GEOSPATIAL IDENTIFICATION OF POTENTIAL ENVIRONMENTAL JUSTICE CONCERNS: PROVIDENCE, RHODE ISLAND

In the early 1990's social activists driven by a concern with the uneven impacts of toxic pollution drew the attention of federal policy makers, establishing an official discourse focused on the issue of environmental justice. The concerns of these activists were supported by a number of statistical and Geographic Information System (GIS)-based studies of demographic patterns and toxic sites (Foreman, 1 998). The concept of environmental justice is based on the premise that disadvantaged groups such as the poor and racial and ethnic minorities bear a disproportionate burden of the negative externalities associated with economic development, including toxic pollution exposure (Buzzelli et al. 2003). Over the past decade and a half, environmental justice, which began as a loosely organized social movement -has become institutionalized in a number of federal, state and local policies and bureaucracies (Holifield, 2001 ). The United States Environmental Protection Agency (EPA), requires the integration of environmental justice into " ... all programs, activities, -consistent with existing environmental laws and their implementing regulations (EPA, 2001)." The implementation of environmental justice policies is intended to establish environmental equity, or an equitable distribution of environmental pollution, health risk, and also access to environmental amenities (Holifield, 2001 ). This study examines and evaluates spatial approaches to identify, and quantify environmental justice concerns existing in the City of Providence, Rhode Island. The study applies geographic information systems (GIS) technology; making use of existing geospatial data for selected toxic sites, and socio-demographic data from the 2000 US Census. Proximity measures are used as a means of quantifying the potential risk associated with the selected hazardous/toxic sites. The distributions of risk across various socio-demographic gradients are examined to highlight disproportionate impacts, or the lack thereof.

Affected and Unaffected Block Groups: Host / Non-Host and HDI 14 Table 2 Correlations among counts of hazards and hazard density indices 23 by census block groups Table 3 Mean socio-demographic characteristics and difference of means 25 t -tests census block groups : host / non-host toxic sites Table 4 Correlations among socio-demographic variables and absolute 27 counts of hazards by census block groups Table 5 Mean socio-demographic characteristics and difference of means 28 t-tests census block groups non -zero / zero hazard density indices Table 6 Correlations among socio-demographic variables and hazard 29 density index score by census block groups lV LIST OF FIGURES In the early 1 990's social activists driven by a concern with the uneven impacts of toxic pollution drew the attention of federal policy makers, establishing an official discourse focused on the issue of environmental justice. The concerns of these activists were supported by a number of statistical and Geographic Information System (GIS)-based studies of demographic patterns and toxic sites (Foreman, 1998). The concept of environmental justice is based on the premise that disadvantaged groups such as the poor and racial minorities bear a disproportionate burden of the negative externalities associated with economic development, including toxic pollution exposure    2005). With this clearly defined mandate the important question of how to identify these populations presented itself (Most et al., 2004).
Over the past decade and a half, environmental justice, which began as a loosely organized social movement -has become institutionalized in a number of federal, state and local policies and bureaucracies (Holifield, 2001  techniques to identify environmental justice concerns and the populations they affect, and to also inform federal, state and local policymakers in their decision making processes (Harner et al. 2002).
This study examines and evaluates spatial approaches to both identify, and quantify environmental justice concerns existing in the City of Providence, Rhode Island. Methods utilized in this analysis make use of geographic information systems (GIS), applying existing geospatial data for selected toxic sites with socio-demographic data from the 2000 US Census. The following analyses incorporate recognized environmental justice parameters with anticipated concerns that have yet to be widely recognized within disadvantaged communities in the City of Providence.
The value of these analyses is viewed to be the establishment of new parameters for spatial analysis which permit the proactive engagement of social issues related to environmental justice. The techniques utilized allow the establishment of essential baseline data, providing the means for environmental justice programmatic evaluation. Analytical tools providing quantitative measures for environmental justice concerns allow for important prioritization of scarce existing federal funding resources dedicated to addressing social concerns at the community level.

Measuring Environmental Justice
Environmental justice research has grown over the past several decades to the point that it is now a "working hypothesis" -that disadvantaged groups face "disproportionate" environmental health hazards (Buzzelli et al., 2003). Acceptance of this working hypothesis has, and will continue to shape environmental policy in the United States (Bowen et al.,1 995) for some time to come.
Even with growing the growing acceptance of existing disproportionate impacts, outcome studies focusing on quantifying the extent and presence of environmental justice issues with regards to disparities in current exposure are frequently challenged. To date, environmental justice researchers have argued over: the optimal scale, spatial units for analysis, selection of socio-economic variables, statistical techniques, and definition of facilities or physical features that pose a toxic threat (Bowen, 2001;Harner et al, 2002). Adding to the clouded picture is the fact that environmental justice continues to be measured in many different ways, with often-contradictory results (Mohai, 1996;Weinburg, 1998;Williams, 1999;Holifield, 2001 ).
Environmental justice researchers interested in measuring risk associated with environmental hazards must deal with a scarcity of measured exposure data for toxic releases . As a result, a number of methodologies have developed to calculate risk measurements including: correlations of social group and hazard co-location or host/non-host studies (Greenburg, 1993); buffering (Glickman, 1 994;Harner et al., 2002) ; plume dispersal modeling (Chakraborty and Armstrong, 2001;Karkazis and Boffey, 2001); toxicity indices (Bowen et al., 1 995;Harner et al., 2002); and proximity to hazards as an estimate of exposure (Bolin et al., 2002;Cutter et al., 2001 ). Holifield (2001) suggests that environmental justice research has progressed to the point at which researchers should no longer be asking: whether or not patterns of disproportionate exposure to environmental hazards exist, but rather: are disproportionably burdened minority and low income communities receiving appropriate attention and resources.
Arguably, an important element in assessing appropriate allocation of attention and resources is the effective quantitative measurement of environmental justice concerns. Measurement of existing environmental justice concerns provides local, state and federal policy-makers with baseline data, valuable information in their decision-making processes (Harner et al, 2002).

The City of Providence
This study will focus on the geographic areas defined by the administrative boundaries for the capital city of Rhode Island, Providence.  The City of Providence's major manufacturing industries: metals, machinery, textiles, jewelry, and silverware were established by 1 830.
These industries have historically played an important role in attracting international immigrants contributing to racial and ethnic diversity (RIEDC, 2005). Unfortunately, Providence's storied manufacturing and industrial heritage has also created numerous toxic and, or sites that are regulated by either, state and / or federal agencies. Toxic sites are common in the post-industrial central city context, and are typically located on former industrial or commercial sites (Miner, 2003). In Providence, many of these sites are located in what were originally prime sites for industrial development -at the core of the city, on waterfronts and close to major transportation routes (Miner, 2003). In Rhode Island regulated toxic sites occur across a wide spectrum of neighborhoods and communities from rural and suburb to the urban core, the issues and concerns of importance in these extremes are very different. In the later contexts, they are commonly seen as community burdens because they may not contribute substantially to the tax base, possess negative aesthetic qualities and pose a possible contamination threat to the water supply; in the former they present the same burdens but are usually linked to a number of wider socio-economic problems (Solitare and Greenburg, 2002). Understanding spatial relationships between toxic sites, the risk associated with them, and those affected is key to addressing a number of socio-economic issues facing the City of Providence today.

OBJECTIVES AND METHODS OF STUDY
In the context of environmental justice literature, this study is to be considered an outcome studyas it focuses on the extent of environmental justice concerns in terms of disparities in current exposure Uerrett et al. 2001 ), for the City of Providence. Analysis will attempt to examine and highlight disproportionate burdens related to quantified measures of toxic risk in the City of Providence; specifically patterns and/or relationships between the spatial distribution of environmental hazards in the form of toxic sites, and low income and ethnic/racial minority residents.
The product of this analysis is a preliminary indicator of possible environmental justice concerns for the city of Providence; revealing inequalities in potential risk based on selected socio-economic variables.
Similar studies in the future will hopefully provide valuable guidance to public and private decision-makers, when they are faced with decisions related to the allocation of funds and resources for neighborhood scale development and / or redevelopment projects. Additionally, baseline and evaluative data provided from similar studies will allow for the monitoring and evaluation of programs and policies designed to address environmental justice issues challenging disadvantaged populations.
Methods used in this study to examine the spatial distribution of risk associated with toxic sites will draw upon recent techniques developed by environmental justice researchers in the absence of detailed data regarding the type and amount of toxic exposure associated with point sources, specifically -proximity measures. Proximity measures provide a geospatial indication and quantification of potential environmental risk and those disproportionately affected; a valuable tool in understanding and addressing environmental justice concerns at the citywide level and valuable data for comparison at the statewide and regional scale.
This study will examine several individual point source toxic site spatial distributions and their relationships to socio-demographic variables. The goals of this study are to address the following questions: • Do different environmental hazards have differing spatial and / or social distributions in the urban context of Providence, RI?
• How does the evaluation of social and spatial distributions of environmental risk change when considering risk density measures from single point sources as opposed to a host / non-host analysis?
• How does the evaluation of the social and spatial distributions of environmental risk change when considering cumulative hazard measures: sum of toxic sites hosted (host / non -host methodology) versus cumulative hazard index (hazard density index methodology)?

Level of Analysis
The unit of analysis for this study takes place at the census block level.
Assessment of risk associated with toxic sites will be analyzed at the census block group level.
The census block group is the smallest unit at which the US Census Bureau reports the desired socio-economic variables of: race and median household income. The census block group allows for aggregation and comparison at several scales including: the census tract; and the neighborhood. Additionally, Most et al. (2004), suggest the appropriateness of smaller spatial units (such as census block groups) in cross-sectionals studies such as this one.
Census block groups are analyzed in context, with reference to each of the City of Providence ' s 25 neighborhoods. The study area is delineated in Figure 2. Residential landuse as interpreted from 1997 aerial photography is provided as referential data, indicating the general pattern of residential development for the City of Providence (RIGIS,

Evaluation and Quantification of Risk
With no comparable measures of risk among the selected toxic sites, all hazardous sites in this analysis will be treated as equally hazardous to those living in proximity. For the purposes of this study relative hazardousness -or risk will increase relative to the number of hazards in a given area. Risk will be considered a proxy measure for the burdens associated with negative environmental externalities associated with hazardous/toxic sites.

Host/Non-host Approach
Initially, risk associated with each of the three classes of toxic sites for each of census block group was analyzed by registering either the presence, or absence of each toxic site class. This host/non-host binary approach classified census blocks containing at least one of the three toxic site classes as host sites and -at risk, while those containing none non-hosts will be considered to be not at risk. Sums of all hazards hosted within the census block groups were also calculated.

Hazard Density Indices
The levels of risk associated with each of the three classes of toxic sites for each census block group were analyzed and measured using the Hazard Density Index (HDI) procedure developed by Bolin et al. (2002).
HDI can be considered an indicator of potential risk for residents of affected census block groups from chronic and acute emissions. No inferences can be made from these indices regarding actual emissions from the toxic sites (Bolin et al., 2002). This density-based approach to measuring risk is based on several assumptions: • All of the environmental hazards (toxic sites) will be considered to produce, process, and/or emit toxic substances regulated by the US EPA/RIDEM and; • Physical proximity to the environmental hazards (

Cumulative Hazard Density Index
The HDI procedure yielded a separate HDI for each toxic site class. The separate HDls for each of the three toxic site classes were summed to create the Cumulative Hazard Density Index (CHDI) for each census block group (Bolin et al. 2002). CHDI measures the agglomeration of all hazard zones within a given census block group; providing an indicator of the compounding risk in each census block group with the inclusion of the proportionate contributions of all proximal toxic sites (Bolin et al. 2002).
Looking for Disproportionate Impacts  However, a consideration of block groups with HDI> zero (or those block groups that intersect with some portion of the 1-mile-radius area for each toxic site) a very different picture emerges. None of the 162 census block groups is untouched by at least one of the hazard zones created by one of the three classes toxic sites (HDl>zero).

Host/Non-Host
Census block groups for the City of Providence hosting one of the three toxic sites were identified. This methodology provided a good picture of how each of the three toxic sites analyzed are distributed throughout the city.
Each toxic site was found to have its own spatial pattern using this approach. CERCLIS sites and TRI sites were found to be concentrated in historically industrial/commercial areas, while LUSTS sites were diffusely distributed throughout the city of Providence; not limited to areas with past or present commercial/industrial and or manufacturing uses. The host/non-host methodology does not, however, take into consideration the aggregate effects of multiple adjacent toxic sites, nor the existence toxic sites located nearby-but not within census block groups.
The sum of all toxic sites hosted by each census block group provided a limited idea of the degree to which block groups are affected by the toxic sites. Analyzed in aggregation, but without data accounting for the magnitude density for toxic sites, this information does not provide detailed quantitative information relating the magnitude of toxic risk.
The results of the sum of all toxic sites analyzed are shown in Figure 3.
This means of measuring risks associated with toxic sites did prove to be a valuable preliminary investigation into the spatial distributions of the   Hazard Density Indices

CERCLIS Hazard Density Index:
HDI values calculated for CERCLIS sites at the census block group level are shown in Figure 5; the values are presented by standard deviations. containing, but spatially proximate to CERCLIS sites are perceivable.

LUSTS Hazard Density Index:
HDI values calculated for LUSTS sites are shown in Figure 6; the values are presented by standard deviations. Perhaps as a result of this wide ranging distribution, calculated HDI values do not exhibit as high a degree of variation as those calculated for CERCLIS sites.
The LUSTS HDI measure does appear provide a better understanding of compounding hazard risk associated with LUSTS, as those census block groups with multiple, and / or are located within close proximity of LUSTS sites exhibit higher index values.

TRI Hazard Density Index:
HDI calculated values for TRI sites are shown in Figure 7; the values are presented by standard deviations.     Correlations among the counts of toxic sites are relatively strongly correlated, indicating the likelihood the coexistence of different toxic site classes within the city of Providence's census block groups. TRI sites and the sum of toxic sites hosted by census block groups show the strongest correlation-this strong correlation is likely due to the consideration of individual chemicals released from TRI sites. For example a TRI site releasing more than one type of regulated chemical is considered for each type of chemical released (e.g. if a census block group were to host a TRI site releasing for example three chemicals -the block group would be considered to host three TRI sites).
Analyzing correlations among HDI scores indicates an overall lower degree of correlation. This may indicate less redundancy in the HDI measures when compared to the counts of hazards by census block group. It is more likely that the HDI measures are measuring different spatial aspects of the toxic sites analyzed; particularly adjacency -or accounting for the compounding effects of multiple proximate toxic sites affecting census block groups.

Evaluating Disproportionate Impacts
The following section investigates some of the differences in the evaluation of the socio-spatial distributions for examined toxic sites when using either the host/non-host, or the hazard density methods.
Specifically this section addresses the questions posed earlier in the objectives of the study: • Do different environmental hazards have differing spatial and/ or social distributions in the urban context of Providence, RI?
• How does the evaluation of social and spatial distributions of environmental risk change when considering risk density measures from single point sources as opposed to a host/ non-host analysis?
• How does the evaluation of the social and spatial distributions of environmental risk change when considering cumulative hazard measures: sum of toxic sites hosted (host/ non-host methodology) versus cumulative hazard index (hazard density index methodology)?

Summed Toxic Sites Hosted
How does this evaluation change when considering the absolute numbers of toxic sites hosted by census block groups? The summary measure created by adding the total number of toxic sites hosted by each census block group did not appear to provide any detectable strong linear relationships to any of the socio-demographic variables examined.

HDI Methodology
How does the evaluation of the relationships between the distribution of hazards and the selected socio-demographic characteristics associated with toxic sites change when using the proximity measure HDI? Of particular interest is how this measure, which considers spatial adjacency, detects disproportionate impacts resulting from multiple point source toxic sites. Individual block group hazard density scores were not considered in these t-tests, but rather: whether or not block groups scored a HDI greater than zero. Do significant differences in the racial / ethnic composition and median household income for block groups exist when evaluated using the HDI methodology that were not apparent using the host / non-host methodology?    With significant correlations at either the 0.01, or 0.05 levels to all of the socio-demographic variables analyzed CHDI shows promise as a summary measure of risk associated with toxic sites and its disproportionate effects on minority racial/ethnic groups and low median income households.

Data Interpolation
To further understand the patterns of risk associated with calculated CHOI scores data interpolation methods were employed. The calculated CHOI values for each census block group were converted to point data.
Each point was assigned to the center of gravity of each census block group, or centroid. Three data interpolation methods were used to examine spatial trends in the CHOI calculated dataset including: an inverse distance weighting function; a Krig prediction map and a triangular irregular network (TIN) generated grid.
The inverse distance weighting function was used to create a risk surface based on the CHOI score for each census block group. The extrapolated risk surface is presented in Figure 9. Neighborhood boundaries are included for spatial and community reference. The second method used to examine the CHOI data was a Krig prediction surface to create a risk surface based on the CHOI score for each census block group. The extrapolated risk surface is presented in Figure 1 0. Neighborhood boundaries are included for spatial and community reference. The final method used to examine the CHDI data was a Triangular Irregular Network (TIN) generated grid. To create a risk surface based on the CHDI score for each census block group. The extrapolated risk surface is presented in Figure 11. Neighborhood boundaries are included for spatial and community reference. All interpolation methods used in this analysis are in consensus with their indication neighborhoods containing areas with the highest levels for interpolated CHDI risk score. These neighborhoods include: • Charles  Bolin et al. (2002). The HDI method produced results that pointed to significant differences related to the racial/ethnic composition (with the exception of Asian) and median household income and census block groups with HDI values greater than zero for all toxic sites analyzed. The summary measure, Cumulative Hazard Density Index was not included in this statistical test since all of Providence's census block groups exhibited a CHOI score greater than zero.
The cumulative hazard density index did exhibit correlations to all of the socio-demographic variables examined. CHOI score for all block groups was significantly positively correlated to the number of racial and ethnic minorities living in a block group, and significantly negatively correlated to the number of whites and increasing median household income for census block groups .
The host / non-host methodology identified no differences among the selected socio-demographic variables and the existence of toxic sites within the census block group. This method did provide a general and preliminary understanding of the spatial distributions of the toxic sites across the extent of the City of Providence examined in this study.
The findings of this and related studies can provide useful data on several levels: First, with incorporation of recognized environmental justice parameters allow for the preliminary identification of environmental justice concerns for disadvantaged communities in the City of Providence. Secondly, the data generated provide baseline information regarding the status of environmental justice concerns for the city of Providence. This baseline data, derived from recent and / or existing conditions permits comparison and evaluative reference for individuals and / or agencies hoping to address environmental justice concerns for the city. Finally, the quantitative measure CHDI, allows for important prioritization of scarce existing federal funding resources dedicated to addressing social concerns at the community level for the City of Providence.