-
The simulations in this study use the WRF model version 3.8.1, employing aerosol-aware microphysics parameterization, open lateral boundary conditions, 3DTKE (Zhang et al., 2018), and a free-slip condition at the lower boundary. Dry aerosols are horizontally advected or locally diffused as cloud particles by model dynamics during the simulation. The model domain is 300 km × 300 km × 20 km, with each storm simulated for 3 h. Horizontal grid spacing is 1 km and vertical grid spacing is 200 m. The model has 101 vertical levels and the levels are equally spaced in height from the surface up to the top of the model. According to Weisman and Rotunno (2000), these resolutions are sufficient to resolve storm-scale features, such as mid-level updraft structure and low-level mesocyclogenesis but are not generally considered sufficient to accurately represent tornadogenesis. Storms are triggered using an ellipsoidal bubble of warm air with a horizontal radius of 10 km and a vertical radius of 1500 m, with a maximum temperature perturbation of 3 K specified at the center of the bubble, decreasing to zero at the edges. The thermodynamic profile used for each simulation is presented in Fig. 1, similar to that used by Weisman and Klemp (1982, 1984, 1986), representing an environment of moderate instability (2667 J kg−1 of Convective Available Potential Energy, CAPE), with moist conditions throughout the troposphere. The vertical wind shear is based on the unidirectional shear profile (or straight shear) in Weisman and Rotunno (2000) to simplify the interpretation of the results, leading to the generation of mirror-image "splitting" supercell storms.
Figure 1. Initial Skew-T diagram showing temperature (black line), dew point temperature (blue line), parcel ascent starting at the Level of Free Convection (red dashed line), and wind speed (bars) profiles used in model simulations.
Two 6-class aerosol-aware BMSs are used to represent aerosol effects in the simulations. Table 1 compares the predicted hydrometeor categories in the Thompson and CLR schemes: cloud water, cloud ice, rain, snow, and graupel; their density and size distribution (SD) are also listed. The Thompson scheme has a one-moment prediction of the mass mixing ratio for snow and graupel, and a two-moment prediction (addition of number concentration) for cloud water, cloud ice, and rainwater. Although the density and size distribution of hydrometeors are represented differently in the two schemes, the current study will show characteristic differences beyond those bulk properties. For aerosols, the types and chemical compositions are simplified into two categories: "water-friendly" aerosols (sulfates, sea salts, and organic matter) for CCN and "ice-friendly" aerosols, i.e., dust, for ice nuclei (IN), each with unimodal lognormal ASD. In comparison, the CLR scheme has two-moment predictions for each type of hydrometeor. The CCN is presumed to be non-coated background aerosols, and ice nuclei (IN) as dust, with a default trimodal lognormal ASD.
Prognostic variable Thompson CLR CCN (ASD) unimodal trimodal IN (ASD) unimodal trimodal Cloud water (moment, ρc, SD) double, 1000, gamma double, 1000, exp Rainwater (moment, ρr, SD) double, 1000, exp double, 1000, exp Cloud ice (moment, ρi, SD) double, 890, exp double, 500, exp Snow (moment, ρs, SD) single, 100, exp & gamma double, 100, exp Graupel (moment, ρg, SD) single, 500, exp double, 400, exp Table 1. The physical characteristics/properties of the prognostic variables in the Thompson and CLR schemes. The aerosol size distribution (ASD) is listed with the size distribution (SD) of each hydrometeor species along with the prognostic moment (s) and density (kg m−3). The exp and gamma stand for inverse exponential and incomplete gamma distributions, respectively.
Droplet nucleation is the most critical step for introducing dry aerosols into clouds. The Thompson scheme activates CCN using a look-up table as a function of five variables: air temperature, vertical velocity, number of available aerosols, pre-determined hygroscopicity (0.4, 0 for hydrophobic, and 1 for hydrophilic particulates), and the aerosol mean radius (0.04 μm). The look-up table is originally derived from an explicit treatment of the Köhler activation theory within a parcel model (Feingold and Heymsfield, 1992; Petters and Kreidenweis, 2007; Eidhammer et al., 2009). When the air is saturated, the condensation is parameterized by saturation adjustment in the Thompson scheme, but by explicit treatment of supersaturation in the CLR scheme. The CLR scheme activates CCN explicitly according to the Köhler equation. When supersaturation occurs, CCN with radii larger than the critical radius is activated. The critical radius is the minimum radius of activated CCN particles depending on supersaturation. According to Cheng et al. (2010), to simulate CCN activation and cloud diffusional growth, higher temporal resolution is needed, because supersaturation can change drastically after being explicitly resolved spatially. Therefore, the scheme uses time splitting integration in a Lagrangian framework, treating the airmass in each grid box as an ascending/descending air parcel, to resolve the changes in CCN and thermodynamic fields caused by activation and diffusional growth. In this way, the maximum supersaturation can be explicitly resolved, and the number of activated cloud drops can be more accurately simulated.
Some of the aerosol species are effective IN, such as mineral dust. IN activation is also parameterized by the two schemes independently of CCN activation. To be successfully activated, IN must compete with hydrometeors (and sometimes CCN) for water vapor. In the Thompson scheme, the deposition nucleation of IN is highly dependent on temperature and IN number concentration following the parameterization of DeMott et al. (2010). Part of the deliquesced CCN can also be nucleated to cloud ice as proposed by Koop et al. (2000). Besides, ice can also be initiated in the freezing processes (IN immersion or contact freezing), and the freezing rate would increase with increasing IN number concentration. In the CLR scheme, the total number concentration of cloud ice is predicted based on supersaturation with respect to the ice following Meyers et al. (1992). When the existing number of crystals exceeds the predicted amount, ice crystals are allowed to remain, but no additional condensation-freezing nucleation is performed. Contact freezing is not parameterized. When the IN is insufficient, cloud ice can be produced by the secondary ice production process (Hallett and Mossop, 1974) before homogeneous nucleation can occur in both schemes. To allow for the fully functional cold-rain process with the maximized IN effects, this study provides a sufficient IN number concentration of 400 per liter for all simulations, a concentration much larger than a typical background IN concentration of 0.1 particles per liter of air referred to by Eidhammer et al. (2010).
A key point of this study is to investigate the relative significance of aerosol number concentration and size distribution in the parameterized aerosol effects. Figure 2a shows the prescribed clean and polluted CCN profiles for the initial conditions of the four baseline experiments. The clean profile has a maximum concentration of 1.38 × 108 m−3 at the surface that exponentially decreases as the height increases, whereas the concentration of the polluted profile is 1000 times greater than that of the clean profile. Figure 2b expands the lognormal aerosol-size spectra presumed by the Thompson scheme (Thompson and Eidhammer, 2014). The ASD in the two baseline experiments assumed by the CLR scheme is remodeled to match the unimodal distribution of the Thompson scheme. A comparison between unimodal and trimodal ASDs is briefly presented in section 3.5.1. The trimodal ASD (including a nucleation mode, an accumulation mode, and a coarse mode) is used only in two additional experiments to compare the droplet activation efficiency by different ASDs. The Trimodal_Marine and Trimodal_Urban use marine surface background and urban average trimodal ASDs, respectively, after Whitby (1978), but are normalized to the same total aerosol number concentration in the clean experiments. The trimodal number concentration, mass, mean radii, and standard deviations used in the experiments are listed in Table 2. Trimodal IN ASD is prescribed in the CLR scheme but not considered in the nucleation of cloud ice in the present study.
Figure 2. (a) Initial vertical profiles of the aerosol number concentration decreasing from the surface to the top of the atmosphere, with a scale height of 2 km for the four baseline experiments. (b) The presumed unimodal and trimodal aerosol probability density functions used in the experiments.
Trimodal_Marine Trimodal_Urban IN Mode Unimodal N A C N A C N A C Number concentration (%) 100 84.37 14.89 0.74 76.8 23.2 0 0 99.88 0.12 Mass (%) 100 0.011 0.289 99.7 1.1 55.46 43.5 0 6.5 93.5 Mean radius (μm) 0.04 0.005 0.036 0.31 0.007 0.027 0.43 0.0045 0.024 0.55 Standard deviation (σ) 0.59 0.47 0.69 0.99 0.59 0.77 0.79 0.53 0.75 0.74 Table 2. Unimodal and trimodal aerosol properties used in our experiments. N as nucleation mode, A as accumulation mode, and C as coarse mode.
Prognostic variable | Thompson | CLR |
CCN (ASD) | unimodal | trimodal |
IN (ASD) | unimodal | trimodal |
Cloud water (moment, ρc, SD) | double, 1000, gamma | double, 1000, exp |
Rainwater (moment, ρr, SD) | double, 1000, exp | double, 1000, exp |
Cloud ice (moment, ρi, SD) | double, 890, exp | double, 500, exp |
Snow (moment, ρs, SD) | single, 100, exp & gamma | double, 100, exp |
Graupel (moment, ρg, SD) | single, 500, exp | double, 400, exp |