VOL. 18, NO. 10 JOURNAL OF APPLIED METEOROLOGY An Investigation of Pollutant Source Strength-Rainfall Relationships at St. Louis Doxa F. Gatz Atmospheric Sciences Secon, [inoisStote Water Sursey, Urbano 61801 (Manuscript received 8 March 1979, in final form 16 June 1979) ABSTRACT. Speculation concerning effects of pollutants on weather, especially rainfall, has intrigued scientists and laymen alike in recent years. Results from Project METROMEX suggest that rainfall is increased after ‘2 moving raincell intersects the urban plume. This paper investigates correlations between rainfall and indices of aerotl pollutant source strength to see whether a causative mechanism might be suggested “Although the correlations between rainfall in certain areas and the source strength of three groups of urban aerosols are significant enough to warrant further investigation, physical relationships do aot support fan arotol involvement in the urban rainfall anomaly. OCTOBER 1979 1, Introduction The perceived rapid increase in annual precipitation at LaPorte, Indiana, during the 1940's was accompanied bya similar jump in the number of smoke and haze days at Chicago (Changnon, 1968). This caused speculation that airborne pollutants might be involved in some way in increasing the precipitation at LaPorte. Since then a number of papers have examined relations between aerosols and rainfall. Both increases and decreases in rainfall have been attributed to increased concentrations of the aerosol ‘or some component of the aerosol, such as condensa- tion nuclei. Warner (1968) found a decrease in rainfall coinciding with an increase in sugar cane production in areas affected by smoke from cane fires in Australia. Conversely, increased concentrations of condensation nuclei in pulp and paper mill plumes were associated with increased rainfall in Washington State (Hobbs at al, 1970). However, a more recent paper (Pittock, 1977) refutes this. Braham (1974) found that increases in the number of small cloud droplets near an area of increased summer rainfall east of St. Louis could be attributed to enhanced condensation nuclei concentrations. Other studies (Vogel and Huff, 1978) have shown that the distribution of summer rainfall in the St. Louis area varies with surface wind direction before the rain and with the direction of cell motion within the rain systems. The locations of the rainfall maxima under various combinations of surface wind direction and storm cell motion suggest a tendency for cell rainfall to peak shortly after ingestion of the “urban plume,” which can often be a plume of relatively warm air, containing high concentrations of particles. Also at St. Louis, Gatz (1978) identified four area-wide sources (including a natural one) and three prominent local sources of airborne trace elements using factor analysis, ‘This method also produced factor scores (indices of source strength) for each of the 48 sources identified at 12 separate filter sampling locations. If pollutants cause urban-related increases in rainfall, then the indices of pollutant source strength downwind. of the sources should be highly correlated with rainfall at locations determined by the surface winds and storm motion. Of course, if significant correlations ‘were found, causation would not be proven; other reasons might be responsible. However, to gain insight into the relationships between pollutants and rainfall that may possibly exist in the St. Louis area, it seemed useful to perform a correlation analysis. ‘This paper reports the results of a correlation anal- ysis between 1) daily factor scores (indices of source strength) and 2) storm rainfall in 21 selected areas (Huff and Vogel, 1978) near St. Louis to test the null hypothesis that there is no correlation between pol- Iutant source strengths and summer rainfall near St. Louis. 2, Methods Rainfall was measured in the 250-gage METROMEX, network of recording raingages. Average rainfall for each storm was calculated for the 16 areas shown in Fig. 1, plus the four quadrants and the whole area cast of the Mississippi River. The factor scores used in the correlation analysis with rainfall data were calculated as part of the factor 1245 (0021.8952/70/101245.07805.75 © 1979 American Meteorological Society1246 JOURNAL OF APPLIED METEOROLOGY Vouvne 18 GRANITE cITY- EDWARDSVILLE REFINERIES \, ee EDWARDSVILLE, SH _ X\ ST. LOUIS ‘URBAN-TNDUSTRIAL GOLLINSVILLE- BELLEVILLE poigtgIND io, 1, Map of rainfall areas (from Huff and Vogel, 1978). analyses performed for an earlier paper (Gatz, 1978). A principal components model was used in the anal- ysis. For any observation, variables are related to the factors by the expression ye onPrbonPet + where 2; is the jth standardized variable, = (aj 2)/s55. Here 2; is the observed value of variable j; # is the ‘mean value of variable j over all observations; 3, is the sample standard deviation of variable j; m is the number of factors common to all the variables; and aj: are the factor loadings. The number of factors m is less than or equal to the number of variables. Gen- erally, only the first few factors, which explain perhaps 80-00% of the variance, are retained. Variables x; in this study are concentrations in air of element j One of the attractive features of factor analysis is that the original variables can be replaced by the more important factors without losing much of the ‘original information. For each case (in this situation, case is a filter sample) a score may be computed for each factor # from Bast bast bist where is the number of variables and the by are the factor score coefficients. The factor scores may be used to replace the original variables in further analyses, ‘As such, if the factors represent sources, the factor scores are the best available index of source strength. In some cases where a certain element is loaded very heavily on a single factor, that element's concentra- tion could be used equally well as a correlation vari- able. However, in the general case, all factors will not have such @ convenient index element, For exam- ple, Gatz (1978) showed that over a limited portion fof the St. Louis area airborne Ph was contributed by two sources: auto exhaust and a local secondary Pb smelter. Here quite clearly the use of the factor score of the ‘‘auto exhaust” factor is preferable to the use of the Pb concentration. In this paper the factor scores are used as indices of the daily “source strength” of the various pollution sources identified at each of several sampling sites. The scores have a ‘mean of zero and a standard deviation of one. Further discussion of factor analysis may be found in Harman (1967) or Rummel (1970). ‘The task of testing the hypothesis that all correla- tion coefficients r between factor scores and rainfall in a subarea equal zero is one of simultaneous in- ference. That is, one must test simultaneously the whole family of statements regarding significance of individual values of r. ‘This simultaneous test may be done by making use of the Bonferroni inequality (Miller, 1966). ‘The Bonferroni relationship says that the probability P that all statements (je, hypotheses regarding the significance of individual values of 7) are correct (ie., r not significantly different from zero) is at least as great as 1 minus the sum of the probabilities that each statement (individually) is incorrect. ThisOcroner 1979 DONALD may be expressed as P2I1-ka, where & is the number of simultaneous hypotheses, being tested and « the usual significance level, Then if & tests of significance are desired with a family significance level a, one should test each statement, at the significance level a/# (Neter and Wasserman, 1974). The choice of which & to use in this particular application varies with the degree of statistical con- servatism one wishes to maintain. The most con- servative approach is to test all 1008 (21 rain areas X48 source strength indices) values of r simulta neously. A less conservative approach would test simultaneously the 21 correlation coefficients ass0- ciated with each source factor. In neither case will the correlation coefficients be independent, but the Bonferroni relationship does not require independence. It is still less conservative, but perhaps to some extent still reasonable, to test simultaneously the values of r associated with each combination of filter site and rainfall area. There were either 3, 4 or 5 coefficients in these groups, namely, the number of ‘meaningful source factors found at each filter site (Gatz, 1978). ‘Thus, the values of & one might choose range from 1008 (most conservative) to 3, 4 or 5 (least con- servative). In this work, significance tests were per- formed for all three degrees of conservatism, and the results are presented and compared. ‘The significance tests were made with normal tables, using the fact that (n—3)! a is normal with mean 0, standard deviation 1, when 1 is the number of pairs son se Cities sre, Sear F. GATZ 1247 in the sample and # is Fisher's s-transformation (Neter ‘and Wasserman, 1974, p. 405); i.e. sed lnl(l+7)/(—r)], ‘The tests performed were the one-tailed type since only increases in rainfall were sought. 3. Results Correlation coefficients between mean storm rainfall and aerosol source factor scores were calculated for every combination of 21 rainfall areas and 48 separate source strength indices identified at the 12 filter sampling sites shown in Fig, 2 (left). The results are given in Table 1, which shows only r values significant at a=1% or a=5M%, for the three different simulta- neous testing procedures described earlier. The number of pairs of data used in each correlation analysis varied from 25-52, reflecting the number of valid filter sam- ples collected at each site during the summers of 1973, 1974 and 1975. Table 1 shows that no correlation coefficients were found to be significant, even at a=10%, when all 1008 correlations were’ tested simultaneously. The table also shows that 2 and 4 correlations, respectively, were significant at @=1%, when testing under suc- cessively less conservative conditions. For these same successively less conservative conditions, an additional 2 and 10 correlations, respectively, were significant at a=5%, Although some may argue that the matter is settled by the result that mo correlations were significant, even at a=10%, when all correlations were tested simultaneously (= 1008), it seems prudent to examine ee tebe + Bt coe eo yore foe Scooany 00 fos PM TNE fe PAINT Pes 1 = Paina Fie, 2, Map of research area, shosring (a) filter sites and (b) some sources of elements in ambient particulate matter collected on filters1248 JOURNAL OF APPLIED METEOROLOGY Vorums 18 ‘TanLe 1. Correlation matrix, showing values significant at 1% and 8% levels for three testing procedures, Filter site number /factor number (number of data pairs) 33/3 49/2 19/5 113/3, 303/2 Rain subarea (32) GI) es, 9) (3) A. Testing all correlations simultaneously: no correlations significant at a= 10%, &= 1008, B, Testing sets of 21 rain subareas simultaneously: #=21. Colinsville-Belleville 0.62 1%) Belleville blufs| 0.62 (1%) Refineries 0.39 (56), 0.82 (8%) C. Testing sets of factors simultaneously: 4=3, 4 oF 5. Colinsvlle-Belleille 0.02 1%) Belleville buts 0.62 1%) Refineries 0.39 (1%) 0.52 (19%) Suburban west 0.483 (5%) East downwind 0.47 (8%) Alton-Wood River downwind 0.36 (89%) Alton: Wood River bottomlands 0.35 (89%) Edwardsville 037 (6%) 0.40 (8%) Granite City-Edwardaville 034 (5%) 0.40 (5%) Alton-Wood River 035 (8%) 0.33 (5%) further some of the correlations found to be the most significant under less conservative approaches. We shall thus consider in more detail the four correlations significant at the 1 or 5% levels in Table 1B, which hhappen to be the same ones found significant at the 1% level in Table 1C. We now examine spatial relationships between the filter sites and rain areas hiving the most significant correlation coefficients. These relationships are shown in Figs. 3, 4 and 5, cach figure corresponding to a separate pollutant source factor. Fis. 3. Spatial relationship between Slter site 33 (vertically hatched circle) and the rain area correlated mest highly with factor 3. The number in the rain area box is the correlation, coefiient. There are several general things to notice about the results. The three factors having the highest cor- relations with rainfall occurred at three different filter sampling locations. The three factors were all different in the sense that a different set of elements was asso- ciated with each. All three of the factors could be classified as “pollutants.” These include auto exhaust Gite 33, Fig. 3), non-ferroust metals (site 49, Fig. 4) and ferrous metals (site 79, Fig. 5). All had positive correlations, indicating that rainfall had increased with enhanced element concentrations. Fig. 3 shows that the rain area of high correlation (refinery area) with factor 3 at site 33 is immediately Fic. 4. Spatial relationship between filter site 49 and the rain ‘area most highly correlated with factor 2. Other conventions asin Fig. 3.Ocronex 1979 adjacent to the filter sampling site. This is an auto exhaust factor, with high loadings on Br and Pb, as shown in the figure; it accounted for 12% of ‘the total variance of element concentrations at site 33. None of the five wind direction categories (four quadrants plus variable) was associated with this factor. That is, all wind direction loadings were less than 0.25, Still, it should be noted that a four-lane divided highway runs north and south ~0.5 km west of the filter sampling site. 4 shows that rain in the refineries area was also highly correlated with factor 2 at filter site 49, immediately adjacent to the southeast. This is an industrial metals factor, accounting for 13% of the variance at site 49. It is strongly associated with southwest winds, which bring the highly loaded metals (Ti, Fe, Zn and Pb) to the sampling site from in- dustrial areas in St. Louis, East St. Louis, Granite City and nearby municipalities. See Fig. 2 (right) for the locations of these sources and Gatz (1978) for further discussion. Enhanced rainfall could conceivably have fallen in the refineries area after cells moving from the southwest had ingested the pollutants near their sources. 5 shows the two rain areas having identical high correlations with factor 5 at site 79. The two areas overlap and are located generally cast and southeast of the filter site. This again is an industrial metals factor, whose source is almost certainly the large steel mill in Granite City, located just a few Kilometers southeast of the filter site. It accounts for 7% of the variance in the element concentrations at site 79. 4. Discussion ‘A few rainfall subareas exhibit significant positive correlation coefficients with pollutant source factors for all but the most conservative approach to signifi- cance testing. We have taken the approach that it is prudent to examine a few of the most significant of these cor- relations to see whether they suggest any mechanisms for pollutant enhancement of urban summer rainfall that are physically reasonable. The rain areas of high correlation are all relatively close to the respective filter sites. Is it physically reasonable that storms that ingested polluted air ob- served at these sites would have’ produced enhanced rainfall in the various areas found most highly cor- related? Are these observations consistant with pre- vious results of Vogel and Huff (1978) that suggest rain enhancement after cells ingest the ‘urban plume?” ‘These questions will be considered next. Fe is present with a moderate loading, but overall the group of metals may be described as non-ferrous DONALD F. GATZ 1249 ‘Fic, 5. Spatial relationship betwoen filter site 79 and the rain areas having highest correlations with factor 8. Other conven- tions as in Fig. 3. The rainfall distributions found by Vogel and Huff (1978) for surface winds from the SE and SW (the ‘most frequent surface wind directions in summer) are shown in Figs. 6 and 7, respectively. For each surface wind direction, the mean rain patterns associated with both SW and NW storm motions are shown. For SE surface winds and SW storm motion (Fig. 6a), the broad rainfall maximum across the northern part of the research circle includes the refineries area where rainfall was found to be highly correlated with two different pollutant factors. The pattern for NW storm motion (Fig. 6b) includes rainfall peaks in the areas where the rain was highly correlated with factor 5, site 79, For surface winds and NW storm motion (Fig. 7b) an area of maximum rainfall near Alton-Wood River again includes the refineries area, For SW storm motion, however, the patterns do not coincide at any point. ‘These comparisons show that all rain areas of sig- nificant correlation are included in areas of high mean rainfall under certain combinations of surface winds and storm motion. However, the converse is not true; many places with high rainfall do not correlate sig nificantly with any source strength indices. Further- more, in each case where there is agreement between patterns for certain combinations of wind direction and storm motion, there is disagreement for other combinations. In addition, the fact that the three pollutant factors found to be significantly correlated with rainfall all represent diferent pollutant sources appears to be a1250 a. WSH-ENE Storm Hovenents JOURNAL OF APPLIED METEOROLOGY Vous 18, b. NM-SE Storm Movenent. Fie. 6. Rainfall patterns for SE surface winds and storms from the WSW and NW ({rom Vogel and Huff, 1978), 07 is the number of storms, and the mean and standard deviation refer to rainfall at 250 gages, crucial observation. One of these factors (auto ex- hhaust) was identified at all 12 sampling sites (Gatz, 1978), but only at site 33 was it correlated with rainfall in any of the rain areas, ‘Thus, a physically ‘reasonable explanation for the few high correlations is not apparent, and we appear to be justified in accepting the verdict of the simul- taneous test of all 1008 correlations; none of the correlations is significantly diflerent from zero, 5. Summary and conclusions Indices of pollutant source strength (factor scores for pollutant sources identified by factor analysis in an earlier study) were correlated against rainfall in 21 separate subareas near St. Louis. This was done as a test of the null hypothesis that there is no cor- relation between the indices of pollutant source strengths and summer rainfall near St. Louis. b. Fro. 7. Rainfall patterns for SW surface winds and storms from the WSW and NW (from Vogel and Huf, 1978). (Other conventions as in Fig. 6.Ccrones 1979 ‘Three pollutant sources were found to have cor- relation coefficients significant enough to warrant examination for a physically reasonable explanation. Examination of the relationships between the filter site where the pollutant was detected and the rain areas of high correlation showed that rain areas were always relatively close to the filter site. However, the relatively weak agreement between rain areas of high correlation with the indices and previously observed patterns of mean rainfall for certain combinations of surface wind direction and storm motion, plus the fact that the three pollution indices having the highest correlations with rainfall all represented different kinds of sources leads to the conclusion that the correlations were chance occurrences. Acknowledgments. 1 thank Mr. 8. A. Changnon, Jr, and Mr. R. G, Semonin for general supervision’ and helpful discussions concerning this work, Mr. J. L. Vogel and Mr. F. A. Huff for supplying the rainfall data used, and Dr. S. L. Portnoy for instruction in the proper statistical methods to use. This work was supported by the Department of Energy, Division of Biomedical and Environmental Research, under Con- tract EY-76-S-02-1199, and by the Research Applied to National Needs (RANN) Program of the National Science Foundation. DONALD F. GaTz 1251 REFERENCES Draham, R. R, 1974: Cloud physics of urban weather modica- iowa preliminary report Ball, Amer. Meteor. Sec, 58, 10-106, CChangoon, 8, A. 1968: The LaPorte anomaly—fact or fiction? Bal Amer Meter. Soe, B, 4 Gata, D, F, 1978 Identncation of serosal soures in the St ‘Louis area using factor analysis. J. Appl. Meteor, 17, (0-08. Barman, H. Hy 1967: Moderm Factor Analysis, 2d. Uni werty of Chicago Pres, 474 pp. Hobbe, P. Vs, Fe Radke and &. E, Shumway, 1970: Cloud ondenseion nice from indosral sources ‘and. thelr p- arent infence on precipitation in Washington State, J Times, Si, 21, 81-89. Buf FA, and J, L. Voge), 1978: Urban, topographic and “una eects on rainfall inthe St. Loa region. J. APE Meteor, 17565577. Miley 1966: Simalloneus Si i, 22 pp Noten, J and W. Wanscrman, 1974: Apid LincorStotisca “fade Richard D. Trin Te, #42 Dp. Pittock, AB, 1977: On the causes of loel climatic anomalies, ‘ith spell reference to precipitation in Washington State. J. Appl. Meteor., 16, 223-230. Rammel, R. J 1970: Applied Racor Anolyris. Northwestern ‘Univesity Press, 617 pp. Voes, J Un and FA. Ha, 1978: Relations between the St Touis urban precipitation anomaly and synoptic weather factors JAP. Meteor, 7, ALLS, ‘Warner, J, 1968: A reduction in rainfall associated with smoke {rom sugarcane fesan inadvertent weather modification? ApH Meter, 7, F-25 eal Inference. McGraw-