University of Southern California Dissertations and Theses

Soil lead contamination from the Exide battery smelter: the role of spatial scale in cleanup efforts

(USC Thesis Other)

Soil lead contamination from the Exide battery smelter: the role of spatial scale in cleanup efforts

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Soil Lead Contamination from the Exide Battery Smelter: The Role of Spatial Scale in Cleanup Efforts

by

Monica A. Finnstrom

A Thesis Presented to the
Faculty of the USC Graduate School
University of Southern California
In Partial Fulfillment of the
Requirements for the Degree
Master of Science
(Geographic Information Science and Technology)

August 2019

Copyright © 2019 by Monica Finnstrom

To my parents, Richard “Rick” A. Finnstrom and Alice M. Finnstrom, for always supporting me.

&

In Loving Memory of my two Grandpas:

Victor P. Avila and Richard “Dick” Finnstrom

iv
Table of Contents
List of Figures ............................................................................................................................... vii
List of Tables ............................................................................................................................... viii
Acknowledgments.......................................................................................................................... ix
List of Abbreviations ...................................................................................................................... x
Abstract .......................................................................................................................................... xi
Chapter 1 Introduction .................................................................................................................... 1
1.1. Lead’s Impact on Human Health ........................................................................................3
1.2. Soil as a Source of Lead Exposure .....................................................................................4
1.2.1. How Lead Behaves in Soil .........................................................................................5
1.2.2. Exposure to Lead from Soil .......................................................................................6
1.2.3. Soil Pb Standards and BLL Guidelines .....................................................................7
1.3. Exide Facility Site Characterization and Background Information ....................................8
1.3.1. Conceptual Site Model ...............................................................................................9
1.3.2. Preliminary Background Lead Levels Study ...........................................................10
1.4. Research Goals..................................................................................................................11
1.5. Study Organization ...........................................................................................................12
Chapter 2 Related Work................................................................................................................ 13
2.1. Geostatistical Approaches to Understanding Soil Contamination ....................................13
2.2. The Effect of Spatial Scale on Analysis ...........................................................................17
2.3. Aggregation Methods........................................................................................................19
2.3.1. Using the Statistical Mean as an Aggregation Method ............................................19
2.3.2. Using Percentage as an Aggregation Method ..........................................................20
2.3.3. Using Risk Assessment to Inform an Aggregation Method ....................................21
Chapter 3 Methods ........................................................................................................................ 24
v
3.1. Soil Sample Data...............................................................................................................25
3.2. Data Preparation................................................................................................................29
3.2.1. Initial Data Preparation ............................................................................................30
3.2.2. Integrating Duplicates by Averaging .......................................................................34
3.3. Data Exploration ...............................................................................................................35
3.3.1. Data Quality Issues Regarding Recorded Locations ...............................................35
3.3.2. Outlier Assessment ..................................................................................................37
3.4. Study Areas and Scales of Analysis..................................................................................39
3.4.1. Study Areas ..............................................................................................................39
3.4.2. Scales of Analysis ....................................................................................................41
3.5. Creating Surfaces through Geostatistical Analysis ...........................................................42
3.5.1. Empirical Bayesian Kriging .....................................................................................42
3.5.2. Determining the Appropriate Model ........................................................................43
3.5.3. Determining the Cell Size ........................................................................................47
3.5.4. Post-Kriging .............................................................................................................47
3.6. Aggregation Methods........................................................................................................48
3.6.1. Aggregation Methods for Block Groups and Blocks ...............................................48
3.6.2. Aggregation Method for Parcels ..............................................................................51
Chapter 4 Results .......................................................................................................................... 52
4.1. Empirical Bayesian Kriging ..............................................................................................52
4.2. Aggregation Results for Block Groups and Blocks ..........................................................55
4.2.1. Northern Study Area ................................................................................................56
4.2.2. 1
st
Southern Study Area ...........................................................................................62
4.2.3. 2
nd
Southern Study Area ..........................................................................................66
4.3. Aggregation Results for Parcels........................................................................................70
vi
Chapter 5 Discussion and Conclusions ......................................................................................... 74
5.1. Discussion .........................................................................................................................74
5.1.1. Block Groups and Blocks Aggregation Discussion .................................................74
5.1.2. Parcels Aggregation Discussion ..............................................................................76
5.2. Limitations ........................................................................................................................77
5.3. Future Research and Implications .....................................................................................79
References ..................................................................................................................................... 83
Appendix A: EBK Model Results................................................................................................. 87

vii
List of Figures
Figure 1 Preliminary Investigation Area (PIA) determined by DTSC for the Cleanup Plan of
soil lead contamination ........................................................................................................... 2
Figure 2 Pictorial Conceptual Site Model for Exide ...................................................................... 9
Figure 3 Summary of Workflow .................................................................................................. 24
Figure 4 Exide Soil Pb Sample Locations ................................................................................... 32
Figure 5 Data Preparation Process ............................................................................................... 35
Figure 6 Outlier Soil Samples ...................................................................................................... 38
Figure 7 Study Area Boundaries .................................................................................................. 40
Figure 8 Empirical Bayesian Kriging in Geostatistical Wizard ................................................... 43
Figure 9 Aggregation Methods Workflow for Block Groups and Blocks ................................... 49
Figure 10 Northern Study Area Interpolation Surface ................................................................. 53
Figure 11 1
st
Southern Study Area Interpolation Surface ............................................................ 54
Figure 12 2
nd
Southern Study Area Interpolation Surface ........................................................... 55
Figure 13 Edge Effect of Boundaries ........................................................................................... 58
Figure 14 Block Group and Block Results for the Northern Study Area – Mean and HQ ......... 60
Figure 15 Block Group and Block Results for the Northern Study Area – Percent .................... 61
Figure 16 Block Group and Block Results for the 1
st
Southern Study Area – Mean and HQ ...... 64
Figure 17 Block Group and Block Results for the 1
st
Southern Study Area – Percent ............... 65
Figure 18 Block Group and Block Results for the 2
nd
Southern Study Area – Mean and HQ .... 68
Figure 19 Block Group and Block Results for the 2
nd
Southern Study Area – Percent .............. 69
Figure 20 Difference in Parcel Values for the Northern Study Area ........................................... 71
Figure 21 Difference in Parcel Values for the 1
st
Southern Study Area ...................................... 72
Figure 22 Difference in Parcel Values for the 2
nd
Southern Study Area ..................................... 73
Figure 23 Boundaries Completely Within Northern Study Area ................................................. 79
viii
List of Tables
Table 1 Main Attributes for Soil Samples and Comments ..................................................... 28-29
Table 2 Polygon Feature Layers Used for Scales of Analysis ..................................................... 42
Table 3 EBK Cross Validation Results for 30 Subset Size K-Bessel Model .............................. 47
Table 4 Minimum and Maximum – Sampled vs. Interpolated Values for Each Study Area ...... 53
Table 5 Comparison between the number of block groups and blocks within each study area
and the number of units selected for potential cleanup ......................................................... 56
Table 6 Summary Statistics for the Northern Study Area ........................................................... 57
Table 7 Summary Statistics for 1
st
the Southern Study Area ....................................................... 62
Table 8 Summary Statistics for 2
nd
the Southern Study Area ...................................................... 66
Table 9 Summary Statistics for Cell Values Extracted from Interpolated Surfaces at Parcel
Scale ...................................................................................................................................... 70
Table 10 Comparisons between EBK model results for the Northern Study Area ..................... 87
Table 11 Comparisons between EBK model results for the 1
st
Southern Study Area ................ 88
Table 12 Comparisons between EBK model results for the 2
nd
Southern Study Area ................ 88

ix
Acknowledgments
I am so grateful for my advisor, Dr. Karen Kemp, for providing me with valuable direction and
assistance. I appreciate all of her time and effort in helping me produce this thesis and am
thankful for all of those extended Blue Jeans meetings. I would also like to express my gratitude
to my committee members, Dr. An-Min Wu and Dr. Su Jin Lee, for providing me with valuable
insight and feedback. Without the inspiration for the topic and early guidance from Dr. Wu, this
thesis would not have come to fruition. I am also fortunate in having had the opportunity to work
with Dr. Lee as an Undergraduate Researcher for two years, helping me cultivate my interest in
GIS. I consider Dr. Kemp, Dr. Wu, and Dr. Lee as mentors and am grateful for all the
opportunities SSI has afforded me.
I would also like to acknowledge Dr. Jill Johnston for providing her expertise and
valuable insight into the subject matter. I would like to recognize the California Department of
Toxic Substances Control for making the data publicly available for use, which provided the
backbone for this thesis.
Most importantly, I want to thank my parents for being so supportive, encouraging, and
patient throughout this entire process. They are my true inspiration and their love and support has
motivated me to constantly be and do my best.

x
List of Abbreviations
AIN Assessor Identification Number
APN Assessor’s Parcel Number
ASE Average Standard Error
bgs Below Ground Surface
BLL Blood Lead Level
CDC Centers for Disease Control
DTSC Department of Toxic Substances Control
Exide Exide Technologies
EBK Empirical Bayesian Kriging
GIS Geographic Information System
HQ Hazard Quotient
IQ Intelligence Quotient
Mg/kg Milligram per kilogram
Pb Lead
PIA Preliminary Investigation Area
Ppm Parts per million
RMS Root Mean Square
SSI Spatial Sciences Institute
UCL Upper Confidence Limit
USC University of Southern California
USEPA United States Environmental Protection Agency
XRF X-Ray Fluorescence
xi
Abstract
Lead is a significant health threat to people, especially for children where elevated absorption of
lead into the bloodstream can cause permanent damage. One site for concern of lead exposure is
the surrounding communities of the retired Exide Technologies lead-acid battery smelter in
Vernon, California. The California Department of Toxic Substances Control (DTSC) is leading
an extensive cleanup effort to remove lead-contaminated soil from affected residences and
eliminate the negative health risks posed by the contamination. Soil sampling conducted for
approximately 8,500 parcels serves as the primary dataset for this research. While DTSC is
currently undertaking the cleanup process on a parcel-by-parcel basis, this thesis works toward
understanding the effect of geographic scale in the estimation of levels of lead contamination. It
also offers alternatives for identifying priority areas for cleanup by using various aggregation
methods and examining how the resulting values may be affected by scale. This research used
Empirical Bayesian Kriging to produce interpolated surfaces of lead concentration values.
Various aggregation methods were then utilized to aggregate the surfaces into easily defined
geographical units of different scales, including block groups, blocks, and parcels. The resulting
aggregation values include the mean, percent area, and a Hazard Quotient, an index value for
determining health risk. The results demonstrate that the larger areas of the block groups
moderate high lead concentration values and thus have lower overall aggregation values for the
block groups. In contrast, blocks have a greater tendency to include these high lead
concentrations in the aggregations resulting in higher overall values and wider ranges of values
for the blocks. This research provides alternative approaches for prioritizing the cleanup of
contaminated sites that could be more effective to address the health risks associated with
contamination and can be applied to other areas faced with the same problem in the future.
1

Chapter 1 Introduction
Lead (Pb) poses a significant environmental health risk to people. Children are especially
vulnerable because of their ability to absorb more ingested lead, which can lead to permanent
neurologic and developmental disorders (Wu et al. 2010). The recently closed Exide
Technologies lead-acid battery smelter in southeast Los Angeles has demonstrated a significant
threat of lead exposure to the community, with many possible health risks affecting the
community.
This 15-acre facility was one of only two west of the Rocky Mountains that could melt
lead from old car batteries for use in the production of new ones and had operated since 1922,
with Exide taking over the facility in 2000 (Barboza 2015). Smelter operations in violation of
numerous environmental regulations resulted in the release of lead, arsenic, and other heavy
metals into the local environment through aerial release. Violations included lead and acid leaks,
large cracks in the floors, hazardous levels of lead in the soil outside, and an overflowing pond of
toxic sludge (Barboza 2015).
Despite local, state, and federal officials citing the plant of these violations several times
over the years, the plant was allowed to remain open. In addition, the state allowed this smelter
to operate for 33 years without a full permit. The California Department of Toxic Substances
Control (DTSC) had known of these violations as well but failed to correct them. It was only
when the company was threatened with a serious federal investigation in 2014 that the company
was forced to sign an agreement to permanently close in 2015 (Barboza 2015).
Currently, the smelter site has become one of the most extensive cleanups of its kind by
targeting the removal of lead-contaminated soil from thousands of homes within an approximate
1.7-mile radius of the facility. The California Department of Toxic Substances Control is now
2

acting as the primary agency overseeing the cleanup efforts. The agency has established a
Preliminary Investigation Area (PIA) where soil sampling was conducted at approximately 8,500
properties within the community for heavy metal analysis, with an emphasis on the lead
concentration results (Figure 1). Figure 2 displays the PIA boundary and its regional location.
Both figures are directly sourced from DTSC’s Final Environmental Impact Report.

Figure 1 Preliminary Investigation Area (PIA) determined by DTSC for the Cleanup Plan of soil
lead contamination (Source: directly from the Final Environmental Impact Report DTSC 2017,
S1-5)

Although a majority of the properties are determined to have lead concentrations that
exceed California’s standard for residential soil (80 ppm), current funding limits the cleanup
efforts to only 2,500 of the impacted properties. The most recent Cleanup Plan prioritizes the
following properties for cleanup (DTSC 2017, ES-4):
3

• “Residential properties with a representative soil lead concentration [95%
upper confidence limit] of 400 ppm or higher; and
• Residential properties with a representative soil lead concentration of less than
400 ppm, but where any soil sampling result of 1,000 ppm or higher is
detected; and
• Daycare and child care centers with a representative soil lead concentration of
80 ppm or higher that have not yet been cleaned up.”
While these values used for prioritization provide sufficient measures of risk for lead
exposure, the parcel-by-parcel approach taken is inefficient. The Exide facility has been closed
for three years now. However, only 496 total parcels have been cleaned as of October 2018, with
166 of these properties being in the recent July 2017 Cleanup Plan. Meanwhile, residents have
grown frustrated with the slow pace of cleanup and have limited their children’s playtime
outside, as they are fearful for their children’s exposure to lead (Barboza 2018). Community
groups and county health officials have become critical of the Cleanup Plan and believe the
continuation of the cleanup on a parcel-by-parcel basis is insufficient, leaving contaminated
properties still intermixed with cleaned properties (Barboza 2018). This thesis attempts to
understand the effect of geographic scale in estimating the level of lead contamination and offers
other approaches for identifying priority areas for cleanup that could be more effective than a
parcel-by-parcel approach.
The following sections discuss important background information on lead’s impact on
human health, soil as a source of lead exposure, further context on Exide, and concludes with the
research goals and organization of this thesis.
1.1. Lead’s Impact on Human Health
With sufficient exposure, lead can wreak havoc on human health, exerting severe and
chronic health effects. Blood-lead levels (BLLs) typically reported as micrograms of lead per
deciliter of blood ( g/dL), is considered to be one of the primary biomarkers for lead exposure
4

(Juberg, Kleiman, and Kwon 1997). While adults absorb roughly 5 to 15% of ingested lead, only
retaining less than 5% of these lead values, children can absorb approximately 30 to 40% of
ingested lead because of metabolic and physiological differences. After absorption in the blood,
lead is “distributed primarily among three compartments – blood, soft tissue (kidney, bone
marrow, liver, and brain), and mineral tissues (bone and teeth)” (Juberg, Kleiman and Kwon
1997, 167). Lead initially absorbed by the bloodstream has the potential to be stored in bone for
years, which then becomes a long-term source of Pb back into the bloodstream (Laidlaw et al.
2017). Initial research has determined an association between elevated Pb in children and lower
Intelligence Quotient (IQ), behavioral problems, and learning disorders. The impacts are also age
dependent, with Pb presence in blood interfering with proper neuron formation in young
children. Recent research provides a greater understanding of chronic health effects with lead
exposure. There is evidence now that specifies strong associations between lead and many
diseases including “motor neuron disease, autism, preeclampsia developmental delays in
children, heart disease, ADHD, dementia, mental illness, and brain cancer” (Laidlaw et al. 2017,
16). With such negative health risks present from lead exposure, it is vital that DTSC address the
health risks posed by the Exide contamination in a timely and effective manner.
1.2. Soil as a Source of Lead Exposure
Although lead occurs naturally in the soil at concentrations ranging from 10 to 50 mg/kg,
human influence has caused lead to become more prominent in soils, especially in urban areas
(Stehouwer 2010). The widespread use of lead-based paint before the mid-1970s and the use of
lead additives in gasoline before the mid-1980s have contributed as major sources of lead in soil.
The peak of lead-based paint usage happened in the 1920s, while the peak of leaded gasoline
usage occurred in the early 1970s (Mielke and Reagan 1998). It is now estimated that leaded
5

gasoline left a residue of roughly 4 to 5 million metric tons of Pb in the environment (Mielke and
Reagan 1998).
Banned in 1978, lead-based paint is estimated to have been used on approximately 75%
of houses built before then (Stehouwer 2010). Although less widespread, airborne lead from
industrial sources such as smelters has significant potential to contaminate nearby residential
soils. Due to these sources, urban soils tend to have higher soil lead concentrations than normal
background levels or soils in rural areas (Markus and McBratney 2001; McClintock 2012;
Stehouwer 2010; Mielke and Reagan 1998). These urban lead concentrations can range from 150
mg/kg to even 10,000 mg/kg, if at the base of a home with lead-based paint (Stehouwer 2010).
1.2.1. How Lead Behaves in Soil
According to Chaney, Mielke, and Sterrett (1989, 2), research has “repeatedly shown that
small Pb-rich particles reaching the surface of a soil profile largely remain on or near the surface
for a prolonged period.” This is because organic matter particles and very fine clay hold onto soil
lead very tightly, in which the lead typically accumulates in the upper 1-2 inches of soil
(Stehouwer 2010). However, the mixing of soil by humans for the health of plants and the lawn
and the movement of soil by creatures such as earthworms allow lead to penetrate deeper into the
soil strata, which can make it more difficult to clean up. In addition, lead becomes most
concentrated in very fine soil particles. These particles tend to form airborne soil dust and can
stick to clothing and even skin.
Furthermore, chemical factors play an important role in the bioavailability of lead, as not
all of the lead is readily absorbed by plants or human bodies. The availability of lead in soil
largely depends on two factors: solubility (how much the lead dissolves in water) and how
tightly it is held by soil particles. Lead is more soluble and held less tightly in acidic conditions
6

(pH < 5), while less soluble and held more strongly in neutral (pH 5-6.5) to basic conditions
(pH > 6.5) (Stehouwer 2010). Since lead is held tightly by soil organic matter, lead availability
decreases as organic matter increases. The bioavailability of lead also has important implications
within the human body. Adsorbed Pb becomes soluble within the human stomach, due to the
acidic environment. When it enters the small intestine, the pH rises, in which soil then adsorbs
the Pb and reduces its solubility. Due to this adsorption equilibrium process, “higher soil Pb
concentration should strongly increase the bioavailability of soil Pb” (Chaney, Mielke and
Sterrett 1989, 1).
1.2.2. Exposure to Lead from Soil
In the past decades, researchers disagreed on soil being an important pathway for lead
exposure to humans, as some researchers argued that lead-based paint was the most important
source of lead exposure, as described by Mielke and Reagan (1998). However, further research
now proves soil to be a relatively significant source for lead exposure and must be considered in
order to have effective primary lead prevention (Mielke and Reagan 1998).
Pathways of exposure commonly emphasized for humans, especially children, are that of
ingestion and inhalation, with ingestion as the primary pathway (Laidlaw et al. 2017). Pb
exposure from soil can occur through various forms including “track-in via soil particles attached
to shoes, pets tracking Pb indoors on their fur, and direct contact with soils when the weather is
favorable and children play outdoors” (Laidlaw et al. 2017, 15). Younger children are also
known to ingest more dirt than adults relative to their body mass. Since children can absorb
greater amounts of lead than adults, Pb exposure from soil is deemed particularly dangerous for
this vulnerable population.
7

In studies done after the use of leaded gasoline was prohibited, it was found that blood Pb
levels declined concurrently with the decrease of air lead from the gasoline sources (Laidlaw et
al. 2017). Similarly, child blood Pb levels declined sharply after the Bunker Hill Pb smelter in
Idaho was closed and Pb aerosols therefore ceased. These studies indicate the importance of
inhalation as a pathway for exposure and not to be overlooked.
A small, yet growing number of studies have examined the spatial relationships between
soil lead and BLLs. One such study in New Orleans utilized 5,467 soil Pb samples spanning 286
census tracts and geo-referenced BLL data for 55,551 children in New Orleans (Laidlaw et al.
2017). The results of the study indicated that the BLLs of the children are spatially associated
with the soil Pb levels, in which 67% of the variation in children’s BLL could be explained by
soil Pb sample location variables.
1.2.3. Soil Pb Standards and BLL Guidelines
Having established soil as an important source for lead exposure, what soil Pb standards
and BLL guidelines are in existence? Current soil standards are inadequately developed, ranging
from 20 mg/kg to over 1,000 mg/kg (Laidlaw et al. 2017). However, the most widely cited soil
Pb standard that a majority of guidelines are aligned with is the U.S. EPA 400 mg/kg value,
which is not a health-based standard. As mentioned by Laidlaw (2017), several studies and
reviews suggest health-based soil Pb guidelines, correlating to target levels of BLL. A prominent
review proposed a standard of <100 mg/kg centered on evidenced-based data from studies and
the assumption that 10 g/dL BLL is safe (this is now considered to be still not safe according to
clinical studies) (Laidlaw et al. 2017). A New Orleans study in 1999 determined a soil Pb
guideline of around 80 mg/kg, with the goal of preventing Pb exposure  10 g/dL for children.
A repeat of this study ten years after Hurricane Katrina revised this soil Pb standard value to  40
8

mg/kg, with the goal of preventing Pb exposure  5 g/dL for children (a version of a revised
CDC BLL reference value).
According to the California Code of Regulations, lead contaminated soil is defined as “…
bare soil that contains an amount of lead equal to, or in excess of, four hundred parts per million
(400 ppm) in children’s play areas and one thousand parts per million (1,000 ppm) in all other
areas” (DTSC 2017, S2-24). DTSC’s screening level for lead in residential properties is 80ppm,
based off a revised toxicity evaluation for lead that now determines the threshold for children’s
blood level concern as 1 g/dL, as opposed to the previous 5 g/dL (DTSC 2017). Because
children are the most sensitive group to lead exposure and are the group used for determining
these standards, the places where they spend most of their time present the greatest risk for
exposure, which can be considered as sensitive land uses. A majority of the parcels within the
Preliminary Investigation Area are residential, parks, schools, child care facilities, and day care
centers, all of which are sensitive land uses and should adhere to the guidelines.
1.3. Exide Facility Site Characterization and Background Information
To better understand the nature of the site and critical background information, this
section provides a description of the conceptual site model as laid out by DTSC in their Final
Removal Action Plan (Cleanup Plan) and describes a preliminary background lead levels study
performed by Exide. The conceptual site model aids in planning efforts by determining exposure
pathways from the facility. The preliminary background lead levels study, carried out by Exide
and enforced by DTSC, ensures that the contribution of background lead concentrations have
been considered and indicates Exide as a primary source for the lead concentrations.
9

1.3.1. Conceptual Site Model
Conceptual site models are important planning tools utilized to support the decision-
making process of impacted properties, such as that of the lead-contaminated properties in the
PIA (DTSC 2017). A conceptual site model provides insight into contaminant sources and
release mechanisms by describing the potential movement of chemicals throughout a particular
area and its exposure pathways to potential affected populations. In the case of the Exide site,
lead-acid battery activities from the years of operation, inadequate waste management practices,
and insufficient air pollution control are the likely causes for the lead contamination on affected
properties (DTSC 2017). The primary sources for this contamination include aerial releases from
the smokestacks and facility in general, as well as trucks transporting the materials (Figure 2).

Figure 2 Pictorial Conceptual Site Model for Exide (Source: DTSC 2017, S2-19). Note: “bgs”
stands for below ground surface.
Within the conceptual site model, it is acknowledged that other lead sources are also
possible, including lead-based paint, leaded gasoline, and historical industrial operations from
other facilities operating in the same industrial area. The key pathways to exposure identified
include particulate inhalation, dermal contact, and accidental ingestion. The main population
identified for lead exposure is the residents within the PIA, including sensitive individuals such
10

as children and pregnant women. The most likely source of ongoing exposure involving lead
from the Exide facility during years of operation is surface soil and soil or dust around plants.
1.3.2. Preliminary Background Lead Levels Study
DTSC had Exide do a background lead study in 2014 to ensure that the selected soil
screening level of 80 ppm for cleanup did not fall below background, or typical, levels of lead in
soil within the area. For this background study, Exide chose a background area located roughly
14 miles south of the Exide facility in the City of Long Beach to serve as a control for the soil
samples. The area was “residential and considered similar to the PIA, but without potential lead
contamination from the former Exide facility, because of its proximity to major freeways, a
historically industrial area, a sizable rail yard with intermodal facility and switching yard, and
housing of similar size and density” (DTSC 2017, 25-26). As a part of the study, 19 residential
properties were sampled. According to the results, surface lead concentrations ranged from 29 to
195 ppm, with a median value of 54.8 ppm and a representative soil lead concentration of 76.6
ppm. The representative soil lead concentration is a property-wide lead concentration calculated
from the soil sample results and is more health protective than an average of the soil sample
concentrations. This term is further explained in Chapter 3. These concentrations are
significantly lower than the lead concentrations sampled from soil in the Initial Assessment
Areas and are below the DTSC screening level for lead. These conclusions mean that the
representative soil lead concentrations within the PIA are unlikely attributable to background
concentrations; rather, they indicate Exide as a primary source for the lead concentrations.

11

1.4. Research Goals
This thesis works toward understanding the effect of geographic scale in estimating levels
of lead contamination to help determine the effectiveness of using a scaled-up approach for
directing the decision-making process for cleanup efforts. Currently, DTSC is undertaking the
cleanup process on a parcel-by-parcel basis, with only parcels where samples exceed defined soil
lead concentration limits are chosen for cleanup. Since the distribution of lead concentrations in
the soil is a continuously varying surface, this research approaches the problem spatially. It
considers alternative ways to delineate areas for cleanup based on estimating single summary
values from an estimated surface of varying lead concentrations in the soil for various sized
spatial units. If larger areas can be identified as having sufficiently higher lead concentrations
than other parts of the study area, it could help decision-makers determine more efficient ways to
identify priority areas for cleanup.
Thus, the research aims to answer these questions:
1. Based on the available sampled data, what is the continuous spatial distribution of
lead concentrations in the soil within the Exide Preliminary Investigation Area?
2. How does scale (i.e. size of the spatial zones) affect the values allocated to various
sized zones that can be used to identify priority areas for cleanup of lead
contaminated soils?
While DTSC is using a parcel level representative soil lead concentration to determine
priority areas for cleanup, this thesis aims to provide alternatives for identifying priority areas
through the use of numeric aggregation methods. These aggregation methods provide other
estimates of lead concentration values on which to base the decision for priority areas of cleanup,
including the statistical mean of values for an area, the percentage of an area where lead
12

concentration values exceed the national standard of lead in soil concentration of 400 ppm, and a
Hazard Quotient, an index value that indicates the level of risk to human health. In addition, this
thesis aims to look at how scale affects these values when assigning them to different
geographical units of differing scales. These geographical units include block groups, blocks,
and parcels, those typically utilized in policy implementation. When looking into how scale
affects these values, the results provide insight into which scale may be best for managing
cleanup of lead contaminated soils. Although DTSC may have already decided how they are
approaching the cleanup for Exide, this research provides other approaches for prioritizing
cleaning up of contaminated sites that could be applied to other areas faced with the same
problem in the future.
1.5. Study Organization
This study includes four additional chapters. Chapter Two provides a literature review
regarding approaches to understanding soil contamination and highlights studies that use
geostatistical approaches. This chapter also presents an overview into how scale affects spatial
analysis and examines various studies that inform the methodologies for this thesis. Chapter
Three provides a detailed description of the soil sample dataset used in this study and presents
the methodology, which includes the use of Empirical Bayesian Kriging to create predictive
surfaces of lead concentrations followed by various aggregation methods used to aggregate the
surface results into the different scales of analysis. Chapter Four provides the results of the
analysis. Chapter Five concludes with further discussion of the results and their implications,
limitations of the study, and suggestions for further research.

13

Chapter 2 Related Work
Studies utilizing GIS and spatial analysis to examine soil contamination of heavy metals,
including lead, have increased through the years. This section starts by discussing literature that
has taken various approaches to understanding soil contamination, with a focus on geostatistical
methods. The next part of this section examines scale by looking at literature concerning how
scale affects spatial analysis. Different aggregation methods for the results of geostatistical
processes into different units of analysis are explored as well. This determination requires
examining literature from different fields of study, those other than contamination studies, to
look for existing frameworks that can be applied to soil contamination.
2.1. Geostatistical Approaches to Understanding Soil Contamination
The use of GIS and geostatistical approaches are increasingly being used to study soil
contamination of heavy metals, including lead. Mapping contaminant distribution, such as lead,
makes detecting patterns within these distributions easier and allows for the identification of
areas that contain potentially hazardous concentrations by showing how the contaminant changes
with space (Markus and McBratney 2001). Knowledge of the spatial distribution of a
contaminant is an essential factor in site assessment and for any succeeding risk assessment. Due
to expensive costs of large sampling efforts and chemical analyses that follow, a technique
known as interpolation has been used to estimate concentrations at locations between sampling
sites. In particular, Kriging, a geostatistical method of interpolation, has become popular within
studies to map the distribution of lead and other heavy metal concentrations in soil (Markus and
McBratney 2001). In addition to providing a predicted surface of concentration values, Kriging
also provides other useful information, including a measure of uncertainty associated with the
14

predicted values. This makes Kriging a desirable method for interpolation in soil contamination
studies.
The following studies provide examples of how Kriging was used in understanding soil
contamination. One study examined soil metal contamination of Cd, Cr, Cu, Ni, Pb, and Zn in
the highly urbanized Kowloon area of Hong Kong, generating geochemical maps via a Kriging
method that shows hot-spots of heavy metal contamination in soils (Li et al. 2004). The
researchers used Principle Component Analysis and Cluster Analysis to determine significant
spatial relationships for the metals. Further overlay analysis, which included comparing the
proximity of hot-spot results to other features, concluded that road junctions, major roads, and
industrial buildings were possible sources for the metals in urban soils.
Another study that utilizes geostatistics focused on the soil heavy metal concentrations in
the rice paddy fields in the Hangzhou-Jiaxing-Huzhou (HJH) Plain (Liu, Wu and Xu 2006). The
study employed ordinary Kriging and lognormal Kriging to map the spatial patterns of heavy
metals including Cu, Zn, Pb, Cr, and Cd. Using 450 soil samples spread through the study area,
contour maps of the heavy metals were produced. The authors use these maps of the spatial
distributions as a way to quantify risk assessment. They argue that for a variety of practical
problems in environmental management concerning heavy metals in soil and their relative
threshold values, information is needed at unsampled sites (Liu, Wu and Xu 2006). The resulting
maps produced using the geostatistical methods serve as a solution and inform risk assessment of
environmental pollution and associated decision-making.
Kriging has also been used as the geostatistical mapping method in studies that center on
smelter-impacted soils. For example, multivariate and geostatistical analysis was used to
investigate the spatial variation of heavy metals in the soils of a mining-smelting area in the
15

Hunan Province of China (Wei, Wang and Yang 2009). Ordinary Kriging for As, Cd, Pb, and Zn
and Inverse Distance Weighting for Cr and Cu were performed, demonstrating hot spots and
similar dispersion patterns that indicated the smelter and mining source areas.
In another study concerning a former coal-mining area in France, ordinary Kriging was
also used in its investigation into the spatial variability of the pseudototal concentrations of Cd,
Pb, and Zn (Pelfrêne, Détriché and Douay 2014). The researchers used exploratory spatial data
analysis to characterize data distribution to aid in the Kriging process. Unlike the previous
studies, this one combined its geostatistical analyses with the incorporation of oral
bioaccessibility to improve the assessment of the population’s exposure to metals in the smelter-
impacted soils. As exemplified in these studies, the use of Kriging has become a primary
approach in understanding the distribution of soil contamination, with ordinary Kriging proving
to be the most commonly used type of Kriging.
In regard to this thesis, Kriging serves as the main method for analyzing the spatial
distribution of lead in the soil of the PIA. Although Kriging is widely used in soil contamination
studies, the ease of implementing Kriging within a GIS environment, without the full
understanding of its various factors, can often produce unreliable and possibly misleading results
(Oliver and Webster 2014). Users must therefore take caution when carrying out the Kriging
process, making sure they understand all of the important data considerations. Sound Kriging
requires a plausible function for the variogram. A paper by Oliver and Webster provides
guidelines for choosing suitable functions, among other important considerations with respect to
Kriging. This paper, as well as other documentation, are key in guiding this project’s appropriate
use of Kriging.
16

In addition to Kriging, a variety of other methods are used to characterize the spatial
distribution of lead in soils and were informative for this thesis. One study investigated the
process of identifying pollution hotspots using Pb concentrations in the urban soils of Galway,
Ireland and determining which factors influence hot spot identification (Zhang et al. 2008). The
researchers in this study used local Moran’s I-test to identify pollution hotspots of Pb
contamination, classifying them into spatial clusters and outliers. Factors influencing the
determination of results included definition of weight function, data transformation, and the
existence of extreme values.
In another study, researchers analyzed the bioavailable soil lead concentrations in Los
Angeles. They sought to understand the contribution of lead to soil from residential lead-based
paint and from cars using leaded-gasoline based on traffic variables (Wu et al. 2010). Utilizing
several variables including house age, traffic index, proximity to freeways and highways, and
land use patterns, the researchers used multi-variable regression models to explain soil lead
concentrations and mapped the spatial distribution of these concentrations. This study provides a
good example of understanding lead contamination in soils in Los Angeles. Although the lead in
the soil near the Exide facility could possibly be explained by a few of these variables, Exide is
considered the primary source of soil lead in the surrounding area. Regression models could be
applied to this study of lead within the PIA to determine the extent of each variable’s
contribution to lead contamination. However, this addition is outside the scope of this thesis
since the focus is on exploring the effect of the scale of spatial units for which values are
aggregated from the soil contamination surface.
17

2.2. The Effect of Spatial Scale on Analysis
Much research in the social sciences focuses on how structural characteristics affect
various outcomes, with one form studying if structural characteristics of neighborhoods affect
various aggregate outcomes (Hipp 2007). According to Hipp, despite “the variety of research
paradigms focusing on the importance of neighborhoods, a commonality of many studies is that
less attention is paid to the appropriate level of aggregation for such neighborhood effects” (Hipp
2007, 659). Although studies attempt to understand the effect of neighborhoods on outcomes, the
definition of neighborhood itself tends to get lost in the methodological details. Studies have
consistently used various geographic units including blocks, block groups, tracts, and zip codes
to test the effects of structural characteristics.
However, this strategy hardly considers if the geographic unit is appropriate for the
outcome of interest (Hipp 2007). The significance and challenge of determining appropriate
aggregation levels is inherent in the modifiable areal unit problem (MAUP). The issue of MAUP
persists in any study that analyzes a difference between scales and affects the decision of
aggregation level. Whereas Hipp attempts to determine the appropriate geographic level of
aggregation using crime and disorder as an example, this thesis seeks to determine the best
geographic level of aggregation for cleanup of soil lead contamination, using Exide as a case
study.
Similar to Hipp’s study (2007), Root (2012) also points out the weakness a lack of
definition of neighborhood can have on health studies. Root (2012) states that it results in little
consideration of the spatial scale at which socioeconomic factors influence a certain health
outcome. While neighborhood studies have been popular in health literature, earlier studies
hardly attempted to understand the limitations of using a single geographic unit for analysis. In
18

the case of Exide, this limitation is present in that DTSC has only considered the parcel scale as
the single geographic unit for determining the prioritization of cleanup. Through her study in
examining the role of spatial scale in neighborhood effects on health, Root explores how
geographic and statistical methods can aid in defining neighborhoods and selecting the
appropriate scale. She accomplished this through a case study that examined the relation of
socioeconomic status (SES) to orofacial clefts at different spatial scales: 4,000m Buffer, census
tract, and census block group (Root 2012). Similar geographic and statistical methods can be
used to explore different spatial scales that have not been considered by DTSC, allowing for the
recommendation of possible better approaches to prioritizing cleanup.
While Hipp (2007) and Root’s (2012) studies emphasize the problem of limiting studies
to a single geographic unit for analysis, other studies have demonstrated how the use of multiple
scales can actually change the analytical results. For example, Su and Ang examine the effects of
spatial aggregation on energy-related CO2 emissions embodied in trade using the input-output
analysis, with China as the study area (Su and Ang 2010). They used three different spatial levels
– China as a whole, 3 regional groups, and 8 regions – for analysis and concluded that the
embodied emissions depend greatly on the spatial aggregation scheme.
Another study explored the modifiable areal unit problem in the relationship between
exposure to NO2 and respiratory health (Parenteau and Sawada 2011). According to the
researchers, previous Canadian population health studies have succumbed to the limitation of
using a single geographic unit for analysis by using census tracts as proxies for neighborhoods.
The researchers attempt to remedy this by providing a study on the relationship between NO2
and respiratory health using three different spatial structures to demonstrate the effects of spatial
units on analytical results. They used Moran’s I and regression analysis for each of the units of
19

analysis; however, found no significant effect of NO2 exposure on respiratory health (Parenteau
and Sawada 2011). Both of these studies demonstrate the importance of using multiple scales for
analysis to account for the differences in results. This thesis attempts to take that into
consideration by using different analytical scales for determining an approach to prioritizing
cleanup.
In a study that assesses soil lead contamination in Oakland, California, multiple scales
were used to determine the extent of contamination being an obstacle to the expansion of urban
agriculture within the city (McClintock 2012). Through mapping soil samples via GIS and
reconstructed land histories, McClintock performed spatial analysis at city, neighborhood, and
site-specific scales. Clusters of Pb contamination at the city and neighborhood levels were found
to be related to land use history, while contamination at the site-scale revealed high variability
(McClintock 2012). Although McClintock’s study on soil contamination employs multiple
scales, it is different from the others in that sampling methods were used to differentiate between
scales, rather than data being aggregated into units.
2.3. Aggregation Methods
Census geographic units of varying sizes were explored in this study to investigate their
use for efficient targeting of cleanup operations. However, summarizing the values across a
continuously varying surface, estimated through interpolation as a raster layer, into a single value
for each unit examined requires determining appropriate methods for this aggregation.
2.3.1. Using the Statistical Mean as an Aggregation Method
Researchers undertaking a quantitative estimation of health risk to residents from
contaminated groundwater and soils in the Slovak Republic utilize municipality, district, and
regional boundaries to map potential risk areas (Fajčíková et al. 2014). In their analysis, they
20

calculated the arithmetic means of each studied compound for each administrative unit of
analysis based on interpolated gridded data. This example serves as one potential method for
aggregating the geostatistical results of soil lead contamination. The researchers also convey that
the use of basic administrative units to map potential risk areas provides for easy discussion with
policy- and decision-makers (Fajčíková et al. 2014). This provides support for the use of census
geographic units for this project.
In a thematically different study using similar approaches, Liang and Weng (2008)
utilized GIS data and remote sensing to perform a multi-scale analysis of the urban heat island
(UHI) in Indianapolis using the census-based units of block, block group, and tract. Selected
variables for the urban landscape and land surface temperatures (LST) were aggregated at the
various census levels, in which correlation analysis and linear regression modeling were then
performed at each level. This aggregation into the various census levels allowed for the
examination of the scale effect, or the sensitivity of the relationship to aggregation. The study
also worked to “establish a method for evaluating urban thermal environment within
geographical (census) units” (Liang and Weng 2008, 129). It was successful by allowing for the
calculation of census-based variations of land surface temperatures. The researchers aggregated
numerous vector and raster variables into the three census levels. For land surface temperatures
provided as raster data, the statistical mean values of LST were calculated for each census zone.
This approach is similar to the previous study and supports using the statistical mean value as
potential method for aggregation.
2.3.2. Using Percentage as an Aggregation Method
In a similar fashion to the area proportions used for grouping vector and raster features in
Liang and Weng’s study, the use of percentages could be used as a potential aggregation scheme.
21

A study carried out by Hanna-Attisha and her colleagues (2016) analyzed the differences in
pediatric blood lead level incidence pre- and post- the Flint drinking water crisis. Percentage of
elevated blood lead levels were assessed from both time periods and identified geographical
locations through analysis (Hanna-Attisha et al. 2016). As part of the methods, ordinary Kriging
was used to create a predicted surface of child blood lead levels throughout Flint. In their map of
the Kriging results, the researchers overlaid the results with Flint City Wards, to determine the
percentage of the ward areas where water samples exceeded 15ppb. This seemed to be used as a
validation method for their predicted surface. However, this approach could be applied to the
aggregation of soil lead distribution by indicating the percent of the area in each geographical
unit where the lead concentrations exceed a certain threshold, such as 80ppm or 400ppm, the
values determined to be significant for lead exposure, based on California and EPA guidelines,
respectively.
2.3.3. Using Risk Assessment to Inform an Aggregation Method
A subset of literature examining contaminated sites uses a combination of spatial
statistics and health risk assessment to aid in remediation decisions. This subset offers insights
into different approaches when tackling contamination and remediation issues. In their study,
Gay and Korre propose a methodology that combines spatial statistical methods with quantitative
probabilistic human health risk assessment of produce an assessment of risk to human health
from contaminated land (Gay and Korre 2006). A population is mapped across the contaminated
area and an adaptation of the Contaminated Land Exposure Assessment model is used to
probabilistically calculate the intake of soil contaminants by individuals. Essentially, exposure
maps are combined with population data to provide risk evaluation and enable efficient risk
reduction measures.
22

The use of geostatistics for helping to characterize risk assessment has also grown and
can be seen in the following studies. Guastaldi and Del Frate utilize geostatistical simulations of
pollutants measured in terrain samples to assess the uncertainty for risk analysis of a
contaminated industrial site in northern Italy (Guasaldi and Del Frate 2011). These geostatistical
simulations quantify the risk through simulation of possible realities and how many exceed
contamination thresholds; they provide a means for visualizing risk. These models are then used
in aiding the decision-making processes of deciding areas targeted for remediation.
A prior study suggests that sample-specific risk can be used in a geostatistical procedure
to produce maps of block-specific risk at the site to aid in decision-making processes with
remediation strategies (Ginevan and Splitstone 1997). The researchers employ probability
Kriging of sample-specific risks, with five indicator levels of risk for the Kriging procedure. This
resulted in the estimation of conditional cumulative distribution functions for each of 3,911
conceptual remediation blocks, which defined the surface soil of the site. Median and 95% upper
percentile risks of the blocks were then mapped according to the indicator levels.
A commonality among many of the risk assessment studies include the use of a hazard
quotient (HQ) and/or a hazard index (HI) to quantify the means of risk. This suggests a third
method for aggregating the geostatistical results of soil lead contamination. Zhao et al. (2012)
examined heavy metal contamination near the Dabaoshan Mine in Southern China. The
researchers produced spatial distribution maps of the various metals and used land uses and
uptake models to create a dose-response model for human health risks (Zhao et al. 2012). The
hazard quotients for the various metals were then mapped on continuous surfaces, in which the
spatial patterns indicated Cd as the primary pollutant contributing to health risk for humans.
Other studies previously discussed also engage the use of hazard quotients and/or hazard indexes
23

in components of their analysis (Fajčíková et al. 2014); (Pelfrêne, Détriché and Douay 2014).
The simple equation for hazard quotient is
Hazard Quotient = Exposure Concentration / Reference Concentration (1)
The reference concentration is typically determined from credible sources, such as the EPA. If
HQ < 1, then there is no risk to human health. If HQ > 1, then some degree of risk exists. Hazard
quotients can be calculated for each parcel and combined with the percentage method as a
combined approach. This would result in values showing the percentage of area with hazard
quotients over 1. However, it is important to note that, like all the other methods, risk
calculations will change as geographic boundaries change (Fajčíková et al. 2014). Using the
municipality, district, and regional boundaries in their study, Fajčíková and colleagues pointed
out that these boundaries bear no relationship to the geochemistry of contamination, in which the
resulting spatial associations of the studied chemicals could differ from actuality. This lends
urgency to the need to consider spatial scale in the assessment of risk.

24

Chapter 3 Methods
The purpose of this study is to provide valuable alternatives for identifying priority areas for
cleanup by using various aggregation methods and examining how the resulting values may be
affected by the scale of spatial units considered. Here, block groups, blocks, and parcels were
used in this analysis to provide insight into what scales are best for planning and managing
cleanup of lead contaminated soils. This chapter begins with a description of the soil sample data
used for this study and details the process of preparing these data for analysis, following with a
section on data exploration that delves into data quality issues and the importance of outliers on
this analysis. Next, the study area boundaries and the scales for analysis are established. The
study then employs Empirical Bayesian Kriging for the development of a suitable Kriging model
to create interpolated surfaces of lead concentration values, using cross-validation as a guide.
These surfaces are then utilized in various aggregation methods, resulting in the mean, percent
area, and Hazard Quotient (HQ) values for spatial units at each scale. Figure 3 displays the
overall workflow for this thesis.

Figure 3 Summary of Workflow

25

3.1. Soil Sample Data
This research utilizes soil sample data collected as part of the Exide Preliminary
Investigation and provided by DTSC. The Preliminary Investigation Area (PIA) covers an area
surrounding the smelter site within an approximately 1.7-mile radius. The dataset was provided
by DTSC in the form of a table in Excel format. In this Excel spreadsheet provided by DTSC,
there are four sheets/tabs: Data Table (contains all of the soil sample data and attributes), Data
Legend (provides additional clarifying information about the data and its attributes), List of
Acronyms (acronyms used in the spreadsheet), and Multiple APN Properties (listing a primary
APN to represent properties that contain multiple parcels).
Soil was sampled on approximately 8,500 properties within the PIA, with a total of 328,
069 soil samples in the Excel table. Roughly 10 to 15 soil samples were taken from each
property at several locations in the front, back, and side yards. The sampling matrixes included
soil and paint, with soil being the focus for this analysis. The samples were either sent to a
laboratory for chemical analysis or analyzed in the field using X-Ray Fluorescence (XRF).
Approximately 20% of the samples from each property were analyzed in a laboratory.
According to the Final Sampling Workplan for DTSC, there were several steps taken to
select the soil sampling areas. This included choosing sampling locations that targeted “bare
exposed soils that have not been recently disturbed and open grassy areas away from structures
or thick trees” and areas “in which maximum deposition and exposure potential are likely”
(DTSC, Final Workplan 2015, S3, 1). It was noted that approximately 15 sample locations would
be taken at each property, marked by pin flags. The sampling distribution criteria is as follows:
“five locations in the front yard; five locations in the back yard; five locations distributed in drip
zones, near downspouts, side yards, and other open bare soils areas; and two additional
26

contingency sample locations if a play area is present” (DTSC, Final Workplan 2015, S3, 2).
Samples for all sample locations were collected at the 0- to 3- inch depth interval, with additional
depth intervals taken (up to 18 inches) for the two locations detected with the highest lead
concentrations. Information regarding how coordinates were captured is provided in the Final
Sampling Workplan as follows: “The location of each sample will be measured from a reference
point at the property and marked on a field sketch. In addition, coordinates of each soil sampling
location will be recorded using a global positioning system (GPS) unit. GPS coordinates of each
sampling location will also be recorded in the field notes” (DTSC, Final Workplan 2015, S3, 3).
Furthermore, the Final Sampling Workplan for DTSC indicates that sampling reports
were also provided for each property. These sampling reports included the following
information: “a description of the property, a map showing the sampling locations, coordinates
of the sampling locations, sampling results in tabular form and electronic format (MS Excel),
screening of the results against criteria established in the Workplan to determine if further action
is required at the property, photographs of the sampling locations, laboratory analysis reports, an
evaluation of the quality of the data, [and] an explanation from any deviation from the
Workplan” (DTSC, Final Workplan 2015, S4, 1).
In addition to providing the locational coordinate information, several attributes are
associated with each of the sample results in the Excel table. A unique property ID assigned by
the sampling contractor and address locations are given for each of the sampled sites, along with
the facility type, such as if it is residential, a child care center, commercial, a park, school, or
other type. The primary Assessor Parcel Number identified for the properties is also listed for
each soil sample. Detailed information about the samples is also provided: Sample ID, dates
samples were collected, their location, depth taken (in inches), sampling matrix, and location
27

codes (specifying location on properties). In addition to these attributes, characteristics of the
analysis are also included: analysis type (in the field or lab), sample type, test type, metal
analyzed, the analysis results and their units (ppm, mg/kg, and mg/cm2) and detection limit. A
list of the main attributes and their descriptors can be found in Table 1, with information taken
directly from the ‘Data Legend’ within the source Excel document.
Using all samples from each property, a Representative Soil Lead Concentration was also
determined by DTSC for each property, summarizing all the lead values into one value assigned
to each property. This value was used by the DTSC in prioritizing cleanup by parcel. The value
was calculated using the USEPA’s ProUCL software and is the 95 percent upper confidence
level limit of the lead concentrations, which is typically greater than the average concentration of
all samples from the property, but below the maximum concentration. It is important to note that
the 95% UCL values were only calculated when there are more than 8 samples per property.
When there are less than 8 samples per property, the maximum sample concentration value in the
0-3-inch depth interval is used as the Representative Soil Lead Concentration. This
Representative Soil Lead Concentration is also listed as an attribute for each soil sample result
and more details concerning this value can be found in Table 1. Although the dataset includes
samples taken at various depths within the PIA and information about multiple metals, this
research used only data on lead in samples collected from a depth of 0-3 inches below ground
surface.

28

Table 1 Main Attributes for Soil Samples and Comments (most of the table is sourced directly
from the ‘Data Legend’ tab in the original source Excel spreadsheet)
Category Field Name Comments
Property
Information
Property ID This number is a unique identifier assigned by the sampling
contractor
Address The address of the property (separated into different columns for
numeric address, street name, city, state, and zip code)
Primary APN Primary Assessor Parcel Number (APN) associated with a
Property ID (Some properties have multiple APNs; for example,
some schools have many APNs but for sampling and cleanup
purposes they are considered as a single property)
Sample
Location
Information
Sample ID Each Sample collected has a unique identification number
(XXXXX)-(XX)-(XX).

Typically, this is the Property ID, followed by the location
number at the property, followed by the depth the sample was
collected (refer to "Depth" column for depth definitions).

The location number is a sequentially assigned number up to the
total number of samples collected at a property.
Sample Location This is the numeric identification number for where samples
were collected at a property (XXXXX)-(XX). The first part is
the Property ID and the second is the location number at the
property.

The location number is a sequentially assigned number up to the
total number of samples collected at a property
Location Code The type of location on the property where a sample was
collected.
BY = Back Yard, Chip = Paint sample, DL = Drip Line, FY =
Front Yard, G = Garden, GCComp = Composite Sample, O =
Other, P = Property Boundary, PA = Play Area, SY = Side Yard
Analysis
Information
Analysis Type Samples were either analyzed in the field (XRF) or sent to a
laboratory. FI = Field Sample (XRF) or Lb = Laboratory Sample
Sample Type Sample type indicates whether the sample was a normal sample
or a duplicate sample.
Metal Analyzed Metal that either the soil or paint was analyzed for: Lead,
Antimony, Arsenic, Cadmium, Copper, Zinc
Result Result of the XRF or laboratory analysis.
Result Units The unit the results are presented in: ppm = parts per million
(soil analyzed by XRF), mg/kg = milligram per kilogram (soil or
paint analyzed in a laboratory), mg/cm2 = milligram per square
centimeter (paint analyzed by XRF)
29

Category Field Name Comments
Representative
Soil Lead
Concentration
Information
Representative
Soil Lead
Concentration
The representative property-wide lead concentration is based on
the 95 percent upper confidence limit (UCL) statistical analysis,
which is typically greater than the average concentration of lead
on the property, but below the maximum concentration.

The 95 percent UCL concentration represents the concentration
that would be greater than the true mean value of the samples on
the property 95 percent of the time. The 95 percent UCL and
maximum lead concentrations are used to determine the level of
exposure to lead at each property.

95 percent UCL value calculated by ProUCL using the lead soil
results for the 0-3" depth. Properties with fewer than 8 samples
from 0-3" do not have 95 percent UCL values calculated. For
these properties the maximum concentration in the 0-3" depth
interval was used as the representative lead concentration.

Some properties have fewer than 8 samples because there are
often size limitations of the surface area of unpaved surfaces at a
property. In instances where properties have small surface areas
from which to collect samples, professional judgment is used to
determine an appropriate number of samples to collect to
represent the conditions at a property.
Coordinate
Information
Longitude and
Latitude
On a map, x,y coordinates are used to represent features at the
location they are found on the earth's spherical surface.

Georeferencing in progress = Using Geographic Information
System (GIS) software to determine the X (longitude) and Y
(latitude) coordinates of sampling locations from the maps
provided in the sampling reports

3.2. Data Preparation
This section discusses how the soil sample data from the Excel spreadsheet were
prepared for use in a GIS and the steps involved in the preparation for their use in the study’s
analysis. Data preparation included the geocoding of some sample points and a reduction in the
dataset to only the attributes that were needed for the analysis. It also included the integration of
duplicate soil samples through a process of averaging. Most of the data preparation steps were
carried out using ArcGIS Pro.
30

3.2.1. Initial Data Preparation
Since the soil sample data were supplied in table format, some initial preparation was
needed so that the data could be used in a GIS. This included converting the Longitude and
Latitude columns from the Excel data table from text to numeric data type and saving the file in
CSV format to be imported into ArcGIS Pro so that it could be converted into vector-based
discrete points.
Out of the 328,069 soil samples, 322,761 samples had coordinates and could be directly
uploaded as points in ArcGIS. However, 5,308 samples of all metal types and depths were
missing coordinates, listed as ‘Georeferencing in progress’ in the original Excel data table, and
some had coordinates that did not fall within the PIA boundary, ranging from some just missing
the boundary to some points appearing in the ocean. The phrase ‘Georeferencing in progress’
was defined in the ‘Data Legend’ as “using Geographic Information System (GIS) software to
determine the X (longitude) and Y (latitude) coordinates of sampling locations from the maps
provided in the sampling reports.” It is unclear what is meant by this term in relation to the
location coordinates. The points with missing coordinates and inaccurate coordinates were
extracted from the main dataset in Excel and geocoding was carried out in ArcGIS for all
samples that could not be directly or correctly mapped, using the addresses associated with each
of the samples. In addition to the geocoding, the rest of the data preparation took place in
ArcGIS Pro.
After being re-geocoded, the original incorrectly located points did fall into the PIA
boundary, demonstrating potential mistakes in the original data table. These mistakes could have
arisen from the transferring of the field data into the Excel table. Aside from these mistakes, the
rest of the coordinates appear to be relatively accurate, given that they were determined by GPS
31

units. The possible uncertainty that arises from these mistakes does not present a concern for this
analysis, as the soil samples were used to create an approximate interpolated surface.
The total number of geocoded points for samples used in the analysis, those originally
with no coordinates and those redone, was 399 samples. While the Esri World Geocoding
Service provided the approximate locations of the sample points, the geocoded sample points
were then manually placed within their properties, respective to the additional location codes
provided for the samples, such as front yard and backyard, etc., and confirming they were
associated with the correct property ID. The geocoded points were then merged with the rest of
the dataset. The complete set of points is shown in Figure 4. The red points are the samples that
were geocoded, while the blue points are the samples that had existing coordinates within the
study boundary. Note that the solid blue areas indicate the density of individual sample locations.
The zoomed in map illustrates the distribution of samples within a small area of the PIA.
32

Figure 4 Exide soil Pb sample locations.
Since the full version of the dataset includes analysis for multiple metals within the soil
and differing matrixes of soil and paint, the dataset needed to be filtered to include the proper
parameters for analysis. In particular, the Selection by Attribute tool was used to reduce the
dataset and create a new feature class with samples only having these necessary attributes: Metal
Analyzed – Lead, Matrix – Soil, and Depth – 3 inches (depth interval from 0-3 inches below
ground surface, considered the surface layer). The surface values are important for this analysis,
33

as lead typically stays towards the surface of soil and does not penetrate too deeply into the soil.
Although other depths were measured, the values are not necessary for this analysis.
Though a Representative Soil Lead Concentration was determined for each property and
could have been used for this analysis to eliminate uncertainty of the sample data, it was
determined that the raw data from the soil samples would be more valuable for this analysis. Dr.
Jill Johnston, Assistant Professor of Preventative Medicine and Director of USC’s
Environmental Health Centers Outreach Program, has become an expert in the situation
concerning Exide and provided this recommendation.
Since this study used the raw data from the soil samples, the dataset was further filtered
to include only samples that were analyzed in the field with XRF, while lab samples were
removed from the dataset for this study’s analysis. This was done to address a concern regarding
the difference in measuring methods and sample selections between field and lab samples. In
addition, the field sample concentration units are parts per million (ppm) and the lab
concentration units are milligrams per kilogram (mg/kg). While these units are ultimately the
same scale, a noticeable difference in the range of lab values versus fields values suggested
important measuring differences. This decision to remove the lab samples was also supported by
Dr. Jill Johnston. The total number of field samples was determined to be 52,947 out of the
combined field and lab sample total of 115,584 samples.
To further prepare the data for use in analysis, the soil sample point data was projected
into the NAD 83 California State Plane coordinate system, Zone V. This coordinate system was
chosen due to its high accuracy in each zone and its high utilization by state and local
governments. Since the soil sampling data is highly localized, the State Plane coordinate system
is appropriate for use in this study.
34

3.2.2. Integrating Duplicates by Averaging
In the initial Excel spreadsheet of the soil sample data, it was noted in the ‘Data Legend’
tab that there was marked duplicate soil samples. This distinction was made known in a column
titled Sample Type that established whether a sample was a normal sample or a duplicate sample.
Duplicate samples arose when two or more samples were taken for a specific location and depth.
This procedure was done to account for variability within the soil and is a standard procedure for
soil sampling in the field. Although duplicate samples are important when analyzing the soil
from a scientific perspective, these redundant points can pose issues when creating interpolated
surfaces through geostatistical means. Hence, each set of original and corresponding duplicate
samples was averaged to produce a single value at each sample location. The averaging of such
duplicates, a common practice with such data, ensures that a good estimate for each sample
location is considered when making the interpolated surfaces.
Upon further examination into the soil sample data, it was noted that there were
additional coincident sample points, beyond the samples marked as duplicates. This was
determined by a one-to-many relationship between sample coordinates and Sample Location, the
unique numeric identifier assigned to each sample location. With this in mind, the Summary
Statistics tool in ArcGIS Pro was used to average the coincident point values with the longitude
and latitude as the case fields, rather than using Sample Location as a case field. The longitude
and latitude case fields meant statistics were calculated for each unique coordinate pairing,
considering the coincident points and those that were marked as duplicates. This resulted in a
table with a count of 50,952 records.
The attributes included in the Summary Statistics table were the following: Property ID
(unique ID for each property), the primary APN (parcel number associated with the property),
Sample Location (unique sample location ID), the mean of the results, the result units, and the
35

representative soil lead concentration. Since the output of the Summary Statistics tool is a table,
the longitude and latitude values in the table were used to create a new feature class of the newly
averaged soil sample data and was projected into the appropriate State Plane coordinate system.
The resulting dataset is the set of soil samples, with any coincident point values averaged, that
was used for all of this study’s subsequent analyses.
The initial data preparation as well as the steps of integrating duplicate samples by
averaging are summarized in the following workflow diagram (Figure 5).

Figure 5 Data Preparation Process

3.3. Data Exploration
After preparing the soil sample data for use in the analysis, further exploration of the
dataset investigated potential data quality issues and provided an assessment of outliers within
the data.
3.3.1. Data Quality Issues Regarding Recorded Locations
One data quality concern is related to the uncertainty associated with the X and Y
coordinates for the sample locations in the Excel file. Although the coordinates were recorded
using a GPS unit and were sampled by property, some sample points appear to be on top of
buildings when visualized in a GIS, indicating the possible use of parcel centroids or
inaccuracies resulting from the GPS unit. To investigate these inaccuracies, building outline data
was acquired from the LA County GIS Data Portal and spatially joined to the soil sampling point
36

data to determine how many points are on buildings. Out of the 50,952 data points, only 2,999
points are on buildings, with the highest number of points on a building being 13 points. This
indicates minor horizontal errors in the coordinates and the chances of the samples being on the
incorrect parcel are slim. These inaccuracies can be attributed to human error in the process of
transferring the locational information from the sampling reports to the Excel file, GPS
positioning accuracy from the devices used, possible errors in the building outline data, or even
errors in the ArcGIS base map data. As mentioned earlier, the use of the phrase ‘Georeferencing
in progress’ could also present confusion in the process and suggests possible error in the
preparation of the Excel table. Despite these uncertainties within regards to the coordinates
associated with the samples, the error resulting from these uncertainties does not significantly
affect this study’s analysis, as the samples were utilized in the creation of an interpolated surface.
Another indicator of potential data quality concern was whether the Assessor ID (AIN) in
the parcel boundaries polygon feature class matched the primary Assessor Parcel Numbers
(APN) associated with the soil sample data on each property. These numbers are formatted
identically so they can be directly compared. The AIN of the parcels and the APN of the
properties should match each other and provides an indication of the accuracy of the soil sample
data. To investigate this concern, the parcel data was spatially joined to the soil sampling data
and a query was performed to indicate the number of instances where the APN from the soil data
did not match the AIN from the parcel data. The resulting record count was 3,185. However,
critical examination into these records revealed that the APN and AIN of these records were only
off due to adjacent parcels being listed as the APN for the soil data.
An explanation for this data quality concern stems from the source Excel spreadsheet,
where the tab ‘Multiple APN Properties’ mentions that a primary APN was selected to represent
37

properties that contain multiple parcels and thus multiple APNs. This explains why the AIN of
some parcels do not match with the APN of the corresponding properties, as the AIN could
correspond to any of the APNs associated with the property instead of the primary APN.
Although the AIN of the parcels do not match exactly with the APN of the properties in the soil
sample data, the explanation for it validates the accuracy of the soil sample point locations and
does not present itself as a primary concern for this analysis.
3.3.2. Outlier Assessment
Although outliers within data could be a cause for concern in most analyses, outliers
serve an important role in this analysis. Typically, outliers should be excluded from the data
before analysis takes place. However, for this study, outliers with higher concentration values
indicate areas with higher levels of lead contamination, signifying areas that should be a health
concern. Even though outliers are important to this analysis, the maximum value in the dataset is
21,854 ppm, which can be considered unreasonably high and likely be due to human error in the
compilation of data. Therefore, the sample with this particular outlier value was excluded to
make for better models in the geostatistical analysis. Excluding the high outlier, the minimum
value of the dataset is 8.82 ppm, the maximum value is 7,348.04, and the mean is 268.53.
Figure 6 displays the remaining outliers in the soil sample data at both ends of the data’s
standard deviation spectrum. The soil sample data was symbolized using standard deviation
classification on a logarithmic scale. The symbology was set to only display outlier values less
than -2.5 standard deviations and greater than 2.5 standard deviations. The distribution of outliers
appears to be evenly scattered within the study areas.

38

Figure 6 Outlier Soil Samples

39

3.4. Study Areas and Scales of Analysis
This section discusses how three separate study areas were determined within the overall
PIA boundary, using the sample points as a guideline, and the decisions made regarding the
different scales of analysis.
3.4.1. Study Areas
The Exide Preliminary Investigation Area, which encompasses an approximately 1.7-
mile radius around the smelter site, aids in determining the study areas for this analysis. The
Preliminary Investigation Area includes parts of the following cities: Los Angeles, East Los
Angeles, Maywood, Boyle Heights, Huntington Park, Commerce, Bell, and Vernon. To capture
the PIA, the PIA boundary was created as a new feature, using a map made by DTSC as a guide.
However, since the sampling points are densely concentrated in particular sections within the
PIA boundary, three separate study area boundaries were created, as shown in Figure 7, which
ultimately was needed for the geostatistical analysis.
40

Figure 7 Study Area Boundaries
To produce these three separate study areas, the Aggregate Points tool, which creates
polygons around clustered point features, was used. The polygons generated from this tool and
the PIA boundary were used to determine and smooth the newly created separate study area
boundaries. One of the three concentrated areas of the sample points is located north of
Washington Boulevard in the northern part of the PIA boundary, while the other two
concentrated areas of the sample points are located in the southern part of the PIA boundary,
41

with a smaller cluster to the left. Corresponding to their respective concentrated areas of points,
the three separate study area boundaries are defined as follows: Northern Study Area Boundary,
1
st
Southern Study Area Boundary (the larger cluster of points in the southern area of the PIA),
and 2
nd
Southern Study Area Boundary (the smaller cluster of points in the southern area of the
PIA). The soil sample points were then clipped to their respective study area boundaries and the
feature classes analyzed separately.
3.4.2. Scales of Analysis
In determining the scales of analysis for this study, spatial units typically used for policy
implementation and that are easily defined geographically were of interest. Hence, it was decided
that census units would form the basis for the scales of analysis for this study, along with parcel
boundaries, which is the current aggregation unit being used to prioritize cleanup of the
contaminated soils. The aggregation units used for this study include block groups, blocks, and
parcels. These zones of differing scales were used to aggregate the results of the interpolated
surfaces created from Kriging through various aggregation methods. These diverse aggregation
units provide for demonstration of how the values assigned to these units vary based on scale and
offer other approaches for delineating priority areas to clean up the lead contaminated soils, in
addition to the parcel-based approach. Table 2 lists the sources of these feature layers.

42

Table 2 Polygon Feature Layers Used for Scales of Analysis
Dataset File Type Data Type Details Source
Census Block
Groups (2018)
Shapefile (.shp) Polygon Feature
Class
Boundaries of
block groups
U.S. Census
Bureau TIGER
Products
Census Blocks
(2018)
Shapefile (.shp) Polygon Feature
Class
Boundaries of
blocks
U.S. Census
Bureau TIGER
Products
LA County
Assessor Parcels
(2016)

Geodatabase
(.gdb)
Polygon Feature
Class
Parcel
boundaries for
LA County
Los Angeles
County GIS
Data Portal

3.5. Creating Surfaces through Geostatistical Analysis
Section 3.5 describes in detail the process for creating the interpolated surfaces using the
Empirical Bayesian Kriging geostatistical method. It first defines and explains what Empirical
Bayesian Kriging entails and then discusses how an appropriate Kriging model was determined,
using cross-validation as a guide. The section then lays out the logic for determining the cell size
of the output surfaces and what was done post-Kriging to prepare the interpolated surfaces to be
used in the various aggregations.
3.5.1. Empirical Bayesian Kriging
The geostatistical method of Empirical Bayesian Kriging (EBK) was utilized in this study
to obtain interpolated surfaces of the soil samples. Unlike other Kriging methods in Esri’s
Geostatistical Analyst, Empirical Bayesian Kriging automates some of the more challenging
aspects of creating a valid Kriging model by using a large number of subsets of the data to create
and compare various simulated models rather than requiring the user to manually adjust the
parameters to find a suitable Kriging model. This latter approach can result in a model based on
arbitrary decisions (Esri 2018). EBK also accounts for a possible error when estimating
43

semivariograms by creating a collection of semivariograms for the simulated models and
averaging them to create a suitable Kriging model. This contrasts to other Kriging methods
which are dependent on a single semivariogram, calculated from all of the known sample points,
to estimate the values at unknown locations. Since these other methods rely on the single
semivariogram, they can underestimate the true standard errors of the predictions (Esri 2018).
3.5.2. Determining the Appropriate Model
Even though EBK significantly aids the process of building a valid Kriging model, there
are still decisions that need to be made when considering some of EBK’s parameters. Esri’s
Geostatistical Wizard helps guide this decision-making process of constructing and evaluating
each model’s performance through a set of interactive pages (Figure 8).

Figure 8 Empirical Bayesian Kriging in Geostatistical Wizard
One of the main considerations in building a successful Kriging model is determining the
type of semivariogram that would be best for the model, depending on if a transformation is
applied. The Empirical Bayesian Kriging method offers several different types of
semivariograms along with multiplicative skewing normal score transformation, with Empirical
44

and Log Empirical as its two base distributions. The normal score transformation transforms the
dataset to closely resemble that of a normal distribution. If the Transformation parameter is set to
None, then only the Power, Linear, and Thin Plate Spline options are available. If the
Transformation parameter is set to Empirical or Log Empirical, then Exponential, Whittle, K-
Bessel and their detrended counterparts are available options. Each of these semivariogram types
have their advantages and disadvantages.
Other considerations in building a successful Kriging model using EBK include deciding
the number of simulations for determining the semivariograms and the subset size for the local
models. The default number of simulations per subset is 100 and was the number used for this
analysis, as the greater number of simulations produce greater precision in the predictions. The
subset size is the maximum number of points in each local model. Depending on the size and
dispersion of the dataset, the subset size can play a crucial role in determining valid Kriging
models.
3.5.2.1. Using Cross-Validation as a Guide
To determine the appropriate subset size and semivariogram model for use in this study, a
technique known as cross-validation was used to evaluate each of the models. Cross-validation
aids in making informed decisions by evaluating which of the models provide the best
predictions. The process of cross-validation involves the removal of one sample point from the
model and then using the rest of the sample points to predict the value at the location of the
omitted point, repeating the process for all of the points in the dataset. The predicted value is
compared to the measured value. The process results in useful information in the form of
statistical diagnostics, which enables the comparison of models.
45

When comparing models, certain guidelines based on the statistical diagnostics are
considered to determine the best model. Four of the most important statistics that pertain to the
guidelines are the standardized mean, the standardized Root-Mean-Square, the Root-Mean-
Square (RMS), and the Average Standard Error (ASE) (Esri 2019). The standardized mean, or
mean of the cross-validation errors, assesses bias in the model, such as if the model predicts too
low or too high. This value should be the closest to zero, indicating a model with the least bias.
The Root-Mean-Square value directly measures the accuracy of the predictions. Generally, a
smaller value signifies greater accuracy of the cross-validation predictions to the measured
values (Esri 2019). The standardized Root-Mean-Squared prediction error should be the closest
to 1. Lastly, the Average Standard Error should be nearest to the Root-Mean-Squared prediction
error (Esri 2019).
These guidelines influenced the workflow for determining the most appropriate Kriging
model to use for all the study areas. After initial testing using the different types of
transformations (None, Empirical, and Log Empirical), the Empirical transformation consistently
proved to have the best results after cross-validation and was the transformation used for
building the Kriging models. It is also important to note that the Log Empirical transformation
was purposely not used due to its sensitivity with outliers, in which resulting predictions could be
orders of magnitudes smaller or larger than the actual values (Esri 2018). Since this study relies
on the importance of outliers, this transformation was not suitable for use.
Next, the ideal range of subset sizes was chosen. Using the Exponential semivariogram
for the Northern Study Area as a control, the subset sizes of 20, 25, 30, 50, and 100 (the default)
were tested, with the range of 20 to 30 performing best in cross-validation. To determine the best
semivariogram model and the ideal subset size, the semivariogram types of Exponential, Whittle,
46

and K-Bessel with the subset sizes of 20, 25, and 30 were tested for each of the study areas
(North, 1
st
South, and 2
nd
South). The full model results can be found in Appendix A.
All of the models performed reasonably well, emphasizing the significance of the
guidelines for the statistical diagnostics resulting from cross-validation. The models that had the
standardized means closest to 0 and the standardized Root-Mean-Squares closest to 1 had a
higher Root-Mean-Square and greater differences between this value and the Average Standard
Error. The models with the lower RMSs and smaller differences between this value and the ASE
had higher standardized means and standardized RMSs not the closest to 1. The top three models
were chosen for each of the study areas by balancing the two outcomes based on meeting the
objectives of the statistical guidelines.
After taking all of the objectives into consideration, the Kriging model determined to be
the best suited for all of the study areas is the K-Bessel semivariogram using a subset size of 30
and it was the chosen model for making the interpolated surfaces. This model produced
standardized means closest to 0, standardized Root-Mean-Squares closest to 1, and relatively
small differences between the RMS and ASE values (Table 3). Although additional factors and
greater scrutiny of the cross-validation results could be considered for determining the best
Kriging model for use, it is important to note that this study only requires the development of
suitable interpolated surfaces of each of the study areas for use in exploration of the impacts of
aggregation of different scales. The interpolated surfaces are a means to the end, rather than the
end itself.

47

Table 3 EBK Cross Validation Results for 30 Subset Size K-Bessel Model
Study
Area
Standardized
Mean
Standardized
Root Mean
Squared
Root Mean
Square
(RMS)
Average
Standard
Error
RMS & Avg.
Standard
Error Approx.
Difference
North 0.00390 0.99888 201.127 184.207 17
1
st
South -0.00199 1.00246 132.653 119.816 13
2
nd
South 0.00519 0.92088 84.120 83.981 1
3.5.3. Determining the Cell Size
For the purpose of the aggregation comparison, the gridded surfaces created through
EBK need to have a cell size that made it possible to capture the variation of the surface at all
scales. Given the wide range of parcel sizes, it was determined there should be at least 2 cells per
parcel. To determine this cell size, the mean, minimum, and maximum parcel area of all the
parcels within the study area boundaries were determined. The mean parcel area was 10,717 sq.
ft. (996 sq. meters), the minimum parcel area was 38 sq. ft. (4 sq. meters), and the maximum
parcel area was 2,117,852 sq. ft (196,755 sq. meters). Thus, a cell size of 50 feet (15.24 meters)
was chosen so that it would create a 2,500 square foot cell (232 sq. meters), fitting all but the few
smallest parcel sizes. It is important to note that the coordinate system used for analysis is in U.S.
feet and was chosen for this study because it is the common coordinate system used by Los
Angeles County.
3.5.4. Post-Kriging
After producing the geostatistical layers for each of the study areas using EBK, the
interpolation output surfaces were extracted to rasters, with the appropriate cell size of 50 feet
(15.24 meters) clipped to the study areas using the mask function.
48

3.6. Aggregation Methods
This section discusses the different aggregation methods used to summarize the
interpolated surface results into each geographical unit of block groups and blocks, which
include taking the mean of surface result values, the percent area, and determining a Hazard
Quotient. It also describes the use of a different method for the parcel unit.
3.6.1. Aggregation Methods for Block Groups and Blocks
To aggregate the interpolated surface results into each analysis unit (block groups and
blocks), the tool Zonal Statistics as a Table was utilized. This tool works by summarizing the
values of a raster into zones defined by polygons that are overlaid on the raster. Results are
stored in the form of a table with a row for each of the overlaid polygons and columns providing
summary statistics within each polygon. Figure 9 displays a workflow for the aggregation
methods used for blocks and block groups.
49

Figure 9 Aggregation Methods Workflow for Block Groups and Blocks

The summary values assigned to each aggregation unit are the statistical mean, the
percent area, and a Hazard Quotient (HQ). All of these values are different ways to identify
priority areas for cleanup of the lead contaminated soils.
The easiest summary value to calculate was the statistical mean as this is determined
directly as one of the statistics produced from zonal statistics. It is the statistical mean of all lead
concentrations found in the set of cells encompassed by each polygon. Using a polygon overlay
algorithm, zonal statistics involves zone features (the geographical units in this case) overlapping
with the cell centers of the value raster (the interpolated surfaces). Thus, zonal statistics works by
providing statistics for the various zones by using the cell values from the input raster where
50

their centers fall within the boundaries of the zones. If zone features do not overlap with any cell
centers of the input value raster, then these zone features will not be represented in the output, as
determined through the algorithm.
To calculate the percent of the area of a polygon where lead concentration values exceed
400 ppm (USEPA standard for lead exposure), the interpolated surfaces were first reclassified
into binary rasters, where any value below 400 became 0 and any value greater than or equal to
400 became a 1. After creating zonal statistics tables from these reclassified rasters, the sum and
count statistics were utilized to calculate the areal percentages. The sum statistic represents the
sum of all the cells in a polygon containing the value of 1, while the count statistic represents the
count of all the cells per polygon. A new percentage field was then calculated using the sum
divided by the count multiplied by 100. This percentage represents the percent of the area of a
polygon (block group or block) where lead concentration values exceed 400 ppm.
The third summary value assigned to each aggregation unit is a Hazard Quotient. The
simple equation for determining a Hazard Quotient is Exposure Concentration / Reference
Concentration. If HQ < 1, then there is no risk to human health. If HQ >1, then some degree of
risk exists. The reference concentration utilized for this study was 400 ppm, the USA EPA
standard for lead exposure. The means of the cells per polygon calculated from the zonal
statistics were used as the exposure concentrations. A new field for Hazard Quotients was
calculated by dividing the means by the reference concentration of 400ppm. This creates an
index value that standardizes the mean values and offers proportions that can be used to easily
identify which areas may present some degree of risk to human health from the lead
contamination.
51

The tables of the resulting summary values – the mean, percentage, and the Hazard
Quotients – were then joined to their respective spatial aggregation units for each of the study
areas. These aggregation units included block groups and blocks.
3.6.2. Aggregation Method for Parcels
For the parcel scale, recognizing that in many cases only small portions of a small
number of grid cells on the interpolated surface would fall within individual parcels, it was
determined that the polygon overlay algorithm in zonal statistics would not provide a valid
result. So, it was decided that it would be useful to compare a single value extracted from the
interpolated surface with the Representative Soil Lead Concentration values currently used. To
do this, centroids were calculated for each parcel polygon, then the Extract Multi Values to
Points tool was utilized to extract the cell values from the interpolated surfaces at each centroid.
These values were then subtracted from the original Representative Soil Lead Concentration
values determined by DTSC to calculate the magnitude of differences.

52

Chapter 4 Results
Chapter 4 presents the results of the analysis described in the previous chapter. The first
section discusses the interpolated surfaces produced from Empirical Bayesian Kriging which
show the spatial distribution of lead concentrations in the soil within the various study areas. The
second section describes the results from the various aggregation methods used to summarize the
interpolated surfaces into the different scales of block groups and blocks. The results indicate
how scale affects the values allocated into these various zones which might be used to identify
priority areas for cleanup. The values are the mean, percent area, and Hazard Quotient (HQ) for
each geographical unit. The results also compare which areas would be chosen for cleanup,
depending on the aggregation value and scale used to base the decision. The last section
discusses the results for the parcels aggregation method and compares the findings to the
Representative Soil Lead Concentration DTSC is using for the current cleanup plan by mapping
the differences in values.
4.1. Empirical Bayesian Kriging
The minimum and maximum interpolated values for each study area, as well as their
comparative minimum and maximum sampled values, can be found in Table 4. The interpolated
surfaces produced by Empirical Bayesian Kriging are displayed in Figures 10 to 12. The surfaces
demonstrate the wide range of lead concentration values distributed throughout the study areas,
with noticeably higher concentrations in the Northern Study Area. The Northern Study Area
maximum value of 7,299.61 determined the upper limit of the classification for the maps, 7,300
ppm, while the significant value of 400 ppm, as the US EPA standard for lead exposure, helped
determine the remaining classification numbers.
53

Table 4 Minimum and Maximum – Sampled vs. Interpolated Values for Each Study Area
Minimum and Maximum – Sampled vs. Interpolated Values
Northern
Study Area
1
st
Southern
Study Area
2
nd
Southern
Study Area
Minimum Sampled
Value (ppm)
8.82 10 19
Minimum
Interpolated Value
(ppm)
13.33 6.64 25.93
Maximum
Sampled Value
(ppm)
7,348.04 4,902.00 1,699.50
Maximum
Interpolated Value
(ppm)
7,299.61 1,973.16 584.59

Figure 10 Northern Study Area Interpolated Surface
A majority of the predicted lead concentration values in the Northern Study Area appear
to be in the 0-250 ppm and the 250-400 ppm ranges, while the 0-250 ppm range is the majority
54

in both of the Southern Study Areas. Both the Northern and the 1
st
Southern Study Areas have
1,000-7,300 ppm as their highest concentration range. The ranges with higher concentrations are
dispersed throughout the two study areas as well, indicating the wide range of area in which
higher lead contamination values may exist. These study areas are also the closest to the Exide
facility, while the 2
nd
Southern Study Area is the furthest. There is a cluster of high lead
concentration values on the west side of the 1
st
Southern Study Area, while the 2
nd
Southern
Study Area exhibits a cluster of values in the 250-400 ppm and 450-600 ppm ranges on the east
side.

Figure 11 1
st
Southern Study Area Interpolated Surface
55

Figure 12 2
nd
Southern Study Area Interpolated Surface
4.2. Aggregation Results for Block Groups and Blocks
As expected, the results of the analysis demonstrate how scale affects the values allocated
to various zones that may be used for determining priority areas for cleanup. Generally, the
values of mean, percent area, and Hazard Quotients assigned to block groups are smaller than
when assigned to blocks. To give a big picture of the difference in scale between block groups
and blocks and how these geographical units may affect cleanup decisions, Table 5 shows the
total number of block groups and blocks that would be selected for cleanup where the statistical
mean value exceeds 400ppm, along with the number of parcels that would be selected for
cleanup within those units. The number of parcels is used as a reference to compare block groups
and blocks, indicating that this would be the number of parcels that would need to be cleaned if
the whole block or block group were slated for cleanup.
56

Table 5 Comparison between the number of block groups and blocks within each study area and
the number of units selected for potential cleanup

Northern Study
Area
1
st
Southern Study
Area
2
nd
Southern
Study Area
Blocks
Block
Groups
Blocks
Block
Groups
Blocks
Block
Groups
Total # of Units 492 47 268 30 18 3
# of Selected Units
for Cleanup
59 2 11 0 0 0
# of Parcels Selected
for Cleanup
795 301 91 0 0 0
Total # of Parcels in
the Study Area
6,619 parcels 4,467 parcels 353 parcels

In the Northern Study Area, 2 out of 47 block groups and 59 out of 492 blocks had mean
values greater than or equal to 400 ppm and are thus selected for potential cleanup. This is
equivalent to 301 and 795 parcels out of 6,619 parcels, respectively. The effect of scale on
deciding priority areas for cleanup is clearly evident in these results, in addition to the fact that
only 11 blocks and 0 block groups were selected for potential cleanup in the 1
st
Southern Study
Area and 0 blocks or block groups were selected in the 2
nd
Southern Study Area. The following
subsections detail the results according to study area.
4.2.1. Northern Study Area
The results for the Northern Study Area are shown in Table 6, as well as in the following
maps. Table 6 displays summary statistics of the result tables for blocks and block groups in the
Northern Study Area, including the mean, minimum, maximum, range, and standard deviation
for each value: mean, percent area, and Hazard Quotient.

57

Table 6 Summary Statistics for the Northern Study Area
Northern Study Area
Analysis Approaches
Scale Summary Statistic Mean Percent
Area Over
400 ppm
HQ
Blocks
Mean 307.03 18% 0.77
Minimum 111.96 0% 0.28
Maximum 759.58 100% 1.90
Range 647.62 100% 1.62
Standard Deviation 90.27 23.6% 0.23
Block Groups
Mean 311.98 19.3% 0.78
Minimum 183.79 0% 0.46
Maximum 526.00 98.1% 1.31
Range 342.21 98.1% 0.86
Standard Deviation 68.48 19.2% 0.17

While all of the statistics provide insight into how scale affects the allocation of values,
the mean statistics is one of the more indicative statistics. The average for all of the mean values
in the blocks is 307.03 ppm, while the average for all of the mean values in the block groups is
311.98 ppm. These results are contrary to the tendency for values to decrease when the area
increases. This is due to an edge effect, where boundaries on the edges of the study area get
assigned values from only a small part of the raster surface that intersects with them (Figure 13).
In addition, some block groups on the edges contain extra blocks that were not used in the
analysis for the block scale (Figure 13). The consequences of the edge effect are discussed in
Chapter 5.
58

Figure 13 Edge Effect of Boundaries
The Hazard Quotient values correspond with the mean values, since these values are
directly calculated using the mean values. The average HQ for blocks is 0.77, while the average
HQ for block groups is 0.78. The average percent area value for blocks is 18%, while the average
percent area value for block groups is 19.3%. This indicates blocks having 18% of their area
where lead concentration values exceed the nationally recommended exposure value of 400 ppm.
The values for percent area on the edge boundaries have also been affected by the edge effect
and are examined more in Chapter 5.
Figure 14 displays the range of mean and HQ values for block groups and blocks in the
Northern Study Area. The Quantile classification method was used to classify the values for all
of the result maps for each study area. Although this classification method is considered to have
limitations, the Quantile classification works well in displaying the results for this analysis. The
breaks of ranges between block groups and blocks using the Quantile classification are so similar
that any differences can be considered negligible. As expected, the block results have a wider
range of values than the block groups. This can be seen in the maps of block groups versus
59

blocks. The block groups that have higher values do however correspond to blocks also with
high values. The white hatched fill demonstrates areas that would be designated for cleanup on
the basis that their mean value is greater than or equal to 400 ppm.
Figure 15 displays the range of percent area values for block groups and blocks in the
Northern Study Area. While the block groups and blocks generally correspond and have a
similar range in values, the maps show the difference a larger or smaller area has when
determining how much of the area is over the 400 ppm lead concentration. The percent area
values are also affected by the edge effect of the boundaries.
60

Figure 14 Block Group and Block Results for the Northern Study Area – Mean and HQ

61

Figure 15 Block Group and Block Results for the Northern Study Area – Percent Area
62

4.2.2. 1
st
Southern Study Area
The results for the 1
st
Southern Study Area are shown in Table 7, in addition to the maps
depicted in the following figures. Table 7 contains summary statistics of the result tables for
blocks and block groups in the 1
st
Southern Study Area, including the mean, minimum,
maximum, range, and standard deviation for each value: mean, percent area, and Hazard
Quotient. The average for all of the mean values in the blocks is 237.65 ppm, while the average
for all of the mean values in the block groups is 225.12 ppm. The average HQ for blocks is 0.59,
while the average HQ for block groups is 0.56. These results are consistent with the tendency for
values to decrease when there is an increase in area. The average percent area value for blocks is
6.2%, while the average percent area value for block groups is 4%.
Table 7 Summary Statistics for the 1
st
Southern Study Area
1
st
Southern Study Area
Analysis Approaches
Scale Summary Statistic Mean Percent
Area Over
400 ppm
HQ
Blocks
Mean 237.65 6.2% 0.59
Minimum 115.06 0% 0.29
Maximum 804.34 100% 2.01
Range 689.28 100% 1.72
Standard Deviation 83.83 15.4% 0.21
Block Groups
Mean 225.12 4% 0.56
Minimum 129.46 0% 0.32
Maximum 313.41 22.4% 0.78
Range 183.95 22.4% 0.46
Standard Deviation 40.18 5.1% 0.10

Figure 16 displays the range of mean and HQ values for block groups and blocks in the
1
st
Southern Study Area. Similar to the Northern Study Area, the block results have a wider
range of values compared to the block group results. Both the block group and block maps show
63

the highly concentrated area of lead on the western side of the study area. The results for the 1
st

Southern Study Area carry the same observations noted in the previous subsection for the
Northern Study Area. It is also important to note that a few of the block groups on the edges of
the study area only have a portion of their zones intersecting with the study area boundary. This
is noticeably observed in the two very large block groups that extend past the 1
st
Southern Study
Area boundary, covering a large portion of the map (Figures 17 and 18). This also demonstrates
the consequences of the edge effect.
Figure 17 displays the range of percent area values for block groups and blocks in the 1
st

Southern Study Area. The maps for this study area clearly demonstrate the notion that larger
areas (block groups) will have a greater number of concentration values used towards the
calculation of the aggregation values to be applied to their areas, while smaller areas (blocks)
will have a smaller amount of concentration values used towards the calculation of the
aggregation values for the zones. This is demonstrated on the eastern side of the study area,
where the larger areas of the block groups likely had a greater number of high concentration
values in determining the percentage of the block groups where lead concentrations are 400 ppm
or higher. In contrast, since the blocks generally have smaller areas compared to the block
groups, certain blocks will not have the high concentration values picked up by the larger area of
the block group that encompasses it. This can be seen in the same area on the eastern side where
there is a greater number of blocks colored the lightest shade, representing 0 percent of the area
having concentrations greater than or equal to 400 ppm. The percent area value also emphasizes
the high lead concentrations on the western side of the study area.
64

Figure 16 Block Group and Block Results for the 1
st
Southern Study Area – Mean and HQ

65

Figure 17 Block Group and Block Results for the 1
st
Southern Study Area – Percent
66

4.2.3. 2
nd
Southern Study Area
The results for the 2
nd
Southern Study Area are shown in Table 8, in addition to the maps
represented in the following figures. Table 8 shows summary statistics of the result tables for
blocks and block groups in the 2
nd
Southern Study Area, including the mean, minimum,
maximum, range, and standard deviation for each value: mean, percent area, and Hazard
Quotient. The average for all of the mean values in the blocks is 134.78 ppm, while the average
for all of the mean values in the block groups is 124.84 ppm. The average HQ for blocks is 0.34,
while the average HQ for block groups is 0.31. These results are also in agreement with the
tendency for values to decrease when there is an increase in area. The average percent area value
for blocks is 0.8%, while the average percent area value for block groups is 0.9%.
Table 8 Summary Statistics for the 2
nd
Southern Study Area
2
nd
Southern Study Area
Analysis Approaches
Scale Summary Statistic Mean Percent
Area Over
400 ppm
HQ
Blocks
Mean 134.78 0.8% 0.34
Minimum 65.77 0% 0.16
Maximum 294.04 10.1% 0.74
Range 228.27 10.1% 0.57
Standard Deviation 71.91 2.9% 0.18
Block Groups
Mean 124.84 0.9% 0.31
Minimum 108.22 0% 0.27
Maximum 141.45 1.8% 0.35
Range 33.23 1.8% 0.08
Standard Deviation 23.5 1.3% 0.06

Figure 18 displays the range of mean and HQ values for block groups and blocks in the
2
nd
Southern Study Area. Due to the overall small size of the 2
nd
Southern Study Area, the maps
clearly demonstrate how values get partitioned into the different scales of geographical units and
67

can change which class/range the units get assigned to, based on how many individual values are
being considered when the overall value is calculated for that unit. It can be easily seen that the
one block group that covers the entire study area gets assigned to a lower class/range than its
block counterparts. The blocks have a more diverse set of ranges, with the eastern side of this
study area containing higher concentration values, thus putting some of the blocks in a higher
class/range. With such a small study area, it is clearly shown how the larger area of a block
group can moderate the high concentration values that would be noticed in the smaller areas of
blocks. It is also important to note that the edge effect is seen within this study area as well,
where only a small portion of some blocks and the one block group intersect with the study area
boundary.
Figure 19 displays the range of percent area values for block groups and blocks in the 2
nd

Southern Study Area. The small size of the study area also provides insight into the previous
observations on the size of areas affecting percent area values. As shown in the map, the one
high value in the one block is accounted for in the whole block group that covers the entire study
area. This is somewhat contrary to the notion previously presented that larger areas moderate
high concentration values. While although that may be true for the mean and HQ values, this
contrast shows a difference in outcomes between the mean and HQ values and the percent area
values. For percent area, a larger area means that all the values in the blocks get accounted for in
the block group, which can affect the outcome of the percent area where lead concentrations
exceed the 400 ppm value.
68

Figure 18 Block Group and Block Results for the 2
nd
Southern Study Area – Mean and HQ

69

Figure 19 Block Group and Block Results for the 2
nd
Southern Study Area – Percent
70

4.3. Aggregation Results for Parcels
For the parcel aggregation, the method utilized a tool that extracted the cell values from
the rasters at the centroid of each parcel. The results for all the study areas are presented as
summary statistics in Table 9. The resulting values are compared to the original Representative
Soil Lead Concentration values determined by DTSC through differences. The difference
calculation is as follows: Representative Soil Lead Concentrations minus resulting values from
the cells. The positive differences indicate that the Representative Soil Lead Concentration is
greater than the cell value, while the negative differences indicate that the cell value is greater
than the Representative Soil Lead Concentration. These differences were then mapped, as shown
in Figures 20-22. Overall, there are more positive differences than negative differences, meaning
in general, the interpolation surface method produced generally lower concentration values.
Table 9 Summary Statistics for Cell Values Extracted from Interpolated Surfaces at Parcel Scale
Summary Statistics for Parcels
Summary Statistic
Northern
Study Area
1
st
Southern
Study Area
2
nd
Southern
Study Area
Mean 328.58 241.51 122.48
Minimum 21.30 30.39 25.93
Maximum 7299.61 1973.16 576.03
Range 7278.31 1942.77 550.10
Standard Deviation 182.41 106.5 88.35
71

Figure 20 Difference in Parcel Values for the Northern Study Area
72

Figure 21 Difference in Parcel Values for the 1
st
Southern Study Area
73

Figure 22 Difference in Parcel Values for the 2
nd
Southern Study Area
74

Chapter 5 Discussion and Conclusions
This study examined the effect of scale on aggregated values into various zones and the
role spatial scale has in cleanup efforts. This chapter concludes this thesis and provides a deeper
discussion of the results, with explanations for the results and the effects of using these
boundaries. The chapter concludes with limitations of the study, overall implications, and
suggestions for further research.
5.1. Discussion
This section provides a deeper discussion of the results from the analysis, offering
explanations for both the results and the effects of using block groups, blocks, and parcels as
aggregation boundaries. It also points out general conclusions from the results, both from the
block groups and blocks aggregation as well as the parcels aggregation.
5.1.1. Block Groups and Blocks Aggregation Discussion
As evidenced by the results, this study clearly demonstrates how scale affects the values
allocated to different zones used for determining priority areas for cleanup. For all of the study
areas, the mean and HQ values for the blocks have wider ranges than those for the block groups.
This can be attributed to the high outlier values being more influential in the block aggregations
than the block group aggregations. Since the block groups are a larger unit and have a greater
area, the same few high values present in the blocks get moderated through the averaging with
more lower values present in the larger area of the block groups. This notion explains why the
mean and HQ values for both Southern Study Areas are higher for blocks and lower for block
groups. This finding is also present in the inherent nature of the dataset. As visualized by the
75

interpolated surfaces, the dataset contains a lot of low to moderate lead concentration values,
with a few high lead concentration values dispersed throughout the study areas.
In addition to scale affecting the mean and HQ values, different scales affect the percent
area values in much of the same way. Although the ranges for the percent area values are similar
for both block groups and blocks among all the study areas, the larger area of the block groups
contains more values in the calculation for how these values spread across the block group areas,
while the blocks have less values in the calculation. Since the values for the percent area were
calculated using binary rasters, where any cell containing a concentration value greater than or
equal to 400 ppm became a 1, the larger areas of the block groups have a greater chance of
containing more cells reclassified as 1’s than blocks. The result, a percentage representing the
percent area of a polygon where lead concentration values exceed 400 ppm, is then reflected in
the percentage values for the differing scales of block groups and blocks.
Although the results are largely indicative of how scale affects the mean, percent area,
and HQ values partitioned into the geographical units, there are some inaccuracies present due to
edge effects that occur with the boundaries. For both the block group and block results, there are
certain block groups and blocks on the edges of the study areas that receive assigned values from
only a small part of the interpolated surface that intersects with them. Essentially, the small part
of the surface that does intersect with these boundaries determines the values for the whole
boundaries.
In addition, some of the same block groups on the edges contain extra blocks that were
not used in the analysis for the block scale, as those blocks were not present when selecting the
block boundaries based on the study areas. These edge effects impact the summary statistics
tables and show in the maps for each of the study areas, particularly in the Northern Study Area.
76

One of the noticeable consequences of the edge effects is in the mean and HQ values in the
summary statistics table for the blocks and block groups in the Northern Study Area. The edge
effects explain why the average values for the block groups are higher than the blocks, contrary
to how it should be. It also causes some blocks and block groups to be selected as potential
cleanup areas, even though no soil samples were taken in those areas.
Despite the consequences of the edge effects, it is a natural outcome from constructing
the study area boundaries around the soil sample points. It is a compounding limit of boundaries
in a sense. The study area boundaries were determined from the soil sample points, then the
block group and block boundaries that intersected with the study area boundaries were selected
using Select by Location. Thus, it is the polygons themselves that impacted the results, as
opposed to decisions made.
5.1.2. Parcels Aggregation Discussion
For the parcel aggregation, the larger number of positive values compared to negative
values suggest that the Representative Soil Lead Concentrations are higher than the polygon
centroid cell values pulled from the raster surfaces. This signifies that the values pulled from the
surfaces tend to be lower than the determined Representative value for that parcel. This could
pose problems if this method was to be used for potential cleanup, as it will miss the higher
outlier values present in the dataset, which are important in determining health effects. This
aggregation method for the parcels proves to be more useful for comparing the surface values to
the Representative Soil Lead Concentrations and can be used for determining the accuracy of the
surfaces, as opposed to an actual method used for decision-making on cleanup.
77

5.2. Limitations
Although this study provides a successful look into the effect of geographic scale in lead
contamination, it was not without limitations. One of the main limitations for this study was
issues in data quality. While most of the soil sample data had coordinates associated with the soil
sample points, there were a handful of samples that needed to be geocoded and another handful
of samples that had been inaccurately geocoded and therefore needed to be geocoded again.
While these problems were fairly simple fixes, they could have produced inaccuracies in the
mapping of the soil samples in terms of where their actual locations may have been in the field.
Data quality could also be improved in the initial phase of data acquisition through choice of
sample sites and methods. More sample sites, or differently chosen sample sites, could lead to
greater accuracy in the interpolated surfaces.
In addition, there is an issue with coordinate accuracy from the original soil sample data.
Some of the coordinates associated with the samples have less than six decimal places, which has
a precision of 7-10 meters. This could cause these samples to be placed in roughly 1-2 parcels
away from the actual sampled parcel, if not in the same parcel, depending on the size of the
parcel. To solve this issue, the coordinates with less than six decimal points would need to have
their addresses geocoded again as well. This issue of coordinate accuracy could explain the data
quality issues seen in the earlier data exploration, where some of the points fell on top of
buildings and some sample’s Primary Assessor Parcel Numbers did not match with the Assessor
IDs for the parcels in the parcel data table.
Another major limitation for this thesis was not being able to use the same aggregation
methods and values used for the block group and block scales, for the parcel scale. Initially, the
aggregation methods and values of mean, percent area, and HQ were intended to be used for all
78

the scales of analysis, including block groups, blocks, and parcels. This approach would have
made the effects of scale on lead contamination and the determination of priority areas for
cleanup even more clear.
However, when attempting to use these methods for the parcel scale, it was discovered
that the Zonal Statistics as a Table tool only produces results that use the center of the cells to
determine which polygon the cell values get partitioned to, as opposed to splitting the cell so that
part of the cell’s value goes to one polygon and the other part goes to another. This would need
to happen if a single cell overlaps two boundaries. At the parcel level, some parcels do not
contain any centroids of the interpolated cells, which created “holes” in the aggregation results,
as the results do not include parcels that do not contain any centroids of cells. Numerous tools
and solutions, such as turning the rasters into polygons or decreasing the cell size used, were
attempted, but the same problem still existed. This indicates that the parcel scale may be too
small to be used for these aggregations, unless the method is found to appropriately apportion the
cell values into the different parcels.
As demonstrated by the edge effects, a third limitation of this study includes the
boundary issues that arise from these edge effects. The limitation being that the areas at the edge
boundaries may have some inaccuracies in values. A solution would be to instead use boundaries
that are only completely within the study areas. However, Figure 23 depicts that if this solution
were to be used, it would severely trim down the number of blocks and block groups used for the
analysis and the boundaries would not cover the whole study areas, which means not using some
of the soil samples provided within the study areas. This would lead to an even greater limitation
in the results. Therefore, it must be accepted that the edge effects will occur when using these
specific geographic units for analysis.
79

Figure 23 Boundaries Completely Within Northern Study Area
Considering the edge effects that occur with the choice of these geographical units for
analysis, it would be better to set the boundary of the study areas to the geographic unit
boundaries, rather than the soil samples. Because blocks and block groups were chosen as the
units for analysis, framing the study area boundaries around these units would provide for less
complications in the analysis, while still including the important locations of the points.
5.3. Future Research and Implications
Although this thesis successfully met its objective in demonstrating how scale can affect
the allocation of values into the various zones that may be utilized to prioritize potential areas for
cleanup, there are several improvements that could be incorporated into this research and ideas
for future studies. One improvement that could be made is to make the interpolated surfaces
80

more accurate by considering the error that is associated with them. Since the aim of this thesis
was to see how scale affects the aggregation of values into different areas, it was only necessary
to develop suitable surfaces for the aggregation, as the surfaces were a means to an end and not
the final result. However, one of the benefits of Empirical Bayesian Kriging is the option to
create error surfaces associated with the predictions, which could be utilized in future studies to
evaluate the interpolated surfaces. The use of different cell sizes for the creation of the surfaces
could provide another insight into how scale may affect the aggregation of values as well.
In addition to making the surfaces more accurate, the use of the Representative Soil Lead
Concentration values determined by DTSC for the creation of the interpolated surfaces rather
than the raw values may help smooth the surfaces. Currently, the surfaces are more rigid and
“bumpy,” a reflection of the raw values and the parcel sizes. The use of the Representative
Concentrations would smooth the surface of the raster, as only one value is associated with each
parcel, as opposed to 10-15 sample values. The addition of the lab samples into the raw values
used for analysis could also improve accuracy in the interpolated surfaces and may provide
differing results.
To improve the accuracy of the aggregations, building outline data could be used to clip
the rasters, thereby eliminating some of the impervious surfaces that get assigned values due to
the nature of raster creation. Since soil lead concentrations are really only prominent in
permeable surfaces, taking out the areas where buildings are in the raster surfaces could better
demonstrate where the distributions of the concentrations are. It would be interesting to see if
high outlier values exist where buildings actually exist. Another future study could employ the
use of different boundaries other than block groups, blocks, and parcels for the aggregations of
81

the values. This would change the results according to the boundaries used. It is a possibility that
there could be better determined boundaries to use for policy implementation.
Furthermore, the soil sample dataset is quite extensive and contains samples and data not
utilized in this analysis. Future research can take these additional data into account. There are
samples taken from paint chips within the parcels. These samples could possibly be utilized in a
correlation analysis with the soil samples to determine how much the concentrations are affected
by the paint from the age of the homes. Samples of different depths were also taken, which could
be utilized in a depth variation analysis. In addition, other heavy metals, such as arsenic, were
sampled in the soil sampling process. The distribution of these heavy metal samples could be
determined and examine how scale may affect the allocation of these values into zones for
cleanup. With this knowledge, it could be determined if the highly concentrated areas of the
other heavy metals overlap with the highly concentrated areas of lead.
While there is much future research that could be accomplished in relation to Exide and
the soil samples gathered, this thesis provides important implications for the cleanup of lead
contaminated soils. This thesis suggests alternatives for identifying priority areas for cleanup
through different aggregation methods. The values resulting from these aggregations – the
statistical mean, the percentage of an area where lead concentration values exceed the nationally
recommended exposure value of 400 ppm, and a Hazard Quotient, an index value that
determines the risk to human health – provide options for deciding which areas should be
cleaned. The examination of scale, using block groups, blocks, and parcels, also provides insight
into which scale would be best for managing the cleanup of areas. There is much that still needs
to be considered in the decision-making process for cleanup, such as specified criteria, budget,
timeline, permissions, etc. However, this thesis may help demonstrate to decision makers that
82

using a spatially informed methodology can help them in the decision-making process. Although
these options may not be considered for the Exide site specifically, the conclusions from this
study can be applied to other areas faced with the same problem in the future.

83

References
Barboza, Tony. 2015. “How a Battery Recycler Contaminated L.A.-Area Homes for Decades.”
Los Angeles Times, 2015. http://www.latimes.com/local/lanow/la-me-exide-cleanup-
story-so-far-20151121-story.html
Barboza, Tony, and Ben Poston. 2018. “The Exide plant in Vernon closed 3 years ago. The vast
majority of lead-contaminated properties remain uncleaned.” Los Angeles Times, 2018.
http://www.latimes.com/local/lanow/la-me-exide-cleanup-20180426-story.html
Chaney, Rufus, Howard Mielke, and Susan Sterrett. January 1989. "Speciation, Mobility, and
Bioavailability of Soil Lead." [Proc. Intern. Conf. Lead in Soils: Issues and Guidelines.
B.E. Davies & B.G. Wixon (eds.)]. Environmental Geochemistry and Health 11
(Supplement): 105-129).
Department of Toxic Substances Control. 2015. "Final Workplan Sampling and Analysis of
Properties in the Vicinity of the Exide Facility (Vernon, California)." Pasadena: Parsons.
https://www.dtsc.ca.gov/HazardousWaste/Projects/upload/Exide_Adv-Grp_Sampling-
workplan.pdf
Department of Toxic Substances Control. 2017. "Final Removal Action Plan (Cleanup Plan)
Offsite Properties Within the Exide Preliminary Investigation Area." Sacramento: State
Clearinghouse. http://www.dtsc.ca.gov/HazardousWaste/Projects/Residential-
Cleanup.cfm
Department of Toxic Substances Control. 2017. "Final Environmental Impact Report: Volume 1
Chapter 1 Through Chapter 2, Section 2.3.4." Pasadena: State Clearinghouse.
https://www.envirostor.dtsc.ca.gov/public/community_involvement/9913273453/Final_E
IR_Vol_1_Chapter_1_Executive_Summary.pdf
Department of Toxic Substances Control. 2018. Soil Sampling Data for the Exide Preliminary
Investigation Area. July 12, 2018 version. Distributed by California Department of Toxic
Substances Control. https://www.dtsc.ca.gov/HazardousWaste/Projects/pia-sampling-
data.cfm
Esri. 2018. “What is Empirical Bayesian Kriging? — ArcGIS Pro | ArcGIS Desktop.”
Pro.Arcgis.Com. https://pro.arcgis.com/en/pro-app/help/analysis/geostatistical-
analyst/what-is-empirical-bayesian-Kriging-.htm.
Esri. 2019. "Comparing Models — Help | ArcGIS Desktop.” Desktop.Arcgis.Com.
http://desktop.arcgis.com/en/arcmap/latest/extensions/geostatistical-analyst/comparing-
models.htm.
Fajčíková, K., V. Cvečková, A. Stewart, and S. Rapant. 2014. "Health Risk Estimates for
Groundwater and Soil Contamination in the Slovak Republic: A Convenient Tool for
Identification and Mapping of Risk Areas." Environmental Geochemistry and Health 36
(5): 973-986. doi:10.1007/s10653-014-9612-9.
84

Gay, J. Rebecca, and Anna Korre. 2006. "A Spatially-Evaluated Methodology for Assessing
Risk to a Population from Contaminated Land." Environmental Pollution 142 (2): 227-
234. doi:10.1016/j.envpol.2005.10.035.
Ginevan, Michael, and Douglas Splitstone. 1997. "Improving Remediation Decisions at
Hazardous Waste Sites with Risk-Based Geostatistical Analysis.” Environmental Science
and Technology 31 (2). doi:10.1021/es972129n.
Guastaldi, Enrico, and Andrea Alessandro Del Frate. 2011. "Risk Analysis for Remediation of
Contaminated Sites: The Geostatistical Approach." Environmental Earth Sciences 65 (3):
897-916. doi:10.1007/s12665-011-1133-6.
Hanna-Attisha, Mona, Jenny LaChance, Richard Casey Sadler, and Allison Champney Schnepp.
2016. "Elevated Blood Lead Levels in Children Associated with the Flint Drinking Water
Crisis: A Spatial Analysis of Risk and Public Health Response." American Journal of
Public Health 106 (2): 283-290. doi:10.2105/ajph.2015.303003.
Hipp, John R. 2007. "Block, Tract, And Levels of Aggregation: Neighborhood Structure and
Crime and Disorder as a Case in Point." American Sociological Review 72 (5): 659-680.
doi:10.1177/000312240707200501.
Juberg, Daland R., Cindy F. Kleiman, and Simona C. Kwon. 1997. "Position Paper of the
American Council on Science and Health: Lead and Human Health." Ecotoxicology and
Environmental Safety 38 (3): 162-180. doi:10.1006/eesa.1997.1591.
"Kriging Interpolation - The Prediction is Strong in this One - GIS Geography." 2018. GIS
Geography. https://gisgeography.com/Kriging-interpolation-prediction/.
Laidlaw, Mark A.S., Gabriel M. Filippelli, Sally Brown, Jorge Paz-Ferreiro, Suzie M. Reichman,
Pacian Netherway, Adam Truskewycz, Andrew S. Ball, and Howard W. Mielke. 2017.
"Case Studies and Evidence-Based Approaches to Addressing Urban Soil Lead
Contamination." Applied Geochemistry 83: 14-30.
doi:10.1016/j.apgeochem.2017.02.015.

Li, Xiangdong, Siu-lan Lee, Sze-chung Wong, Wenzhong Shi, and Iain Thornton. 2004. "The
Study of Metal Contamination in Urban Soils of Hong Kong Using a GIS-Based
Approach." Environmental Pollution 129 (1): 113-124.
doi:10.1016/j.envpol.2003.09.030.
Liang, Bingqing, and Qihao Weng. 2008. "Multiscale Analysis of Census-Based Land Surface
Temperature Variations and Determinants in Indianapolis, United States." Journal of
Urban Planning And Development 134 (3): 129-139. doi:10.1061/(asce)0733-
9488(2008)134:3(129).
Liu, Xingmei, Jianjun Wu, and Jianming Xu. 2006. "Characterizing the Risk Assessment of
Heavy Metals and Sampling Uncertainty Analysis in Paddy Field By Geostatistics and
GIS." Environmental Pollution 141 (2): 257-264. doi:10.1016/j.envpol.2005.08.048.
85

Markus, Julie, and Alex B. McBratney. 2001. "A Review of The Contamination of Soil with
Lead: II. Spatial Distribution and Risk Assessment of Soil Lead." Environment
International 27 (5): 399-411. doi:10.1016/s0160-4120(01)00049-6.
McClintock, Nathan. 2012. "Assessing Soil Lead Contamination at Multiple Scales in Oakland,
California: Implications for Urban Agriculture and Environmental Justice." Applied
Geography 35 (1-2): 460-473. doi:10.1016/j.apgeog.2012.10.001.
Mielke, Howard W., and Patrick L. Reagan. 1998. "Soil Is an Important Pathway of Human Lead
Exposure." Environmental Health Perspectives 106 (1): 217-229. doi:10.2307/3433922.
Oliver, M.A., and R. Webster. 2014. "A Tutorial Guide to Geostatistics: Computing and
Modelling Variograms and Kriging." CATENA 113: 56-69.
doi:10.1016/j.catena.2013.09.006.
Parenteau, Marie-Pierre, and Michael C Sawada. 2011. "The Modifiable Areal Unit Problem
(MAUP) in the Relationship Between Exposure to NO2 and Respiratory
Health." International Journal of Health Geographics 10 (1): 58. doi:10.1186/1476-
072x-10-58.
Pelfrêne, Aurélie, Sébastien Détriché, and Francis Douay. 2014. "Combining Spatial Distribution
with Oral Bioaccessibility of Metals in Smelter-Impacted Soils: Implications for Human
Health Risk Assessment." Environmental Geochemistry and Health 37 (1): 49-62.
doi:10.1007/s10653-014-9629-0.
Root, Elisabeth Dowling. 2012. "Moving Neighborhoods and Health Research Forward: Using
Geographic Methods to Examine the Role of Spatial Scale in Neighborhood Effects on
Health." Annals of the Association of American Geographers 102 (5): 986-995.
doi:10.1080/00045608.2012.659621.
Stehouwer, Richard. 2010. "Lead in Residential Soils: Sources, Testing, and Reducing
Exposure." Penn State Extension. https://extension.psu.edu/lead-in-residential-soils-
sources-testing-and-reducing-exposure.
Su, Bin, and B.W. Ang. 2010. "Input–Output Analysis of CO2 Emissions Embodied in Trade:
the Effects of Spatial Aggregation." Ecological Economics 70 (1): 10-18.
doi:10.1016/j.ecolecon.2010.08.016.
Wei, Chaoyang, Cheng Wang, and Linsheng Yang. 2009. "Characterizing Spatial Distribution
and Sources of Heavy Metals in the Soils from Mining-Smelting Activities in
Shuikoushan, Hunan Province, China." Journal of Environmental Sciences 21 (9): 1230-
1236. doi:10.1016/s1001-0742(08)62409-2.
Wu, Jun, Rufus Edwards, Xueqin Elaine He, Zhen Liu, and Michael Kleinman. 2010. "Spatial
analysis of bioavailable soil lead concentrations in Los Angeles, California."
Environmental Research 110 (4): 309-317. https://doi.org/10.1016/j.envres.2010.02.004.
86

Zhang, Chaosheng, Lin Luo, Weilin Xu, and Valerie Ledwith. 2008. "Use of Local Moran's I and
GIS to Identify Pollution Hotspots of Pb In Urban Soils of Galway, Ireland." Science of
the Total Environment 398 (1-3): 212-221. doi:10.1016/j.scitotenv.2008.03.011.
Zhao, Huarong, Beicheng Xia, Chen Fan, Peng Zhao, and Shili Shen. 2012. "Human Health Risk
from Soil Heavy Metal Contamination Under Different Land Uses Near Dabaoshan
Mine, Southern China." Science of The Total Environment 417-418: 45-54.
doi:10.1016/j.scitotenv.2011.12.047.
87

Appendix A: EBK Model Results
Table 10 Comparisons between EBK model results for the Northern Study Area (The red colored
model results indicate 1
st
best choice, while those colored in blue indicate 2
nd
and 3
rd
best
choices.)
Northern Study Area
Semivariogram
Type
Subset
Size
Standardized
Mean

Standardized
Root Mean
Squared
Root
Mean
Square
Average
Standard
Error
RMS &
ASE
Approx.
Difference
Objective: Nearest to 0 Nearest to 1 Lowest
value
Lowest
difference
Exponential 20 0.01394 0.95593 198.592 184.016 14
Exponential 25 0.01426 0.94336 197.772 185.177 12
Exponential 30 0.00699 0.99130 200.489 179.566 21
Exponential 50 0.00186 1.02557 200.505 176.432 24
Exponential 100 -0.00721 1.09192 201.016 171.059 30
Whittle 20 0.01014 0.96454 197.523 181.316 16
Whittle 25 0.01192 0.94652 196.099 183.357 13
Whittle 30 0.00579 0.99109 200.556 181.266 19
K-Bessel 20 0.00825 0.96884 196.519 182.353 14
K-Bessel 25 0.01094 0.94974 197.573 186.959 11
K-Bessel 30 0.00390 0.99888 201.127 184.207 17

88

Table 11 Comparisons between EBK model results for the 1
st
Southern Study Area (The red
colored model results indicate 1
st
best choice, while those colored in blue indicate 2
nd
and 3
rd
best
choices.)
1
st
Southern Study Area
Semivariogram
Type
Subset
Size
Standardized
Mean

Standardized
Root Mean
Squared
Root
Mean
Square
Average
Standard
Error
RMS &
ASE
Approx.
Difference
Objective: Nearest to 0 Nearest to 1 Lowest
value
Lowest
difference
Exponential 20 0.00332 0.97220 133.279 122.782 10
Exponential 25 0.00432 0.96033 133.495 122.005 11
Exponential 30 -0.00003 0.99747 134.175 119.453 15
Whittle 20 0.00144 0.97616 132.465 122.493 10
Whittle 25 0.00372 0.95872 132.138 121.983 10
Whittle 30 -0.00123 1.00310 133.258 118.816 14
K-Bessel 20 0.00083 0.97739 131.627 123.761 8
K-Bessel 25 0.00256 0.96184 130.917 121.109 10
K-Bessel 30 -0.00199 1.00246 132.653 119.816 13

Table 12 Comparisons between EBK model results for the 2
nd
Southern Study Area (The red
colored model results indicate 1
st
best choice, while those colored in blue indicate 2
nd
and 3
rd
best
choices.)
2
nd
Southern Study Area
Semivariogram
Type
Subset
Size
Standardized
Mean

Standardized
Root Mean
Squared
Root
Mean
Square
Average
Standard
Error
RMS &
ASE
Approx.
Difference
Objective: Nearest to 0 Nearest to 1 Lowest
value
Lowest
difference
Exponential 20 0.01995 0.85843 83.840 83.432 1
Exponential 25 0.01732 0.89699 83.069 82.977 1
Exponential 30 0.00391 0.91733 84.799 83.121 2
Whittle 20 0.01788 0.86920 82.646 80.403 2
Whittle 25 0.01084 0.91876 82.068 81.163 1
Whittle 30 0.00465 0.90616 83.822 84.533 1
K-Bessel 20 0.02061 0.86413 82.807 79.532 3
K-Bessel 25 0.01879 0.90930 81.724 81.599 1
K-Bessel 30 0.00519 0.92088 84.120 83.981 1

Abstract (if available)

Abstract Lead is a significant health threat to people, especially for children where elevated absorption of lead into the bloodstream can cause permanent damage. One site for concern of lead exposure is the surrounding communities of the retired Exide Technologies lead-acid battery smelter in Vernon, California. The California Department of Toxic Substances Control (DTSC) is leading an extensive cleanup effort to remove lead-contaminated soil from affected residences and eliminate the negative health risks posed by the contamination. Soil sampling conducted for approximately 8,500 parcels serves as the primary dataset for this research. While DTSC is currently undertaking the cleanup process on a parcel-by-parcel basis, this thesis works toward understanding the effect of geographic scale in the estimation of levels of lead contamination. It also offers alternatives for identifying priority areas for cleanup by using various aggregation methods and examining how the resulting values may be affected by scale. This research used Empirical Bayesian Kriging to produce interpolated surfaces of lead concentration values. Various aggregation methods were then utilized to aggregate the surfaces into easily defined geographical units of different scales, including block groups, blocks, and parcels. The resulting aggregation values include the mean, percent area, and a Hazard Quotient, an index value for determining health risk. The results demonstrate that the larger areas of the block groups moderate high lead concentration values and thus have lower overall aggregation values for the block groups. In contrast, blocks have a greater tendency to include these high lead concentrations in the aggregations resulting in higher overall values and wider ranges of values for the blocks. This research provides alternative approaches for prioritizing the cleanup of contaminated sites that could be more effective to address the health risks associated with contamination and can be applied to other areas faced with the same problem in the future.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Increase in surface temperature and deep layer nitrate in the California Current: a spatiotemporal analysis of four-dimensional hydrographic data

PDF

Spatial narratives of struggle and activism in the Del Amo and Montrose Superfund cleanups: a community-engaged Web GIS story map

PDF

Evaluating the MAUP scale effects on property crime in San Francisco, California

PDF

An evaluation of soil sampling methods in support of precision agriculture in Northeastern North Carolina

PDF

Modeling nitrate contamination of groundwater in Mountain Home, Idaho using the DRASTIC method

PDF

Estimating at-risk population for lead service lines induced lead exposure and their correlation to socioeconomically disadvantaged neighborhoods in Milwaukee, Wisconsin

PDF

Evaluating surface casing depths of oil & gas operations in an effort to protect local groundwater: a GIS enabled process

PDF

Social media to locate urban displacement: assessing the risk of displacement using volunteered geographic information in the city of Los Angeles

PDF

Surface representations of rainfall at small extents: a study of rainfall mapping based on volunteered geographic information in Kona, Hawaii

PDF

A spatial analysis of beef production and its environmental and health impacts in Texas

PDF

Testing LANDIS-II to stochastically model spatially abstract vegetation trends in the contiguous United States

PDF

Estimating populations at risk in data-poor environments: a geographically disaggregated analysis of Boko Haram terrorism 2009-2014

PDF

Effect of spatial patterns on sampling design performance in a vegetation map accuracy assessment

PDF

Population disaggregation for trade area delineation in retail real estate site analysis

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

Developing and implementing a GIS-based framework to identify optimal locations for clean water wells in sub-Saharan Africa

PDF

Temporal analysis of soil degradation in San Joaquin County, California: a close examination of soil erosion using RUSLE

PDF

Building better species distribution models with machine learning: assessing the role of covariate scale and tuning in Maxent models

PDF

The role of precision in spatial narratives: using a modified discourse quality index to measure the quality of deliberative spatial data

PDF

Assessing the transferability of a species distribution model for predicting the distribution of invasive cogongrass in Alabama

Asset Metadata

Creator Finnstrom, Monica A. (author)

Core Title Soil lead contamination from the Exide battery smelter: the role of spatial scale in cleanup efforts

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 06/06/2019

Defense Date 05/15/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag aggregation,cleanup,concentration,contamination,DTSC,empirical Bayesian kriging,Exide,GIS,health,health threat,Lead,lead concentration,OAI-PMH Harvest,Pb,risk,scale,smelter,Soil,spatial,spatial sciences

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen (committee chair), Lee, Su Jin (committee member), Wu, An-Min (committee member)

Creator Email monicafinnstrom@alumni.usc.edu,monicafinnstrom@yahoo.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-170581

Unique identifier UC11660628

Identifier etd-FinnstromM-7459.pdf (filename),usctheses-c89-170581 (legacy record id)

Legacy Identifier etd-FinnstromM-7459.pdf

Dmrecord 170581

Document Type Thesis

Format application/pdf (imt)

Rights Finnstrom, Monica A.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA