Meteorology and Atmospheric Physics

, Volume 118, Issue 3, pp 199–214

Investigation of aerosol indirect effects on simulated flash-flood heavy rainfall over Korea

Authors

  • Kyo-Sun Sunny Lim
    • Pacific Northwest National Laboratory
    • Department of Atmospheric Sciences and Global Environment Laboratory, College of SciencesYonsei University
    • Department of Atmospheric Sciences and Global Environment Laboratory, College of SciencesYonsei University
Original Paper

DOI: 10.1007/s00703-012-0216-6

Cite this article as:
Lim, K.S. & Hong, S. Meteorol Atmos Phys (2012) 118: 199. doi:10.1007/s00703-012-0216-6

Abstract

This study investigates aerosol indirect effects on the development of heavy rainfall near Seoul, South Korea, on 12 July 2006, focusing on precipitation amount. The impact of the aerosol concentration on simulated precipitation is evaluated by varying the initial cloud condensation nuclei (CCN) number concentration in the Weather Research and Forecasting (WRF) Double-Moment 6-class (WDM6) microphysics scheme. The simulations are performed under clean, semi-polluted, and polluted conditions. Detailed analysis of the physical processes that are responsible for surface precipitation, including moisture and cloud microphysical budgets shows enhanced ice-phase processes to be the primary driver of increased surface precipitation under the semi-polluted condition. Under the polluted condition, suppressed auto-conversion and the enhanced evaporation of rain cause surface precipitation to decrease. To investigate the role of environmental conditions on precipitation response under different aerosol number concentrations, a set of sensitivity experiments are conducted with a 5 % decrease in relative humidity at the initial time, relative to the base simulations. Results show ice-phase processes having small sensitivity to CCN number concentration, compared with the base simulations. Surface precipitation responds differently to CCN number concentration under the lower humidity initial condition, being greatest under the clean condition, followed by the semi-polluted and polluted conditions.

1 Introduction

The aerosol effect on precipitation processes, also known as the cloud lifetime effect (Albrecht 1989), is complex especially for mixed-phase convective clouds. Khain (2009) reviewed previous modeling studies (Lynn et al. 2007; Tao et al. 2007; Lee et al. 2008; Xue et al. 2008; Fan et al. 2007) and observational studies (Givati and Rosenfeld 2004; Yum and Hudson 2002) concerning the cloud lifetime effect. He concluded that much of the disparity among the reported results concerning precipitation response to aerosols can be attributed to cloud type and atmospheric conditions, which varied among the evaluated studies. His principal conclusions are as follows. (1) Aerosols seem to decrease precipitation from isolated deep clouds developing in very dry, unstable atmospheres. (2) Condensate generation increases with increasing humidity. As a result, under high humidity typical of tropical convection, aerosols increase precipitation from deep convective clouds. (3) As aerosol loading increases, evaporation of the precipitation mass increases and leads to the acceleration of downdrafts, which foster the formation of secondary clouds. In short, the aerosol-induced dynamic effect increases condensate generation and, consequently, increases precipitation in the zone of convection.

More recently, several modeling studies have utilized bulk microphysics schemes to research aerosol–cloud precipitation interactions (Seifert and Beheng 2006b; Seifert et al. 2006; Lee et al. 2008; Khain and Lynn 2009). Khain and Lynn (2009) showed that the Thompson bulk scheme (Thompson et al. 2008), which only predicts number concentrations of cloud ice, is much less-sensitive to aerosols than the bin microphysics scheme (Khain et al. 2004; Khain and Lynn 2009). While the bulk scheme predicts a decrease in precipitation in clean air under both low and high humidity, the bin scheme predicts a decrease in precipitation in polluted air under low humidity and an increase in precipitation under high humidity. One weak point in the study of Khain and Lynn (2009) is the treatment of the aerosol effects in the bulk scheme. They modify the gamma shape parameter of the size distribution for cloud water to incorporate aerosol indirect effects. More specifically, for a clean maritime (dirty continental) cloud condensation nuclei (CCN) distribution, a very high (small) value of the gamma shape parameter is introduced. Meanwhile, bulk microphysics schemes have achieved greater flexibility in the description of the particle size distribution by predicting additional moments such as number concentration of cloud water or CCN (Seifert and Beheng 2006b; Phillips et al. 2007; Lim and Hong 2010). Seifert et al. (2006) addressed that both the bin (Khain et al. 2004) and bulk (Seifert and Beheng 2006a, b) modeling approaches show similar results regarding the vertical structure of clouds, updraft velocities and surface precipitation as well as the sensitivity of these parameters to changes in CCN characteristics. Lee et al. (2008) also showed that the bulk microphysics scheme (Phillips et al. 2007) is capable of simulating different precipitation responses in an imposed large-scale environment supporting cloud development. Previous studies showed the usefulness of a bulk microphysics scheme in modeling studies of aerosol indirect effects (Seifert et al. 2006, 2011; Lee et al. 2008; Noppel et al. 2010), but some of studies have been limited to simple configurations tested on idealized testbeds (Seifert et al. 2006; Lee et al. 2008). The two-dimensional cloud-system model affords substantial computational advantages over the three-dimensional (3D) model, but as Phillips and Donner (2007) note, some aspects of cloud system dynamics and microphysics differ between the two- and three-dimensional models.

The Weather Research and Forecasting (WRF) Double-Moment 6-Class (WDM6) microphysics scheme is a double‐moment bulk‐cloud microphysics scheme based on the WRF Single‐Moment Six‐Class (WSM6) microphysics scheme (Lim and Hong 2010). In addition to predicting the mixing ratios of the six water species (water vapor, cloud droplets, cloud ice, snow, rain, and graupel), which is afforded by WSM6, the WDM6 scheme also predicts the number concentrations of cloud droplets and raindrops together with the prognostic variable of CCN number concentration. Thus, in WDM6, the impact of aerosol concentration on simulated precipitation can be evaluated by varying the initial CCN number concentration. Evaluation of the WDM6 microphysics scheme on precipitating moist convection in 3D platforms has confirmed that the skill statistics of precipitation forecasts over Korea from June to August 2008 are generally better with the WDM6 scheme than with the WSM6 scheme (Hong et al. 2010).

The purpose of this study is to investigate aerosol indirect effects on the development of heavy rainfall near Seoul, South Korea, on 12 July 2006, using a bulk-type WDM6 microphysics scheme. This study will allow us to examine the ability of WDM6 scheme to simulate aerosol effects not only on single cloud systems (Lim and Hong 2010; Lim et al. 2011) but also on multi mesoscale convective systems. We also evaluate environmental effects on precipitation response under different aerosol number concentrations. This paper is organized as follows. Section 2 describes the selected 3D moist convection case and the numerical experimental setup. The results and summary are presented in Sects. 3 and 4, respectively.

2 Case description and numerical experimental setup

The 3D convective case associated with flash-flooding and heavy rainfall near Seoul, South Korea, on 12 July 2006 was selected to investigate aerosol effects on the development of a convective storm. A significant amount of precipitation was recorded in Korea on 12 July 2006, with a local maximum value of 239.0 mm at Goyang, Gyeonggi province (Fig. 1a). The rainfall period can be divided into three sub-periods; the first rainfall lasted from 1700 UTC 11 to 0400 UTC 12 July, the second, 9-h rainfall period continued from 0400 UTC 12 to 1300 UTC 12 July, and the third rainfall period occurred between 1300 UTC and 2300 UTC on 12 July (Fig. 1b). The maximum rainfall intensity was 77.5 mm h−1, which occurred over the Goyang area at 2300 UTC 11 July. The first rainfall period was the most intense of the three. In addition to being concentrated in time, the difference in intensity between the maximum point value and the area-average exceeds about 40 mm h−1, indicating that the rainfall during the first period was highly localized at Goyang. However, the difference in intensity between the maximum point and area-average values is relatively smaller during the second and third rainfall periods.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig1_HTML.gif
Fig. 1

a The 35-h accumulated rainfall (mm) valid at 11 UTC 12 July 2006 from the Korea Meteorological Administration (KMA) Automatic Weather Station (AWS) observation. b Same property as a but for the simulated rainfall from the CNTR_M experiment. The open circle and triangle in b indicate Goyang and the maximum precipitation region for the model simulation, respectively. c The time series of observed hourly rainfall at Goyang (thick grey line) and near the central districts (black line), marked as a rectangular box in a. d Same as c but for the simulated hourly rainfall. In d, thick grey line indicates the simulated rainfall at the maximum precipitation region marked as an open triangle in b. The three solid boxes in c and d indicate the first, second, and third rainfall periods, respectively

Figure 2 shows the rain rate that is calculated from the composite constant-altitude plan position indicators (CAPPIs) at 1.5-km altitude of radar reflectivity. The CAPPIs data are a standard product of the Korea Meteorological Administration (KMA) that covers South Korea and the adjacent oceans. Typhoon “Bilis” moving toward Taiwan and the southeastern part of China supplies abundant moisture to the Korean peninsula during the heavy rainfall period analyzed herein. During this period, the stationary front moves from the south coast of the Korean peninsula to north of it. A more detailed synoptic overview for the selected case can be found in Hong and Lee (2009). A convective line formed first over the Jeollabuk province before heavy precipitation occurs over the Goyang area (Fig. 2a). At 2100 UTC 11 July (first rainfall period), a strong echo covered the Gyeonggi province including Goyang, but not the Yellow Sea (Fig. 2b). There was a mountain-induced mesoscale high on the Gaema Heights, and a ridge developed south to north, along the Taebaek mountain range in the first rainfall period. Also during the first period, a low-pressure trough existed over the Yellow Sea and moved east toward land (not shown). Numerous flashes of lightning and a corresponding precipitation system, which developed from isolated thunderstorms to cloud cluster types, occurred near Goyang. At 0600 UTC 12 July (second rainfall period), an echo extended southwestward from the border between North and South Korea to the Yellow Sea (Fig. 2c). In the third rainfall period, scattered convective activity extended from the Yellow Sea to the east (Fig. 2d).
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig2_HTML.gif
Fig. 2

Radar reflectivity at 15 UTC 11 (a), 21 UTC 11 (b), 06 UTC 12 (c), and 18 UTC 12 July 2006 (d), obtained from the observed 1.5 km CAPPI radar image of rain rate (mm h−1)

The numerical experiments were executed using the WRF version 3.2 (Skamarock et al. 2008), which is a fully compressible non-hydrostatic model with an Arakawa-C grid system. The model physics packages include the WDM6 microphysics scheme (Lim and Hong 2010), the simplified Arakawa–Schubert (SAS) deep cumulus scheme (Arakawa and Schubert 1974; Han and Pan 2011), the Yonsei University planetary boundary layer (YSUPBL) (Hong et al. 2006; Hong 2010), a simple cloud-interactive radiation scheme (Dudhia 1989), and rapid radiative transfer model (RRTM) longwave radiation (Mlawer et al. 1997) schemes. In contrast to the previous version, WDM6 microphysics scheme uses a semi-Langrangian advection scheme for falling hydrometeors (Juang and Hong 2010) replacing the Eulerian advection method with sub-time steps. Initial and boundary conditions were derived from National Centers for Environmental Prediction (NCEP) final analysis data (FNL) on 1° × 1° global grids every 6-h (available at http://dss.ucar.edu/datasets/ds083.2/data/), without specific assimilation of observational data. All experiments employ a prognostic sea surface skin temperature scheme (Zeng and Beljaars 2005), revised roughness length formulation (Donelan et al. 2004), and daily SST, forced by the observed SST. The SST data are based on the optimally interpolated sea surface temperature (OISST) on a 1° × 1° grid. The weekly OISST was linearly interpolated in time to derive daily values during the integration period. The experiments were performed for 48 h starting at 0000 UTC 11 July 2006.

All experiments consist of one-way interactive triple-nested domains with a Lambert conformal map projection (Fig. 3). A 3-km domain covering the Korean peninsula (Domain 3, 205 × 205) is nested in a 9-km domain (Domain 2, 118 × 118), which in turn is nested in a 27-km domain (Domain 1, 58 × 58). All domains have 28 vertical layers with a terrain following sigma coordinate, and the model top is 50 hPa. Note that the cumulus parameterization scheme was not used in the 3-km grid model where convective rainfall generation is assumed to be explicitly resolved. The three simulations were initiated with different initial CCN number concentrations based on the observed values on the west coast of the Korean peninsula (Yum et al. 2005), specifically 100, 2,000, and 7,000 cm−3, respectively. We applied the open boundary condition for CCN number concentration over the inflow region of air. Background aerosol interacts with clouds through the activation and evaporation process from cloud drops. For the details of the source and sink terms of CCN particles, the reader can refer Lim and Hong (2010). Any other process related to CCN number concentrations is ignored in the current WDM6 scheme. Meanwhile, ice nuclei number is treated as a diagnostic variable depending on temperature. The impact of ice nuclei on clouds and precipitation remains in question (Demott et al. 2010). Efforts on adding aerosol properties into the ice nucleation process is ongoing in the conventional bulk microphysics development area, which will lead to better understanding of aerosol effects on the mixed-phase and ice clouds. The authors will keep this development issue essential for the improvement of WDM6 microphysics scheme as a future study. The CNTR_M (maritime type of the CCN number concentration; clean condition) experiment employs the initial number concentration of CCN of 100 cm−3. The CNTR_C (continental type of the CCN number concentration; semi-polluted condition) and the CNTR_EC (extreme continental type of the CCN number concentration; polluted condition) uses 2,000 and 7,000 cm−3, respectively. We will call these three experiments as the CNTR experiments hereafter. We also design several sensitivity experiments that employ 5 % reduced relative humidity through the whole vertical layer at the initial time to investigate precipitation response with respect to varying aerosol concentrations under different environments. The DRY_M, DRY_C, and DRY_EC experiments were executed with three different CCN number concentrations being 100, 2,000, and 7,000 cm−3, which are identical to those of the CNTR experiments. These three experiments will be named as the DRY experiments. All experiments, being analyzed in the 3-km inner domain, are summarized in Table 1.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig3_HTML.gif
Fig. 3

Model domain with terrain heights contoured every 200 m. Terrain heights >600 m are shaded. A 3-km domain covering the Korean peninsula (Domain 3, 205 × 205) is nested in a 9-km domain (Domain 2, 118 × 118), which in turn is nested in a 27-km domain (Domain 1, 58 × 58)

Table 1

Description of experiments conducted in this study

Experiments

Description

Initial CCN number concentration (cm−3)

CNTR_M

The WDM6 microphysics scheme that includes a prognostic variable of CCN number concentration

100

CNTR_C

Same as the CNTR_M

2,000

CNTR_EC

Same as the CNTR_M

7,000

DRY_M

Same as the CNTR_M run but with the 5 % reduced relative humidity at the initial time

100

DRY_C

Same as the DRY_M

2,000

DRY_EC

Same as the DRY_M

7,000

3 Results

3.1 Aerosol effects on the development of heavy rainfall over Korea

The CNTR_M experiment well captures the localized precipitation core, even though the modeled rainfall amount near Goyang is about 40 mm less than the observed total, and the maximum precipitation core is located south of the observed core (cf. Fig. 1b, a). The observed precipitation maximum near 36°N at the western coast is weakened in the model result, relative to the observed value. Figure 1d shows the time series of simulated rainfall near the central part of the Korean peninsula including Goyang. The model well captures the first rainfall peak over Goyang shown in Fig. 1b; however, the model shows surface precipitation initiating earlier than it was observed. The model accurately reproduces the amount and time of precipitation at the second peak, which was observed at 0600 UTC 12 July.

Radar reflectivity simulated by the CNTR experiments at 1.5-km altitude, which corresponds to the rain rate calculated from the CAPPIs of radar reflectivity (Fig. 2), is shown in Fig. 4. All three CNTR experiments trigger convection earlier than observed and show spurious convection across the Jeollabuk province at 1300 UTC 11 July. This inaccuracy causes the model to miss the intense observed precipitation maximum near 36°N at the western coast. At 1900 UTC 11 July (first rainfall period), the model simulated strong echoes over the Goyang area about 2 h earlier than the observation (see Fig. 2b). We therefore compare the corresponding echoes for the first rainfall period. The model result shows a strong echo that covers the Gyeonggi province including Goyang, but not the Yellow Sea, which is similar to the observation (cf. Figs. 2b, 4b). The model also simulates considerable convection echoes over the eastern part of the Gyeonggi province, implying that the model exaggerates the eastward movement of the convective cells that developed along the coast line to the west of Goyang. The exaggerated movement produces considerable rainfall in the eastern part of the Korean peninsula (cf. Fig. 1a, b). The features of the echo extending southwestward from the border between North and South Korea to the Yellow Sea are well captured in the simulation during the second rainfall period. However, the model develops the convection line southward toward what was observed and shows some deficiencies such as an excessive convection line over the Yellow Sea and secondary convective cells over the eastern part of the Korean peninsula (cf. Figs. 2c, 4c).
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig4_HTML.gif
Fig. 4

Simulated radar reflectivity (dBZ) at 1.5 km for the CNTR_M (ac), CNTR_C (df), and CNTR_EC (gi) experiments. a, d, g Reflectivity at 1300 UTC 11; b, e, h reflectivity at 1900 UTC 11; c, f, i reflectivity at 0500 UTC 12 July 2006

Figure 4 shows the effects of aerosol on convection developments. In general, the three CNTR experiments, which have different initial CCN number concentrations, show similar features of convection cell evolution. However, the CNTR_C and CNTR_EC runs develop stronger convective cells over the central eastern part of the Korean peninsula at 1900 UTC 11 July. At 0500 UTC 12 July, the second rainfall period, the CNTR_C and CNTR_EC runs simulate intense convective lines over the northwestern part of strong convective echoes extending southwestward from the northeastern part of South Korea to the Yellow Sea, where we marked with the dashed black circles in Fig. 4c, f, i. The CNTR_C run shows a broader area of intense convection than the CNTR_EC run. As shown above, the intensive precipitation events occurred primarily during the first and second periods and the model well captured the observed characteristics of the convective cells in terms of their intensity and location during those periods. Thus, the analysis of the aerosol effects on the development of the convective storm will focus on the period of the first and second peaks ending at 1100 UTC 12 July 2006 in the model.

Figure 5 shows accumulated surface precipitation from the CNTR_M, CNTR_C, and CNTR_EC experiments and the differences in the time sequences of the domain-averaged surface rainfall rates. The simulated sea-level pressure and wind vector at 850 hPa are also shown in Fig. 5. Excessive precipitation, relative to that in the CNTR_M experiment, is reproduced over ocean in the CNTR_C experiment. Relative to the CNTR_C experiment, the CNTR_EC experiment narrows the region of intense precipitation. The model accurately reproduces observed mesoscale feature such as a ridge developed south to north along the Taebaek mountains range and the low-pressure trough on the Yellow Sea approaching from the west (compare Fig. 6a in Hong and Lee 2009), which combine to effect moisture convergence over the Gyeonggi province. Even though some changes in wind fields are shown over the inflow and outflow regions of the intense precipitation region, with respect to the CCN number concentrations, the overall mesoscale features of sea level and wind fields do not change.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig5_HTML.gif
Fig. 5

Accumulated surface precipitation from a CNTR_M, b CNTR_C, and c CNTR_EC experiments. The simulated sea-level pressure (solid lines, hPa) and wind vectors at the level of 850 hPa at 0100 UTC 12 July from the CNTR_M experiment are also shown in a. In addition, the differences in simulated sea-level pressure and wind vectors from the CNTR_M experiment are represented in bc. d The differences in time series of precipitation rate over the analysis region (black rectangular region) in ac. Solid line is for the CNTR_C minus CNTR_M experiment and dotted line is for the CNTR_EC minus the CNTR_M experiment

During the initial stage, rain suppression in the CNTR_C and CNTR_EC experiments is evident when we compare the three experiments (Fig. 5d). During the mature stage of the simulations, from 0100 UTC to 0600 UTC 12 July (see black solid line in Fig. 1d), the difference between the surface rain amounts of the low and high CCN cases increases relative to that of the initial stage. The CNTR_C experiment produces the largest amount of surface precipitation at 0600 UTC 12, followed by the CNTR_M and CNTR_EC experiments. The domain-averaged surface rainfall amounts over the analysis region and the entire 3-km domain region are shown in Table 2. The CNTR_C experiment shows the highest surface rainfall amount among the three experiments and the CNTR_EC experiment gives the lowest rate over the entire 3-km domain region. Khain and Lynn (2009) showed the similar response of precipitation with respect to the aerosol changes for the supercell storm case using a bin microphysics scheme. However, we note that domain-averaged precipitation amounts do not differ considerably among the experiments. Even though the differences in averaged precipitation amounts are not significant among the experiments, here we try to investigate what makes the small differences in the surface precipitation amounts according to the CCN number concentrations through detailed analysis of the cloud microphysical budget, mainly by focusing on the microphysics processes.
Table 2

Domain-averaged surface rainfall amount (mm/35 h) over the entire 3-km domain and the heavy precipitation region for all of the experiments conducted in this study

 

CNTR_M

CNTR_C

CNTR_EC

DRY_M

DRY_C

DRY_EC

Over the entire 3-km domain

26.99

27.30

26.93

21.40

20.88

20.78

Over the heavy precipitation region

48.42

48.52

47.90

45.11

43.91

42.51

The average cloud microphysical budgets for the analysis region, during the initial and mature stages, are shown in Fig. 6. The warm rain processes play a more important role than the ice or mixed phase processes in the initial stage (Fig. 6a). Auto-conversion (Praut) is a major process affecting the raindrop budget that shows its largest value in the CNTR_M experiment. Cloud droplets are larger in size in the CNTR_M experiment (not shown), thus cloud water is more easily converted through raindrop formation that is faster than that in the high-CCN runs. The auto-conversion process is a very efficient way to produce surface precipitation in the early developing stage of a cloud system (See Fig. 5d). Less evaporation of cloud water and more evaporation of rain in the CNTR_M experiment than other experiments are another distinct features shown in the cloud microphysical budgets analysis during the initial stage.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig6_HTML.gif
Fig. 6

Cloud microphysical budgets averaged over the analysis region during the initial stage (left panel) from 0700 to 1200 UTC 11 July 2006 and mature stage (right panel) from 0100 to 0600 UTC 12 July 2006. Units for microphysics budget are g kg−1. a The CNTR_M experiment. b, c The CNTR_C and CNTR_EC experiments, respectively

The stronger updrafts during the mature stage than during the initial stage cause clouds to extend vertically to higher levels, thus mixed phase processes start to produce ice phases including ice, snow, and graupel. Even though the amount of ice phases rapidly increases with vertical extension, warm rain processes still play an important role in producing surface rainfall by the auto-conversion process during the mature stage. Accretion of cloud water by rain (Pracw) increases in the high CCN case because this accretion process depends greatly on the cloud droplet number concentrations in the WDM6 scheme (Lim and Hong 2010). While warm rain processes drive rain production in the low CCN case, the efficient ice and snow production in the CNTR_C experiment increases surface precipitation through the melting of snow and graupel (Psmlt and Pgmlt). Meanwhile, surface rainfall is lower in the CNTR_EC experiment than in the CNTR_C experiment. As with the CNTR_C experiment, abundant ice phases including ice and snow are generated in the CNTR_EC experiment by enhanced mixed phase and ice-phase processes, which can be invoked by abundant super-cooled cloud water droplets. However, in the CNTR_EC experiment, less surface rain is induced than in the CNTR_C experiment due to extremely suppressed auto-conversion and enhanced evaporation of rain due to a larger number of small raindrops (See Figs. 6c, 7). The list of symbols used to describe the cloud microphysical budget analysis in Fig. 6 is provided in Table 3.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig7_HTML.gif
Fig. 7

Vertical profiles of time-domain-averaged mixing ratios of hydrometeors under three different CCN number concentrations with the CNTR_M (blue), CNTR_C (green), and CNTR_EC (red) experiments for a cloud water, b rain, c ice, d snow, and e graupel. All outputs were averaged during the mature stage

Table 3

List of symbols used in cloud microphysical budgets analysis

Symbol

Description

SI units

Pcond

Production rate for condensation/evaporation of cloud water

kg kg−1 s−1

Pgacr

Production rate for accretion of rain by graupel

kg kg−1 s−1

Paacw

Production rate for accretion of cloud water by graupel (or snow)

kg kg−1 s−1

Pgdep

Production rate for deposition/sublimation rate of graupel

kg kg−1 s−1

Pgmlt

Production rate for melting of graupel to form rain

kg kg−1 s−1

Piacr

Production rate for accretion of rain by cloud ice

kg kg−1 s−1

Pidep

Production rate for deposition/sublimation rate of ice

kg kg−1 s−1

Pracs

Production rate for accretion of snow by rain

kg kg−1 s−1

Pracw

Production rate for accretion of cloud water by rain

kg kg−1 s−1

Praut

Production rate for auto-conversion of cloud water to form rain

kg kg−1 s−1

Prevp

Production rate for evaporation/condensation rate

kg kg−1 s−1

Psaci

Production rate for accretion of cloud ice by snow

kg kg−1 s−1

Psacr

Production rate for accretion of rain by snow

kg kg−1 s−1

Psaut

Production rate for auto-conversion of cloud ice to form snow

kg kg−1 s−1

Psdep

Production rate for deposition/sublimation rate of snow

kg kg−1 s−1

Psmlt

Production rate for melting of snow to form cloud water

kg kg−1 s−1

Figure 7 shows the vertical profiles of time-domain-averaged mixing ratios of hydrometeors for the CNTR_M, CNTR_C, and CNTR_EC experiments. The three experiments show significantly different results in terms of the distribution of cloud water and rain. The CNTR_EC experiment simulates the highest mixing ratio of cloud water, followed by the CNTR_C and CNTR_M experiments. In contrast, the highest mixing ratio of rain is shown in the CNTR_M experiment due to the effective auto-conversion process from cloud water to rain. Meanwhile, the CNTR_C and CNTR_EC experiments show similar distributions of ice phases and simulate more ice and snow amounts, compared to the CNTR_M experiment. This is because of the enhanced processes of deposition from water vapor and accretion of cloud water by graupel or snow (Paacw). Paacw is enhanced under the high CCN condition, whereas accretion processes of rain by graupel and ice (Pgacr and Piacr) are suppressed. These differences in mixed-phases cloud processes result in similar graupel amounts among the three experiments, as seen in Fig. 7e.

Figure 8 shows the probability distribution functions (PDFs) of simulated rainfall intensity from the CNTR experiments, which are normalized by total precipitation events. The left panel shows the results over land with different initial CCN number concentrations and the right panel shows the results over the entire 3-km domain. Observed precipitation is only available over land from the KMA Automatic Weather Station (AWS). By setting the precipitation event threshold value to 0.001 mm h−1, we see that fewer precipitation events occur in the CNTR_C and CNTR_EC experiments than in the CNTR_M experiment. This is due to the CNTR_C and CNTR_EC experiments having auto-conversion processes that are inefficient under abundant CCN number concentrations. As compared with the CNTR_C and CNTR_EC experiments, the CNTR_M experiment shows a greater number of light and moderate precipitation activities. Meanwhile, the PDFs from the CNTR_C and CNTR_EC experiments show greater agreement with observed values than do those of the CNTR_M experiment. The PDFs over the entire region also show that the CNTR_M experiment typically increases light and moderate precipitation activities relative to the other experiments. Meanwhile, heavy precipitation activity is shown to increase slightly in the CNTR_EC experiment, which can be explained by the enhanced mixed phase and ice-phase processes in the vicinity of the heavy precipitation cores located over the ocean area (see Fig. 5a–c).
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig8_HTML.gif
Fig. 8

Probability distribution functions (PDFs) of the observed and simulated surface rainfall intensities from the CNTR_M, CNTR_C, and CNTR_EC experiments. a and b are obtained over land and over the entire 3-km domain, respectively

3.2 Sensitivity experiments under dry environment

Figure 9 shows the spatial distribution of accumulated surface precipitation and the differences among the time sequences of the domain-averaged surface rainfall rate for different initial CCN number concentrations under a drier environment. The location of the maximum precipitation core in the DRY_M experiment moves northeastwards with the enhanced southwesterly winds, compared with the CNTR_M experiment. Concurrently, the precipitation intensity decreases irrespective of CCN number concentrations in the DRY experiments, as compared with the CNTR experiments forced by FNL data. Also unaffected by the CCN number concentrations, and consistent with the CNTR experiments, the DRY experiments maintain the south to north ridge along the Taebaek mountain range and the low-pressure trough over the Yellow Sea (not shown). From the DRY experiments, we can notice that a 5 % reduction in relative humidity can considerably affect the amount and spatial distribution of precipitation.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig9_HTML.gif
Fig. 9

Accumulated surface precipitation from the a DRY_M, b DRY_C, and c DRY_EC experiments. The differences in simulated sea-level pressure (solid lines, hPa) and wind vectors at the level of 850 hPa at 0100 UTC 12 July are also shown in ac. a The difference between the DRY_M and CNTR_M experiments (DRY_M − CNTR_M). b, c The differences from the DRY_M experiment. d The differences in time series of precipitation rate over the analysis region (thick rectangle) in ac. Solid line is for the DRY_C − DRY_M experiment and the dotted line is for the DRY_EC − DRY_M experiment

As in the CNTR experiments, CCN number concentration affects the amount of surface precipitation even though it does not affect the mesoscale features of sea level and wind fields (see Fig. 9b, c). During the initial stage, the experiments do not significantly differ by CCN number concentration. However, during the mature stage from 0100 UTC to 0600 UTC 12 July, the DRY_M experiment produces the largest amount of surface precipitation, followed by the DRY_C and DRY_EC experiments (Fig. 9d). Note that under low humidity, maximum precipitation does not occur in the semi-polluted condition. More rain accumulated in the DRY_M experiment than in the DRY_C and DRY_EC experiments. Thus, we can see that precipitation responds differently to aerosols under different environmental conditions. This finding is in agreement with those of previous studies, which have mentioned that environmental conditions can affect precipitation response to aerosol concentrations for supercell storms (Khain and Lynn 2009), single convective clouds (Khain et al. 2005, 2008) and squall lines (Tao et al. 2007). Table 2 shows the domain-averaged surface rainfall amount for the DRY experiments. Here, we can notice that precipitation is more strongly affected by a global change in relative humidity than by global changes in aerosol concentrations.

The cloud microphysical budget for the DRY experiments is shown in Fig. 10. Auto-conversion (Praut) is a major process affecting the raindrop budget, as in the CNTR experiments. Even though the difference in rain amounts is not significant among the DRY experiments during the initial stage, the DRY_M run produces the highest rain amount among the experiments, because it most effectively converts cloud water into rain by Praut. An increased process of Pracw in high CCN environment mainly contributes to the increase in rain amounts in the DRY_C and DRY_EC runs, which consequently results in an insignificant difference in rain rate among the experiments (See Fig. 9d). Meanwhile, because all of the dry runs have limited water vapor available during the initial stage, relative to the CNTR experiments, they generate comparatively fewer cloud droplets and result in less surface precipitation (compare Figs. 5 and 9).
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig10_HTML.gif
Fig. 10

Same as Fig. 6 but for the DRY experiments. a The DRY_M experiment. b, c The DRY_C and DRY_EC experiments, respectively

As in the CNTR experiments, mixed-phase or ice-phase processes start to produce ice phases during the mature stage. Since the DRY runs sublimate more ice than the CNTR runs, they yield less snow from ice, such as through accretion of ice by snow (Psaci) and auto-conversion of ice to snow (Psaut). Even though differences among the three DRY experiments are relatively small, compared with the base experiments, the DRY experiments produce more snow in the DRY_C and DRY_EC runs, which is attributed to more effective processes of Paacw, Psaci, and Psaut. The dry environment experiments show a greater range of liquid yields and a lesser range of ice yields than the CNTR experiments (Fig. 11), indicating that CCN number concentrations influence warm rain processes to drive changes in the amount of surface precipitation.
https://static-content.springer.com/image/art%3A10.1007%2Fs00703-012-0216-6/MediaObjects/703_2012_216_Fig11_HTML.gif
Fig. 11

Same as Fig. 7 but for the DRY_ experiments

4 Summary and concluding remarks

Aerosol effects on the development of heavy rainfall near Seoul, South Korea, on 12 July 2006 were investigated, focusing on precipitation amount. The impact of the initial aerosol concentration on the development of a convective storm was evaluated by varying the number and mass of aerosols, reflected in the CCN number concentration, within the WDM6 microphysics scheme. The observed 3D heavy rainfall structure, documented in previous observation studies, was well captured by the WDM6 microphysics scheme. Under the polluted condition, our model shows reduced surface precipitation during the initial stage due to the inefficient conversion of cloud water to rain. The experiment showed the greatest amount of surface precipitation during the mature stage of the simulations, under the semi-polluted condition, due to enhanced ice-phase processes. However, surface precipitation was lower during the same stage, under the polluted condition, due to suppressed auto-conversion and enhanced rain evaporation processes. Thus, the greatest amount of surface precipitation was seen in the experiment under semi-polluted air, followed by those under clean and polluted air. It is worth noting that the change in averaged precipitation amount under different aerosol concentrations is not significant under the testbed of 3D mesoscale convective system configured in this study. Aerosol concentration does not change the mesoscale environment, but it does affect cloud microphysical budgets. Recent studies also addressed that there are no significant differences in cumulative precipitation between simulations, which varies aerosol number concentrations in multi-cloud systems. This is known as a buffering mechanism among microphysical processes and various cloud types (Lee 2011; Stevens and Feingold 2009).

It has been shown that aerosol conditions over the Korean peninsula may be considerably different than that in other areas. Yum et al. (2005) observed CCN number concentrations using the the Desert Research Institute (DRI) instantaneous CCN spectrometer at the Korea Global Atmosphere Watch (GAW) Observatory (KGAWO) (30°32′N, 126°19′E) on the west coast of the Korean peninsula from May 1–22, 2004. They classified data into maritime or continental types according to air mass back trajectories. The average maritime and continental CCN concentrations at 1 % supersaturation were 2,406 and 5,292 cm−3, respectively. They reported that maritime CCN number concentration over the Korean peninsula exceeds even the continental values of the other countries observed by special observation projects. Thus, we feel it important to stress that our simulations under the semi-polluted and polluted air conditions show good agreement with the observed PDFs of rainfall intensity.

To evaluate effects of the environment on precipitation response under different aerosol number concentrations, we executed sensitivity experiments in which simulations were initiated with the 5 % reduced humidity. The decrease in humidity was shown to effect dramatic changes in the amount and distribution of precipitation. Notably, the experiment under clean air reproduced the greatest amount of surface precipitation, followed by the semi-polluted and polluted air experiments. The clean air experiment showed the most efficient rainfall production by auto-conversion from cloud water to rain. The amounts of ice were comparable across all experiments. Warm rain processes were responsible for the difference in surface precipitation amounts between the dry environment experiments. Our study indicates that humidity can change the precipitation response under the varying CCN number concentrations for the 3D heavy rainfall case and have greater influence on the characteristics of simulated convective system than aerosol number concentrations.

Meanwhile, Tao et al. (2012) reviewed past efforts and summarized current understanding of aerosol effects on convective precipitation processes and proposed future research directions to achieve a better understanding of aerosol–cloud-precipitation interactions. They pointed out that idealized CCN concentrations and uniform horizontal distribution are the weakness of current cloud-scale modeling in dealing with aerosol information and stressed that inclusion of aerosol chemical composition is the way to improve understating of aerosol–precipitation interaction. In general, a bulk microphysics approach suffers from an inaccuracy in representation of microphysical processes because of its simple treatment of hydrometeor size distribution, compared with a bin approach. Further improvement and evaluation of the WDM6 bulk microphysics scheme through a comparison of bin microphysics scheme and observed microphysics properties, which can be availed from a special observation, should be a next step for an accurate assessment of aerosol indirect effect using bulk microphysics schemes.

Acknowledgments

This work was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (2012-0000158). The authors would like to express their gratitude to Jong-Jin Baik, Seong Soo Yum, Jhoon Kim, and Jimy Dudhia for their comments in the review of a thesis from which this work has been derived.

Copyright information

© Springer-Verlag Wien 2012