Effects of aerosols on precipitation from orographic clouds



[1] Spectral (bin) microphysics was coupled to the Weather Research Forecast model to investigate the effect of aerosols (i.e., air pollution) on precipitation in the Sierra Nevada Mountains. Two-dimensional simulations were produced using either maritime (“clean-air”) or continental (“dirty-air”) aerosols. The simulation with clean air produced more precipitation on the upwind mountain slope than the simulation with continental aerosols. After 3 hours of simulation time, the simulation with maritime aerosols produced about 30% more precipitation over the length of the mountain slope than the simulation with continental aerosols. Sensitivity tests demonstrated the importance of relative humidity and vertical velocity on cloud microphysical structure and precipitation amount. Greater differences in precipitation amounts between simulations with clean and dirty air were obtained when ice microphysical processes were included in the model simulations.

1. Introduction

[2] Recent observational and numerical studies demonstrate a significant effect of aerosol particles on precipitation amount and spatial distribution [e.g., Rosenfeld, 1999; Ramanathan et al., 2001; Andreae et al., 2004; Givati and Rosenfeld, 2004; Khain et al., 2005; Lynn et al., 2005a, 2005b; Jirak and Cotton, 2006]. Effects of anthropogenic aerosols produced in urban areas on precipitation are of special interest. Studies have found that air pollution from industrial and urban areas can act to suppress precipitation [Rosenfeld, 2000; Borys et al., 2000]. Yet some work has shown precipitation enhancement around heavily polluted urban areas such as Houston [Shepherd and Burian, 2003] and Tokyo [Ohashi and Kida, 2002]. The difference in the results is possibly related to different environmental conditions in the zones investigated in the studies. As shown by Khain et al. [2005], aerosol effects on precipitation from deep convective clouds strongly depend on the thermal stability of the atmosphere, air humidity, and magnitude of the dominating wind shear. Since urban zones affect both thermal stability and aerosol concentration, the aerosol effects on precipitation can change from location to location. Moreover, since many factors affect precipitation formation in urban areas, it is difficult to reveal and to quantitatively evaluate effect of aerosols in these areas.

[3] In this sense, investigation of precipitation from topographically produced clouds located downwind of urban areas could provide better opportunity to reveal and evaluate aerosol effects. For instance, Givati and Rosenfeld [2004] examined the effects of air pollution on short-lived shallow clouds, forming over the mountains of California (and Israel) during the cold season. Jirak and Cotton [2006] focused their study on cold season clouds forming at elevated sites downwind of urban areas along the Front Range of California. Each found decreases in precipitation associated with polluted air relative to stations in pristine air of around 30%. Urban areas affect many meteorological parameters in zones located downwind (e.g., temperature, atmospheric stability, wind speed, etc.) that can affect precipitation regime. The main question addressed in the study is whether production of anthropogenic aerosol can be the mechanism of the decrease in precipitation reported by Givati and Rosenfeld [2004] and Jirak and Cotton [2006].

[4] This paper uses a spectral (bin) microphysics model (SBM) coupled with the Weather Research Forecast (WRF) model to reveal the sensitivity of precipitation from orographic clouds over the Sierra Nevada Mountains to aerosol concentration. Section 2 presents the conditions of numerical experiments, whereas section 3 examines the results. The discussion and conclusions are presented in section 4.

2. Design of Numerical Experiments

[5] To investigate aerosol effects on precipitation, an SBM scheme has been used that is based on solving an equation system for size distribution functions of drops, three types of ice crystals (dendrites, columns, and plates), snow, graupel and hail/frozen drops, as well as aerosol particles. This scheme has been described in detail in the work of Khain et al. [2004]. It has been used for investigation of aerosols effects on precipitation from single clouds under continental and maritime conditions using the Hebrew University cloud model (HUCM) [e.g. Khain et al., 2001, 2004; Khain and Pokrovsky, 2004]. A modified version of this scheme was coupled to a mesoscale model, MM5 and used for simulation of a rain event accompanied by squall line formation [Lynn et al., 2005a, 2005b]. It was found that the SBM allows better prediction of both spatial distribution and amount of precipitation as compared to commonly used bulk-parameterization schemes. Here, the full SBM scheme has been coupled to WRF [Skamarock et al., 2005] using the same approach for embedding the microphysics within the model dynamic time step as in Lynn et al. [2005a, 2005b].

[6] A significant advantage of the SBM is that it simulates cloud–aerosol interactions through the calculation of aerosol effects on the size distribution of cloud hydrometeors and hence on precipitation. At the same time, many SBM schemes artificially broaden the droplet size distribution (DSD) during the process of diffusional growth by remapping of the DSD on a regular mass grid at each time step [Khain et al., 2000, 2004]. This broadening can mask the effect of aerosols on the DSD, especially in the case of high aerosol concentration, when in fact a quite narrow DSD should form. In the present study, a new scheme of remapping has been used that conserves three moments of the hydrometeor size distributions (concentration, mass, and radar reflectivity) as compared with the commonly used scheme of Kovetz and Olund [1969] that conserves only concentration and mass during the remapping. The conservation of the radar reflectivity (6th moment) leads to the simulation of a narrower spectrum, more consistent with in situ formation of clouds under both low and high aerosol concentrations. In fact, supplemental simulations of well-documented convective clouds measured in Smoke, Aerosols, Clouds, Rainfall, and Climate (SMOCC) [Andreae et al., 2004] indicate that the DSD produced with the new scheme provided better agreement than the original scheme with observed DSD at all levels of measurements [BenMoshe et al., 2006].

[7] The coupled model was used to simulate the development of orographic clouds observed during 7 Dec 2003 (local sidereal time) over the Sierra Nevada Mountains located a few hundred kilometers inland from the California (Pacific) coastline. They are nearly parallel to the coastline. All simulations were produced using a single (nonnested) two-dimensional domain, oriented west to east. Simulations were run for 3 hours, which was the sufficient time for clouds to form on the upslope side of the mountain and to advect over the far mountain peak. The model was run at 6-s time steps using 1-km grid resolution in the horizontal and about 200-m grid resolution in the vertical. The simulation domain extended over 494 km, but results are presented for only the inner part of the domain, which was defined as beginning at 150 km and ending at 350 km. Radiation and surface model physics were not included. The clouds were generated by the vertical winds that resulted from the strong upslope (westerly) winds. Moreover, the turbulent kinetic energy (TKE) boundary layer scheme employed artificially included a TKE drag coefficient of 0.003 and ground sensible heat flux of 100 W m−2. The temperature, humidity, and wind are constant at the left boundary. There is a flux of temperature, humidity, and wind velocity from the left boundary into the domain. Within the domain, these fields are modified by the development of precipitation along the mountain ridge. At the right boundary, we used open boundary conditions; that is, zero horizontal gradients of these quantities were assumed.

[8] A satellite picture showing observed clouds is shown in Figure 1.

Figure 1.

Satellite picture of the cloudiness during polluted conditions (7 Dec 2003). One main peak of cloudiness and the sharp eastern boundary of the cloudiness are seen. Clouds are small convective or stratocumulus clouds.

[9] Initial atmospheric conditions were obtained from a sounding at Oakland, CA (16 LST 07 Dec 2003), and then interpolated to the WRF grid domain using the WRF standard interpolation program. The initial atmospheric conditions prior to the formation of clouds within the model domain (obtained from a “control” simulation at 30 min) are shown in Figure 2. We used the component of the vector wind perpendicular to the mountain ridge and specified this as the u component in the west-to-east WRF domain. The vertical velocity generally increased as the strong westerly winds transverse the highest mountain peaks, ranging from about 0.5 m s−1 to greater than 2.5 m s−1 (upper panel). Local topography peaks led to concentration of positive vertical velocities at western slopes of these peaks and negative velocities over eastern slopes. This structure of the vertical velocities fosters the formation of cumulus and stratocumulus orographic clouds. The temperatures (middle) range from about 284 K at the base of the mountain slope to about 260 K at the mountaintop at 3.3 km, with the freezing level located at a height just less than 1.5 km. The relative humidity was a minimum of between 50 and 60% at the base of the mountain slope, but generally increased as the moist air ascended and cooled on the mountain slope (bottom).

Figure 2.

West-to-east cross sections of vertical velocity and horizontal wind (upper panel), temperature (middle panel), and relative humidity (lower panel) 30 minutes after the start of the simulations. The figures show 201 grid elements, stretching from x = 150 to x = 350 km.

[10] The sensitivity of simulated precipitation to aerosols was tested using two distinct aerosol concentrations, referred to as either “maritime” (Mar) or “continental” (Con). The first represents “clean” air, whereas the second represents “dirty” air. The fields of cloud condensational nuclei (CCN) were initially (t = 0) assumed to be spatially homogeneous. The initial size distribution of CCN was calculated using the method described by Khain et al. [2000]. Initial dependence of cloud nuclei of supersaturation was given by a well-known expression: NCCN = N0Sk, where S is the supersaturation in % (maritime: N0 = 250 cm−3, k = 0.462; continental: N0 = 1250 cm−3, k = 0.308). The maximum size of dry CCN particles in the continental case was 0.4 μm, which roughly corresponds to a 2-μm radius nucleated droplet. The maximum size of dry aerosol particles in maritime air was assumed equal to 2 μm, which can produce nucleated droplets with radius of about 10 μm. Coefficients N0 chosen for the experiments provide realistic droplet concentrations in continental (several hundred to thousand per cubic centimeter) and maritime (∼100 cm−3) conditions. At t > 0, the size distribution of aerosols was modified through nucleation scavenging and advection.

[11] The size distributions of aerosol particles reaching the mountains most likely form as a result of a mixture of maritime aerosol (coming from the sea) and anthropogenic aerosols produced near the coastline. The maritime air contains small aerosol concentrations, but the size distribution is wider (with larger aerosol particles than in continental). The question arises, what is the effect of the tail of large maritime aerosols in the aerosol size distribution on precipitation? To clarify this problem, we performed a simulation referred to as (M + C), in which aerosol size distribution was obtained as a sum of the maritime and continental size distributions.

[12] Sensitivity tests were produced that included increasing the relative humidity from the surface to 2 km to 90%, and from 2 to 5 km to 50%. These were referred to as Mar-RH90 and Con-RH90, respectively. Two more simulations were produced using a reduced value for the sensible heat flux of 10 W m−2, referred to as Mar-SHF10 and Con-SHF10. Finally, simulations were done to simulate the effect of background wind on precipitation under both maritime and continental aerosol conditions, referred to as Mar-3/4 and Con-3/4. In these last simulations, the profile of the horizontal wind speed was set equal to three fourths of its initial value in the control.

3. Results

3.1. Comparison of Mar-Control With Con-Control

[13] The differences in aerosol concentration led to important differences in the microphysics of the simulated orographic clouds. Figure 3 shows that the maximum droplet concentration in Con-Control reached greater than 1000 cm−3, whereas in Mar-Control, the droplet concentration did not exceed 100 cm−3. Figure 4 shows that Con-Control produced much more cloud liquid water (LWC) on the upwind slope, but less rainwater content (RWC) than Mar-Control (Figure 5).

Figure 3.

West-to-east cross sections of cloud droplet concentration simulated with (a) MAR-Control and (b) Con-Control at 3 hours.

Figure 4.

West-to-east cross sections of cloud water content simulated with (a) MAR-Control and (b) Con-Control at 3 hours.

Figure 5.

Same as Figure 3, but for rainwater mass.

[14] Khain et al. [2004, 2005] and Lynn et al. [2005a, 2005b] simulated deep convective clouds and obtained larger LWC in “polluted” clouds. As explained in these papers, the number of droplets forming in continental air masses is quite large, but because these droplets are relatively small, they do not fall as precipitation, but remain suspended in large numbers in clouds and continue growing by diffusion. The same mechanism appears to be effective for relatively shallow orographic clouds as well. Note that cloud droplets in the Con-Control reach higher heights than in Mar-Control because the smaller size of the droplets in the Con-Control allows them to escape accretion onto precipitation. In comparison, large raindrops formed in Mar-Control fall down before reaching even 1.5 km above the surface.

[15] The Con-Control produced more ice crystals (Figure 6) and snow (Figure 7) than Mar-Control, especially downwind on the mountain slope (and even beyond the highest peak). The higher production of ice crystals and snow content in the Con-Control can be attributed to several factors: first, the process of droplet freezing is not efficient, as stated, in the Con-Control case since most liquid droplets remain quite small. Thus, in Con-Control, most droplets ascend to levels of about −10 to −20°C temperatures. Here, they reach sizes large enough (larger than 10 μm in radius) to be collected by ice crystals (formed by primary ice nucleation, which at these heights reach sizes exceeding ∼50 μm, through depositional grown). The collision of ice crystals then leads to formation of snow.

Figure 6.

Same as Figure 3, but for ice crystal mass (dendrites, columns, and plates).

Figure 7.

Same as Figure 3, but for snow mass.

[16] In contrast, the Mar-Control simulation produced much more graupel mass (and large frozen drops—not shown) on the first half of the upwind side of the slope than Con-Control (Figure 8). The formation of graupel in the Mar-Control at between x ∼ 50 and x ∼ 90 km is related to the freezing of raindrops at comparatively high temperatures (−5 to −8°C). In the Con-Control, the drops are small and their freezing is ineffective. At the same time, these drops are collected by large crystals to produce graupel. Thus, the production of graupel in the Con-Control is caused mainly by process of riming of ice crystals and snow and is concentrated in the area of the high LWC, snow, and ice contents. In both cases, graupel falls on the upwind slope because of significant sedimentation velocity, but it forms and falls farther upwind in Mar-Control than Con-Control.

Figure 8.

Same as Figure 3, but for graupel.

[17] Figure 9 shows accumulated precipitation (warm + ice) obtained from the Mar-Control and Con-Control for the 3-hour simulation period. The figure shows that the maritime simulation produced more precipitation upwind (toward the western boundary or sea) than the simulations with continental aerosols. In fact, the precipitation accumulated in Mar-Control experiment began about 40 km upwind of the starting point of accumulation in Con-Control. Also, the highest amount of precipitation in Mar-Control fell to just to the west of the highest peak, whereas in Con-Control, the largest amount of precipitation fell downwind of the highest peak.

Figure 9.

Accumulated precipitation on the mountain slope for 3 hours for both Mar-Control and Con-Control.

[18] In total, there are three maxima in the precipitation distribution in Mar-Control, whereas there are two peaks in the precipitation distribution in Con-Control. Each occurred near local maxima in topography. On the basis of our analysis of the figures above, the first maximum in Mar-Control's precipitation peak occurred because of warm rain processes. The second maximum occurred mainly because of graupel. Ice crystals and snow induced the third maximum in the precipitation peak. In Con-Control, warm rain processes did not contribute to precipitation. Rather, the first maximum was formed by graupel, but the amount of graupel in Con-Control was apparently less than in Mar-Control. Both cloud ice and snow processes led to the formation of the second maximum over the highest peak, which was somewhat larger than obtained in Mar-Control.

[19] In both simulations in the third peak, there was sedimentation of crystals, graupel, and snow in downdrafts over the eastern side of the peak from stratocumulus-like clouds with cloud base located at or near the surface. Yet the masses of snow and ice crystals in the Con-Control case were larger and advected farther eastward than in the Mar-Control; thus the precipitation peak is larger in the Con-Control and was shifted farther downwind than in the Mar-Control case. The value of the excess was, however, significantly smaller than the deficit in the precipitation in Con-Control over the upwind slope. As a result, the accumulated precipitation in the Mar-Control case turned out to be larger than in the Con-Control case by about 30% (see Table 1), which corresponds well to the observations by Givati and Rosenfeld [2004] and Jirak and Cotton [2006].

Table 1. Accumulated Precipitation (mm) Obtained During 3 Hours of Simulations
Model RunMar (3 hours)Con (3 hours)
  1. a

    The data were averaged from 150 to 350 km in the simulation domain.

Control runs0.440.32
SHF 10 W m−20.530.34
3/4 Wind0.1580.04
RH = 90%3.623.78

[20] The west-to-east cloud structure obtained in both simulations was punctuated by convective elements west of the highest topographical peak, with stratiform cloud over the peak and downwind. This type of cloud structure is similar to that observed and shown in Figure 1. Moreover, both simulations produced a sharp cutoff in precipitation amount and cloud mass downwind of the highest peak (as implied by the satellite observation). Also, the Con-Control simulation produced many supercooled droplets at cloud top, with ice particles present in maximum amount below this level (about 3 km). According to aircraft observations on the same day of these simulations, cloud tops of polluted clouds contained a large amount of supercooled droplets, whereas a large amount of ice particles was located below.

[21] The difference in the accumulated rain is related to higher precipitation loss in the Con-Control. For instance, ice crystals and snow penetrating eastward until 180 km contributed to precipitation only slightly, as noted, because of high evaporation within the range 150 < x < 180 km, where the relative humidity is low (see Figure 10a) because of downdrafts leading to air heating. Thus the important factor in the decrease in the accumulated precipitation in the Con-Control is the higher loss of precipitating mass by ice sublimation in the dry air farther eastward beyond the highest peak and over the downwind slope. The concept of the higher loss of precipitating mass in clouds developing in dirty air was also the major mechanism by which aerosols can decrease precipitation from deep clouds [as discussed by Khain et al., 2005 in detail].

Figure 10.

Relative humidity fields in simulations (a) Con-Control and (b) Con-RH.

[22] The fields of supercooled water and ice crystals and snow indicate that cloud tops in the Con-Control are higher than in the Mar-Control so that aerosols invigorate the orographic clouds. This result corresponds to finding by Khain et al. [2004, 2005] and Lynn et al. [2005a, 2005b] obtained in simulations of deep convective clouds, and reflects the dynamical effects of aerosols. In the Con-Control case, droplets continue growing by diffusion leading to higher latent heat release as compared with the Mar-Control. The formation of larger amounts of ice also leads to higher latent heat releases. This leads to higher vertical updraft velocities in the Con-Control as compared with the Mar-Control (Figure 11). Another reason for higher cloud tops in the Con-Control is that both droplets and ice particles are small and escape accretion onto precipitating particles. Therefore, having low sedimentation velocity, they are able to ascend to higher levels than in Mar-Control.

Figure 11.

Vertical velocities in the Con-Control and Mar-Control cases.

[23] There was a very high similarity of the results obtained in the new (M + C) simulation and in C-case performed earlier. The differences both in the amount and spatial distribution of precipitation turned out to be negligible. Accumulated rain was 0.34 mm in the (M + C) simulation as compared with 0.32 mm in the Con-Control. We attribute this effect to the compensation of the effects of a general increase in concentration of aerosols compared to the increase in the possible effect on precipitation of the amount of large aerosols. Thus results of the (M + C) simulation indicate that the existence of relatively small amount of large maritime aerosols cannot change the effect of anthropogenic aerosols. Note that no ultragiant CCN were assumed in the aerosol spectrum either in the M and (M + C) runs.

3.2. Comparison of Liquid-Only and Mixed-Phase Microphysics Simulations

[24] Figure 12 shows rainfall obtained from Mar-Control-Liq and Con-Control-Liq. Comparing with the corresponding graph in Figure 9, one notes that the aerosol-induced differences in accumulated precipitation are much larger when ice processes are included. Significant difference in precipitation amounts in the liquid-only and mixed-phase microphysics with continental aerosols is seen at x ∼ 100 km (local topography maximum). Figure 5b indicates that at x ∼ 100 km, a small amount of warm rain occurred in Con-Control, even when ice microphysics was included. This indicates that collisions between drops start to be efficient to produce warm rain over the first large topographical peak. However, the formation of ice particles by drop-ice collisions actually eliminated warm rain in the Con-Control when ice microphysics was included. Since graupel particles formed do not collect any ice particles (as assumed in the microphysical scheme), the ice particles, including graupel, remain comparatively small and have lower sedimentation velocity and are advected downwind. Thus, simulation of liquid-only processes (without liquid/ice interaction) increases precipitation over the upwind slope and decreases it over the downwind slope. Thus, aerosols leading to narrowing of the DSD affect significantly not only warm but also ice cloud microphysics and, accordingly, precipitation distribution and amount.

Figure 12.

Same as Figure 8, but for simulations with liquid-only microphysical processes.

3.3. Sensitivity Tests

[25] To investigate the effects of air humidity on orographic precipitation, sensitivity tests were produced that included increasing the relative humidity from the surface to 2 km to 90%, and from 2 to 5 km to 50%. These were referred to as Mar-RH90 and Con-RH90. Cloud microstructure also depends on vertical velocities and wind speed. Respectively, we performed the sensitivity experiments in which we study effect of the air humidity, effect of surface heat fluxes, and effect of the background wind speed.

3.3.1. Effects of Air Humidity

[26] Increasing the relative humidity led to a decrease in the height of cloud base and to a shift of cloud formation westward (Figure 13) to x ∼ 15 km. Clouds formed near the underlying surface (Figures 10a and 13b). Since the slope was quite gentle at x ranged from 0 to 30 km, the vertical velocity at cloud base was less than ∼1 m s−1. This led to a relatively low droplet concentration even in Con-RH90 that varied from several tens near cloud base to several hundred at 1–1.5 km above the underlying surface (Figure 13a). Thus, high air humidity led to large LWC (Figure 13b). Such cloud structure fosters droplet collisions since the droplet concentration was not high and the DSD contained large droplets nucleated near the cloud base. Moreover, high humidity dramatically decreased the evaporation of falling droplets. As a result, the increase in the air humidity led to warm rain formation in Con-RH90. In Mar-RH90, warm rain also started at smaller x (Figure 14).

Figure 13.

Droplet concentration and LWC in simulation Con-RH90 at t = 3 hours.

Figure 14.

Fields of RWC in simulations Mar-RH90 and Con-RH90 at t = 3 hours.

[27] However, this spatial difference in the initiation of precipitation was not accompanied by the decrease in accumulated rain in Con-RH90 as compared with Mar-RH90. The production of larger ice precipitation in the Con-RH90 (not shown) fully compensated for the decrease in the warm precipitation. As seen in Figure 15, the increase in humidity assumed in these sensitivity simulations largely eliminated the aerosol effects on precipitation amount. Moreover, the accumulated rain in Con-RH90 was even larger than in Mar-RH90 (see Table 1). We suppose that situations with such high air humidities are relatively rare so that they do not contribute significantly to the statistics presented by Givati and Rosenfeld [2004] and Jirak and Cotton [2006].

Figure 15.

Same as Figure 8, but for simulations Mar-RH90 and Con-RH90 at t = 3 hours.

3.3.2. Effects of Surface Heat Fluxes

[28] Figure 16a shows results from the Mar-SHF10 and Con-SHF10. Comparing to the appropriate graph in Figure 9, one sees that the 90% decrease in sensible heat flux led to a small decrease in accumulated precipitation, while increasing somewhat the difference between Mar-SHF10 and Con-SHF10 on the upwind slope. In general, the effects of surface heat fluxes turned out to be not very important as concerns the precipitation amount and distribution. We attribute this effect to the fact that the characteristic timescale of the air heating due to the fluxes is of several hours, so the fluxes could not change temperature and relative humidity significantly during the 3-h period of simulations.

Figure 16.

Same as in Figure 15, but for simulations with a sensible heat flux of (a) 10 W m−2 and (b) smaller (3/4) background wind speed.

3.3.3. Effects of Wind Speed

[29] Decreasing the wind to three fourths of its initial value reduced the amount of precipitation in both Mar-3/4 and Con-3/4 (Figure 16b). This is because a decrease in the horizontal velocity led to a corresponding decrease in the vertical velocity over the upwind slope. Small vertical velocities produce very weak and shallow clouds at the upwind slope. Because of low cloud depths, ice processes are ineffective. Each simulation produced two peaks in precipitation, and Con-3/4 produced precipitation even near the beginning of the upwind slope. Because there were smaller vertical velocities, the droplet concentration was smaller in Con-3/4 than in Con-Control (not shown). However, because of small vertical velocities and low supersaturation, droplets grow slowly. Respectively, precipitation at the surface forms via cloud “sedimentation” (drizzling) on upwind mountain slope.

4. Conclusions

[30] SBM was coupled with a two-dimensional version of the WRF model and used to investigate aerosol effects (pollution) on amount and spatial distribution of precipitation in the Sierra Nevada Mountains. Two microphysical situations were simulated: the first with low concentration maritime aerosols (clean air), and the second with high concentration continental aerosols (dirty-air), using initial sounding data from 7 Dec 2003. The continental aerosol simulation was designed to reproduce the effect of anthropogenic aerosols produced in upwind urban areas on downwind precipitation forming on the mountain slope. The maritime aerosol simulation reproduced precipitation forming in pristine air. Simulations were performed with a relatively high grid resolution (1 km in the horizontal and 200 m in the vertical). After 3 hours of simulation, precipitation amount in the maritime simulation was about 30% larger on the upwind slope than in the simulation with continental aerosols.

[31] The maritime simulation produced warm rain near the beginning of the upwind slope, whereas the continental simulation did not produce warm rain anywhere on the slope. The maritime simulation also produced graupel precipitation in a second maximum occurring further up the mountain slope above a topographical peak. The continental simulation produced less graupel in the same location and less precipitation. This simulation, however, produced more cloud ice and snow than the maritime simulation, which accumulated in larger amounts on the highest peak and downwind. Owing to the greater production of cloud ice and snow, the maximum amount of precipitation on the highest topographical peak was shifted downwind from the location of maximum precipitation in the maritime simulation. Both simulations had convective-type precipitation on the upwind slope, which transitioned to stratiform precipitation farther up the mountain slope. Evaporation of cloud ice and snow in atmospheric downdrafts beyond the highest peak led to a sharp cutoff in precipitation downwind of this peak, similar to what was shown in an observed satellite photograph [in more detail, the strong effect of ice sublimation on precipitation is discussed by Khain et al., 2005].

[32] As noted, the simulation with continental aerosols produced more cloud ice and snow particles than the simulation with maritime aerosols without producing warm rain. Clouds forming in the continental aerosol air turned out to be more vigorous and reached higher heights than those formed in clean air with ice crystals, and snow within that had lower sedimentation velocities than raindrops and graupel. This led to a shift of ice precipitation downwind in the simulation with continental aerosols compared to warm rain precipitation and graupel type precipitation in the maritime simulation. These cloud ice and snow particles were advected by the background wind and, as noted, evaporated on the downwind side of the highest mountain peak. Because cloud ice and snow particles were evaporated, the simulation with continental aerosols produced less precipitation over the whole mountain slope, owing to the greater prevalence of these types of precipitation particles in this simulation than in the maritime simulation. The results obtained using the aerosol size distribution equal to the sum of maritime and continental aerosol size distributions indicate that the existence of relatively small amount of large maritime aerosols does not change the effect of anthropogenic aerosols.

[33] According to statistical analysis [Givati and Rosenfeld, 2004; Jirak and Cotton, 2006], precipitation over the mountain regions located downwind from coastal urban areas decreased during past several decades by about 30%. Our simulations show that the effects of anthropogenic aerosols can be the reason for such a decrease. It is clear that the results of statistical analysis include a wide range of meteorological situations. Partially, we show the possible variability in our sensitivity tests. It is shown that anthropogenic aerosols decrease precipitation under specific conditions, e.g., for comparatively dry environmental conditions. We believe that these conditions play a dominating role in the total statistics.

[34] The mechanism of aerosol-induced convection invigoration was first analyzed in detail by Khain et al. [2004, 2005]. Now the aerosol-induced convection invigoration has been reported in many numerical studies [e.g., Lynn et al., 2005a, 2005b; Wang, 2005; Teller and Levin, 2005] and observations [Koren et al., 2005]. It seems that it is a common effect that clouds in environments with higher aerosol concentrations become deeper and more vigorous with more glaciation as a consequence.

[35] Supplemental simulations with ice microphysics excluded revealed a crucial role of ice formation in the aerosol effects on precipitation. Without simulated ice processes, the simulation with continental aerosols produced more precipitation in the location of the first topographical peak than it did when ice processes were included. Yet the precipitation amount in the maritime simulation did not show similar sensitivity to the inclusion or exclusion of ice processes. This further emphasizes the importance of drop size distribution on the size distribution and types of ice particles that formed in each simulation. Thus, ice formation significantly intensifies the effects of aerosols on the precipitation amount and its spatial distribution.

[36] In sensitivity tests, we identified relative humidity and wind speed as critical environmental factors that determined both precipitation amounts and relative differences between simulations in clean and dirty air. Higher humidity decreased the cloud base level and triggered the cloud formation farther upwind on the mountain slope where the vertical velocity was smaller than farther downwind on the slope. As a result, droplet concentration turned out to be relatively small, and droplet spectra distributions were able to develop to produce raindrops. Effective warm rain formation occurred even in the continental aerosol case (but with some time delay and spatial shift in the downwind direction). Also, high relative humidity reduces precipitation loss caused by drop and ice evaporation. Thus, the increase in air relative humidity decreased the difference in precipitation amounts between the clean- and dirty-air simulations, and even changed the sign of this difference. Regarding our simulations with reduced sensible heat flux, it is possible that over a longer time period in which atmospheric humidity and temperature could be modified, fluxes of larger magnitude or fluxes with spatial and nonuniform variability in time could each have had a stronger effect on the precipitation.

[37] A decrease in the speed of the background wind led to a decrease in the vertical velocity and to a delay in the cloud and precipitation formation. Even so, the maritime precipitation formed earlier and in greater amounts than in the continental simulation. Low vertical velocities lead to formation of narrow clouds, which precipitate by cloud “sedimentation” (drizzling) on the upwind mountain slope.

[38] The main result of these idealized simulations is the revealing of aerosol effects on precipitation formation and distribution from orographic clouds, as well as revealing the most important microphysical and environmental factors that can enhance or inhibit the aerosol effects.

[39] Future studies will be focused on the quantitative evaluation of aerosol effects with a three-dimensional version of the same model, and include forcing from radiation processes.


[40] The authors express their deep gratitude to Dr. J. Dudhia for consulting related to utilization of the WRF model and implementation of the SBM in WRF. The authors are indebted particularly to the Public Interest Energy Research (PIER) Program of the California Energy Commission (CEC) for its support of the study. They thank especially Mr. Guido Franco of PIER/CEC for his support and advice. This study was supported also by NSF (grant 0503152) and by the Israel Water Company (grant 162/03).