Empirical test of the association between gross contamination of wells

Empirical test of the association between gross contamination of wells with toxic substances and surrounding land use. Michael. Greenberg, Richard. An...
1 downloads 0 Views 839KB Size
Environ. Sci. Technol. 1982, 16, 14-19

Reucroft, P.J.; Simpson, W. H.; Jonus, L. A. J.Phys. Chern. 1971, 75, 3526. Dubinin, M. M.; Plavnik, G. M. Carbon 1968, 6, 183. Lange, N. A. “Handbook of Chemistry”; McGraw-Hik New

York, 1961. Timmermans, J. “Physico-Chemical Constants of Pure Organic Compounds”; Elsevier: Amsterdam, Netherlands,

1950.

(10) Washburn, E. W. “International Critical Tables”; McGraw-Hill: New York, 1926;Vol. 3. (11) Sansone, E. B.; Tewari, Y. B.; Jonas L. A. Environ. Sci. Technol. 1979,13, 1511.

Received for review October 6,1980. Revised manuscript received February 17, 1981. Accepted September 25, 1981.

Empirical Test of the Association between Gross Contamination of Wells with Toxic Substances and Surrounding Land Use Mlchael Greenberg,* Rlchard Anderson, Jennifer Keene, Annye Kennedy, G. William Page, and Sandy Schowgurow

Departments of Urban Studies and Urban Planning, Rutgers University, New Brunswick, New Jersey 08903

To begin to understand where particular groups of toxics are found in the environment, the 10 groundwater wells most contaminated with light chlorinated hydrocarbons were identified from a 408-well sample in New Jersey. Thirty other wells were selected: ten each with the highest levels of pesticides and heavy metals and, as a control group, a clean group of ten wells with nondetectable or the minimum detectable levels of 45 toxic pollutants. Twenty-one categories of land use drawn from aerial photographic surveys were measured for the 10 mi surrounding each site. The pesticide wells showed a relative excess of mixed and evergreen forests and agricultural land uses within 1 mi of the well sites. The light chlorinated hydrocarbon wells exhibited a surfeit of urban land uses within 1 mi of the well sites. The land-use profiles of the heavy-metal wells were not distinct from the clean group. W

Introduction The combination of the failure to consider the long-term impacts of using the environment for a dumping ground for hazardous substances and deliberate criminal activity has become the most serious environmental problem of the 1980s ( I ) . The USEPA and the states are trying to cope with this legacy and at the same time plan for future acceptable sites. Included among USEPA actions are a “cradle-to-grave”hazardous-waste tracking and regulatory process under the Resource Conservation and Recovery Act, litigation under the “imminent-hazard” provision of existing Federal environmental laws, a “Superfund” to provide monies for the cleaning up of the many dangerous sites that have been uncovered, and implementation of emergency control of toxic chemicals threatening navigable waters. At the state level, government activities to combat illicit dumping, particularly when criminal prosecution results, command wide publicity. New Jersey, the study area for the research reported in this paper, exemplifies state government activity against illegal dumpers. A multigovernment strike force has been set up to aid in the detection, investigation, and prosecution of violators (2). This program, which has received heavy USEPA funding because it is viewed as a model for the remainder of the nation, uses numerous legal and scientific methods, including photographic surveillance to detect buried drums and to catch midnight dumpers. Cutting into illicit dumping is a necessary, but not sufficient, means of accounting for all cases of gross hazardous waste contamination. To get a broader perspective on the spatial distribution of hazardous substances in the 14

Environ. Sci. Technol., Vol. 16, No. 1, 1982

Table I. Forty-Five Chemical Substances Identified in Groundwater Samples light chlorinated heavy chlorinated heavy hydrocarbons (17 ) metals ( 9 ) hydrocarbons (19) arsenic methylene chloride BHC-a beryllium methyl chloride BHC-B’ cadmium lindane methyl bromide copper aldrin chloroform chromium dieldrin bromoform nickel heptachlor bromodichloromethane + lead heptachlor 1,1,2-trichloroethylene selenium epoxide 1,1,2,2-tetrachloroethane zinc 1,1,2-trichloroethane toxaphene dibromochloromethane trifluoromethane carbon tetrachloride p,p’-DDT 1,2-dibromoethane meth y oxychlor 1,2-dichloroethane mirex 1,1,1drichloroethane endrin vinyl chloride 1,1,2,2-tetrachloroethylene y -chlordane polychlorinated dichlorobenzene biphenyls trichlorobenzene diiodomethane

environment, it is necessary to step back from the highly visible dumping cases and to explore the relationship between land use and the presence of hazardous substances in the environment. This paper follows this broad-perspective approach and tests three hypotheses: (1)gross organic pesticide contamination of groundwater is associated with agricultural, forest, and horticultural land uses; (2) gross light chlorinated hydrocarbon pollution of groundwater will be found in industrial and commercial areas; and (3) gross heavy-metal contamination of groundwater is found in industrial, commercial, and agricultural areas. To the best of our knowledge this is the first reported attempt to test empirically these hypotheses over a large geographical area.

Data and Methods Two data sets were available from the New Jersey Department of Environmental Protection. The toxic-substances data were selected from a 408-well sample ( 3 , 4 ) . Briefly, the 408 samples are distributed relatively uniformly across the state’s 21 counties. Each county has about 20 samples. Forty-five chemicals were sampled including nineteen heavy organic substances (pesticides), seventeen light chlorinated hydrocarbons (LCHs), and nine heavy metals (Table I). Statistical analyses (factor analyses) of the data disclosed that the chemicals strongly associated by type and each

0013-936X/82/0916-0014$01,25/0

0 1981 American Chemical Society

Table 11. Illustrative Factor Matrix for Toxic Substances in New Jersey Groundwater‘ factors chemicals chloroform bromodichloromethane t 1,1,2-trichloroethylene dibromochloromethane 1,1,2,2-tetrachloroethylene BHC-a lindane aldrin dieldrin heptachlor heptachlor epoxide o,p’-DDE o,p’-DDT mirex endrin y -chlordane p,p’-DDD p,p’-DDT BHC-R copper lead zinc

F1

F2

F3

hb

0.53

0.45

0.69 0.46 0.61

0.69 0.26 0.18 0.27 0.63 0.59 0.54 0.48 0.39 0.58 0.71 0.21 0.78 0.49 0.72 0.64 0.67 0.37 0.41 0.37

0.47 0.63 0.74 0.70 0.62 0.61 0.74 0.75 0.44 0.79 0.68 0.75 0.70 0.55 0.50 0.53 0.48

a The squared multiple correlation coefficients were used as communality estimates, and only factors with an eigenvalue greater than 1.0 were extracted. N = 408. Communality. Only factor loadings greater than 0.40 are presented.

well tended to associate with only one of the three above chemical groups or with an uncontaminated group (5,6). Many factor-analysis runs were made. One is presented as illustrative and typical. The illustrative factor analysis extracted three factors using the eigenvalue greater than one criteria (Table 11). The fourth and fifth factors explained 9.9% and 7.5%, respectively, of the variation within the data, but neither factor had any of the chemical substances with a factor loading (correlation between each chemical and the chemical factors) greater than 0.40. Without any strong factor loadings, these factors cannot be interpreted. The first three factors explained 48.7%, 21.2%, and 12.7% of the total variation. Factor 1 is a pesticide factor. All 14 of the chemical substances loading strongly on factor 1 are pesticides (Table 11). All of these pesticides are heavy chlorinated hydrocarbons. Members of this group of chemically related substances are noted for their low volatility and high persistence in the environment. This factor explains the greatest percentage of variation and is strongly associated with the variables having the highest communalities. The communalities measure the extent of association of a variable to all of the other variables in the analysis. Factor 2 is an organic chemical factor. All of the chemical substances highly associated with this factor are light chlorinated hydrocarbons. The light chlorinated hydrocarbons are more volatile and less persistent in the environment than the pesticides. Of the 14 light chlorinated hydrocarbons in the data set, 4 load strongly in factor 2. None of the remaining light chlorinated hydrocarbons are highly associated with any of the other factors. All of .these substances are commonly used in industrial products including industrial solvents, gasoline additives, disinfectants, and cleaning agents. Some of these products are also widely used in the home. The trihalomethanes are a group of organic chemicals within the light chlorinated hydrocarbon group. Members

of this small group of substances have received wide notoriety because of their carcinogenic potential and their presence in finished potable water. They are also prevalent in the effluent of sewage treatment plants. The trihalomethanes (chloroform, dichlorobromomethane, dibromochloromethane, and bromoform) result from a variety of causes: natural reactions, direct pollution of the substances, pollution of chlorine stimulating the reaction, or others. Chloroform and dibromochloromethane are two of the trihalomethanes which load strongly in factor 2. The two remaining chemical substances with significant associations with factor 2 are connected by a possible biological pathway. 1,1,2,2-tetrachloroethyleneunder anaerobic conditions may be metabolized into the biological base equivalent of 1,1,2-trichloroethylene.“One” of the 45 chemicals in this study is bromodichloromethane + 1,1,2-trichloroethylene,which are combined as one chemical because of the analytic difficulties of distinguishing between them. While this explanation may account for some of the covariation of the chemicals, their presence in New Jersey groundwater may result from other sources. These two substances have the highest associations with factor 2. Factor 3 is a heavy-metal factor. All of the chemical substances loading strongly in factor 3 are heavy metals. Of the nine heavy metals in the factor analysis, three of them load strongly in factor 3. Heavy metals may find their way into groundwater from the natural weathering of rock formations and soil, but they may find their way into groundwater from industrial activities, urban runoff, municipal waste treatment, or abandoned mines. Overall, factor analysis was used to reduce a set of 45 chemicals into a more manageable set of common factors. Each factor is the basis of a group in subsequent statistical analyses. As it was not economically feasible to obtain land-use information about the areas of each of the 408 wells, a spot sample of each of the four statistically distinct groups was obtained. The 10 wells most heavily contaminated with pesticides were chosen to represent gross pesticide pollution. These wells generally contained between 1and 5 ppb of one or more pesticides. Similarly, 10 LCH and 10 heavy-metal wells were selected as the worst cases of LCH and heavy-metal contamination. The 10 LCH wells had between 4 and 300 ppb of one or more light chlorinated hydrocarbons. The 10 heavy-metal wells had extremely high concentrations of one metal, some exceeding 1000 ppb. The fourth group comprised 10 wells that had trace or no monitored levels of any of the 45 toxic substances. New Jersey’s Land Oriented Reference Data System (LORDS) provided 21 categories of land use for the entire state (7,B). LORDS was developed to store data in grid cells that can be overlayed on 17 state topographic atlas sheets at a scale of 1 in. to 1 mi (1:63360). The grid cell system is nested. Each of the 17 atlas sheets is subdivided into a uniform rectangular grid of 25 blocks. Each block contains 34 mi.’ Each block, in turn, is subdivided into nine rectangles of about 3.8 mia2These 3.8-mi2blocks can be subdivided twice more: into nine squares of approximately 0.5 mi2and, in turn, nine squares of about 30 acres. LORDS provides 21 land-use categories from 1972 and 1974 aerial photography (Table 111). Some of these are precise (e.g., industrial, commercial); others are imprecise (e.g., mixed urban, other urban). The 40 wells were located in the smallest LORDS unit, the 30-acre cell. Then the land use for a 10-mi radius surrounding the center of the cell was measured by overlaying a grid on the LORDS land-use maps. Assuming that Environ. Sci. Technol., Vol. 16, No. 1, 1982

15

Table 111. Twenty-one Land Use Categories Found in LORDSa 1. Residential: ranges from high-density, multiple-unit structures to low-density, single-unit structures situated on 1 or more acres. 2. Commercial and Services: areas used predominantly for the sale of products and services. Components of this land-use type include central business districts, shopping centers, commercial strip development, junkyards, etc. Institutional land uses such as educational, health, correctional, etc., are subsumed under this category.

3. Industrial: the wide array of light and heavy manufacturing plants. Surface structures associated with mining operations are included in this category. 4. Transportation, Communications, and Utilities: transportation includes highways, railways, airports, and seaports with their associated shipyards and drydocks. Communication and utilities include areas involved in processing, treatment, and transportation of water, gas, oil, and electricity as well as areas used for airwave and radar stations. 5. Industrial-Commercial: includes those industrial and commercial land uses that typically occur together or in close functional proximity. These areas are also called “industrial parks” because manufacturing, warehousing, wholesaling, and some retailing may occur simultaneously, 6. Mixed Urban: this category comprises a mixture of residential, industrial, commercial, etc., land uses which are mixed such that individual mapping would be extremely difficult.

7. Other Urban: includes land areas with the following uses:zoos, urban parks, cemeteries, golf driving ranges, waste dumps, ski areas, undeveloped land within an urban setting. 8. Cropland and Pasture: these agricultural land uses are broadly defined as land used primarily for production of food and fiber. Cropland consists of harvested areas, areas of bush fruit, cultivated summer-fallow and idle cropland, land on which crop failure occurs, cropland in soil-improvement grasses and legumes, and cropland used only for pasture in rotation with crops. Pasture consists of land which is more or less permanently used for that purpose.

9. Horticultural Areas Including Orchards, Groves, Vineyards, and Nurseries: orchards, groves, and vineyards pertain to lands which produce various fruit and nut crops. Horticultural and nursery areas include floricultural and seed-and-sod areas, and some greenhouses.

IO. Deciduous Forest L a n d : all forested areas having a predominance of trees that lose their leaves at the end of the frostfree season, or at the beginning of a dry season. 11. Evergreen Forest L a n d : all forested areas in which the trees are predominantly those which remain green throughout the

year; both coniferous and broad-leaved evergreens are included in this category. 12. Mixed Forest Land: all forested areas where both evergreen and deciduous trees are growing and neither predominates.

13. Streams, Canals, and Lakes: the streams and canals category includes rivers, creeks, canals, and other linear water bodies, Lakes include nonflowing, naturally enclosed bodies of water, including regulated natural lakes but excluding reservoirs. 14. Reservoirs: areas of artificial water impoundments used for irrigation, flood control, municipal water supplies, recreation, hydroelectric power generation. 15. Bays and Estuaries: only those bays and estuaries classified as inland water are included in this category.

16. Forested Wetland: wetland dominated by woody vegetation includes reasonably flooded bottomland hardwoods, mangrove swamps, shrub swamps, and wooded swamps including those around bogs.

17. Nonforested Wetland: areas dominated by wetland herbaceous vegetation, or those which are nonvegetated. This category also included wetlands of the following types: tidal and nontidal freshwater, brackish, salt marshes, nonvegetated flats, freshwater meadows, wet prairies, and open bogs.

18. Beaches and Sand: beaches consist of smooth sloping accumulations of sand and gravel along shorelines. Sand areas are distinct from beaches and primarily comprise dunes.

19. Bare Exposed R o c k : areas of bedrock exposure, desert pavement, scarps, talus slides, volcanic material, rock glaciers, and other accumulations of rock without vegetative cover.

20. Strip Mines, Quarries, and Gravel Pits: areas of extractive mining activities. 21. Ocean: areas abutting the shoreline which are part of the Atlantic Ocean. a

Source ref 7.

propinquity is an important factor, the data were gathered in three concentric rings: 0-1, 1-5, and 5-10 mi. After initial testing the 1-5- and 5-10-mi rings were merged because their land-use profiles were highly correlated. Specifically, rank correlations were made between the percentage of land classified in particular land uses (e.g., percent residential 0-1 mi vs. percent residential 1-5 mi and percent residential 5-10 mi). The Spearman rank correlations were at least 0.75 between 16 of the 17 land uses of special interst measured over the 40 sites in the 1-5and 5-10-mi zones. In contrast, only 3 of the 17 land uses correlated at 0.75 or greater when the 0-1- and 1-10-mi zones were compared. 18

Environ. Sci. Technol., Vol. 16, No. 1, 1982

Since pollution plumes spread along specific aquifers, aquifer location and shape are another important set of factors that should have been taken into account. Unfortunately, sufficiently precise aquifer maps were not available. Summarizing, the data set consists of observations of 40 places each classified as grossly contaminated with pesticides, LCHs, heavy metals, or uncontaminated. Twenty-one land-use types surrounding each of these sites to 10 mi constitutes the data base for the tests. Two important weaknesses of the data set are imprecise definitions of some of the land uses and the absence of aquifer information.

Two mathematical methods were employed to test the three general hypotheses: the Mann-Whitney test and discriminant analysis. The Mann-Whitney test is a difference of means test for ranked data. A difference of means statistic was used to determine whether any of the four groundwater-well groups had significantly different pergentages of any of the 21 land uses. Six tests were made with the Mann-Whitney test: pesticide vs. clean; LCH VS. clean; heavy metal vs. clean; pesticide vs. LCH; pesticide vs. heavy metal; and LCH vs. heavy metal. Alternatively, the Kruskal-Wallis test, a nonparametric alternative to parametric analysis of variance tests, could have been used as initial protection against type I errors. If significance had been found, then individual pairs could have been tested with Mann-Whitney test. We skipped the Kruskal-Wallis test because we wanted to review each of the pairwise combinations. The Mann-Whitney test was used instead of the parametric difference of means test because each group had only 10 samples. The parametric difference of means test may be used with small samples if normality assumptions are not badly violated. In this case some of the land-use data are skew. Since the MannWhitney test has approximately 95% of the power efficiency of the Student’s t test, it was decided to be conservative and use the nonparametric test. The second method was discriminant analysis, a widely used multivariate method used to isolate the most important differences between groups (9,IO). Discriminant analysis was used to repeat the above-noted tests between the three contaminated-samplegroups and the clean group. In addition, it was used in efforts to discriminate between groups taken simultaneously. Since the two methods are quite different, a brief explanation of why they were used is in order. The MannWhitney statistic is a more conservative test in this project than discriminant analysis because it used rank-order data while the second used parametric data. The MannWhitney test should isolate the strongest and most consistent differences between the four sets of wells. Land uses that are minor when measured by percentage of occupied land but important when measured by pollution potential should play more of a role in the discriminant analysis, which uses parametric data, than in the MannWhitney results. Normally, the researcher tries to prevent minor categories from strongly influencing the results. The normal practice was modified in this study because land uses (for example, landfills) occupying extremely small areas can cause gross contamination of a water supply. We were hesitant about discarding any land-use types. Overall, the Mann-Whitney test should have produced and did produce fewer land uses that discriminate between the four types of wells. Despite our conservatism regarding the premature disgarding of data, 4 of the 21 land uses were eliminated because preliminary analyses suggested that they were unimportant: beach/sand, bare rock, forested wetlands, and oceans. The results focus on 17 land uses which comprise more than 95% of the area within 1 mi of the 40 sample sites and 85% of the area within 10 mi of the 40 sites. Discriminant analysis has a major advantage of producing predictive equations. The land-use profiles of areas that have not been sampled can be substituted into the equations produced by this investigation to yield estimates of which of the four contamination groups the site would probably identify with. Such a step, while exceptionally attractive as a method of roughly identifying potential problem areas, is not advised for two reasons. First, there

are not enough water samples to justify the study as more than a pilot. If, through repeated samples, sites are consistently identified as grossly contaminated or clean, then a predictive model would be more justified than it is with the present water-quality data. Second, the LORDS land-use classification system is not ideal from the perspective of water-quality studies. A classification based on more precise land-use categories with as recent data as possible is suggested before embarking upon a predictive use of discriminant analysis. Another problem with the discriminant-analysis results is that some of the land uses are correlated. For example, many of the areas with a large percentage of industrial land use also have a large percentage of commercial land use. The preferred methods for treating this problem are principle components analysis, or ridge regression. They would produce land-use factors that are statistically independent (principle components (IO)) and regression coefficients that will be stable given small changes in the data (ridge regression (11)). After preliminary testing, it was concluded that neither preferred method could be used because there were as few as 20 observations in many of the statistical operations. Discriminant analysis is also a complex statistical method that suffers when the number of observations is limited. The discriminant analyses made for this study were all examined to be sure that a few observations were not responsible for the results. We tried two methods to control the problem: aggregation and stepwise analysis. A variety of land-use-type aggregations were tested. Each sacrificed too much information. The results were obfuscated rather than clarified. The stepwise method is a widely used means of reducing the problem of intercorrelations between variables. The method reduces the problem by accepting variables into the solution in order of their ability to accurately predict group classification. Insignificant land uses are not incorporated into the solution. Usually, only one of a group of highly intercorrelated land uses will be selected, even if all members of the group are significantly associated. Unfortunately, the method is not perfect and intercorrelated land uses were selected in some of the solutions. The result is that the predictive equations cannot be used. All that can be reported from the discriminant-analysis results are the following: the significant land uses that were p&t of the final stepwise solution, the percentage of the sites correctly identified by the stepwise model, and the standardized canonical discriminant coefficients. The last are statistically standardized weights that, for purposes of this paper, should be used only to identify how strongly each land use identifies with each group.

Results Distinctive land-use profiles were found for the l-miradius areas surrounding the pesticide and LCH wells. The heavy-metal- and clean-well land-use profiles were similar to one another. The results for the area between 1 and 10 mi from the sampling sites were not always significant for both statistical methods. Accordingly, the Results section will focus on the 0-1-mi-ring-pesticide-, LCH-, and clean-well results. The results will be presented in three parts: (1)comparison of the four groups by studying their group means; (2) comparison of the clean group with the pesticide and LCH groups; and (3) simultaneous comparison of the three groups. Comparison of Group Means. The mile surrounding the pesticide sites is dominated by croplands, pastures, forests, and residential areas (Table IV, column 1). The Envlron. Sci. Technol., Vol. 16, No. 1 , 1982

17

Table IV. Comparison of Percentage of Land Use Surrounding the Sampling Sites of the Four Water-Quality Groups €or the 0-1-mi R i n e

LCH

pesticide

water-quality group heavy metal

cleanest

grand mean

% -.

land use 1. residential

(1) 19.6

0.64

41.1

1.35

26.8

0.88

5.4

0.81

4.6

0.69

10.7 -

1.60 __

4.0

0.75

7.9

-

1.49

4.7

0.89

0.59

5.4

3.18 -

0.1

0.06

0.2

0.12

1.7

0.33

10.0

-

2.78

0.9

0.25

2.3

0.64

3.6

0.00

0.0

0.0

0.4

4.00 -

0.0

0.00

0.1

0.51

7.0

-

2.00

3.9

1.11

1.4

0.40

3.5

__

2. commercial 3. industrial 4. utility

1.0

5. indusicomm

1.2

6. mixed urban

0.0

7. urban land

1.8

1.12

6.2

0.93

4.6

0.87

.-

30.4 6.7

__

5.3

8. crop pasture

36.9

1.61

10.3

0.45

25.0 -

1.10

19.2

0.84

22.8

9. horticulture

0.1

0.50

0.1

0.50

0.5

2.55

0.50

0.2

10. deciduous forest

-

0.1

0.0

0.00

5.1

0.62

14.9 -

1.82

12.8

11. evergreen forest

6.7

1.63

1.4

0.34

2.2

0.54

15.3 1.9 0.0

2.89 -

1.5

0.28

3.8

13. streams/lakes 14. reservoirs

1.12 0.00

1.5 0.0

0.88 0.00

2.2 0.1

15. bays/estuaries

0.3

1.00

0.0

0.00

0.0

16. nonforested wet. 17. quarries/pits

2.6 0.1

1.37 0.07

1.0 1.7

0.53 1.13

2.8 1.0

12. mixed forest

sum of 1-17

96.9

__

rural profile of the pesticide group stands in strong contrast to the urban profile of the LCH group. Residential, industrial, utility, commercial, and urban land are dominant in the areas surrounding the LCH sites (Table IV, column

1.56

8.2

1.46

4.1

0.72

0.15

5.3

1.29 4.00

1.3 0.0

0.76 0.00

1.7