In-cloud and below-cloud numerical simulation of scavenging processes at Serra Do Mar region, SE Brazil

In-cloud and below-cloud numerical simulation of scavenging processes at Serra Do Mar region, SE Brazil

Atmospheric Environment 36 (2002) 5245–5255 In-cloud and below-cloud numerical simulation of scavenging processes at Serra Do Mar region, SE Brazil F...

634KB Sizes 0 Downloads 6 Views

Atmospheric Environment 36 (2002) 5245–5255

In-cloud and below-cloud numerical simulation of scavenging processes at Serra Do Mar region, SE Brazil F.L.T. Gonc,alves*, A.M. Ramos, S. Freitas, M.A. Silva Dias, O. Massambani ! Department of Atmospheric Sciences, IAG-USP, Rua do Matao, * 1226-Cidade Universitaria-USP, 5508-900 Sao * Paulo, Brazil Received 11 February 2002; received in revised form 11 June 2002; accepted 26 June 2002

Abstract Atmospheric scavenging processes have been investigated, taking into consideration a numerical simulation through the model Regional Atmospheric Modeling System (RAMS), the below-cloud scavenging model, local atmospheric conditions and local emissions in the Serra do Mar region in southeastern Brazil. The RAMS modeling was coupled with a one-dimensional (1-D) below-cloud scavenging model in order to simulate the in-cloud and below-cloud scavenging processes. RAMS modeling was also used in order to simulate the cloud structures. The aim of the modeling  + was to predict the average concentration of three chemical species found in rainwater: SO= 4 , NO3 and NH4 , scavenged from the atmosphere. The concentrations of gases and particles in the samplings, as well as the meteorological parameters obtained during the March 1993 Campaign, were the input data in both models. Another objective was to compare the modeled and the observed rainwater and determine the variability in concentration. Rainwater was obtained by using fractionated rain samplers. Variability was determined through chemical analysis. Urban and rural aerosol spectra modeling were also used in order to compare the rainwater concentration species variability. When both in-cloud and below-cloud processes are included, the general result of the March 1993 events presents a better agreement between modeled and observed data sets than only below-cloud. Preliminary results lead us to conclude that the rainwater variability of nitrate is explained by the scavenging of particles from urban spectrum size distribution, whereas rural spectra explain ammonium rainwater variability—indicating the different sources of those species. Comparing to the March 1992 events, these case studies present a significant contribution from the in-cloud scavenging, supported by the Weather Radar maps and RAMS modeling. In particular, the almost constant rainwater concentrations on 16 March (an indication of strong in-cloud contribution) are related to the rainfall event, which crossed the study area on that day. These results add an important understanding to the atmospheric wet removal processes in the region studied. r 2002 Elsevier Science Ltd. All rights reserved. Keywords: Wet removal; Numerical modeling; In-cloud scavenging; Below-cloud scavenging; Atmospheric pollution

1. Introduction During the past few years, in-cloud and below-cloud scavenging processes modeling of air pollutants have been evaluated, in relation to their impact on ecosystems, by many authors. Those articles indicated the *Corresponding author. E-mail address: [email protected] (F.L.T. Gon,calves).

relevance of reservoir transference from the atmosphere to the hydrosphere. The transference impact might be analyzed in the rain forest ecosystems, mainly those that are in close proximity to industrial regions. The ENV-3 Project, between Germany and Brazil, described in Vautz et al. (1995) and Klockow et al. (1996), had the task of mapping the sensitivity of ecosystems to this pollution impact. The ENV-3 Project was entitled: ‘‘Air Pollution and Vegetation Damage in the Tropics—The Serra do

1352-2310/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 2 - 2 3 1 0 ( 0 2 ) 0 0 4 6 1 - 2


F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

Mar as an example’’ and was promoted in Cubat*ao, in the southeastern region of Brazil. The main objective was to study the effects of topography and the physicochemical parameters of the vegetation, soil and climate. Cubat*ao is an industrial complex close to the Serra do Mar mountain range ridge, which is covered by tropical rainforest. There are several industrial facilities located in this area, giving the atmospheric pollution dispersion a very unfavorable characteristic. The soil and vegetation have suffered from the pollution impact, mainly due to the atmosphere/hydrosphere transference. The high rainfall amount per year, which can reach up to 4500 mm, contributes to the impact. From 1990 to 1996, several observational field experiments were conducted in order to investigate the rain forest damage. The results were presented in the final report of the project (Klockow et al., 1996). Numerical modeling studies have also been promoted in order to simulate the reservoir transference in the Cubat*ao region. Gonc-alves et al. (2000) present a numerical study of the belowcloud scavenging modeling during the March 1992 Campaign. The results show that there was a belowcloud scavenging process dominance for the rain events studied. As for scavenging process studies in tropical regions, Cautenet and Lefeivre (1994) have also used a numerical modeling to evaluate the gas and aerosol scavenging + processes. For instance, SO2, SO= 4 and NH4 have been studied in convective rainfall in the African equatorial forest. The model examined the relationship between liquid-water content and the trace elements in the convective precipitation. The model results compared favorably with the observed data set evaluated by ABLE 2B (Amazonian forest) and DECAFE (African forest) experiments. The modeling also verified, in particular, the reduction of aerosol scavenging efficiency with an increase in rain intensity and a strong impact of vertical profile of atmospheric trace elements on the ground rain concentrations. Gonc-alves et al. (2002), at Amapa! State (Amazonian region), presented a scavenging modeling with similar results, in particular, aerosol vertical profiles are quite important to rainwater concentrations. Recently, Orb (1999) showed a numerical modeling of the ammonium/ammonia scavenging processes, derived from a field campaign at Mt. Rigi (Switzerland), using CLARK and DESCAN models. The main results are: the microphysics modeling predicted the cloud droplet distribution; the prediction of the pH value from the cloud chemistry modeling was improved—as was the precipitation formation. More recently, Kasper-Giebl et al. (2000) calculated the scavenging efficiency of aerosols at Mt. Sonnblick (Austria), showing that the non-carbonate carbon was scavenged less efficiently than the sulfate aerosols. Barth et al. (2001) differentiate tracers of varying solubilities under different assumption of microphysical

processes. The authors found that below-cloud scavenging of soluble tracers is much smaller than in-cloud scavenging. For highly soluble tracer, the below-cloud scavenging is three orders of magnitude smaller than the in-cloud scavenging. For a moderately soluble tracer, the below-cloud scavenging is 30% of the in-cloud scavenging because this tracer reaches Henry’s law equilibrium between the gas and the liquid phases. The authors conclude that wet deposition parameterizations that assume Henry’s law equilibrium for soluble species in rain are in error. At this point, it is important that the authors discuss the consequences of such findings on their results and eventually moderate their conclusions. Yin et al. (2001) point out the effect of assuming gas– liquid equilibrium in chemistry models, which can lead to overestimation of scavenging of soluble gases by raindrops. The authors highlight the effect of the microphysics structure of the cloud on the scavenging efficiency. The authors used a 2-D cloud model with detailed microphysics and spectral treatment of gas scavenging, simulated trace gas vertical profile in clouds, presenting similar results as that of Barth et al. (2001). Mari et al. (2000) stress the impact of entrainment of environmental air in the cloud on the efficiency of incloud scavenging. The same authors underscore the role of ice in the scavenging process. To aid in the evaluation of the in-cloud scavenging, the Regional Atmospheric Modeling System (RAMS) model, by Cotton and Anthes (1989), can be used. RAMS can address the space and time evolution of the cloud microphysics in rainfall systems. In particular, the cloud liquid-water content, with its respective cloud droplet spectrum and vertical dimensions, can be analyzed. Therefore, the numerical modeling RAMS provides a useful tool in investigating scavenging processes, contributing to the understanding of wet deposition. It is important to note that atmospheric wet deposition is quite relevant to reservoir transference, which has significant consequences to local, as well as global, modeling studies.

2. Experimental site Paranapiacaba Station (PS) is located at approximately 23.71S and 46.31W at an altitude of 800 m, near the Atlantic coast (see Figs. 1a and b). The data from the meteorological instruments and fractionated rain samplers at the ground station, for detecting the rainwater chemical composition, are described in Vautz et al. (1995) and Gonc-alves et al. (2000). PS is about 10 km northeast of the industrial complex, which is in the Mogi Valley in the Serra do Mar mountain range. The continuous monitoring of the rainfall structures was accomplished by the use of S*ao Paulo Weather Radar. The S*ao Paulo Weather Radar is S-band radar, which

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255


3. Model description 3.1. RAMS modeling The RAMS (Walko et al., 2000) was used in order to simulate, diagnostically, the following parameters: the rainwater and cloud liquid-water contents, their vertical and horizontal profiles, the cloud base and the cloud spatial distributions. RAMS is a non-hydrostatic model with a full set of equations for microphysics (Walko et al., 1995) including the continuity equations for water vapor, cloud water, rainwater and seven ice categories. The RAMS cloud physics parameterization can be applied to any water phase, including the precipitation process. The model simulates all phase changes, including latent heat exchanges. Flatau (1989) developed two schemes in order to simulate cloud physics. The first scheme is applied to the water category discretization and the second to the volumetric water, with a continuous distribution. The hydrometeor size distribution is based on Marshall and Palmer (1948), the hydrometeor RAMS scheme is, in general, divided into four categories: cloud droplet, raindrop, frozen hydrometeors and pristine ice crystals. The main RAMS parameterizations, for our goal, are described as follows: *

Fig. 1. (a) S*ao Paulo Weather Radar (R), Paranapiacaba Station (PS) and the Metropolitan Area of S*ao Paulo (MASP). (b) This figure shows the regional topography.

generates Constant Altitude Plan Position Indicator (CAPPI) reflectivity maps, at a height of 3 km, every 10 min. The radar is located at Ponte Nova (231360 0000 S and 451580 2000 W), and covers an area with a radius of about 180 km—presenting a pixel resolution of 2 km  2 km. The Metropolitan Area of (the city of) S*ao Paulo (MASP) is located to the northeast of PS. The MASP is quite a large area with about 6,000,000 vehicles and many industries. Figs. 1a and b show the study area, both the locations of the stations and the radar (Fig. 1a) and topography (Fig. 1b). Radar CAPPI maps were used to validate the RAMS modeling, as used in many works such as Bernadet and Cotton (1998). These authors used radar PPI maps to validate simulated rainwater spatial and temporal distributions.




The homogeneous initialization used the radiosonde data acquired at Congonhas Airport, which is inside the city of S*ao Paulo. The simulation ran for 24 h, starting at 12:00UTC. Three 3-D grids were used around PS, which is situated about 30 km southeastern of Congonhas Airport. The coarse grid specification was defined with 20 km horizontal resolution, the other two with 4 and 1 km horizontal resolutions, respectively. The vertical grid stretches from 100 m of vertical spacing close to the surface increasing by a factor of 1.2 until reaching 500 m of vertical spacing. This is kept constant up to the model top, which is located above the tropopause at 14 km. Time step was assumed to be 20 s for grid 1, and 4 and 2 s for grids 2 and 3, respectively. For the physical parameterization of clouds, the type 3 full microphysical modules from Walko et al. (1995), including ice phase, was used in the three grids. For cumulus parameterization, Molinari (1985) parameterization was used for grid 1 only. Topography and vegetation were used as the modeling input to characterize the Serra do Mar, according to Mahler and Pielke (1977). Type 6 (evergreen broadleaf trees) vegetation parameterization was used. A soil model was assumed, using seven levels with 60% saturation humidity for all depths (Tremback and Kessler, 1985).

5248 *

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

The vertical and horizontal profiles of the cloud water content were used in order to obtain the cloud droplet spectra to integrate the in-cloud scavenging processes along their vertical profiles.

3.2. In-cloud and below-cloud scavenging modeling The numerical modeling was primarily divided into two main mechanisms: below-cloud and in-cloud scavenging processes. Below-cloud modeling was based on Gonc-alves et al. (2000) and on physical and mathematical descriptions generally described in this work. The in-cloud modeling was added to this scavenging modeling. The main modeling assumptions are: *


The spatial structure of the scavenging modeling processes was based upon a vertical 1-D closed box: between the ground and cloud base for below-cloud modeling and between cloud base and cloud top for in-cloud modeling. The gas-scavenging modeling presents similar equations for both processes. Particulate matter (PM) scavenging modeling is also similar and includes nucleation processes. The main goal of all assumptions is the maximization of the scavenging processes.




For fog formation, the nucleation scavenging modeling was assumed to be 0.7 for sulfate and 0.8 for nitrate and ammonium, based on Pandis et al. (1990). The remaining interstitial aerosol (PM) is scavenged through the Brownian diffusion mechanism. For the rural and urban distributions, the particle matter size distribution is characterized by a trilognormal function, based on Whitby (1980) and Jaenicke and Davies (1976) and falls between 0.01 and 40 mm (radius), divided into 73 class sizes of mass (mg/m3) (see Fig. 2a). The distribution in mass and number were assumed constant in time with respect to particle diameter, i.e., hygroscopic growth was not considered in the modeling. The in-cloud modeling was developed using the mesoscale numerical modeling RAMS. This was done in order to evaluate the following parameters: a. The vertical profile of the cloud water content was used in order to obtain the cloud droplet spectra and to integrate the in-cloud scavenging processes along its vertical profile. The cloud spectra based on Levine and Schwartz (1982) as well as RAMS parameterization, with both spectra presenting similar results. From this parameter, the total mass (gas and particle) removed per square meter is

Fig. 2. (a) Rural and urban spectra in mass per volume vs. particle diameter, used as input data from Whitby (1980); (b) sulfur dioxide vertical profile, as input data for 15 March event, in the in-cloud modeling based on Freitas et al. (2000). The other gas species and PM profiles follow a similar pattern.

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

calculated (in Eqs. (1)–(3) as described subsequently). b. The cloud base and top, which are used in Eqs. (1)–(3) as described subsequently. c. The in-cloud scavenging time is used for the time integration and in Eqs. (1)–(3) as described subsequently. d. For low-resolution grids of RAMS, the vertical profile concentrations were obtained with a parameterization of sub-grid scale convective transport of gases and aerosol particles associated with deep and moist convection systems. Aerosol was considered without mass. The parameterization was based on the ‘top-hat’ method, has been coupled to the cumulus parameterization scheme of RAMSCSU model and has been used for long-range transport studies of emissions associated to biomass burning over South America (Freitas et al., 2000). The input vertical profile is shown in Fig. 2b, comparable to the initial concentrations shown in Barth et al. (2001), for CO. *



For below-cloud modeling, the existence of a uniform raindrop size distribution (DSD) was based on a given rainfall rate. No splitting, break-up events or other changes of the DSD took place during the event. Different function relations could express the DSD. The DSD used herein was divided into 50 classes of raindrop size, varying from 0.3 to 5.2 mm in diameter. The raindrop size distributions were also assumed to follow a Gamma function distribution as proposed in Gonc-alves et al. (2000), relative to the measured rainfall rate for each event. The main chemical modeling reactions analyzed from gases absorption were also based on Gonc-alves et al. (2000). SO2, NH3 and HNO3 were the gases and +  SO= 4 , NH4 and NO3 were the PM, both used as input data (atmospheric concentrations). The prog+ nostic variables were SO= and NO 4 , NH4 3 in rainwater. Drop evaporation is not considered due to the high relative humidity (more than 70%).

The following steps were used for the rainwater concentration modeling. For rainwater concentrations originated from particle scavenging, the equation used was Cpw ¼ Cp0  Cp0 expðLp tÞ;


where Cp0 is the initial PM concentration in air, Cpw is the compound concentration from the scavenging of particles (in- plus below-cloud), found in the rainwater, LðDp Þ or Lp is the particles scavenging coefficient and t is the time of the rain event for below-cloud modeling and the cloud formation time period, obtained from RAMS modeling, for in-cloud scavenging. Similar equations for gas scavenging were used.


The bulk concentration scavenged by the drop(let)s was calculated via Cw ¼ Cgw þ Cpw ;


where Cw is the total concentration scavenged by the raindrops. After the incorporation of the chemical species into the raindrop, no chemical reaction was allowed and no release of these species via evaporation of raindrops on the ground was allowed. And the amount of scavenged or deposited mass on the ground was obtained using Cm ¼ Cw H;


where Cm is the deposited mass in mg/m2 (or mg/m2) and H is the height interval in meters between the cloud base and the surface for below-cloud scavenging modeling and between cloud base and top for in-cloud modeling. Four numerical modeling runs have been developed: (a) only below-cloud scavenging modeling using rural spectrum distribution; (b) only below-cloud modeling using urban spectrum distribution; (c) below- plus incloud modeling scavenging using rural spectrum distribution; and (d) below- plus in-cloud modeling scavenging using urban spectrum distribution. For the coupling between particle and gas scavenging, it was assumed that no gas–particle conversion occurs and that all chemical processes are irreversible.

4. Results and discussion 4.1. RAMS simulation and rainfall description The Weather Radar CAPPI maps are shown in Figs. 3 and 4, at 3 km, for the March 1993 events. Both rain cells were considered convective. RAMS event simulations for both events presented spatial distribution and rainfall intensity similar to that shown on the weather radar maps. During the 15 March event (Fig. 3), a large rainfall pattern was advected toward the east, through the Metropolitan Area S*ao Paulo (MASP). The study area was not directly affected by the heaviest rainfall (presenting rates of about 2 mm/h), although the rainfall core reached more than 30 mm/h. On the other hand, during the 16 March event (Fig. 4), a heavy rainfall core was observed crossing the PS study area. This was a smaller convective rain cell, advected from the NW, and presented rainfall rates higher than 20 mm/h, also according to disdrometric data set. Compared to Gonc-alves et al. (2000), these events had a different trajectory pattern. March 1992 events presented no heavy rainfall cores crossing either the study area or the MASP, which implies a different scavenging pattern, as will be discussed later.


F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

Fig. 3. Weather Radar CAPPi maps at 14:53 LT for 15 March event, at 3 km height, including the surface wind (narrow arrow) and rain system direction (large arrow).

Fig. 4. Weather Radar CAPPi maps at 15:53 LT for 16 March event, at 3 km height, including the surface wind (narrow arrow) and rain system direction (large arrow).

The simulated vertical water profiles (cloud+rain), for the 15 March event, presented lower water concentration, reaching a maximum value of only 1.5 g/kg (though in a deeper layer), while the 16 March event shows higher amounts, up to 4.5 g/kg (though in a shallower layer). The 15 March event presents a broader horizontal area of higher concentration, confirmed by Radar maps, while the 16 March event presents a smaller area (see also Figs. 3 and 4). Adding to the simulation results that means the convective system in

the 16 March event was stronger yet smaller. Water content simulated profiles (over PS) are also in agreement with the observed rainfall rates measured by disdrometer at PS. The 15 March event presented a mean rainfall rate around 1–2 mm/h, while the 16 March event presented higher values, reaching 20 mm/h. The total rainfall amount for the 15 March event was 1.6 mm and for the 16 March event was 4.5 mm, almost three times greater. When both events are compared, that result also agrees with the integration of the liquid-water

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255


content profiles in the highest spot (see Fig. 3 where the rainfall core crossed the studied area). As the scavenging processes is a function of the droplet spectra and that parameter is, for its part, a function of liquid-water content, the 16 March event presented a higher scavenging coefficient. This result will be discussed subsequently. 4.2. Below-cloud vs. in-cloud scavenging processes Table 1 presents the results of the correlation coefficients between modeled and observed rainwater species concentrations for both events. The analyzed events took place in o1 h for the in-cloud scavenging, simulated through RAMS, with 1 km grid resolution, and the time periods for below-cloud modeling were E1 h (16 March event) and 1.5 h (15 March event). As a general result, the numerical modeled rainwater concentrations display values at the same scale when compared to the observed ones (see Figs. 5–8 and Table 2). Nevertheless, comparing to results obtained in Gonc-alves et al. (2000), there was a significant increase in the modeled concentrations

Fig. 6. Rainwater concentration of nitrate in the 15 March event, with urban spectrum. Each sample time interval presents a rainfall amount about 0.2770.01 mm, and square correlation coefficient is 0.57. Significance level is 2.5% for seven variables.

Table 1 Correlation coefficient of the linear regression between modeled and observed rainwater concentrations with urban and rural aerosol spectra—all with 97.5% statistical significance Species 15 15 16 16

March March March March

urban rural urban rural

SO= 4

NO 3

NH+ 4

0.48 0.33 0.15 0.13

0.57 0.20 0.25 0.11

0.44 0.60 0.11 0.09

Fig. 5. Rainwater concentration of ammonium in the 15 March event, with rural spectrum (modeled 1) and with urban spectrum (modeled 2). Significance level is 2.5% for seven variables and square correlation coefficients are 0.44 (modeled 1) and 0.63 (modeled 2). Each time interval presents a rainfall amount of about 0.2770.01 mm.

Fig. 7. Rainwater concentration of nitrate in the 16 March event, with urban spectrum. Each time interval presents a rainfall amount about 0.3070.03 mm (per sample) and square correlation coefficient is 0.25.

Fig. 8. Rainwater concentration of sulfate in the 16 March event, with urban spectrum, in-cloud and below-cloud scavenging modeling (modeled 1) and only below-cloud scavenging modeling (modeled 2). Each time interval presents a rainfall amount of about 0.3070.03 mm (per sample). Significance level is 10% for 10 variables and square correlation coefficient is 0.15.

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255


Table 2 The average of rainwater concentrations (mg/l) in both March events 15 March

SO= 4 NH+ 4 NO 3

16 March











9.31 0.23 0.27

1.50 0.31 0.21

34.54 4.63 5.31

17.38 1.51 3.93

5.03 0.46 4.29

7.95 0.47 1.26

0.63 0.15 0.11

18.02 2.93 7.08

16.41 4.59 2.94

7.45 1.69 8.19

Note: (a) only below-cloud modeled rural spectrum; (b) only below-cloud urban spectrum; (c) in- and below-cloud rural spectrum; (d) in- and below-cloud urban spectrum; and (e) observed data.

because the in-cloud contribution was included. The increase is evidenced in Table 2, which show sulfate rainwater concentrations—with and without in-cloud contribution. The in-cloud contribution oscillates between 79% and 97% of the total rainwater concentration, using urban spectrum and 56% and 95%, using rural spectrum, compared to Mari et al. (2000) which found that 77% of the nitric acid was scavenged by deep convection. On the other hand, Barth et al. (2001) found that below-cloud scavenging of soluble tracer gases is three orders smaller than in-cloud scavenging. As for rainwater concentration variability, the 15 March event (Figs. 5 and 6) presented a correlation coefficient between observed and modeled rainwater concentration curves, between 0.20 and 0.60, with both PM spectra (Table 1). In Fig. 5, the modeled 1 (with rural spectrum) result displays the highest square correlation coefficient (0.63) using rural spectra for ammonium. This result also includes the in-cloud contribution occurring before the rainfall reaches the ground. Additionally, there were huge discrepancies between the modeled and the observed rainwater concentrations in the 16 March event (Figs. 7 and 8), with very low correlation coefficients (r2 ranging from 0.11 to 0.25 with urban spectrum and ranging from 0.02 to 0.09 with rural spectrum). The discrepancies become quite clear particularly when it is compared with other events and species (15 March event in Table 1), and also including the March 1992 events presented in Gonc-alves et al. (2000). The modeling assumptions explain part of the discrepancies, also discussed in Gonc-alves et al. (2000), which are optimized (see particularly sulfate in Table 2). In this work, the March 1992 events presented a statistically significant below-cloud scavenging dominance, with r2 higher than 0.70, between observed and modeled (below-cloud modeling only) rainwater concentration curves. The high correlation coefficient is due to the exponential decrease in the observed rainwater concentration data, also verified by the modeled rainwater concentration. Both March 1992 events had a smaller in-cloud contribution, notably the 17 March event, which presented a stratiform rain system.

On the other hand, observed rainwater concentrations in the 16 March event (Figs. 7 and 8) present almost constant values, instead of the typical decreasing values usually attributed to the below-cloud process. In that event, this result was also verified in many other rainwater species, such as calcium, manganese, aluminum and iron. This behavior can also be seen in the observed rainwater concentrations in the 15 March event, though not as clearly defined as it is in the 16 March event. Consequently, both March 1993 events show a type of in-cloud contribution. According to Naik et al. (1994), when nitrate in rainwater concentration presents constant behavior in values, this implies an in-cloud process contribution to the rainwater chemistry. As seen in Figs. 6 and 7, rainwater nitrate concentration highlights that behavior in both events. The almost constant rainwater concentration variability can be primarily explained by the deep convection of those events, which generated an intense removal process inside the cloud. The in-cloud process happened before the rain event itself through the microphysical processes of cloud droplet and raindrop formation (nucleation and gas absorption, for example), which led to the almost constant concentrations thereafter. Additionally, according to the weather Radar data (Fig. 3) during the 15 March event, a large rainfall pattern with deep convection was advected toward the study area, but the heaviest rainfall did not occur in the area. However, the heavy rainfall amounts probably enhanced the removal inside of the cloud over MASP. On the other hand, in the 16 March event (Fig. 4), the convective rain cell was advected from the NW and presented heavy rainfall cores, with high rainfall rates, observed crossing the study area. Consequently, both rainfall systems brought to the ground, in almost constant concentrations, the PM and gas removed through the previous in-cloud processes. To emphasize that result, the average of the belowcloud scavenging modeled concentrations (Table 2) shows a general underestimation of the observed rainwater concentration for both aerosol types and for both

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

events. The exception is seen in modeled sulfate rainwater concentration on 15 March. That exception could be partially explained by the optimization of many modeling assumptions. However, species sources (Section 4.4) could be the main explanation for these discrepancies. Therefore, with the incorporation of in-cloud modeling contribution, both events present a generally better average agreement to the observed values, particularly for ammonium and nitrate. However, the two species probably have different sources, taking into account PM scavenging processes, also discussed below. As observed in Gonc-alves et al. (2000), gas scavenging plays a secondary role in rainwater chemistry variability which present a softer decrease compared to PM scavenging. 4.3. Rural vs. urban aerosol spectra The urban aerosol spectra also present a generally better comparison between the observed and modeled rainwater concentrations (see Table 1), for both in-cloud and below-cloud modeling. Rural spectra modeling present a higher rainwater concentration due to the fact that there are a greater amount of large aerosol (with more mass) than urban spectra (see Fig. 2a). Consequently, the rural aerosol spectrum is fast removed, generating higher rainwater concentration at the first samples. After the second modeling samples, the rural spectra run shows also a significant decrease (see modeled 1 at Fig. 5) which was not generally present in the observed rainwater concentration in the March 1993 case studies (see modeled 2 at Figs. 5–8). On the other hand, in the urban PM spectra, the rainwater concentrations presented a gentler decrease, which fits better with the observed curves. The result could be once again related to the origin of each rain system: the 15 March rainfall system formed over MASP, and the 16 March system formed nearby (although it crossed the area afterwards). Therefore, it can be assumed that both rain systems could have formed with urban, rather that rural, spectrum size. Additionally, in both events, the rural spectrum presented smaller correlation coefficients for nitrate (see Table 1). Sulfate presented an intermediary case. On 15 March, ammonium presented an opposite result, which is discussed in the following. 4.4. Different sources of species For the 15 March event, the correlation coefficients with urban aerosol spectrum were 0.44, 0.57 and 0.48 (ammonium, nitrate and sulfate, respectively), in both observed and modeled concentrations (see Table 1). The 16 March event presented correlation coefficients of 0.25, for nitrate, 0.15 for sulfate and 0.11 for ammonium


(also with urban aerosol spectrum)—all not statistically significant. The slightly higher correlation for nitrate during both events could probably be attributed to its high solubility in rainwater, which induces a better simulation by the numerical modeling. Additionally, Figs. 6 and 7 (for NO 3 ) show mean observed values higher than the modeled concentrations for both events (see Table 2). Opposite behavior has appeared for SO= 4 —also for both events (Fig. 8). NH+ 4 concentrations present an intermediary comparison for mean rainwater concentration (see Fig. 5 and Table 2). The overall results show constant rainwater concentrations as well. This is rather different than the usual exponential decreases for below-cloud process, as it has been described before (particularly for NO 3 ). Consequently, the higher observed nitrate rainwater concentrations and higher correlation coefficients (see Table 2) could also be attributed to the vertical atmospheric nitrate profile over MASP. Therefore, input data from the Serra do Mar ground station gives an underestimation, misrepresenting the actual nitrate vertical profile in the atmosphere over S*ao Paulo. On the other hand, the explanation for the highmodeled sulfate rainwater concentrations could be the optimization of the assumptions in the modeling, as explained before. Some assumptions include mainly: the highest concentrations in the vertical profile, in-cloud pH acidity, high nucleation scavenging rate (for fog formation), no other chemical transformations, the highest scavenging coefficient calculations, etc. All of these assumptions maximize the result, increasing the modeled rainwater concentrations value, as expected. For ammonium, however, with rural spectrum, the correlation coefficient rises from 0.44 (urban) to 0.63 (rural), which is statistically significant. This behavior is counter to the previously described behavior of the other two species. 16 March shows almost no variation, 0.11 with urban spectrum and 0.09 with rural spectrum. This result could be attributed to ammonium sources, as opposed to nitrate sources. Ammonium, supported by the local industry (fertilizer) source near PS, could have also had a particle size distribution closer to the rural spectrum. Also, additionally observed ammonium rainwater concentration average (16 March), modeled with rural spectrum, is closer than that modeled with urban spectrum (see Table 2). Summarizing, the modeling result shows a general agreement with the observed data, e.g., the same magnitude of rainwater concentration and similar concentration curve behavior. Sulfate and ammonium presented an overestimation of the average observed rainwater concentrations and nitrate presented the opposite. The main explanation lies on the species sources.


F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255

5. Conclusions *





The results show the importance of modeling of both processes (in-cloud and below-cloud) and of the PM spectra. The vertical input PM/gas profile, inside the cloud, based on Freitas et al. (2000) present reasonable results, as has been demonstrated in Gonc-alves et al. (2002). The rural and urban aerosol spectra point out the differences in the sources and their importance to the chemistry in the rainwater from wet removal. The rainwater chemistry of ammonium indicates particles scavenged by rural size distribution spectrum sources (local fertilizer industries) and nitrate indicates particles scavenged by urban spectrum sources (MASP). Sulfate presented an intermediary case. In general, the urban spectrum presents the best agreement due to the origin of the rainfall system over MASP. Nitrate presented higher observed mean rainwater concentration compared to the modeled rainwater concentration while sulfate presented the opposite. Ammonium presented an intermediary case. The differences could be attributed to local species sources. The wet deposition in the Cubat*ao and Serra do Mar, during deep rainfall cells, which crossed the MASP, could also be attributed to the clear influence of the in-cloud contribution (between 56% and 97% of total rainwater concentration), also in accordance with the other works. The results indicate a significant in-cloud process, particularly when the rainfall core also crossed over the sample ground station.

In conclusion, the overall results increase understanding of the wet removal in the Serra do Mar region.

Acknowledgements Financial support from the German Ministry for Education and Research, the Ministry of Science and Research of Northrhine-Westphalia, Frankfurt Environment Center-ZUF (Zentrum Unwelt Frankfurt), ISAS (Institut fur . Angewandte Spektrochemie und Spektroskopie), Dr. V. Pedroso from UFBAHIA, Dr. ! C. Solci from UFLONDRINA, Centro Tecnologico e Hidra! ulico, the Secretaria do Meio Ambiente, CETESB and the Conselho Nacional de Desenvolvimento Cient!ıfico e Tecnologico, ! Brazil is gratefully acknowledged. We also thank Prof. Dr. Klaus Beheng of IMK (Germany), Dr. Maria C. Solci of U.E. Paran!a (Brazil), Dr. Vania Tavares of U.F. da Bahia (Brazil) and Drs. W. Vautz, M. Schilling and Prof. Dr. D. Klockow of ISAS-Dortmund (Germany).

References Barth, M.C., Stuart, A.L., Skamarock, W.C., 2001. Numerical simulations of the July 10 stratospheric–tropospheric experiment: radiation, aerosols, and ozone/deep convection storm: redistribution of soluble tracers. Journal of Geophysical Research 106, 12381–12400. Bernadet, L.R., Cotton, W.R., 1998. Multiscale evolution of a derecho-producing mesoscale convective system. Monthly Weather Review 126 (11), 2991–3015. Cautenet, S., Lefeivre, B., 1994. Contrasting behavior of gas and aerosol scavenging in convective rain: a numerical and experimental study in the African equatorial forest. Journal of Geophysical Research 99 (D6), 13013–13024. Cotton, W.R., Anthes, R.A., 1989. Storm and Clouds Dynamic. Academic Press Inc, New York. Flatau, P.J., 1989. The CSU RAMS cloud microphysical module: general theory and code documentation. Colorado State University, Department of Atmospheric Science. Fort Collins, CO 80523. Atmospheric Science, p. 451. Freitas, S.R., Silva Dias, M.A.F., Silva Dias, P.L., 2000. Modeling the convective transport of trace gases by deep and moist convection. Hybrid Methods in Engineering 3, 317–330. Gonc-alves, F.L.T., Massambani, O., Beheng, K.D., Solci, M.C., Rocha, V., Schilling, M., Vautz, W., Klockow, D., 2000. Modelling and measurements of below scavenging processes in the highly industrialized region of Cubat*ao— Brazil. Atmospheric Environment 34, 4113–4120. Gonc-alves, F.L.T., Andrade, M.F., Forti, M.C., Astolfo, R., Ramos, M.A., Massambani, O., Melfi, A.J., 2002. Preliminary estimation of the rainfall chemical composition evaluated through the scavenging modeling for northeastern Amazonian region (Amap!a State, Brazil). Environmental Pollution, in press. Jaenicke, R., Davies, C.N., 1976. The mathematical expression of the size distribution of atmospheric aerosols. Journal of Aerosol Science 7, 255–259. Kasper-Giebl, A., Koch, A., Hitzenberger, R., Puxbaum, H., 2000. Scavenging efficiency of ‘‘aerosol carbon’’ and sulfate in supercooled clouds at Mt. Sonnblick (3106 a.s.l., Austria). Journal of Atmospheric Chemistry 35, 33–46. Klockow, D., Targa, H.J., Vautz, W., 1996. Air pollution and vegetation damage in the Tropics—the Serra do mar as an example: Final report 1990–1996. German/Brazilian Cooperation in Environmental Research and Technology, 207pp. Levine, S.Z., Schwartz, S.E., 1982. In-cloud and below-cloud scavenging of nitric acid vapor. Atmospheric Environment 16, 317–320. Mahler, Y., Pielke, R.A., 1977. A numerical study of the airflow over irregular terrain. Beitrage zur Physik der Atmosph.are 50, 98–113. Mari, C., Jacob, D.J., Bechtold, P., 2000. Transport and scavenging of soluble gases in a deep convective cloud. Journal of Geophysical Research 105 (17), 22255–22267. Marshall, J.S., Palmer, W.M., 1948. The distribution of raindrops with size. Journal of Meteorology 5, 165–166. Molinari, J., 1985. A general form of Kuo’s cumulus parameterization. Monthly Weather Review 113, 1411–1416.

F.L.T. Gon,calves et al. / Atmospheric Environment 36 (2002) 5245–5255 Naik, M.S., Khemani, L.T., Momin, G.A., Rao, P.S.P., Safai, P.D., 1994. Variation in the ionic composition in convective showers in Pune, India. Tellus 46B, 68–75. Orb, J., 1999. Modeling in-cloud scavenging a comparison of measurements and modeling results. Dissertation submitted to the Swiss Federal Institute of Technology, Zurich . (ETH), Switzerland, 103pp. Pandis, S.N., Seinfeld, J.H., Pilinis, C., 1990. Chemical composition differences among droplets of different sizes. Atmospheric Environment 24A, 1957–1969. Tremback, C.J., Kessler, R., 1985. The use of ice–liquid water potential temperature as a thermodynamic variable in deep atmospheric models. Monthly Weather Review 109, 1094–1102. Vautz, W., Schilling, M., Gonc-alves, F.L.T., Solci-Palhares, M.C., Massambani, O., Klockow, D., 1995. Preliminary


analysis of atmospheric scavenging processes in the industrial region of Cubat*ao, southern Brazil. Water, Air, and Soil Pollution 85, 1973–1978. Walko, R.L., Cotton, W.R., Meyers, M.P., 1995. New RAMS cloud microphysics parameterization 0.1. the single-moment scheme. Atmospheric Research 38 (1–4), 29–162. Walko, R.L., Band, L.E., Baron, J., 2000. Coupled atmosphere–biophysics–hydrology models for environmental modeling. Journal of Applied Meteorology 39 (6), 931–944. Whitby, K.T., 1980. The physical characteristics of the sulfur aerosols. Atmospheric Environment 12, 135–159. Yin, Y., Parker, D.J., Carslaw, K.S., 2001. Simulation of trace gas redistribution by convective clouds—liquid phase processes. Atmospheric Chemistry and Physics 1, 19–36.