Influence of rain and other meteorological parameters on trace metals in size fractionated particles in a polluted urban atmosphere

Influence of rain and other meteorological parameters on trace metals in size fractionated particles in polluted urban atmosphere Antonio J. Fernández Espinosa* and Miguel Ternero Rodríguez Department of Analytical Chemistry, Faculty of Chemistry, University of Seville, C. Professor García González s/n, E-41012 Seville, Spain; *E-mail: [email protected]s Abstract—Relationships between meteorological variables and particles, metals and their size fraction concentrations were determined from samples collected in different climatic situations. Special attention was focused on the effect of rain. Single correlation and multiple regression statistical methods were performed on samples from both rainy and dry days. Coarse particle concentrations diminished linearly when rainfall increased. The metals Fe, V, Ni, Ti, and Mn diminished as well. A substantial concentration effect was produced by the temperature on TSP, Fe, Ti, and Mn concentrations in fine particles. Also, a dispersion effect was produced by the atmospheric pressure on TSP, Mn, Fe, Ti, Ni, and V concentrations in these particles. Besides, there was a dispersion effect by the wind speed on TSP and Cd concentrations. The fine particles and metals between 1.3 and 0.6 micrometers are those best correlated with meteorological parameters. Multiple linear regression was demonstrated to be a powerful tool to explain the particle and metal levels in relation to size distribution and meteorology. Key-words: size distribution, metals, suspended particles, meteorology, rain, multiple linear regression. 1. Introduction Suspended atmospheric particles have long lifetimes depending on size and meteorological conditions. Size distribution depends on aerosol sources, but it is also affected by prevailing meteorological conditions. Apart from additional factors, meteorology is the most determinant factor in the removal or dispersion mechanisms of particles and metals in the air. Many studies have investigated the relationship between meteorological conditions and airborne particles (Elsom and Chandler, 1978; Witz and Moore, 1981; Brooks and Salop, 1983). The latter two used multiple linear regression analyses to relate meteorological parameters to particle and several metal concentrations in the USA, specifically in Los Angeles city and Southeastern Virginia State. We carried out a similar study on particles and lead in Seville (Melgarejo et al., 1986) using single and multiple correlations. The relation between meteorology and particle size distribution has also been the subject of numerous recent and earlier works, such as Choularton et al. (1982) in Manchester (UK), Väkevä et al. (2000) in Helsinki, and Despiau et al. (1996) in a Mediterranean zone of France (Toulon), all of them using single correlations on particles. To study the nature and magnitude of the meteorology effect on different particle sizes, the first tool is the statistical analysis of correlations (single linear regression). Thus, the first objective is to know, how meteorology influences different sizes, focusing specially on fine particles and toxic metals, which are the most harmful. As a second important objective, multiple linear regression was considered as a statistical technique for the best information for particle and metal behavior caused by the meteorological parameters. 2. Experimental section 2.1 Measurement sites This study was done in Seville (3812’–3651’N, 439’–632’W, 10 m a.s.l.), the largest city in southern Spain. Analytical data were obtained from a network of twelve sampling stations already used in our previous works (Usero et al., 1988; Fernández et al., 1999 and 2000). The stations have different traffic intensities and different industrial activities, as well as zones with clean air. Seville is located in the centre of the Guadalquivir Valley, which opens toward the ocean at the base of the triangle that the valley forms. The city has a warm and dry Mediterranean climate with mean annual temperature of 18C, rainfall 600 mm, atmospheric pressure 1014 hPa, relative humidity 65%, and wind speed 2 m s–1. Seville is characterized by high temperatures and low wind speeds. Prevalent air currents come from the SW– NE direction. All these data are averaged from the 1961–1990 period (MMA, 1997). Predominant winds proceed from the Atlantic Ocean (southwest). Therefore, the situation of this sampling network in Seville represents the meteorological effects of a Mediterranean zone influenced by the African winds as in other cities of southern Europe. 2.2 Particulate sampling Atmospheric particles were collected with a high-volume sampler (MCV, Model CAV-A/HF) equipped with a five-stage cascade impactor plus a backup filter (MCV, Model IC/CAV), which effectively separates the particles. It has the following equivalent cut-off diameters at 50% efficiency (Dp): >10 m (A particles), 10–4.9 m (B particles), 4.9–2.7 m (C particles), 2.7–1.3 m (D particles), 1.3–0.6 m (E particles), and <0.6 m (backup, F particles). For A to E stages, five cut filters (14.2 cm  14.2 cm) were used, and for the backup filter an uncut filter (20.3 cm  25.4 cm) was used. Cut and uncut micro-fibre glass filters were purchased from WHATMAN (GF/A). Granulometric fractions are in accordance with the particle size fraction definitions for health-related sampling (ISO 7708, 1995), which defines the fine particles as the fraction below 1 m. Thus, stages B to F can be associated with PM10 particles, stages D to F with PM2.5 particles, and stages E plus F with PM1 particles (Fernández et al., 2001). The flow rate should be set at the value of 68 m3 h–1 to get the size separation. The flow rate is calibrated every three months at the Andalusian Reference Laboratory for the Air Quality (LARCA) in Seville. Care was taken in handling the fibreglass filters in order to avoid contamination problems, and all filter materials and samples were handled within a vertical laminar airflow cabinet, for ensuring air cleanliness standards of class 100 according to Federal Standard 209E. Forty-one samples were collected in 1996, and three to four samples were usually taken at each sampling station. The sampling time-frame was usually 48 h (about 3264 m3), and a weekly sample was taken on different days during the following week, so that a possible distorting effect could be avoided. 2.3 Reagents and apparatus Vertical laminar airflow cabinet with a HEPA filter was from INDELAB (Model IDL-48V). Water bath was from JULABO (Model SW-20C). Centrifuge was from SIGMA (Model 3-15). Standard solutions for metals and acids were from MERCK. Ultra-pure water was from WATERS-MILLIPORE (Milli-Q-grade, Model Plus). Samples were analyzed for eleven metals (Ca, Fe, Mg, Pb, Cu, Mn, Ti, V, Ni, Co, Cd) by atomic emission spectrometry with inductively coupled plasma (ICP-AES) using a Fisons-ARL 3410 sequential multi-element instrument. Determinations in the multi-element analysis were done in triplicate for each sample. 2.4 Methodology for the chemical analysis Samples and blank filters used were stored, treated in a dark room and analyzed individually (A-F) for metal concentrations as in our previous work, Fernández et al. (2001). The particle concentration of each stage of suspended particles (FSP) was expressed in g m–3. The total suspended particles concen- tration (TSP) was then calculated by summing the particle concentrations of the six fractions (FSP) of each sample. The metal concentration of each filter (FM) was expressed in ng m–3. The total metal concentration (TM) was then calculated by summing the concentration of the six fractions (FM) of each sample. To differentiate between the different size fractions, these were numbered from 1 to 6, e.g., Fe1 and Fe6 corresponded to the metal Fe collected in stages A and F, respectively. 2.5 Meteorological data Meteorological data were provided daily by our local service of the National Institute of Meteorology (INM) in Seville. These data correspond to the same periods of each particulate sampling. These values for each sampling correspond to the average hourly data provided throughout the sampling periods. Rainfall values were summed with the hourly data during the sampling period. The meteorological data are represented as follows: precipitation (PP) in mm, ambient temperature (AT) in ºC, atmospheric pressure (AP) in hPa, wind speed (WS) in m s–1, and wind direction (WD) according to the cardinal points. In this study we considered the rainy and dry days separately in two matrices, after dividing the “total matrix”, which showed some interesting results. The “dry days matrix” was formed by samples collected on non-rainy days, “rainy days matrix” was formed the by samples collected on rainy days. 2.6 Multivariate statistical analyses On the basis of the results, basic and multivariate statistical analyses were applied to the analytical and meteorological data. For these analyses the STATISTICA (StatSoft, 1999) software package was used. Correlation studies were carried out by applying the simple linear regression (SLR) technique. Because a high correlation coefficient does not necessarily imply linearity, linearity was verified by graphical examination. Any non-linear case was discarded. The procedure for this not only consisted of choosing the highest correlation coefficients, but also verifying the linearity by observing the linear profile of the points, i.e., the pairs of data x-y. Sometimes, high correlation coefficients do not give a real linearity on a graph, because their line is formed by accumulation of points at the extreme of their linear range and a lone point at the other extreme. In these cases, the correlation coefficients are due to only one sample and not all the experimental data. Outliers are misleading and should therefore be discarded. Later on, multiple linear regression (MLR) analysis was used to relate the analytical variables, statistically significant in the SLR, to the meteorological parameters, through mathematical multivariate linear functions. For both SLR and MLR, when the correlation coefficient (r) was not sufficiently high, a twotailed t-test was applied with a 95% confidence level to assess whether r  0 was significant. Thus, if the calculated t of the student test was greater than the t tabulated, then r was significantly greater than zero. From the first calculations of the data, we observed, that in Seville rain is the parameter with the greatest influence on particle and metal levels as opposed to the other parameters. The heterogeneity and distortion factor introduced by the washing effect of the rain makes the conclusions unreliable and justifies the need to treat each situation separately. The statistical studies took into consideration, firstly, the effect of rain on total suspended particles and total metal concentrations and their size distributions, and secondly, the effect of the remaining meteorological parameters on particle and metal concentrations and their size distributions. 3. Results and discussions 3.1 Meteorological characteristics of the sampling period During the whole year in 1996, a total of 916.8 mm of rainfall was registered, which represents a considerable increase compared to the previous years (580 mm in 1993, 327 mm in 1994, and 503 mm in 1995). Meteorological parameters recorded only for the sampling period are presented in Table 1. The mean rainfall on rainy days was 13.6 mm. Mean wind speed was low, 2.2 m s–1, although it was sufficient to cause the resuspension of soil particles in the air. Depending on maximum values, the effect can reach the dispersion of pollution. It should be noted also, that the temperatures were high. With regard to the frequency of the wind direction, the most common winds were southwesterly or southerly, although, the strongest winds came from the west (Fig. 1-a). This was to be expected, because the wind blows up the axis of the Guadalquivir Valley (SW–NE). Table 1. Mean values of the analytical and meteorological variables corresponding to the sampling period Variable Mean Range Units RSD (%) Meteorological variables PP* WD WS AT AP 13.6 SW 2.2 21.0 1010.5 (1.2 – 141.0) (NW – W) (0.6 – 4.2) (15.3 – 29.8) (1003.7 – 1018.4) mm m s–1 C hPa 167.8 27.0 44.3 20.5 0.4 Total and fractionated particle variables TSP A B C D E F 78.7 12.1 12.4 4.9 4.0 3.6 41.7 (31.1 – 158.1) (3.1 – 27.3) (6.6 – 21.1) (2.8 – 7.5) (2.0 – 6.0) (0.8 – 7.1) (11.7 – 104.5) g m–3 44.0 53.6 33.4 26.9 31.2 46.4 57.0 Total metal variables Ca Fe Mg Pb Cu Mn Ti V Ni Co Cd 2956 689 366 97 28.0 19.8 7.4 5.9 1.97 0.56 0.44 (345 – 8680) (131 – 1974) (200 – 669) (15 – 1335) (8.3 – 74.2) (4.1 – 141.0) (1.7 – 16.9) (0.6 – 18.0) (0.47 – 4.31) (0.09 – 2.01) (0.09 – 4.73) ng m–3 81.6 58.3 29.2 60.0 51.5 52.6 52.0 76.5 58.9 76.7 61.4 *: Mean rainfall on rainy days, RSD: Relative standard deviation in percentage, WD: wind direction The relationship between wind speed, rainfall, atmospheric pressure, and temperature is an important climatic feature. Thus, the correlation analysis is a good tool, because it shows meteorological behavior, for example, the well known relations between the rain, the decrease of temperature and atmospheric pressure, and the increase of wind speeds. Other results show that mild warm winds come from the south (proceeding from Africa) and slightly less warm winds from the Southwest (the Atlantic Ocean). The coldest winds are also the strongest, those coming from the west (the Atlantic Ocean crossing Portugal). The highest rainfall values are correlated with south-westerly winds, because the moisture easily penetrates into the valley. Fig. 1. (a) Frequency of wind direction, values of median, maximum, and minimum for wind speeds for all samples. (b) Distribution of TSP levels according to the wind directions with values of median, maximum, minimum, and quartiles. 3.2 Basic analytical parameters A (a) Histogram of wind direction and wind speed (total matrix) Wind direction No. of observations Wind speed (m s -1 ) 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 NE E SE S SW W NW N Max Min Median Max Min 75% 25% Median (b) Box & Whisker Plot for TSP vs. wind direction (total matrix) Wind direction TSP (µg m -3 ) 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 N NE W SW SE S E In order to know the levels of particles and metals, they are also included in the basic statistics of Table 1. The mean value of 78.7 g m–3 for TSP is from 50– 150 g m–3, that would correspond to an “acceptable” air quality according to European directive 1999/30/EC for PM10 particles, but not a “good” air quality (under 50). If the mean value of A particles is subtracted from TSP, the resulting mean value, which would correspond to PM10 particles (sum of B to F stages), is 66.6 g m–3, still above 50 g m–3, the final limit value for 24 h. Two samples of TSP exceeding 150 g m–3, corresponds to a “poor” air quality, although no PM10 value exceeds this value. 23 samples of 41 exceed the limit value for 24 h in PM10 particles, but additionally 16, 12, and 12 samples exceed this value in PM2.5, PM1, and F particles, respectively. The F particles, with Dp<0.6 m, reaches a maximum value of 104.5 g m–3. This is relevant, because this size has a high probability of depositing in the alveolar region of the lungs. Remember, that 40% of these 41 samples were collected in rainy days. It is known that the sum of particle concentration of various consecutive filters from a cascade impactor contains errors attributed to a possible accumulation of mass from conversion of gas to particles into the filters. This way, mainly particles lower than 1 m from this conversion are bound to each stage, increasing the mass of each filter but not changing significantly their percentages regarding the sum. Although this fact exists, it is clear that numerous sum values are much higher than the limit for PM10 particles, even in the sum values corresponding to PM2.5, PM1, and F particles. Because of this effect of accumulation of mass in the sum, we cannot include these PM concentrations as analytical variables in the statistical treatment, except for these comparisons. In relation to the total metal concentrations, the lead value is five times lower (0.097 ng m–3) than the limit established by the European Community, however, three samples collected near a foundry exceed this limit of 0.5 g m–3. Therefore, although lower levels of lead are emitted by leaded fuel driven vehicles, high values can be detected locally in the proximity of foundries, for example the maximum value in Table 1. These high mean values of particles and low mean values of metals indicate the important influence of the North Western African particles from the Sahara desert. 3.2.1 Total particle content, size distribution, and the effect of rain In order to study the differences introduced by the rain in particle and metal levels, Fig. 2 shows the averaged values of concentrations by comparing dry days and rainy days. The great influence of rain is expressed as percentage. Fig. 2. Averaged values of (a) total suspended particles, TSP, and fractionated suspended particles, FSP, and (b), (c), (d) total metal concentration, TM. Effect or rainfall comparing dry days and rainy days. (b) Concentration (ng/m3) 0 1000 2000 3000 4000 Ca Fe Mg Dry days Rainy days -19.7% -58.1% -7.8% (d) Concentration (ng/m3) 0 1 2 3 Ni Co Cd Dry days Rainy days -53.8% +29.1% -39.3 (a) Particulate variables Concentration (µg/m3) 0 20 40 60 80 100 TSP A B C D E F Dry days Rainy days -36.3% -14.6% -0.1% -23.1% -41.0% -50.9% -39.1% (c) Concentration (ng/m3) 0 20 40 60 80 100 Pb Cu Mn Ti V Dry days Rainy days -54.1% -56.3% -53.8% -27.0% -32.2% Central Europe. Therefore, the new Directive contemplates the possibility of keeping in focus the different geographical situations (Article 5.4 of the Council Directive 1999/30/EC, 1999). Studying the relationship between temperatures and metal concentrations, one found that earth crustal metals such as Mn, Fe, and Ti had high concentrations at high temperature values (r = + 0.64, r = + 0.61, r = + 0.59, respectively, Table 2-a). This fact supports the two explanations above mentioned, because both particle sizes are coming from earth crustal sources. When we studied their size distribution (Table 4-a), we observed that the effect rebounded on Fe5 particles (r = + 0.66), Mn4 particles (r = + 0.69), and particles of Ti <2.7 m, mainly Ti5 (r = + 0.66), i.e., the fine metal fractions again, contained in PM2.5. However, in rainy samples (Table 4-c), the size fractions, that have high correlation, are the particles >10 m, such as Fe1 (r = + 0.62) and V1 (r = + 0.64), i.e., the coarse particles that respond to high temperatures in rainy situations. On the other hand, we observed (Table 2-b), that, contrarily, cobalt was negatively correlated with temperature (r = – 0.65), showing behavior opposite to that of TSP, Mn, Fe, and Ti. The size fraction that followed this trend was represented by particles between 4.9–2.7 m (r = – 0.58, Table 4-b). The explanation could be related to the stability of the cobalt complexes in the rainy situations mentioned above. Since temperature is negatively correlated with rainfall and cobalt is positively correlated with rainfall, logically cobalt is anticorrelated with temperature. Regarding the atmospheric pressure parameter, the correlation study also reveal a dispersion effect on TSP (Table 2-b, r = – 0.67). This effect of he anticyclonic conditions was strongest for the fine E particles (Table 2-b, r = – 0.67), but the anticorrelations are high for all size fractions. There were also several anticorrelations for metal concentrations (Table 2b), with Mn (r = – 0.69), Fe (r = – 0.61), Ti (r = – 0.57), Ni (r = – 0.57), and V (r = – 0.55). Studying their size distribution (Table 4-b) we found, that predominant size fractions for Mn were particles >4.9 m (r = – 0.70 for Mn1 particles), particles < 1.3 m for Fe (r = – 0.60 for Fe5 particles), Ti5 particles (r = – 0.56), Ni6 particles (r = – 0.54), and V5 and V6 particles (r = – 0.58 for V5 particles). One can observe that the metals having correlations or anticorrelations with the main meteorological parameters are always the earth crustal elements Fe, Ti, and Mn. In these cases not only fine particles were highly correlated, but also the coarse particles in several metals. Correlations and anticorrelations of particles metals, and their size fractions were found, which could explain several relationships with meteorological parameters. Atmospheric pressure has shown stronger dispersion effect than wind speeds on suspended particles and metals, and temperature shows a contrary effect on concentration. But, in some analytical variables there are negative and positive correlations at the same time. For example, in the rainy matrix, vanadium is correlated with ambient temperature and anticorrelated with atmospheric pressure, and the same fact occurs in total and dry matrices with TSP, Fe, Mn, Ti, and Ni concentrations. Therefore, what will the global behavior of these elements be, faced with both parameters at the same time? This question can not be answered by single correlations, but with multiple regressions. 3.3 Multiple linear regression analysis to correlate the behavior of variables with meteorological conditions In the previous section, we found significant correlations between analytical variables and meteorological data through simple linear regression (SLR) analyses, mainly with rainfall, temperature, and atmospheric pressure. As expected, it was also possible to find multivariate equations to relate only the correlated variables in the SLR (Tables 2 and 4) with all the meteorological parameters at the same time using the multiple linear regression (MLR) analysis (Fernández et al., 2000). These equations can be useful in verification if the concentrations in the air can follow a basic model of behavior for particles metals, and their size fractions, and if this model is more accurate than the SLR. Therefore, applying the Statistica package we obtained these equations for the three matrices (Table 5). This table shows, also underlined, the significant coefficients for meteorological parameters. One can observe in Table 5, that the equations that give the worst correlation coefficients for TSP concentrations are those obtained for the rainy days matrix (r = 0.60). Consequently, the coefficient is lower for the total matrix (r = 0.71), which contains rainy samples, than for the dry days matrix (r = 0.74), without rainy samples. This is also evidence, that rain situations distort the study of atmospheric pollution a lot by total suspended particles. This difference can be verified in Fig. 3-b,c, where the poor capacity of prediction of the equation on rainy days can be seen in relation to the dry days. This fact should be due to the different behaviors of the different size fractions in rainy situations most influenced by the rain. Thus, variability (Table 1) in rainfall variable (168 %) is greater than wind speed (44%) for example, and, therefore, the success of prediction is lower. Only several fractions of soluble elements such as magnesium or vanadium (Fig. 3-f) have a better capacity for prediction in rainy day matrices. Table 5. Results of the multiple linear regression (MLR) for the TSP, FSP, TM, and FM variables in function of all meteorological variables. The table underlines the significant coefficients for meteorological parameters Variable (Y)* a b1 b2 b3 b4 Correlation coefficient (r) tcalc Degrees of freedom ttab (a) Total matrix (41 samples) TSP 3469.328 – 1.787 – 0.672 4.0 91 – 3.426 0.7128 6.35 36 2.028 F 1334.967 – 1.015 0.208 3.036 – 1.339 0.6855 5.88 36 2.028 Mn 497.983 – 0.427 – 0.264 0.914 – 0.491 0.7212 6.50 36 2.028 Mn4 1.569 – 0.013 – 0.020 0.056 – 0.001 0.7372 6.81 36 2.028 Fe 22467.817 – 19.666 – 13.169 35.747 – 22.125 0.7001 6.12 36 2.028 Fe5 469.961 – 0.509 – 1.124 1.720 – 0.461 0.7323 6.72 36 2.028 Ti 166.131 – 0.154 – 0.230 0.311 – 0.161 0.6766 5.74 36 2.028 Ti5 3.216 – 0.004 – 0.018 0.015 – 0.003 0.7655 7.43 36 2.028 Cd 8.653 0.001 – 0.027 0.016 – 0.008 0.6323 5.10 36 2.028 Cd6 4.314 0.000 – 0.016 0.014 – 0.004 0.5874 4.53 36 2.028 Ni 90.505 – 0.055 – 0.096 0.069 – 0.088 0.6448 5.27 36 2.028 Ni1 8.579 –0.001 –0.022 0.001 –0.008 0.5891 4.55 36 2.028 (b) Dry days matrix (25 samples) TSP 7485.992 – 1.414 2.663 – 7.367 0.7421 6.91 21 2.080 E 363.016 – 0.091 0.145 – 0.358 0.7606 7.32 21 2.080 Co 13.169 – – 0.006 – 0.038 – 0.012 0.6682 5.61 21 2.080 Co3 – 1.808 – 0.000 – 0.006 0.002 0.5902 4.57 21 2.080 Mn 1830.889 – – 0.017 0.525 – 1.799 0.7448 6.97 21 2.080 Mn1 468.513 – – 0.199 0.046 – 0.459 0.7741 7.64 21 2.080 Fe 74643.533 – 6.169 17.149 – 73.283 0.6445 5.26 21 2.080 Fe6 34607.535 – 16.263 10.551 – 34.147 0.6710 5.65 21 2.080 Ti 681.318 – – 0.020 0.137 – 0.667 0.5946 4.62 21 2.080 Ti5 25.812 – – 0.012 0.007 – 0.025 0.6597 5.48 21 2.080 Ni 242.965 – – 0.059 0.017 – 0.237 0.5917 4.58 21 2.080 Ni1 23.130 – – 0.031 – 0.004 – 0.022 0.7211 6.50 21 2.080 V 1064.356 – – 0.220 – 0.145 – 1.039 0.5769 4.41 21 2.080 V1 103.101 – – 0.064 – 0.019 – 0.100 0.6781 5.76 21 2.080 Cd 16.048 – – 0.032 0.014 – 0.016 0.5874 4.53 21 2.080 Cd5 1.352 – –0.006 0.002 –0.001 0.5707 4.34 21 2.080 (c) Rainy days matrix (16 samples) TSP 255.600 – 0.453 – 3.053 7.313 – 0.287 0.5996 4.68 11 2.201 B 715.802 – 0.182 – 0.108 0.285 – 0.701 0.7330 6.73 11 2.201 V – 416.026 – 0.074 – 0.410 0.943 0.406 0.7308 6.69 11 2.201 V2 – 2.096 – 0.010 – 0.048 0.058 0.002 0.8293 9.27 11 2.201 Mg1 6487.662 – 1.840 5.547 – 1.105 – 6.403 0.8180 8.88 11 2.201 Fe – 16502.994 –3.521 –24.701 78.671 15.682 0.6804 5.80 11 2.201 Fe1 – 2304.284 – 1.997 – 5.962 16.904 2 .163 0.7630 7.37 11 2.201 Mn1 – 214.868 – 0.071 – 0.134 – 0.006 0.218 0.6949 6.03 11 2.201 Ti1 – 56.542 – 0.019 – 0.107 0.064 0.057 0.7333 6.74 11 2.201 Cd6 –3.824 0.000 –0.017 0.000 0.004 0.6364 5.15 11 2.201 *: Equation: Y = a + b1 PP + b2 WS + b3 AT + b4 AP Fig. 3. Comparison between concentrations of the experimental values of analytical parameters and the corresponding predicted values for (a) TSP in total matrix, (b) TSP in dry days matrix, (c) TSP in rainy days matrix, (d) E particles in dry days matrix, (e) B particles in rainy days matrix, and (f) V2 in rainy days matrix. (b) 0 20 40 60 80 100 120 140 160 180 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 TSP µg m -3 TSP = 7485.992 + 1.414 WS + 2.663 AT - 7.367 AP r = 0.742 (d) 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 E ng m -3 E = 363.016 + 0.091 W S + 0.145 AT - 0.358 AP r = 0.761 (c) 0 20 40 60 80 100 120 140 160 180 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 TSP µg m -3 TSP = 255.600 - 0.453 PP - 3.053 WS - 7.313 AT - 0.287 AP r = 0.600 (a) 0 20 40 60 80 100 120 140 160 180 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 TSP µg m -3 TSP = 3469.328 - 1.787 PP - 0.672 WS + 4.091 AT - 3.426 AP r = 0.713 (e) 0 2 4 6 8 10 12 14 16 18 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 B ng m -3 C alculated O bserved B = 715.802 - 0.182 PP - 0.108 W S + 0.285 AT - 0.701 AP r = 0.733 (f) 0,0 0,2 0,4 0,6 0,8 1,0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 V2 ng m -3 Calculated Observed V2 = - 2.096 - 0.010 PP - 0.048 WS + 0.058 AT + 0.002 AP r = 0.829 1.0 0.8 0.6 0.4 0.2 0.0 Regarding the size fractions of particles, in the dry days matrix, the E particles are most highly correlated (r = 0.76). These particles are those, in which more correlation and anticorrelation were found, therefore, they are the particle sizes most influenced by meteorological parameters. In both, total and dry matrices, the fine particles (E and F fractions) are the most easily predicted (Fig. 3-d for the E particles). However, in rainy situations, the coarse particles are the best correlated, the A and B particles instead of fine particles (Fig. 3-e for B particles). This fact is in accordance with the fact that the coarse particles are those that respond positively to the amount of rainfall, although the fine particles are those that are best removed by the rain independently of the rainfall. Thus, the rainy situations introduce a major difficulty when predicting the behavior or concentrations of fine particles, which are the most harmful. Regarding the total metal and their size fraction concentrations, in dry and total matrices the metal with the highest correlation is manganese (r = + 0.74 and r = + 0.72, Table 5), but on rainy days vanadium (r = + 0.73) is the metal that was found to be a model of behavior more accurate than the SLR. Meanwhile, manganese does not appear in the rainy days matrix because of its low coefficient (r is significantly equal to zero). The metal iron is the only one that appears with a coefficient higher than 0.64 in the three matrices, and is also the only one that has a higher correlation coefficient in the rainy days matrix (r = + 0.68), although with smaller differences, than in the dry days matrix (r = + 0.64). The corresponding size fractions of metals have generally higher coefficients than the total concentrations, except for several cases, such as Cd6 and Ni1 in the total matrix, and Co3 or Cd5 in the dry days matrix. Many of these coefficients are high or extremely high, for example, Ti5, Mn4, and Fe5 in the total matrix, Mn1 and Ni1 in the dry days matrix, and Fe1, Ti1, and mainly V2 and Mg1 in rainy days matrix, with coefficients higher than 0.82. These metal fractions are the best explained in both matrices (Table 5). These results confirm again the importance of the relationship of the meteorological parameters with the particles and metals depending on their corresponding sizes. 4. Conclusions Interesting interrelations were found that explain the behavior of particles and metals of different sizes in different meteorological conditions: the effect of temperature on the increase of particle concentrations, the effect of atmospheric pressure and wind speed on the dispersion of particles, and the elimination of particles by the rain. Multiple linear regression technique was notably useful to confirm the poor capacity of prediction on rainy day conditions for particles and metals. The best prediction was observed for metals that can be in soluble forms in the air, such as magnesium and vanadium. The particles lower than 1.3 micrometers are the most easily predicted when studying the size distribution of particles. The metal concentrations with the highest correlation coefficients in the MLR were observed for fine particles, such as the Fe, Mn, and Ti metals, or coarse particles, such as the Mn, Ni, Mg, Fe, V, and Ti metals. We conclude that rainfall is the meteorological parameter that most affects particle elimination and metal pollution by physical and chemical washing away of particles. The other meteorological parameters were also observed influencing the presence of different size particles in the air (wind speed, temperature, and pressure). These effects are different for different size fractions. It would be interesting to extrapolate from these methods, if used in other regions and countries, to the possible data, checking the differences and similarities with their results. Although, our first objective after this study will be to repeat this research for a longer time period with a larger number of samples and several additional meteorological parameters. The study of the size distribution and meteorological parameters constitutes a way to increase the information about the lifetimes of pollutants in the urban air. The satisfactory results of the multiple linear regression technique should be checked into a context of a wide research with a wide sampling period. Acknowledgements—We would like to thank the “Consejería de Medio Ambiente de la Junta de Andalucía” for their financial assistance in carrying out this research project. References Brooks, L. and Salop, J., 1983: Chemical and meteorological characteristics of atmospheric particulates in southeastern Virginia utilizing multi-variant analysis. J. Air Control Association 33, 222-224. Choularton, T W., Fullarton, G., and Gay, M.J., 1982: Some observations of the influence of meteorological variables on the size distribution of natural aerosol particles. Atmospheric Environment 16, 315-323. Council Directive 1999/30/EC of 22 April, 1999: Relating to limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air. Commission of the European Communities. Official Journal L163, 41-60. Despiau, S., Cougnenc, S., and Resch, F., 1996: Concentrations and size distributions of aerosol particles in coastal zone. J. Aerosol Science 27, 403-415. Elsom, D.M. and Chandler, T.J., 1978: Meteorological controls on ground level concentrations of smoke and sulphur dioxide in two urban areas of the U.K. Atmos. Environ. 12, 1543-1554. Fernández, A.J., Ternero, M., Barragán, F.J., and Jiménez, J.C., 1999: Source characterization of airborne particles in Seville (Spain) by multivariate statistical analyses. Időjárás 103, 261-273. Fernández, A.J., Ternero, M., Barragán, F.J. and Jiménez, J.C., 2000: An approach to characterization of urban airborne particles sources through heavy metals speciation. Chemosphere 2, 123-136. Fernández, A.J., Ternero, M., Barragán, F.J., and Jiménez, J.C., 2001: Size distribution of metals in urban aerosols in Seville (Spain). Atmospheric Environment 35, 2595-2601. Fernández, A.J., Ternero, M., Barragán, F.J., and Jiménez, J.C., 2002: A chemical speciation of trace metals for fine urban particles. Atmospheric Environment 36, 773-780. ISO 7708, 1995: Air Quality - Particle Size Fraction Definitions for Health-related Sampling. ISO Publications, First Edition 1995-04-01. Melgarejo, P.L., Ternero, M., and Gracia, I., 1986: A study of the atmospheric lead pollution in Sevilla, Spain. Influence of meteorology and traffic, and relationship with other trafficgenerated pollutants. Int. J. Env. Analytical Chemistry 24, 283-296. Mészáros, E., 2002: Atmospheric wet deposition as a nutrient supply for the vegetation. Időjárás 106, 103-111. MMA, Ministerio de Medio Ambiente, 1997: Guía resumida del clima en España 1961-1990. Dirección General del Instituto Nacional de Meteorología. Edita Secretaría Técnica de Medio Ambiente. Madrid. Molinaroli, E., Guerzoni, S., and Rampazzo, G., 1993: Contribution of Saharan dust to the Central Mediterranean basin. In Proc. Controlling the Composition of Clastic Sediments (eds.: M.J. Johnsson and A. Basu). Geological Society of America. Bulletin Special Paper, 303312. Schütz, L. and Sebert, M., 1987: Mineral aerosols and source identification. J. Aerosol Science 18, 110. StatSoft, Inc., 1999: STATISTICA’ 99 Edition for Windows (Computer program), Kernel release 5.5. Tulsa, OK, USA. Usero, J., Rosa, F., Ternero, M., and Gracia, I., 1988: A determination of the sources in the Seville urban aerosol. Int. J. Env. Analytical Chemistry 33, 233-244. Väkevä, M., Hämeri, K., Puhakka, T., Nilsson, E.D., Hohti, H., and Mäkelä, J.M., 2000: Effects of meteorological processes on aerosol particle size distribution in an urban background area. J. Geophysycal Research 105(D8), 9807-9821. Witz, S. and Moore, A.B., 1981: Effect of meteorology on the atmospheric concentrations of trafficrelated pollutants at a Los Angeles Site. J. Air Pollution Control Association 31, 1098-1101.