You are on page 1of 14

Natural Gas Pipeline Leaks and Environmental Injustice in Worcester, Massachusetts

Ahna Knudsen and Ariana Nicholson | aknudsen@clarku.edu & anicholson@clarku.edu


Advanced Vector GIS | December 12, 2019
I. Abstract
Environmental injustice is the disproportionate exposure of marginalized populations to
pollutants, health hazards, and climate change. This analysis focuses on environmental injustice
through the effect of natural gas pipeline leaks on marginalized populations in the diverse and
industrial city of Worcester, MA. We utilize the variables of race and household income to
signify marginalization. We utilize the spatial statistics tool, Optimized Hot Spot, and the spatial
analyst tool, Kernel Density, to see if repaired and unrepaired gas leaks are concentrated within
the city. We then use the demographics of race and income to compare inside and outside
concentrated areas. We found that there are higher percentages of people of color and lower
household incomes in gas leak hot spot areas, and some high-density areas. These findings
indicate that gas-leak related environmental injustice exists in Worcester.
II. Introduction
Many research studies have demonstrated that marginalized communities suffer from
increased exposure to pollutants causing a myriad of health and social disparities, often referred
to as environmental injustice. Environmental justice is defined by the EPA as the fair treatment
of people, regardless of socioeconomic factors, concerning environmental laws, policies, and
regulations (EPA). Hence, environmental injustice is when people are not given equal say in the
decision making process and/or experience disproportionate environmental and health hazards.
The purpose of our project is to explore environmental injustice with a focus on natural gas
pipeline leaks in Worcester, utilizing Geographic Information Systems (GIS) and publicly
accessible data. There has been much public discussion in Worcester about the negative effects
of gas leaks due to explosion risks, health effects, and climate impacts (Mothers Out Front 2018).
To approach this timely issue, we will analyze data compiled by the non-profit Home Energy
Efficiency Team (HEET) as well as the ESRI Demographic Data. We used Hot Spot Analysis
and Kernel Density to find clusters of gas leaks and areal interpolation to assess population
demographics. We then compared socioeconomic demographics of high and low concentrations
of repaired and unrepaired gas leaks. We expected that gas leak hot spots will be in areas of more
people of color and lower economic status because of systemic injustice that concentrates
environmental problems in these communities.
III. Problem Statement
Natural gas pipeline leaks are harmful to communities and the environment. Natural gas
leaks can cause dangerous explosions (HEET), but can also cause more daily health risks
through the gas they emit. Five percent of the natural gas emitted through leaks is made of
volatile organic chemicals, which are foul smelling and can lead to adverse symptoms in some
people (Associated Press, 2016). The remaining 95% of the gas is methane, which contributes to
ground level ozone, a toxin that increases premature mortality (West 2006). Additionally,
methane is a greenhouse gas that is over 100-fold greater in absorbing heat than carbon dioxide,
contributing to global climate change – a major public health crisis (Howarth 2015).
Eversource, Worcester’s natural gas provider, reports in their Annual Service Quality
Report all of the documented natural gas pipeline leaks. Grade one and two leaks are considered
potentially explosive in the context of their surroundings and thus are fixed relatively quickly.
Grade three leaks are not considered explosive, so they are rarely fixed no matter how much gas
is coming from them (Schulman 2016).
The city of Worcester has numerous gas leaks every year. We aim to answer: Are there
areas where gas leaks are concentrated in Worcester, and where? How do the demographics of
these concentrated areas compare to areas with low concentrations? Are natural gas pipeline
leaks in Worcester, Massachusetts more highly concentrated in areas containing higher
percentages of marginalized people? To define marginalization, we are using two measures of
privilege and oppression for the purposes of this analysis – race and median household income -
recognizing that people of color and low-income persons are often disenfranchised in our
society. Therefore, we are expecting more gas leaks in non-white, lower income communities
and more repairs in wealthier, white communities.
This work will be built off a project that Nicholson completed in their Intro to GIS class
in the Fall of 2016. They compared the number of gas leaks per person to census blocks to the
demographics of average non-white population and average median household income. The
study found no significant difference between the number of gas leaks in areas with higher than
average and lower than average demographics. However, with hotspot analysis, kernel density,
areal interpolation, and a more advanced understanding of GIS techniques we hope to shed
further light on potential environmental injustices in Worcester.
We were also drawn to this topic from other projects that focus on the umbrella of
environmental injustice. “​Climate Change, Environmental Justice, and Vulnerability: An
Exploratory Spatial Analysis” by Sacoby Wilson (2008) and “Temporal and spatial changes in
social vulnerability to natural hazards”​ by Susan Cutter ​(2008)​ both examine spatial patterns in
social and environmental vulnerability​. Cutter ​developed the Social Vulnerability Index (SoVI)
based on socioeconomic demographics data collected from the US Census. Instead of using a
vulnerability index our study will compare the two demographics of race and household income
of different concentration of gas leaks.
IV. Data Collection and Preprocessing
We utilized three datasets from three sources for our analysis (see Table 1). First, an
expansive geodatabase of Esri demographic data for 2017. This database is a combination of
public and private demographic data, and has been found to be highly accurate (Esri 2012). We
retrieved this database from the IDCE Clark University fileshare. It contains block group
geography for many variables, including average household income and race. We also utilized a
shapefile from MassGIS of 2005 land cover use to create a mask for areal interpolation.
Finally, we utilized a dataset from the nonprofit Home Energy Efficiency Team, a
non-profit that has compiled all of the reported gas leaks (both repaired and unrepaired) from
Eversource’s Annual Service Quality Report between 2015 and 2018 for all counties in
Massachusetts. We will be utilizing the data from Worcester in 2018. The data are stored in
Google Maps, and are downloadable in a KML file (HEET).
To begin our preprocessing, we first isolated the Worcester demographic data from Esri
using clip. Then, we imported our KML files of 2018 pipeline leaks in the City of Worcester
(unrepaired and repaired) in QGIS to obtain a CSV of intersections where the gas leaks are. This
is necessary because Google maps automatically geocodes the data, and therefore the KML
contains no spatial geography. Then, utilizing the geocoding function in ArcPro we fit these
intersections to addresses in a shapefile for repaired and unrepaired. We clipped this point data to
Worcester to account for incorrectly geocoded points. Finally, we added the shapefile of Land
Use data as a layer, and projected all layers to Massachusetts State Plane Mainland 1983
(Meters, FIPS 2001).
V. Methodology
For methodology, please reference Table 7 and 8: Methodology Flowchart (Hot Spot and
Kernel Density). We started with cleaning the 2017/2022 ESRI US Demographic Data by
removing all fields besides total population (TOTPOP_CY), white population (WHITE_CY),
and average household income (AVGHINC_CY). Then we created a new field called
Nonwhite_pop and used the field calculator function to subtract the white population from the
total population to get all non-white residents. This served as our demographic for people of
color. Next, we created mask of uninhabited areas by utilizing the Worcester MassGIS Land Use
data. We clipped Worcester from the data, then selected by attribute for land cover classes that
were described as residential (10-13, 31, 38). We switched the selection to only areas where
people do not live and called the output LandUse_Mask.
To begin our analysis, we ran Optimized Hot Spot tool which creates a map of
statistically significant hot and cold spots of using the Getis-Ord Gi* statistic (Esri). Our input
was the 2018 repaired and unrepaired gas pipeline leaks from HEET MA. This created a fishnet
and revealed significant hot and cold spots of gas leaks. From the output, we selected only
fishnets above 95% confidence level (2 gi_bin_fixed) and made a new polygon layer from the
selection. We utilized the Dissolve tool based on shape length to achieve a uniform polygon of
hot spots.
For the next step, we imported the Areal Interpolation Tool Kit. We exercised Mask Area
Weighting which aggregates data from a source polygon layer to another layer, while masking
out areas not of interest. The source layer was the ESRI Block Group Data and the target layer,
where attributes are transferred to, was the Dissolved Hot Spot Polygon. The Field was the
non-white population. Our mask was the LandUse_Mask which eliminated uninhabited areas.
The output was a single number of non-white people inside the Hot Spot polygon. We divided
this number by the total population from this area to get the percentage of non-white people.
Next we repeated areal interpolation to have outputs for inside and outside of the hot spot areas
for repaired and unrepaired gas leaks. Hence, we were able to calculate percentage of non-white
people inside and outside of hot spots (Table 2: Race Analysis Hot Spot).
Next, we analyzed Average Household Income. We could not run areal interpolation
because it is not population data. Instead, we selected by location based on if the centroid was
inside or outside of the hot spot polygon. We found the average of the average household income
field with field statistics (Table 3: Income Analysis Hot Spot). We did this for both repaired and
unrepaired gas leaks.
We decided to compare our Hot Spot results with Kernel Density, a tool that calculates
point density and outputs a raster file (ESRI). The Kernel Density algorithm could further
highlight significant areas and serve to validate the hot spot areas. Kernel Density was run on
repaired and unrepaired gas leaks. The Contour tool outlined contours of densities which allowed
us to select high density areas. We chose areas that included the highest density areas while also
maintaining an area large enough for analysis. For unrepaired leaks, we decided upon a density
of above 0.000023 per square meter and for repaired gas leaks a density of above 0.000008 per
square meter. Line to polygon tool created a polygon for high density areas. In this analysis, we
decided not to run Areal Interpolation because 1) the results were similar with and without areal
interpolation in our Hot Spot Model and 2) t-tests could not be run on an areal interpolation
output. We selected Block Groups with a centroid inside the high density polygon then with a
centroid outside to get the sum of non-white people. Then we calculated the percentage of non
white people in high density and low density areas for repaired and unrepaired gas leaks (Table
4: Race Analysis Kernel). The same process was repeated for Average Household Income as
done previously; this time selecting inside and outside of the kernel high density areas. (Table 5:
Income Analysis Kernel).
To test significance, Independent T-tests were run on our outputs using SPSS software.
To do this, we reclassified each Block Group as one (1) for hot spot/high density or two (2) for
cold spot/low density. Two fields were made, with these classifications using select by location,
one for repaired leaks and the other for unrepaired. The table was exported to Excel then saved
as CSV. In the SPSS software, two tailed independent t-tests were run for the respective leaks.
To visualize, a choropleth based on race was created with the hot spot polygon outline of
repaired and unrepaired leaks overlayed. The choropleth was repeated based on household
income for both repaired and unrepaired leaks. The same choropleths were used with kernel high
density outlines overlayed.
VI. Results
In our hotspot analysis results, we found that there was a significant difference between
the percentage of people of color in the population inside and outside of the hotspot. For
unrepaired leaks, there is a 10.24% jump in the population of people of color inside the gas leak
hotspot, and an 11.08% jump inside the hotspot for repaired gas leaks. A full explanation of the
population statistics are available in Table 2: Race Analysis Hot Spot. We created a visualization
of the unrepaired hotspot and repaired hotspot, and overlaid it over a choropleth map including
percentage of people of color by block group. This map does not fully portray the results as it
does not visualize the areal interpolation process, but it is another way to understand the
disparities inside and outside the hotspot (Figure 1: Non-White Worcester Residents and Gas
Leak Hotspots).
For our average 2017 household income analysis, we found a jump of $15,067.03 for
households located outside of the unrepaired hotspot and a jump of $3,969.59 for households
located outside of the repaired hotspot. A full explanation of the average household income
statistics are available in Table 3: Income Analysis Hotpot. All of our hotspot statistics were
found to be significant at the 99% confidence interval (Table 6: Significance Test). Again, we
created a visualization of the unrepaired hotspot and repaired hotspot, and overlaid it over a
choropleth map including average 2017 household income by block group (Figure 2: Average
2017 Household Income and Gas Leak Hotspots).
In our kernel density analysis results, we found that there was a significant difference
between the percentage of people of color in the population inside and outside of the
high-density area for unrepaired leaks. For unrepaired leaks, there is a 7.36% jump in the
population of people of color inside the high-density area. This was found to be significant to the
95% confidence interval (Table 6: Significance Test). Repaired leaks found an 5.23​% jump
inside the high-density area for repaired gas leaks, however these results were found to be
insignificant (Table 6: Significance Test). A full explanation of the population statistics are
available in Table 4: Race Analysis Kernel. We created a visualization of the unrepaired
high-density area and repaired high-density area, and overlaid it over a choropleth map including
percentage of people of color by block group (Figure 3: Non-White Worcester Residents and
Kernel High Density Gas Leaks).
For our kernel average 2017 household income analysis, we found a jump of $12,272.01
for households located outside of the unrepaired high-density area and a jump of $6,287.93 for
households located outside of the repaired high-density area. However, only the unrepaired
analysis was found to be significant at the 99% confidence interval, the repaired was found to be
insignificant (Table 6: Significance Test). A full explanation of the average household income
statistics are available in Table 5: Income Analysis Kernel. Again, we created a visualization of
the unrepaired high-density area and repaired high-density area, and overlaid it over a choropleth
map including average 2017 household income by block group (Figure 4: Average 2017
Household Income and Kernel High Density Gas Leaks).
VII. Conclusion
Our results informed us that there are higher percentages of people of color and lower
household incomes in areas with more gas leaks, both repaired and unrepaired. These findings
were all significant except for the repaired gas leaks kernel output. Based on this analysis, we
can confirm our hypothesis that environmental injustice exists in Worcester. However, there are
limitations to this research. According to HEET MA, “Independent researchers typically find 1.5
- 3 times as many leaks as utilities report” (HEET). Therefore, there are potentially numerous
unrecorded gas leaks in unknown locations. With the recorded leaks, there is still some
inaccuracy of geocoded locations due to the vagueness of street corner locations. Additionally,
the data from our land use mask was from over ten years before our other data, which may have
compromised the accuracy of areal interpolation. Continuing this research, we are interested in
exploring trends over time with the additional available years (every year since 2015), or
creating a social vulnerability index (similar to Susan Cutter’s methods) to go beyond our two
variables of oppression and be more intersectional. This would require further research on
vulnerable populations to calibrate the parameters of an index. We are in the process of
contacting local advocacy groups, such as Mothers Out Front, to share our research with them.
We hope to connect and better understand the research needs of local activists to lend our
support in the resistance movements against environmental injustice.

VIII. Figures and Tables


Table 1: Data Sources

Data Source Link Format

2017/2022 Esri US Esri IDCE Fileshare, Clark University: Geodatabase


Demographics X:\ESRIdata\2017-2022 Demographics

Land Use (2005) MassGIS https://docs.digital.mass.gov/dataset/m Shapefile


assgis-data-land-use-2005

Repaired and Home Energy https://heetma.org/gas-leaks/gas-leak- KML file


Unrepaired Gas Efficiency maps/
Leaks (2018, Team (HEET)
Worcester)

Table 2: Race Analysis Hot Spot

Analysis Area Unrepaired Repaired

Non-White Population In 24210.914295 39123.338121


Hot Spot:

Total Population In Hot Spot: 55693.078473 93525.935462

Hot Spot Population 43.47% 41.83%


Percentage

Non-White Population Not In 43924.085648 29011.661855


Hot Spot:

Total Population Not In Hot 132193.921319 94361.064487


Spot:

Non-Hot Spot Population 33.23% 30.75%


Percentage:
Table 3: Income Analysis Hot Spot

Analysis Area Unrepaired Repaired

Outside Hot Spot Average: $68167.80198 $57,170.25

Inside Hot Spot Average: $53100.770833 $53,200.662338

Difference: $15,067.031147 $3,969.587662

Table 4: Race Analysis Kernel

Analysis Area Unrepaired Repaired

Non-White High Density: 9843 7817

Total Population High 23018 19062


Density:

High Density Population 42.72% 41.01%


Percentage:

Non-White Low Density: 58292 60318

Total Population Low 164869 168825


Density:

Low Density Population 35.36% 35.78%


Percentage:

Table 5: Income Analysis Kernel

Analysis Area Unrepaired Repaired

Outside Kernel Average: $52,035.55 $48,757.37

Inside Kernel Average: $39,763.54 $42,469.44

Difference: $12,272.01 $6,287.93


Table 6: Significance Test

Unrepaired Repaired Unrepaired Repaired


Non-White Non-White Income Income

Hot Spot .000* .000* .002* .000*

Kernel .028* .130** .002* .299*

*Equal variances assumed because Levene’s Test for Equality of Variances is below 0.05
**Equal variances not assumed because Levene’s Test for Equality of Variances is above 0.05
Significant values under 0.05 highlighted in ​bold​.

Table 7: Methodology Hot Spot Flowchart


Table 8: Methodology Kernel Density Flowchart

Figures:
Figure 1: Non-White Worcester Residents and Gas Leak Hotspots
Figure 2: Average 2017 Household Income and Gas Leak Hotspots

Figure 3: Non-White Worcester Residents and Kernel High Density Gas Leaks
Figure 4: Average 2017 Household Income and Kernel High Density Gas Leaks

IX. References

Associated Press, 2016. Porter Ranch gas leak: Invisible disaster drives people from their homes.
Tampa Bay Times,​ January 9,
http://www.tampabay.com/news/nation/porter-ranch-gas-leak-invisible-disaster-drives-pe
ople-from-their-homes/2260678 [Accessed November 17, 2019].

Cutter, Susan et. al., 2008. Temporal and spatial changes in social vulnerability to natural
hazards​. ​Proceedings of the National Academy of Sciences​ Feb 2008, 105 (7) 2301-2306.

EPA. ​Environmental Justice.​ Environmental Protection Agency. Available from:


https://www.epa.gov/environmentaljustice [Accessed 7 December 2016].

Esri, 2012. ​And the Study Says, "Esri Demographic Data Most Accurate".​ Esri. Available from:
https://www.esri.com/news/arcwatch/0712/and-the-study-says-esri-demographic-data-mo
st-accurate.html​ [Accessed 15 November 2019].
Esri. ​Optimized Hot Spot Analysis​. Esri. Available from:
https://pro.arcgis.com/en/pro-app/tool-reference/spatial-statistics/optimized-hot-spot-anal
ysis.htm [Accessed 11 December 2019]

HEET. ​Gas Leak Maps​. Home Energy Efficiency Team, Available from:
https://heetma.org/gas-leaks/gas-leak-maps/. Accessed 16 October 2019.

Howarth, Robert, 2015. Methane emissions and climatic warming risk from hydraulic fracturing
and shale gas development: implications for policy. ​Energy and Emission Control
Technologies​. Available at:
http://www.eeb.cornell.edu/howarth/publications/f_EECT-61539-perspectives-on-air-emi
ssions-of-methane-and-climatic-warmin_100815_27470.pdf [Accessed November 17,
2019].

Mothers Out Front. ​Help Mothers Out Front Worcester Open a Dialogue​. Mothers Out Front,
Available from: ​https://ma.mothersoutfront.org/worcester_mof_petition​ [Accessed 21
October 2019].

Schulman, Audrey, 2016. ​Home Energy Efficiency Team.​ November 28. Personal
Communication.

West, Jason et. al, 2006. Global health benefits of mitigating ozone pollution with methane
emission controls. ​The National Academy of Sciences of the USA,​ March 14, 2006 103
(11) 3988-3993. Available from: ​http://www.pnas.org/content/103/11/3988.full
[Accessed November 17, 2019].

Wilson, Sacoby and Finch, Christina, 2008. Climate Change, Environmental Justice, and
Vulnerability: An Exploratory Spatial Analysis. ​Environmental Justice​, ed. Sylvia Hood
Washington, Volume 3, Issue 1, 2010, pages 13-19.

X. Team effort

Ari brought their experience of having worked with this dataset before and excitement to
examine it utilizing better methods. Ahna analyzed outside sources for gauging methods used
with overlapping research topics. We conducted most GIS analysis together, attending office
hours or working on analysis side-by-side in the library. We split up writing the final paper, final
presentation, and final map outputs. Throughout the process we gave each other feedback and
made edits. We met several times to explore the possibilities and limitations of our analysis.

You might also like