On 20 January 2022, an airborne glaciogenic cloud seeding experiment was conducted in Hebei Province over the North China Plain. A persistent supercooled stratiform cloud was documented in the flight area before it was seeded. No surface precipitation was observed, and no radar echo was detected by the ground-based S-band radar, suggesting minor or no natural ice formation in the cloud. Weak low-level baroclinicity was present, with colder air to the north, and a weak southerly flow provided sufficient water vapor, resulting in a high ambient relative humidity (Fig. a). There was warm air advection in this area as suggested by the wind veering (Fig. a and b). A high-level (500 hPa) cloud system was controlled by dry westerly flow around a weak ridge that dominated in the flight area (not shown). A stratiform cloud deck was present at 1.4–1.9 km above the ground level (Figs. c and b). This cloud was decoupled from the surface, only ∼ 500 m deep, non-precipitating (Fig. ), and with a cloud top temperature of about -16 °C (as illustrated in Fig. a of the following subsection). The environment was synoptically quiescent and stably stratified as seen from the positive Brunt–Väisälä frequency at the cloud layer (Fig. b) and the potential temperature profile in Fig. a.

Seeding was conducted at the cloud top at about 08:00 UTC, as shown by the red lines in Fig. c. The true air speed of the aircraft was approximately 100 ms-1, and the release rate of AgI particles was 10¹⁴ s-1 as estimated based on the mass burned per second using burn-in-place pyrotechnic flares. The aircraft flew one and a half north–south-oriented legs while seeding (Fig. c). Any seeding signatures would have to reveal the same spatial pattern in the cloud/precipitation field (advected downwind), as illustrated vividly in photographs of the first airborne cloud seeding experiments (Schaefer, 1949) and reproduced in many meteorology textbooks (e.g., Rogers and Yau, 1989; Lutgens et al., 2006). In areas unaffected by seeding, the cloud top was fairly flat as seen from the visible images detected by the FY4A satellite at 08:45 UTC (Fig. c), and the brightness temperature at 12 µm varied between -12 and -14 °C (Fig. d). Clear seeding signatures were detected downwind of the seeding line on both the visible and infrared images. The IR brightness temperature increased by about 2 °C along the advected flight track, indicating that the clouds become thinner, and the cloud tops descend due to the consumption of supercooled liquid water by the growth of ice crystals. The more resolved visible image reveals a reduction in cloud brightness along the same track, indicating a reduction in droplet concentration and an enhancement in reflected solar radiation along the northeast flank of the cloud top depression (at 08:00 UTC the sun is in the southwest in the flight area). The displacement of the satellite signature relative to the flight track is consistent with the wind speed and direction at the cloud top level.

The seeding signatures were also detected by the ground-based S-band radar (Fig. ), which is located north of the flight area (Fig. a). The radar has a range resolution of 250 m and operates in a volume-scanning mode with nine elevation angles. The scan at 1.5° best captured the enhanced reflectivity since the cloud depth was only about 500 m. As seen in Fig. a, the radar echo appeared about 10 min after seeding. This is consistent with radar data collected during airborne AgI seeding of a shallow stratus cloud deck under quiescent synoptic conditions in Switzerland (Henneberger et al., 2023) and continuously strengthened till 08:42 UTC. Since the seeding started from the eastern flight leg, the radar reflectivity exhibited sequential enhancement. The reflectivity varied between -10 and 10 dBZ, and it gradually weakened after 08:48 UTC. Although changes in radar reflectivity and brightness temperature indicate ice crystals forming, no surface precipitation from cloud seeding was observed. Furthermore, since the cloud was mostly liquid and since no surface precipitation was observed in areas unaffected by seeding either, we are not able to investigate the downwind effect of precipitation using observations. However, we may conclude that the LWC that is available for precipitation downwind of the target area has been reduced by the seeding operation. The downwind effect will be discussed using simulations in Sect. .

The LES mode in the Weather Research and Forecasting model is used to conduct idealized numerical simulations for the case described above. To reproduce the entire seeding trajectory, we use a domain size of 80 km × 80 km × 3 km with periodic lateral boundaries. The surface is assumed to be flat. The model has a horizontal resolution of 100 m and 90 levels in the vertical direction. The idealized simulation with this resolution can resolve eddies larger than 600 m, and this is sufficient to reveal the influence of the seeding effect based on our analysis. The seeding trajectory is the same as that performed in the field experiment, in which the aircraft flew at a speed of 100 ms-1 along the red line in Fig. c and released the AgI particles near the cloud top. The AgI particles are assumed to be homogeneously mixed in a grid box as soon as they are released from the aircraft. The physics schemes used in the simulation include the fast spectral bin microphysics scheme (Khain et al., 2004), the Revised MM5 surface layer scheme (Jiménez et al., 2012), the Noah Land Surface Model (Tewari et al., 2004), and the Rapid Radiative Transfer Model (Mlawer et al., 1997). We use the default natural ice nucleation parameterizations in the fast spectral bin microphysics scheme, including deposition and condensation nucleation, contact nucleation (Meyers et al., 1992), and immersion freezing (Bigg, 1953). Cumulus and boundary layer parameterization are turned off in the LES. The cloud condensation nuclei (CCN) concentration is expressed by NCCN=N0Swk, where N0 refers to the CCN concentration at a supersaturation level of 1 %, Sw represents the supersaturation with respect to water (%), and k is the slope of the CCN size distribution. For the continental area of China, N0 = 4000 cm-3 and k=0.9 at the surface are assumed. This is a polluted condition. We also tried a lower CCN concentration (N0 = 2000 cm-3), but the results are similar (not shown). The reason is that in this case there was no warm rain process, and the ice particles grew through the WBF process (discussed later), which is primarily controlled by the LWC rather than the droplet concentration. For other cases in which the microphysics schemes are sensitive to the droplet concentration, the seeding effect could be different between cases with clean and polluted environments. This is out of the focus of this study, but it would be interesting to investigate the aerosol impact in the future.

The parameterization of AgI nucleation implemented in the fast spectral bin microphysics scheme was developed by Xue et al. (2013), including four nucleation modes: deposition nucleation, condensation nucleation, contact freezing, and immersion freezing. The fraction of AgI aerosols that can nucleate is confined to a specific range of temperature and supersaturation ratio, and the sum of the four nucleation mode fractions cannot exceed one. The contact and immersion freezing modes require distinct consideration of the proportion of AgI particles removed by droplets and the other non-activated fractions immersed in the droplets. Droplets collect AgI particles through Brownian diffusion, turbulent diffusion, and phoretic effects (Xue et al., 2013). The majority of the AgI particles remain in the droplets after being removed from the air, while the remainder are converted into AgI-containing hydrometeors via contact and immersion freezing. The activation process of AgI particles acting as CCN is not considered in the model.

A single real sounding is used to drive the model, i.e., the LES contains no horizontal heterogeneity initially, and large-scale synoptic conditions do not evolve during the model time. Although the simulated results using accurate 3D reanalysis data might be more consistent with observations, the idealized LES is more effective for attributing seeding effect variations to turbulence, as it explicitly excludes influences from larger-scale dynamics. The sounding was launched at the Luancheng station, which was located in the research area shown in Fig.  at 00:00 UTC on 20 January 2022. Thermodynamic and wind profiles are shown in Fig. . The atmosphere is saturated at the altitude of 1372–1893 m. Above 1893 m, a strong inversion layer was present. We implemented an initial LWC profile which increases from 0 to 0.2 gkg-1 from the cloud base to 1893 m and decreases rapidly to 0 at the cloud top (grey shaded area in Fig. b). With the presence of the saturated layer and the inversion layer, supercooled liquid water can persist, enabling the simulated cloud to be maintained in the model. The original wind speed is weak (solid lines in Fig. c), which results in a weak turbulent environment. To investigate the effect of different turbulence intensities on the seeding effect, we follow the method in Hill et al. (2014), which shows that enhancement of vertical wind shear in LES can intensify the modeled turbulence. The wind shear between 1519 and 1733 m height is enhanced five times (dashed lines in Fig. c); this causes a decrease in Richardson number in this layer from 16.81 to 0.67, indicating favorable conditions for turbulence development.

Figure a and b show the average turbulence intensities and turbulent kinetic energy (TKE) from the simulations with default and enhanced wind shear components. It is seen that the turbulence intensity and TKE increase in the first hour, and there is no clear difference between the two simulations at this stage. Therefore, a spin-up time of at least 1 h is needed. The differences in the turbulence intensity between the two simulations can be clearly seen after 1 h, with the maximum difference around 03:00 model time (MT). In this study, to ensure the sufficient development of the cloud, a 2 h spin-up period was chosen. The cross-sections of vertical velocity (Fig. c and d) prove that the updrafts and downdrafts are enhanced due to the stronger wind shear, and it is expected that the enhanced turbulence will influence the droplet activation, ice nucleation, and growth of hydrometeors.

In addition, to test whether turbulence plays the same role for varying AgI concentrations, we enhanced the AgI release rate by 10 times (i.e., 10¹⁵ s-1) in both experiments with default and enhanced shear components. The abbreviations for different experiments are listed in Table . To investigate the seeding effect, we will compare the cloud microphysics and precipitation in SEED and NOSEED areas. In this study, SEED and NOSEED areas are defined as the areas affected and unaffected by the seeding plumes at each moment, respectively. Thus, the SEED and NOSEED areas moved with time along the direction of the prevailing wind. Since the observed cloud was mostly liquid before seeding, we turned off natural ice nucleation when validating the model results (control experiment). However, to better understand the differences in ice generation and growth in both SEED and NOSEED areas, we show analyses from experiments with natural ice nucleation turned on after the model validation section. The natural ice nucleation and the seeding of the experiments in Table  start at 02:00 MT. Since the seeding signature is unambiguous (as shown in Fig.  in Sect. ), analysis between SEED and NOSEED areas can inherently provide the necessary contrast between seeding and no seeding simulations, making experiments with no seeding unnecessary in this study.

The modeled reflectivity from the control simulation without natural ice nucleation is shown in Fig. . Since we use idealized LES, it is within expectation that there are inevitably some differences between the modeled and observed results. Therefore, we do not directly compare the model with observation here, but we focus on evaluating the magnitude and temporal variation of radar reflectivity. It is seen from Fig.  that the spatial distribution of enhanced radar reflectivity is controlled by the seeding trajectory and moved northeasterly with time. With higher resolution, the model can produce finer structures of reflectivity distribution than the observation. The magnitude of modeled reflectivity varies between -12 and 12 dBZ, generally consistent with the observation shown in Fig. , but it seems that the horizontal dispersion of the seeding plume was weaker in the modeled than the observed data. Consistent with observations, the seeding signatures can be seen in the model approximately 10 min after seeding (not shown). The seeding signatures kept strengthening for about 1.5 h before turning weaker, which is slightly longer than observed (Fig. ), suggesting that in the actual cloud the cloud glaciation is faster than in the model. This is consistent with the findings of Omanovic et al. (2024), who also reported a slower WBF process in LES than in the observation.

Due to the consumption of the supercooled liquid water, the cloud top height decreased (identified using a threshold of total water content greater than 0.001 gkg-1), and in most of the SEED areas the cloud top temperature increased by about 2 °C after 2 h (Fig. ), which is generally consistent with the observations. In the NOSEED area, the cloud top persisted at about 1.9 km (Fig. e–h). In the observation, cloud thinning was clearly seen almost in the entire seeding plume 15 min after seeding, while in the model the increase in cloud top temperature was seen only in a small fraction of the SEED area within 1 h after the seeding was performed. This difference again suggests that cloud glaciation was faster in the actual cloud than in the model.

Such a difference could be due to multiple reasons. From the perspective of dynamics, large-scale forcing is not considered in the model, and we use a single sounding measurement to drive the simulation, while in the real cloud the wind field and cloud top stratification may change with time. From the perspective of microphysics, the study lacks measurements of CCN concentration, so the droplet size and concentration may have uncertainties. In addition, the ice growth rate in the model may be underestimated as the crystal shape is not sufficiently considered in the model. A recent study by Yang et al. (2024b) showed that the ice growth rate would be higher at -15 °C if assuming a plate-like shape rather than a spherical particle. Regardless of the uncertainties, the model can reasonably reproduce the magnitude and spatial distribution of radar reflectivity, as well as the increase in cloud top temperature, providing confidence for us to investigate the impacts of turbulence on the glaciogenic seeding effect using the model simulations.

In this section, we investigate the changes in cloud microphysics after seeding was performed in the four different numerical experiments (NI, NI_WS, NI_AgI, and NI_AgI_WS), in which natural ice nucleation is allowed. Figure  shows the cross-sections of radar reflectivity, where the background radar reflectivity is the result of natural ice nucleation. It is seen that cloud seeding can induce enhancement in radar reflectivity. At 02:30 MT, the maximum reflectivity at a lower AgI release rate was about 15 dBZ (Fig. a and e), which is slightly larger than in the simulation without natural ice (Fig. ). With a larger AgI seeding rate, the maximum radar reflectivity reached 22 dBZ at 02:30 MT, when there was still sufficient LWC to support the ice growth (Fig. i and m). The stronger turbulence enhanced the horizontal dispersion of the AgI particles, so the horizontal spread of seeding plumes was larger in the NI_WS and NI_AgI_WS experiments than the other two. In NOSEED areas, the radar reflectivity was higher in the experiments with stronger turbulence, suggesting higher ice concentrations and larger ice particles. Later at 03:00 MT, the radar reflectivity in SEED areas became larger in the NI_WS and NI_AgI_WS experiments than the other two. However, the positive seeding effects attenuated more rapidly in the experiments with stronger turbulence (Fig. g and o). At 04:00 MT, the NI_WS and NI_AgI_WS experiments obtained a negative seeding effect, and the liquid layer top (cyan lines) disappeared. However, in the NI and NI_AgI experiments, enhancement of radar reflectivity in SEED areas can last for a longer time (Fig. d and l).

The positive impact of turbulence on ice nucleation is also evident in the time–height diagrams of averaged ice concentration and ice water content (IWC) (Fig. ). It can be seen that with stronger turbulence the cloud obtained a higher ice concentration in SEED (color-shaded) areas soon after seeding (Fig. a, d, g, and j), with the maximum value reaching 7 L-1 in the NI_AgI_WS experiment, which is 1.4 times higher than that in the NI_AgI experiment. However, in the SEED areas, the ice concentrations in the NI_WS and NI_AgI_WS experiments decreased rapidly with time and became similar to that in the NI and NI_AgI experiments after 03:20 MT. At the same time, stronger turbulence also promotes the nucleation of natural ice crystals in NOSEED areas (grey contours). The modeled natural ice concentrations in the experiments with enhanced turbulence have similar temporal variation patterns as the turbulence intensity (Fig. a), and a maximum value exceeding 1 L-1 was found between 02:50 and 03:20 MT. In the experiments with default turbulence intensity, the ice concentration was lower than 0.5 L-1 and changed little with time after 02:20 MT. The higher concentrations of ice crystals tend to consume LWC more rapidly as crystals grow, so it is seen that the decrease in LWC in the SEED area is fastest in the NI_AgI_WS experiment (Fig. l), resulting in a higher IWC (up to 0.053 gkg-1) before 02:50 MT. Although enhanced turbulence produced more supercooled liquid water, it is evident that the cloud glaciation rate was faster after seeding and that the clouds rapidly became thinner. SEED areas completely glaciated when liquid water remained in NOSEED areas. This process involves competition among liquid water generation, turbulent mixing, and cloud glaciation, which is analyzed in detail below.

To better understand the ice growth and LWC consumption at different turbulence intensities, the depositional growth rate of ice crystals and the condensation rate of droplets in SEED areas are plotted in Fig. , which were calculated based on mass change per unit time. The ice growth is dominated by the WBF process rather than riming and aggregation in this case (not shown). Compared to the diffusional growth rate, the riming and aggregation rate are rather minor for such a thin cloud. This is consistent with dual-polarization radar measurements for seeded wintertime stratiform clouds (Jing et al., 2015; Jing and Geerts, 2015). It can be seen that turbulence contributed significantly to the growth of ice crystals (Fig. a and b), and overall, the experiments with enhanced turbulence had a higher deposition rate before 03:00 MT; then it became lower than that in clouds with weaker turbulence due to the insufficient liquid water supply. The experiment with higher AgI concentration produced greater changes in mass because of more ice crystals. Figure c and d show the condensation rate of liquid water. In weak turbulence with relatively low AgI concentration, the condensation rate varied mostly between 0 and -2 × 10⁻⁵ gkg-1s-1, suggesting a relatively weak water consumption (Fig. c). While in a stronger turbulent cloud, the condensation rate was greater between 02:00 and 03:00 MT, indicating that the generation of liquid water was significantly slower than its consumption by ice growth. With more AgI particles seeded, the condensation rate shifted to negative in both simulations, indicating a faster cloud glaciation (Fig. d). At the cloud tops, turbulence can promote the evaporation of liquid water due to the entrainment of dry air and detrainment of vapor, so cloud glaciation was faster near cloud tops, even though entrainment of dry air may also suppress ice growth to some extent (Chu et al., 2018). Significant entrainment is unlikely in this case because of the very strong inversion just above the cloud top (Fig. ).

In short, based on the analyses of Figs. –, it is seen that stronger turbulence enhances the ice nucleation and ice growth in the cloud. Even though stronger updrafts can provide more liquid water, the cloud in the SEED areas glaciated more rapidly as the water consumption is faster than the water supply, and the turbulence is not able to maintain the cloud in a mixed phase.

Due to the faster cloud glaciation induced by stronger turbulence, it is expected that the precipitation may be enhanced only within a short time in the SEED areas after seeding is performed, and the seeding effect may be negative after the cloud is glaciated. Figure  shows the difference in the precipitation rates and the cumulative precipitation between the SEED and NOSEED areas. The precipitation regions for SEED and NOSEED at each moment are determined based on cumulative precipitation characteristics. Note that the surface precipitation rate was rather low because of the strong sublimation caused by the dry sub-cloud layer (Fig. ). However, the differences in precipitation between the SEED and NOSEED areas are clear. It is seen from the figure that the positive seeding effect was significant after the turbulence was enhanced. The enhancement in precipitation rate in NI_WS was even greater than that in the NI_AgI experiment. With both turbulence and AgI enhanced, the maximum enhancement in precipitation rate was about 10 times greater than that in the NI experiment, resulting in greater cumulative precipitation in the SEED areas (Fig. b). However, due to the fast depletion of LWC, the precipitation rates in the NI_WS and NI_AgI_WS experiments were lower in the SEED area than in the NOSEED area after 04:10 MT. In clouds with default turbulence intensity, the seeding effect was positive most of the time if using less AgI, and the cumulative precipitation generally increased with time. With more AgI seeded, the seeding effect became negative, and the cumulative precipitation decreased after 04:40 MT. Due to the slower cloud glaciation process, the seeding-induced enhanced cumulative precipitation in the NI_AgI experiment exceeded that in the NI_WS experiment (Fig. b).

The negative seeding effect induced by turbulence after 04:10 MT resulted in a decrease in precipitation in the downwind SEED areas. This can be intuitively seen from the maps of differences in cumulative precipitation compared to the average natural precipitation (Fig. ). In the NI simulation (Fig. a), the SEED area obtained more precipitation than NOSEED areas all the time. With enhanced AgI concentration (Fig. c), the precipitation in the SEED area obtained more enhancement, especially between 40–50 km along the x distance. However, as the cloud moved further downwind, the magnitude of the seeding effect became similar in the NI_AgI and the NI experiments. With turbulence enhanced (NI_WS), the accumulated precipitation increased rapidly in the SEED area and then decreased as the cloud moved northeastward. The seeding effect became negative in the downwind areas, indicating the presence of the “robbing Peter to pay Paul” effect. Such a transition from positive to negative seeding effects was more substantial in magnitude in the NI_AgI_WS experiment than in the NI_WS experiment (Fig. d), but the area with a negative seeding effect is similar to that with a positive effect in both simulations. On the contrary, under weaker turbulence, even though the cumulative amount of precipitation enhanced by seeding is less than that with stronger turbulence, it generates more sustained precipitation enhancement and is more beneficial if one wants a positive seeding effect in a larger target area. The absolute (relative) increases in water volume in the areas affected by seeding (black boxes in Fig. ) are 976.3 m3 (8.0 %), 1291.2 m3 (5.1 %), 2042.7 m3 (16.7 %), and 2234.1 m3 (8.7 %) in the NI, NI_WS, NI_AgI, and NI_AgI_WS experiments, respectively, which were calculated by comparing the average cumulative precipitation inside and outside the boxes. The results demonstrate that both seeding and enhanced turbulence can induce significant precipitation changes. Although uncertainties associated with the microphysics scheme and the unresolved smaller turbulent eddies may affect the magnitude of precipitation changes, they are unlikely to explain such clear precipitation variations.

Turbulence helps the clouds maintain a mixed-phase state by enhancing mixing and liquid condensation, but it also promotes the dispersal and activation of AgI INPs and the growth of ice crystals. For the case presented in this paper, the latter dominated in the SEED area. This section further quantifies the competition among turbulent mixing, liquid condensation, and cloud glaciation.

Figure  shows the characteristic times of turbulent mixing and cloud glaciation in the SEED area, which were calculated using the formulas in Korolev and Milbrandt (2022): 1τmix=L2ε1/3,2τgl=ρi4πcSi(T)9π2131ρi23LWCNi23×Li2kRvT2+RvTEi(T)D, where L is the spatial scale (in m), and ε is the turbulence energy dissipation rate (in m2s-3), which is proportional to turbulent kinetic energy (e.g., Pokharel et al., 2017). ρi is the density of ice (in kgm-3), c is the ice particle shape factor characterizing capacitance (0 < c≤ 1, c=1 for sphere), Si(T)=Ew(T)Ei(T)-1 is the supersaturation over ice, and Ew(T) and Ei(T) are the saturation vapor pressures with respect to liquid and ice at temperature T, respectively. Ni is the ice particle concentration, Li is the latent heat for ice sublimation, k is the coefficient of air heat conductivity, Rv is the specific gas constant of water vapor, and D is the coefficient of water vapor diffusion in the air.

In all our simulations, as the AgI particles disperse, the scales of the seeding plume broaden, resulting in a larger L. Therefore, the mixing timescale τmix increases with time, indicating more time is required to refill the SEED area with liquid water through mixing only. On the other hand, the cloud glaciation timescale τgl decreases with time, because the LWC is continuously reduced in the SEED areas (Fig. c, f, i, and l).

By comparing the right and left panels in Fig. , it can be seen that the enhanced turbulence accelerated the cloud glaciation, and the characteristic time of mixing became larger due to the enhanced spread of the seeding plume. Applying more AgI aerosols had negligible impacts on τmix but clearly enhanced the cloud glaciation (Fig. c). The intersection of the curves, which indicates the time when the rate of cloud glaciation exceeded the turbulent mixing, was significantly advanced by the stronger turbulence and stronger seeding rate in NI_AgI_WS (Fig. d). The differences among the four panels better explain why turbulence enhanced the cloud glaciation. In the NI experiment, in which the mixed-phase cloud maintains in the SEED area for a relatively long period, τgl is smaller than τmix after 03:00 MT, meaning that the turbulent mixing was not fast enough to fill the seeding plume with liquid water, and new liquid formation must be more important to maintain the cloud in a mixed phase after this time.

Although the cloud glaciation was accelerated by turbulence, data do not indicate that droplet condensation was entirely prevented in the SEED areas, because there were updrafts strong enough to force droplets to grow. The threshold of vertical velocity (w∗) for which both droplets and ice can grow (Korolev and Mazin, 2003; Hill et al., 2014) is 3w∗=es-eieiηNiriVf, where η is a coefficient dependent on temperature and pressure, Vf is the ventilation factor, and Ni and ri are the number concentration and mean radius of ice crystals, respectively. In our simulations, condensation occurred in the area with a vertical velocity greater than w∗, while evaporation took place when the vertical velocity is lower than w∗, regardless of whether turbulence or AgI amount was enhanced (Fig. ). Therefore, liquid water condensation can still occur in SEED areas even after the cloud is glaciated. However, the fractional area with a positive condensation rate substantially decreased with time (Fig. c), especially in the experiments with enhanced turbulence. Liquid water that forms in such a small area can be rapidly consumed by ice growth, and there is no chance for them to fill the glaciated area through turbulent mixing. Therefore, the cloud ultimately glaciated in SEED areas. It should be noted that w∗ in Fig. a and b is calculated under constant temperature and pressure conditions, whereas temperature and pressure within clouds are not constant, and the ice shape can also influence the condensation rate. Consequently, there could be a low occurrence of negative condensation rates even when vertical velocities are slightly above w∗. Nonetheless, this does not alter the overall distribution of condensation rates when vertical velocities are above or below w∗.

In short, for the case presented here, stronger turbulence can enhance the mixing and updrafts in which condensation can occur in SEED areas, but the mixing is too slow and the area where droplets can grow is too small to maintain the mixed-phase clouds, resulting in fast cloud glaciation and the decrease in precipitation in downwind SEED areas.