads to precipitation enhancement of 2–3 times, while the change in the aerosol concentration from 100 cm−3 to 1500 cm−3 changes precipitation in simulated storm by ∼10% in SBM and even less in the Thompson scheme.

[45] Both schemes indicate an increase in cloud water content in dirty air in agreement with observations [e.g., Andreae et al., 2004], and a great number of simulations elsewhere [e.g., Khain et al., 2003, 2004, 2005, 2008b; Lynn et al., 2005a; van den Heever et al., 2006; Wang, 2005; Lee et al., 2008a; Tao et al., 2007].

[46] At the same time, there were important differences between the SBM simulations of the supercell storms and the Thompson schemes simulations of such storms.

[47] In spite of the fact that the storm splitting into two branches was successfully simulated by both schemes, the SBM predicts stronger strength and precipitation in the right hand branch, the bulk-parameterization scheme predicts more intense the left-hand branch with more precipitation within. It is known [Klemp and Wilhelmson, 1978; Cotton and Anthes, 1989] that the relative strengths of right- and left-moving storms depend on the change of the wind shear vector with height. If the wind shear vector veers with height (i.e., turns clockwise in the northern hemisphere), the development of a cyclonic, right-moving storm is favored. In contrast, if the wind shear vector backs with height (i.e., turns counterclockwise), then an anticyclonic, left-moving storm prevails [Cotton and Anthes, 1989]. Analysis of the vertical wind profile (Table 2) indicates that the clockwise rotated wind shear vector was used, so that the right-moving storm branch should prevail in our case. Thus, the SBM predicts the storm dynamics correctly.

[48] The Thompson scheme simulated much stronger convection with maximum vertical updrafts up to 65 m s−1 versus 35 m s−1 in SBM. Analysis of the CAPE values shows that the SBM simulated vertical velocity seems more realistic. The investigation of the reasons of such overestimation of maximum vertical velocities is beyond the framework of the study. It is possible that too high vertical velocities form because to much freezing or freezing in a two narrow temperature range. We suppose that the differences in the vertical velocities is one of the reasons that the Thompson scheme simulated about twice as much precipitation as the SBM with maximum rainwater content of 15–20 g m−3 versus 5–10 g m−3 in the SBM simulation (the opposite is also possible: the overestimation in the rain production means the higher atmosphere heating in the Thompson scheme, which can result in the overestimation of updrafts). The differences in rainwater contents and precipitation are closely related. In contrast to the SBM, the Thompson scheme produced the (horizontally averaged) maximum LWC located below 3.5 km independently on drop concentration and humidity, which indicates that the autoconversion rate is, probably, too intense in this scheme. In spite of much higher vertical updrafts, the Thompson scheme simulates lower snow and graupel contents as compared to SBM. It is of interest that the Thompson scheme predicts a decrease in graupel mass content, while the SBM indicated increase in the graupel and hail mass contents in the polluted air. The increase of the graupel and hail mass in the polluted air is the expected result of the increase in supercooled water in the polluted clouds. Since graupel and hail determine precipitation in an hour after storm beginning, the increase in the production of graupel and hail determine the increase in precipitation (in the case of high humidity) in the polluted case. The production of comparatively low amount of graupel with mass lower or equal to the mass of snow (see Figure 10) is quite unexpected result in the bulk scheme: under so high vertical velocities one could expect the formation of significant masses of graupel and hail.

[49] Another important difference between the SBM and the Thompson scheme is in the impact of aerosols on spatial distribution of precipitation. While the spatial distribution of precipitation in the Thompson scheme was insensitive to the droplet concentration, the precipitation distribution simulated by the SBM was crucially sensitive to the AP concentration. This sensitivity concerns both area covered by precipitation and the intensity of precipitation branches of the storm. The sensitivity is especially pronounced in the case of low humidity. We attribute this sensitivity in the SBM simulations to the combined effect of aerosols and wind shear. The level of the condensate mass maximum elevates with the increase in the CCN concentration and the condensate was advected eastward by the western background wind in the upper troposphere. When there was low aerosol concentration, the precipitation formed at lower levels and fell within the lower troposphere where the western component of the background wind was comparatively low. Besides, an increase in the cloud ice production at high levels in the case of polluted air leads to shift the cloud ice northward by the southern component of the background flow.

[50] Contrary to these SBM results, the spatial precipitation distribution simulated by the Thompson scheme turns out to be insensitive to drop (i.e., aerosol) concentration. The negligible sensitivity of can be attributed to the fact that the scheme simulates nearly similar vertical structure of hydrometer masses in all simulations (see Figures 4a and 4b).

[51] As mentioned, both schemes indicate a decrease in precipitation with increase in the CCN concentration in relatively dry atmosphere. At the same time, the SBM simulation indicated that at the intermediate aerosol concentration the accumulated rain in the high-humidity case is higher than that in clean air. At the later stage of storm evolution accumulated rain in the polluted case also increases faster than in the clean case, so that one can expect that both in the intermediate and polluted cases precipitation in the high-humidity case are higher than in the clean air case.

[52] The effect of the air humidity on the magnitude and even sign of the precipitation response to aerosols was discussed earlier by Khain et al. [2004, 2005, 2008b], Lynn et al. [2005a, 2005b, 2007], Lynn and Khain [2007], Tao et al. [2007], Fan et al. [2007b] for deep convective clouds and squall lines. Khain [2006] and Khain et al. [2008b] proposed the scheme allowing one to classify aerosol effects on precipitation under thermodynamical conditions with high freezing level (about 4 km), which can now be extended to supercell storms and, supposedly, to other cloud systems with warm cloud bases (Figure 16).

Figure 16.

Classification scheme of aerosol effects on precipitation. The results obtained in this study are denoted by pink boxes.

[53] Precipitation P at the surface during some time period can be represented as the difference between hydrometeor mass production G due to condensation and deposition integrated and hydrometeor mass loss L due to evaporation and ice sublimation integrated over the computational area and over time during which cloud system either decays or reaches a stationary state. In more detail the validity of definition of P as the difference of G-L is discussed by Khain [2009]. Consider an initial situation characterized by a relatively small aerosol concentration. This initial situation is schematically denoted by point A in Figure 16. An increase in the aerosol concentration leads, as it was shown by Khain et al. [2003, 2004, 2005, 2008b], Khain [2009], both to the increase in the condensate production ΔG (due to extra condensation in cloudy updrafts) and the condensate loss ΔL (due to extra evaporation or sublimation of precipitating mass as it falls out and the evaporation of the cloudy air as it mixes with the environment). The diagonal line in Figure 16 separates two zones. The upper zone corresponds to the condition ΔL > ΔG, i.e., to the precipitation decrease (scenario 1), in the zone below the line ΔL < ΔG; that is, the precipitation increases (scenario 2). The realization of the first or the second scenario depends on the environmental conditions.

[54] Any increase in humidity increases the condensation gain, but also decreases the evaporative loss by a certain fraction. Since the evaporative loss is greater for polluted air, the absolute magnitude of its reduction due to the increase in humidity is also greater, relative to the results in clean air. Hence, the precipitation is increased more in the continental aerosol case than in the maritime aerosol case for this increase in humidity and can exceed the maritime precipitation at high enough humidity (e.g., 90%). Such a situation was reported by Khain et al. [2005, 2008b] for tropical deep clouds in dirty air, in supercell storms (this study), as well as in case of squall lines [Lynn et al., 2005b; Tao et al., 2007]. This consideration allows us to attribute the increase in precipitation in dirty air reported by Wang [2005] for deep tropical convection, to high air humidity in the corresponding areas (these cases are marked by corresponding boxes in the scheme).

[55] On the other hand, a decrease in precipitation from orographic clouds under an increase in the aerosol loading [Givati and Rosenfeld, 2004; Jirak and Cotton, 2006] can be attributed to the low humidity, especially, over downwind mountain slopes, leading to a dramatically rapid evaporation of the condensate [Lynn et al., 2007]. Simulations of small warm rain cumulus clouds, as well as stratocumulus clouds [Magaritz et al., 2007, 2008] show a dramatic decrease in precipitation in dirty dry conditions. The aerosol-induced inhibition of precipitation in these clouds can be attributed to the following: an increase in the AP concentration leads to the formation of a large amount of small drops and drizzle that fall slowly and evaporate much more quickly than the rapidly falling raindrops typical of deep cumulus clouds. These results explain the decrease in the precipitation from small cumulus and stratocumulus clouds in dirty air reported in many studies [Albrecht, 1989; Rosenfeld, 1999, 2000; Rosenfeld and Woodley, 2000; Feingold et al., 2005], as well as inhibiting precipitation in dirty air for an Oklahoma warm cloud system found by Cheng et al. [2007].

[56] It should be noted that the relationship between ΔG and ΔL also depends on other thermodynamic parameters, such as the atmospheric instability (lapse rate) and the vertical wind shear. Instability increases the vertical velocity in clouds and transports more drops to the upper levels increasing both ΔG and ΔL, similarly to aerosol effects. The latter makes the separation of the aerosol and instability effects a difficult problem [Williams et al., 2002]. It is especially true, because in many cases (e.g., this study as well as those by Lynn et al. [2005b], Khain et al. [2008b], Tao et al. [2007], and van den Heever et al. [2006]) the changes in accumulated rain caused by aerosols over a mesoscale area usually do not exceed 10–30%. This, however, does not mean that such changes are not important, because the changes in the anthropogenic aerosols can determine a long-range trend in precipitation in some geographical regions.

[57] As it was shown by Khain et al. [2003, 2005] and Lynn et al. [2005a] and recently by Lee et al. [2008a, 2008b], clouds forming in dirty air create stronger downdrafts because of higher atmospheric cooling caused by stronger evaporation. These downdrafts increase the convergence in the boundary layer and foster the formation of secondary clouds, squall lines, etc. An increase in the boundary layer instability can significantly affect the rate of secondary cloud formation. The wind shear also affects ΔG and ΔL. We suspect, however, that its contribution is not unique. According to Lee et al. [2008a, 2008b] aerosols in a sheared flow foster formation of convective downdrafts (by evaporative cooling) and secondary cloud, leading to self-organization of deep convection. As a result of these dynamic effects, aerosols in the presence of wind shear and comparatively high humidity increase precipitation via triggering new cloud cells (ΔG > ΔL). Tompkins [2000] found that deep convection tends to be clustered preferentially in the part of the domain where the humidity is higher. On the other hand, very strong wind shear used in the present study for simulations of supercell storm in highly polluted air leads to detrainment of a lot of ice into the zone of dry air and to strong evaporation of condensate downwind. We suppose that extremely strong wind shear used in the present study is responsible for the effect that in case of high humidity maximum precipitation is reached in semipolluted, but not in the polluted cases. The role of wind shear in case of supercell and multicell storms requires further investigation.

[58] Besides, both ΔG and ΔL depend on the convection structure. Small single clouds experience intense mixing with the environment, which increases ΔL. Contrary to it, large cloud clusters, supercell storms and squall lines have large ΔG and relatively small ΔL, because humidity is high within the convection zone. Thus, such systems should reveal a stronger tendency to a precipitation increase with an increase in the aerosol concentration. At the same time, these systems require significant sources of aerosols to maintain aerosol concentration within the zone of intense convection.

[59] Figure 16 classifies the results obtained in different studies according to simulated (or expected) relationship between ΔG and ΔL for different conditions and cloud types. The results obtained in the present study are shown by boxes marked pink. It should be stressed that the SBM version used in this work does not allow reproduction of hail with the diameters exceeding 1 cm. We suppose that this limitation led to an underestimation of hail mass and hail size in the polluted cases. We believe that the use of the mass grid containing larger hail sizes would increase precipitation in polluted cases more substantially (especially in the high humidity case) than it was reported in the study. Thus, implementation of the possibility of the reproduction of big hail is one of the important future tasks to further improve the SBM.

[60] Note that the purpose of the scheme shown in Figure 16 is to separate different systems as regards to the sign of precipitation response to increase in the AP concentration. In this sense this scheme is only qualitative and does not allow deriving quantitative information about the precipitation increase/decrease. In addition, it concerns mainly cloud systems with a warm cloud base and high freezing level. Aerosol effects on the clouds with cold cloud base, as well as on cloud systems require special investigations. Teller and Levin [2005, 2006] found a significant sensitivity of precipitation from single Mediterranean cumulus clouds to aerosols. The effects of aerosols on midlatitude cloud ensembles, and mesoscale phenomena remain largely unknown. First results were reported only recently [Khain et al., 2008c, 2008d].

[61] Aerosols can influence not only the precipitation from single clouds and supercell storms, but precipitation from large mesoscale systems, such as tropical cyclones. The latter are caused not only by direct effect of aerosols on the microphysics of individual clouds, but via aerosol effects on the tropical cyclone intensity [Cotton et al., 2007; Rosenfeld et al., 2007, Khain et al., 2007, 2008a]. Further investigations of different types of clouds and thermodynamic parameters are required to make the presented classification scheme quantitative. This would involve real case studies under different environmental conditions using SBM type models or improved bulk schemes, perhaps with explicit prediction of drop concentration and aerosol budget included.

Acknowledgments

[62] This study was supported by the European grant ANTISTORM, as well as by the Israel Science Foundation, grant 140/07. It was also supported by a NASA contract (PNL-Battelle contract 46417).