University of Southern California Dissertations and Theses

Exploring the relationship between urban forests and ethnicity in West Covina using GIS and object-based image analysis

(USC Thesis Other)

Exploring the relationship between urban forests and ethnicity in West Covina using GIS and object-based image analysis

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
EXPLORING THE RELATIONSHIP BETWEEN
URBAN FORESTS AND ETHNICITY
IN WEST COVINA
USING GIS AND OBJECT-BASED IMAGE ANALYSIS

by

Kathy Maria Ulloa

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)

December 2015

Copyright 2015 Kathy Maria Ulloa
ii

DEDICATION
This work is dedicated with thanks and love to Ismael, Eliana, Leonardo, parents and entire
family.

iii

ACKNOWLEDGEMENTS
To the various authors mentioned at the end of this work, professors at the University of
Southern California Spatial Sciences Institute including my Thesis Guidance Committee
members Dr. Jennifer Swift and Dr. Robert O. Vos, and to Dr. John P. Wilson for his invaluable
guidance, support and input throughout.

iv

TABLE OF CONTENTS
Dedication ii
Acknowledgements iii
List of Tables vii
List of Figures ix
List of Abbreviations xi
Abstract xiii
Chapter One: Introduction 1
1.1 Urban Ecosystems 1
1.2 Humans in the System 2
1.2.1 Landscape Ecology 2
1.2.2 Incorporating Culture: Exploiting the Spatial Component 4
1.2.2.1 Conceptualizing Human Impact on an Ecosystem 4
1.2.2.2 Increasing our Understanding by Using the Spatial Dimension 5
1.3 Thesis Goals 6
1.4 Thesis Organization 7
Chapter Two: Background and Related Work 8
2.1 Social Capital in Mexico to US Migration Patterns 9
2.2 Agro-related Cultural Capital in Rural to Urban Migration 11
2.2.1 The Mesa Central Migration Route 11
2.2.1.1 The “Mesa Central” Region 12
v

2.2.2 Private Urban Agriculture: Mexican Immigrants & Residential Green
Spaces 13
2.3 Urban Forests in Los Angeles 14
2.3.1 Environmental Benefits of Urban Trees 14
2.3.2 Social Benefits of Urban Trees 15
2.3.3 Managing Urban Forest 16
2.4 Similar Work 17
2.4.1 Urban Tree Canopy Assessments 17
2.4.1.1 Image Analysis Methods and Classifications 18
2.4.2 Individual Urban Tree Detection, Crown Delineation & Species
Differentiation 21
2.4.3 Studying Urban Forests from a Socio-ecological Perspective 22
Chapter Three: Methods 23
3.1 Study Area 23
3.2 Data and Sources 23
3.3 Remote Sensing Methods 27
3.3.1 Extracting UTC 28
3.3.2 Individual Tree Identification 33
3.4 GIS Methods 35
Chapter Four: Results 40
4.1 Spatial Patterns 40
4.2 Summary of Regression Results 43
4.3 Research Questions 50
vi

Chapter Five: Discussion and Conclusions 52
References 56

vii

LIST OF TABLES
Table 1 List of dependent variables aggregated to the US Census
block group level 36
Table 2 List of exploratory variables aggregated to the US Census
block group level 38
Table 3 Exploratory regression results after applying OLS regression to all
possible combinations of explanatory variables:
average Tree Density per Parcel (TDP) 45
Table 4 OLS regression coefficients and significance levels: average TDP 46
Table 5 OLS regression diagnostics: average TDP 46
Table 6 Exploratory regression results after applying OLS regression to all
possible combinations of explanatory variables: average Percent
UTC Cover (PUTCC) per parcel 46
Table 7 OLS regression coefficients and significance levels: average PUTCC
per parcel 47
Table 8 OLS regression diagnostics: average PUTCC per parcel 47
Table 9 Exploratory regression results after applying OLS regression to all
possible combinations of explanatory variables: Average Tree Height
(ATH) per census block group 48
Table 10 OLS regression coefficients and significance levels: ATH per census
block group 48
Table 11 OLS regression diagnostics: ATH per census block group 48

viii

Table 12 Exploratory regression results after applying OLS regression to
all possible combinations of explanatory variables: Trees per
Person (TPP) per census block group 49
Table 13 OLS regression coefficients and significance levels: TPP per census
block group 49
Table 14 OLS regression diagnostics: TPP per census block group 50

ix

LIST OF FIGURES
Figure 1 Schematic showing evolution of urban ecology 3
Figure 2 Illegal Mexican immigration to California from 1963 to 1973 10
Figure 3 Ratio of Mexico per-capita GDP from 1960 to 2000 10
Figure 4 Source regions of illegal Mexican immigrants in 1973 12
Figure 5 The Mesa Central region and major source states of Mexican
immigrants into the US 13
Figure 6 Map of Los Angeles County, Hispanic populations and West Covina
study area location 24
Figure 7 Map of the City of West Covina and major points of interest 25
Figure 8 Example of multiresolution segmentation 28
Figure 9 Example of leaf-off tree being identified as a tree in image analysis 30
Figure 10 UTC Extraction Workflow 31
Figure 11 Resulting UTC extraction 32
Figure 12 Resulting UTC vector layer 32
Figure 13 LiDAR data overlaid over a 1 ft orthoimage 33
Figure 14 Individual tree extraction workflow 34
Figure 15 Extracted individual tree locations 35
Figure 16 Data layers used to produce each variable 39
Figure 17 Maps of Population Density and Median Age by census
block group 41
Figure 18 Maps of Percent Hispanic and Percent Asian by census block group 41

x

Figure 19 Maps of Median Household Income and Average Parcel Size
by CBG. 42
Figure 20 Maps of Average Building Age and Average Parcel Size by census
block group 42
Figure 21 Maps of average TDP and average PUTCC per parcel 44
Figure 22 Maps of ATH and TPP by census block group 44
Figure 23 Front yard of a home in a Hispanic neighborhood in study area 54

xi

LIST OF ABBREVIATIONS
AL Alabama
ATH Average Tree Height
CBG Census Block Group
DC District of Columbia
FTCP Fruit Tree Candidates per Person
GIS Geographic Information System(s)
IL Illinois
INS Immigration and Naturalization Service
LA Los Angeles
LAC Los Angeles County
LAI Leaf Area Index
LARIAC Los Angeles Region Imagery Acquisition Consortium
LiDAR Light Detection and Ranging
NIR Near Infrared
NY New York
NYC New York City
OLS Ordinary Least Squared
PPAP Potential Planting Area per Parcel
PPPAP Percent Potential Planting Area per Parcel
PUTCC Percent Urban Tree Canopy Cover per Parcels
OBIA Object-Based Image Analysis
SFHP Single-family Home Parcels
xii

TD Tree Density
TH Tree Height
TPP Trees per Person
US United States
UTC Urban Tree Canopy
VA Virginia

xiii

ABSTRACT

The relationship between culture and urban forests is explored by analyzing residential urban
trees within the privately owned residential lots of City of West Covina residents in Los Angeles
County, CA. Because the largest percentage of Hispanic immigrants in Los Angeles have
historically come from rural, often agriculturally fertile areas in Mexico, urban forest structure
was studied to identify possible differences in the management practices of privately owned
residential trees in Hispanic neighborhoods; looking for the possibility of increased private urban
agriculture. The second largest minority group in the city, Asians, were incorporated into the
analysis as the second largest minority group and to compare two sets of results. Object-based
image analysis was applied to extract urban forest structure data and OLS regression was
employed to explore these relationships. When controlling for several factors like parcel size,
property values, and income levels, a statistically significant relationship at the 90% confidence
level was found between Hispanic and/or Asian populations and all three dependent variables
describing urban forest structure. An inverse relationship between higher tree densities and the
height of trees and Hispanic populations was found, however, the coefficients were small. Asian
populations were found to have positive associations between all forest structure metrics: a
statistically significant and positive relationship was found between large Asian populations, tree
density, tree height and urban tree canopy cover. Although results showed some connection
between culture and urban forest structure variables, further research and additional methods are
needed to explore the validity, strength and complexity of any relationships found.

1

CHAPTER ONE: INTRODUCTION
According to the US State Department’s Special Representative for Global Intergovernmental
Affairs, “… fifty-two percent of the earth’s population now live in cities and every week one
million new people move to one” (Lewis 2013). She contends that continued rapid urbanization
will lead to three billion new urban dwellers, and that cities, rather than states, are becoming the
focus of governance in the new world order. Landscapes undergo significant land use and land
cover changes under urbanization and concern is growing over how these changes are impacting
the health and life quality of urban residents (McPherson et al. 2011). These landscape
alterations can have enormous costs. One 2006 study found that health costs connected with the
contamination of beaches in southern California totaled $21 to $51 million per year (Given,
Pendleton, and Boehm 2006). As cities expand and population density grows, local governments
are increasingly having to address these issues and are continually being challenged to find
solutions to these problems. Trends, such as the revitalization of downtown areas that can bring a
new influx of people back to the center of a city, are creating new challenges as well as new
opportunities to improve urban environments.
1.1 Urban Ecosystems
There are vast differences between highly disturbed urban ecosystems and those less subjugated
by humans. Human-dominated landscapes have unique biophysical characteristics caused by the
human redistribution of organisms, materials and energy fluxes, and these changes can be
obvious or subtle as well as immediate or long-term (Alberti et al. 2006). Cities are visibly
transformed from their original state and dominated by buildings, roads and other built
structures. But these collaged landscapes also hold a rich array of green spaces in the form of
yards, parks, and commercial landscaping; all playing an integral role in an urban environment.
2

Native vegetation is scarce and is replaced in part by the non-native plants, weeds and trees
growing within the gardens of urban dwellers. Open spaces where native vegetation does have an
opportunity to grow, are often dominated by invasive species of flora and fauna. These
landscapes are not only highly disturbed, they are also highly unstable, only adding to their
ecological complexity.
1.2 Humans in the System
The body of work examining the relationship between urbanization and ecosystems continues to
grow and garner attention as environmental and social challenges in most cities mount. However,
little research has focused on questions surrounding the human and ecological patterns that
emerge from interactions between socioeconomic and physical processes (Alberti et al. 2003;
Wu 2010) or between sociocultural ones and the latter. According to Alberti et al. (2003),
ecological scholars who study urban areas have found themselves challenging ecological theory
to explain the ecology of cities and some have even argued that important theoretical revisions
are needed in order to include human action.
1.2.1 Landscape Ecology
Landscape ecology has brought a broad interdisciplinary and transdisciplinary approach some
would argue is necessary to understand an especially complex urban environment. Evolving out
of an integrative ecosystem approach (Figure 1), it incorporates both bio- and socio-ecological
perspectives to study as well as to influence the relationship between spatial pattern and
ecological processes; at multiple scales including time (Wu 2008; Turner 2001). Landscape
ecology explicitly addresses spatial configuration, composition and form as it approaches process
(Wu 2008) and understandably, many advancements in the field have largely been fueled directly
and/or indirectly by the increased adoption of GIS (Geographic Information Systems), remote
3

Figure 1 Schematic showing evolution of urban ecology
Source: Wu (2010)

sensing and other spatial technologies and techniques, as well as the ever-greater availability of
spatial data including remotely sensed imagery captured at higher and higher resolutions, with
greater coverage and in different spectral ranges. Highly relevant to urban sustainability,
landscape ecology’s contributions have become essential for land use planning and management
(Turner 2001). However, despite the discipline’s explicit humanistic and holistic approach, most
4

research has not focused on the human element. Instead, it has tended to place humans in the
context Turner (2001, p. 7) would suggest: as “but one of many factors creating and responding
to the spatial heterogeneity in an ecosystem”, and not as major active agents of change that are
complex and sometimes unpredictable.
1.2.2 Incorporating Culture: Exploiting the Spatial Component
To restrain human impacts we must learn to understand them. Urban landscapes are both
affected by and developed by culture. In her introduction to “Placing Nature: Culture and
Landscape Ecology”, a collection of essays by researchers of diverse disciplines like philosophy,
ecology, geography, history and landscape architecture, Joan Iverson Nassauer states that in a
world dominated by humans, “we cannot stand apart from nature and now nature as we know it
cannot stand apart from us” (Nassauer 1997, p. 3-4). In order to advance ecological health, she
states, “we must use culture, or we risk removing ourselves altogether from the ecosystems we
know” (Nassauer 1997, p. 3-4). But how do we best conceptualize the relationship between
culture and ecosystem in order to use culture to our ecological advantage? Nassauer (1997)
suggests that “we must formulate ecological questions by considering cultural possibilities and
we must formulate cultural questions by considering ecological processes.”
1.2.2.1 Conceptualizing Human Impact on an Ecosystem
Culture drives the socioeconomic phenomena that integrative disciplines like landscape ecology
and sustainability science study, but these relationships are often difficult to conceptualize. Other
researchers in similar fields have also turned their attention to the human element. Some
conservation biologists have recognized the need for better and more complex representations of
human impact when implementing key concepts like habitat fragmentation in the landscape
matrix (McIntyre and Hobbs 1999). Geography has a long history of studying human-
5

environmental relationships. The term “cultural landscape” for example, has been a fundamental
concept in geography for over a century. Here, the landscape is viewed as one being formed from
a natural landscape by a particular culture, wherein culture is the agent and the natural is the
medium. Wu (2010) argues that the concept of cultural landscape can be useful and effective in
landscape ecology, especially when used in the context of a landscape modification gradient.
However he also notes that “no single perspective is sufficient to understanding human-
environmental relationships and pluralistic approaches are needed to effectively bridge research
cores of different perspectives” (Wu 2010, p. 1148). The spatial sciences play an integral role in
bridging these gaps.
1.2.2.2 Increasing our Understanding by Using the Spatial Dimension
The cultural dimension has largely been ignored in mainstream contemporary landscape ecology
research despite its centrality to the theory (Wu 2010; Alberti et al. 2001; Nassauer 1997).
However, the subject has gained in popularity in recent years, resulting in a surge in research.
This in part can be attributed to the increased manifestation of the ecological pressures our world
faces (Wu 2010), but the surge may also be attributed to an increased integration with the spatial
sciences, largely in the form of an increased adoption of GIS, remote sensing and other
geospatial tools and techniques by researchers. This coupled with an enormous amount of spatial
socioeconomic, demographic and even behavioral data increasingly becoming available offers
great potential. Spatial science theory, methods and tools offer a unique opportunity to support
interdisciplinary and transdisciplinary approaches that can facilitate meaningful new insights
informing the development of these new concepts. The culturally diverse demographics of a city,
for example, may be viewed as just another added layer of variability as we attempt to study an
6

already complex urban environment, but by incorporating and exploring the spatial dimension, it
can afford an opportunity to garner new insights. This study aimed to explore such insights.
1.3 Thesis Goals
The overarching goal of this research was to explore the relationship between culture and urban
trees by analyzing the relationship between Hispanics and urban forests in the City of West
Covina, CA. With the use of high resolution multi-spectral imagery and LiDAR (Light Detection
and Ranging) point cloud data, an Object-Based Image Analysis (OBIA) approach and
regression analysis within a GIS environment that incorporates census and cadastral data, this
study aimed to answer the following questions:
1. Are there quantifiable differences in urban forest structure within predominantly
Hispanic and non-Hispanic neighborhoods in West Covina?
2. If so, what do these structural differences suggest about urban tree management
practices taking place in Hispanic neighborhoods?
3. How does urban forest structure in Hispanic neighborhoods compare with that in
neighborhoods of a different makeup?
4. Are there other socioeconomic, demographic or physical influences affecting
urban forest structure in West Covina?
The City of West Covina was chosen as the study area because it is representative of many of the
cities and neighborhoods in Los Angeles County in terms of its large and increasing Hispanic
population. It was also chosen because of its wide-ranging income levels and the varied
educational attainment of its residents. With declining urban tree canopy cover, West Covina
faces many of the same challenges that other urban areas face (Lee 2012). By studying its urban
forest, this study aims to further our understanding of the dynamics behind urban tree canopy
7

changes and how GIS and remote sensing technologies can be used to unravel and help to
explain them.
1.4 Thesis Organization
The next chapter summarizes related work and starts by exploring the relationship between
Hispanic immigrant populations and urban forests in the Los Angeles metropolitan area. The
chapter provides a brief background on the history of Mexican immigration into the US and
explores questions surrounding how culture may affect the way Mexican and other Hispanic
immigrants view and thereby modify and manage private green spaces, and more specifically,
urban trees. The chapter also discusses the state of our urban forests and why this issue is critical
for the health and well-being of our cities. Examples of other urban forest research are discussed
in the chapter, as well as some of the most common tools and techniques currently being
employed to evaluate the state of our urban forests. Chapter 3 describes the GIS and remote
sensing methodologies used in this study as well as the data used to explore these relationships.
Chapter 4 details the results for this study and the final chapter offers a discussion of the broader
significance of these results and some suggestions for future research as well as the central role
GIS and remote sensing technologies will likely take as we carry our attempts to decipher
human–ecosystem relationships forward.
8

CHAPTER TWO: BACKGROUND AND RELATED WORK
In 1986, the US Immigration Reform and Control Act (IRCA) granted amnesty to nearly three
million illegal immigrants; 2.3 million of which were Mexican-born. Waves of Mexican
immigration had been taking place for decades, and though for many the move had been
temporary, others had chosen to stay, setting roots in their reluctant host-country. During the
first part of the 20
th
Century, Mexican migrants enjoyed somewhat preferential treatment under
US immigration law. At a time when the US was implementing strict quotas limiting
immigration from even some western European countries like Spain and Italy, labor recruiters
were seeking out workers in western Mexico for the rail and agricultural industries (Fussell
2004). Quotas for Mexican immigrants were not put in place until the end of the 1920s, and by
then, the demand and recruitment for low-skilled labor and higher wages had given rise to a
population of approximately 1 million Mexicans living in the US; a population that mixed freely
between the two countries at the time (Anon. 1928; Library of Congress 2015; Borjas 2007).
Things would soon change. The rise in unemployment that began just before and was
accelerated by the Stock Market crash of 1929 led, in part, to changes in the law that restricted
the once steady influx of Mexican low-skilled laborers into the country (Anon. 1928). Anti-
immigrant sentiment grew with job losses and initiatives were put in place by the US
government to encourage, and some would argue push, Mexican migrants to return to their
country of origin (Borjas 2007); mostly men, but some with their American-born families. It is
estimated that hundreds of thousands of Mexicans migrated back to their country of origin during
the 1930s (Library of Congress 2015), but by then, enduring cultural linkages had been formed,
knowledge and treasure had already been gained, and connections had already been established,
9

and this social, cultural and economic capital would inevitably have a lasting impact on the
futures of both countries.
2.1 Social Capital in Mexico to US Migration Patterns
Subsequent waves of Mexico to US migration would largely follow the historical migration
patterns already established in the 1910s and 1920s, patterns that remain strong even today and
that exemplify an interpretation of cumulative causation theory: past migration creates social ties
to the destination and these ties facilitate further migration (Fussell 2004). When in 1942 the US
government began implementing a guest worker program in response to labor shortages
occurring mainly in the agricultural sector, most of the migrants would largely originate from the
same places: predominantly from rural agricultural areas in the high plateau region of “Mesa
Central” in west central Mexico. Many from this new generation of migrants profited from the
social capital gained by prior ones. Often this meant having an upper hand when obtaining the
necessary documentation from farm owners that would allow them to enter and work legally in
the US. The guest worker program came to a halt in 1967, but both legal and illegal immigration
by Mexicans continued to rise (Figure 2), accelerating in the 1980s when the economic
restructuring and ensuing crisis in Mexico made the US labor market more attractive for rural
Mexicans (Figure 3). Most were migrating to urban centers in California and Texas as well as to
an enclave in Chicago where a Mexican population had been established since the early 20
th

century. By 2010, approximately 50 million Hispanics were living in the US, 32 million of them
were of Mexican origin and many of them would track their origins back to these historical
migration patterns (US Census Bureau 2010).

10

Figure 2: Illegal Mexican immigration to California from 1963 to 1973
Source: Dagodag (1975)

Figure 3: Per capita income in Mexico relative to US from 1960 to 2000
Source: Borjas (2007)
11

2.2 Agro-related Cultural Capital in Rural to Urban Migration
Given the rapidly shifting demographics occurring in the US, largely due to the growing
Hispanic and especially Mexican population, identifying and understanding the cultural capital
Mexican and other rural immigrants are transferring to many US urban centers can help us better
identify challenges and opportunities for improving urban environments. Considering what we
know about Mexico-US migration patterns, what can we assert about the role of rurality and/or
agro-related cultural capital in the way Mexican immigrant’s value, manage and modify private
urban green spaces? Do patterns portray challenges or opportunities? The highly concentrated
Hispanic communities that largely exist in urban areas in the US have inadvertently inhibited and
delayed assimilation by even legal Mexican residents and sometimes their American-born
children (Valdez 2005; Telles and Ortiz 2011). This can prolong cultural effects on urban
ecosystems in positive as well as negative ways.
2.2.1 The Mesa Central Migration Route
Although changes in Mexico-US migration patterns have taken place especially since the mid-
1990s, with more skilled, educated, metropolitan Mexican individuals migrating to the US in
recent years, and although large numbers of immigrants from differing states, demographics and
socioeconomic backgrounds have come to the US throughout history, the single largest historical
Mexico-US migration pattern of the 20
th
century still remains that of rural Mexicans from west
central Mexico migrating to the southwestern US (Fussell 2004; Dagodag 1975). The states of
Jalisco, Michoacán and Zacatecas have historically contributed the largest numbers of Mexican
immigrants into the US (Figure 4), with one 1975 study finding that 48 percent of illegal
immigrants apprehended by the INS at the time had originated from the states of Jalisco and
Michoacán alone (Fussell 2004; Dagodag 1975), (Figure 5). Most of these first-time migrants
12

were mestizo and not native-Mexican (Dagodag 1975), and this meant that among rural
communities, they tended to have more social, economic and cultural capital than indigenous
peoples.

Figure 4: Source regions of illegal Mexican immigrants in 1973
Source: Dagodag (1975)
2.2.1.1 The “Mesa Central” Region
The “Mesa Central” is a high plateau region in west central Mexico that has high volcanic
activity, a temperate climate and abundant rainfall. The rich alluvial and volcanic soils create
fertile agricultural areas where even the manufacturing and service sectors of the economy are
strongly tied to agriculture (Fussell 2004). Most of the mestizo rural residents of this area live in
small colonial-style villages surrounded by the ranches and farmland they tend. Cattle grazing is
extensive and the land lends itself to wide-ranging agriculture with major crop production in
corn, sugar cane, avocados, berries and others (Encyclopedia Britannica 2015).
2.2.2 Private Urban Agriculture: Mexican Immigrants and Residential Green Spaces
If we consider the strong agro-related cultural and even agro-related social capital that a large
13

Figure 5: The Mesa Central region and major source states of Mexican
immigrants into the US

number of Mexican immigrants bring to our urban centers, then we may expect that for many
Mexicans, the sustainability practices strongly associated with a rural agricultural way of life
may resonate in their new urban environment thereby affecting the way Mexican immigrants
view, value, and therefore modify and manage private urban green spaces, and in particular,
urban trees. Furthermore, for Mexican immigrants, who are often afforded the ability to require a
lesser degree of assimilation in many urban places, there may be a further intensification and
prolongation of such effects when compared to immigrants from differing cultures. Strong urban
gardening practices in Hispanic neighborhoods predate most recent urban agriculture trends.
Fruit trees and vegetable gardens have lined the front and back yards of Hispanic neighborhoods
14

for decades and the European and now American practice of maintaining green lawns can be a
completely new experience for the rural Hispanic immigrant. Urban trees are some of the most
vital and highly influential flora in urban ecosystems and urban fruit trees tend to be some of the
most common plants of consumption in residential green spaces; can we find differences in how
rural Mexican and other Hispanic immigrants manage them? Are Hispanics more likely to plant
and maintain fruit bearing trees given cultural influences, and inversely, are they more likely to
dispose of trees with no consumption value because of the same reasons? Our urban forests are
on the decline, and for reasons that are poorly understood. Understanding the dynamics behind
urban forest changes and the possible socioeconomic and sociocultural influences affecting them
is important.
2.3 Urban Forests in Los Angeles
The Los Angeles metropolitan area has one of the largest concentrations of Hispanics in the
nation with approximately 48% of the population in the county being Hispanic, and
approximately 34% of Mexican origin. Like other urban areas, Los Angeles faces its own set of
environmental challenges: air and water quality, storm runoff, local flooding, and water
shortages, among others. Urban forests play an integral role in mitigating many of these and
other environmental problems. However, data show that loss of urban tree canopy (UTC), and
the consequent decline of urban green cover is a widespread problem across the US, and the Los
Angeles metropolitan area is no exception. Between the years 2000 and 2010, one study found
that approximately 305,000 trees (and 12% in general green cover) were lost from single-family
neighborhoods in 20 of the largest cities in Los Angeles County (Lee 2012). In an already
stressed environment, losing the ecosystem services provided by established urban trees can be
of great negative impact.
15

2.3.1 Environmental Benefits of Urban Trees
Urban trees provide invaluable benefits to urban residents as well as to the urban ecosystem as a
whole, and by consequence, to those ecosystems that surround it. Increasing UTC, for example,
is one of the most effective ways of reducing temperature; especially important in the “heat
islands” that many urban areas have become. Trees serve as a natural air conditioner; the
evaporation from a single tree can produce the cooling effect of 10 room-sized air conditioners
operating 20 hours a day (Evans 2014). This reduction in local air temperatures can result in less
need for residential cooling, lowering fossil fuel and water consumption at power plants
(McPherson et al. 2011; Shashua-Bar and Hoffman 2000). Trees can also produce a significant
positive effect on storm water runoff rates and volume, helping to prevent local flooding (Dwyer
and Miller 1999): As urbanization increases, so do the impermeable surfaces covering the
landscape. Rainfall will tend to flow quicker to storm water drains and sewers, resulting in more
frequent and severe floods (Dwyer and Miller 1999). An urban tree canopy intercepts this rain
water, slowing its flow. Trees also provide shade, reduce noise, and trap dust and other pollution
with their leaves essentially serving as air filters in a polluted urban environment.
2.3.2 Social Benefits of Urban Trees
Urban trees play a significant role in the social and ecological welfare of cities and their
residents. Increased tree cover has been associated with an increase in respiratory health for
example, despite some tree species being allergic triggers (McPherson et al. 2011). More urban
trees are associated with higher property values, reduced stress and improved general well-being
(McPherson et al. 2011). Troy, Grove, and O’Neil-Dunne (2012) found a strong and inverse
relationship between urban tree canopy and crime rates: a 10% increase in tree canopy cover was
16

linked to a 12% decrease in crime. Trees define space, provide privacy and prompt significant
emotional responses that can be beneficial.
2.3.3 Managing Urban Forests
The environmental and social benefits of urban trees have been well documented and municipal
governments are beginning to take notice. Over the last few years, several tree-planting
campaigns have been launched in cities across the US in an attempt to increase tree canopy.
“MillionTreesNYC” is one such initiative where government agencies, in their efforts to increase
the number of trees in the city of New York are even digging up road surfaces in some cases
(Vermont Monitoring Cooperative 2014). Similarly, the City of Los Angeles launched the
“MillionTreesLA” campaign in 2006 after a UTC assessment determined LA’s tree canopy cover
of 21% was below the national average (Los Angeles Department of Water and Power 2015).
However, a large percentage and in some cases the larger percentage of trees in a city are
managed not by local governing agencies, but by residential property owners, and efforts to
preserve and promote the growth and vitality of urban forests should also focus there. In 2010,
“MillionTreesLA” became “City Plants”, refocusing efforts from a campaign that in large part
concentrated on planting trees on public land, to a more focused effort to increase tree canopy in
low canopy residential neighborhoods. The program provides LA residents with up to seven free
shade trees to plant on their residential properties, holds “adopt a fruit tree” events monthly, and
plants parkway trees (the area between sidewalk and street) at the request of City residents at no
charge (Los Angeles Department of Water and Power 2015). The evolution of this Los Angeles
municipal program highlights both the importance of increasing and preserving urban tree
canopy in residential neighborhoods, including on private property, but also the importance of
identifying and understanding the cultural and socioeconomic dynamics behind differing
17

management practices of urban green spaces. Understanding these dynamics can help better
shape conservation efforts, policy and lead to more informed and effective urban design.
2.4 Similar Work – Studying Urban Forests
A substantial body of work has developed around the quantification of the ecosystem services
urban trees can provide, and in many cases, actual monetary values are being assigned to some of
these benefits. This level of information coupled with an increased adoption of geospatial
technologies by local governments has led to an increased allocation of resources towards spatial
scientific studies of urban forests, in most cases, in the form of urban tree canopy (UTC)
assessments that map and measure urban tree canopy area, and that when repeated, can facilitate
monitoring. Researchers are also making great strides in being able to map individual trees and
delineate individual tree crowns more accurately, as well as in differentiating between specific
tree species, difficult in highly altered and variable urban environments. Some researchers have
taken advantage of the many advancements and increased derivation of urban forest data by
focusing on questions that surround the socio-ecological relationships and processes involving
urban forests; however, this research still remains limited in scope. The three most frequently
pursued research areas are discussed in more detail below.
2.4.1 Urban Tree Canopy Assessments
UTC, the percentage of a site covered by trees, has become a commonly used metric for
assessing the size and structure of urban forests. One of its advantages is the concept’s simplicity
so that it can be more easily understood by the general public (McPherson et al. 2011). During
the past two decades, dozens of UTC assessments have been performed in metropolitan areas
across the US including in major cities like Atlanta, GA, Washington, DC, Baltimore, MD, New
York, NY, and Los Angeles, CA (e.g. McPherson et al. 2011; Troy, et al. 2007; McGee et al.
18

2012). Within the framework of urban landscape ecology, these assessments analyze urban trees
from a landscape level by employing high-resolution aerial or satellite imagery, and in more
recent years, LiDAR, and some form of semi-automated classification technique to extract and
map areas of urban tree canopy in order to quantify and analyze it spatially (McPherson et al.
2011; Walton, Nowak, and Greenfield 2008). Because local governments interested in assessing
their urban forests are usually interested in improving them, being able to identify and then
prioritize land suitable to plant new trees has become a common part of these assessments,
providing local governments with a fairly accurate and cost-effective way to evaluate, monitor
and improve the health and management of urban forests.
2.4.1.1 Image Analysis Methods and Classification Techniques
A large percentage of scholarly urban forest research revolves around the image analysis and
classification techniques used to extract and quantify urban trees and urban tree canopy. Pixel-
based classification methods have traditionally been implemented by researchers to differentiate
between different land use and land cover types, as well as to extract urban tree canopy cover.
McPherson et al. (2011) used a moving masks method in conjunction with a supervised and
unsupervised pixel-based classification technique to map trees and shrubs in the City of Los
Angeles, and found that it had a UTC of 21%, below the national average. Irani and Galvin
(2003) and Nowak, Kuroda, and Crane (2004) also applied pixel-based remote sensing
techniques to assess tree canopy cover over Baltimore, MD by utilizing the near infrared (NIR)
and red bands to discern tree canopy from other green areas and estimating that between 1999
and 2001, Baltimore’s urban forest experienced a 4.2% annual net loss in tree cover. More
recently, McGee and colleagues chose a pixel-based approach to perform a UTC assessment of
the City of Winchester, VA over other approaches “due to the fact that the processing is more
19

easily understood and would be more easily implemented in the future by local jurisdictions”
(McGee et al. 2012, p. 277).
The advancements in the quantity, quality and availability of remotely sensed data,
however, offer a greater capacity for analysis and require more advanced methods to better
exploit these opportunities. The reduction in data acquisition costs has led to more repeatability
and therefore more opportunities for richer information extraction. Many researchers are
beginning to look to other methods because pixel-based image analysis techniques are falling
short of expectations in terms of consistency and efficiency (Chubey, Franklin, and Wulder
2006) and because significant advances in classification techniques are greatly limited by the fact
that they are pixel-based (Blaschke 2010). Unlike humans, the ultimate image interpreters, these
methods are incapable of using photo-interpretative elements such as shape, texture, or spatial
relationships and instead place groups of pixels into classes that are primarily based on their
spectral values (Gao and Mas 2008). In order to extract the spectral characteristics of geographic
features, these methods treat each pixel individually, and assume every pixel to cover an area
relevant to the landscape scale (Gronemeyer 2013). When we consider that there could be
hundreds of millions of pixels to process in a single image, the limitations of pixel-based
methods become increasingly apparent (Blaschke 2010).
OBIA (Object-based Image Analysis) methods, which work differently and more closely
resemble the way humans perform image interpretation, have increasingly gained popularity
among researchers. In this approach humans are the expert system, offering, in theory, endless
potential due to their intuitive nature. Instead of considering the spectral characteristic of each
individual pixel, OBIA begins by creating relatively homogeneous and meaningful objects that
are made up of groups of pixels by considering one or more characteristics explicitly defined by
20

the user; characteristics that can be simple or highly complex. These characteristics can include
spectral value but also, shape, pattern, context, and other cognitive information, placing this
technology at the interface between remote sensing and geographic information science
(Blaschke 2010). OBIA methods are increasingly gaining popularity in urban forest research as
expectations quickly shift toward greater accuracy and finer-scale inventorying of urban forests:
Grove et al. (2006) measured tree canopy over the city of New York using object-oriented
classification methods that included the use of LiDAR data; Moskal and Zeng (2011) measured
tree canopy over Seattle, WA using the RGB and NIR bands in publicly available remote sensing
data and achieved accuracies of over 80%; and Lehrbass and Wang (2010) presented a semi-
automatic, object-based method for urban tree cover extraction that was applied to London,
Ontario, Canada with user’s and producer’s accuracies for trees of 76% and 86%, respectively.
There are further advantages to OBIA, because the spatial relationship information
contained in image objects, for example, allows for more than one level of analysis, it is useful
when making landscape level observations that require multiple scales (Gronemeyer 2013).
OBIA methods also facilitate the integration of different data types, including vector data, and
can create layers to incorporate into the analysis virtually on the fly. OBIA techniques have been
shown to be more accurate at higher resolutions than pixel-based ones (Gao and Mas 2008).
However, despite these advantages, many researchers have been dissuaded from utilizing OBIA
methods since their inception in the 1970s, largely because of their level of difficulty as well as
their dependence on user knowledge and experience (Gao and Mas 2008). Other challenges
include intensive processing that often requires enterprise software, higher software costs, and
limited choices in both software and freeware, and because OBIA methods have only recently
gained traction among researchers, knowledge and resources on which to build new applications
21

are still limited. Despite these limitations, their popularity has grown in recent years and many of
the advancements in image vegetation classification and feature extraction techniques including
those for urban trees can be attributed to the use of these methods.
2.4.2 Individual Urban Tree Detection, Crown Delineation and Species Differentiation
Researchers are developing more accurate methods to collect more specific information about
trees besides general tree canopy cover, and many have tackled individual tree detection,
individual tree crown delineation and even tree species differentiation with a wide variety of
techniques, data and platforms, with good results. However, research in urban tree detection
specifically remains very limited: Bacher and Mayer (2000) performed an automatic extraction
of leafless deciduous trees from 4 cm high resolution aerial imagery taken in spring in Tamm,
Germany. They made use of the dark shadow of the tree as well as the fact that the vertical trunk
is imaged as a straight line and they were able to determine the trunk base, height, and width of
trees with the results showing good potential for their method. Tiede, Hochleitner, and Blaschke
(2005) presented a methodology to extract and delineate single trees in an urban environment
from small footprint high intensity laser scanning point data (LiDAR) in a GIS environment,
employing any additional image data for visualization and accuracy assessment purposes only (a
commonly seen method in current forest research). Dominant trees were detected with an
accuracy of 72%; however, the overall tree detection rate was 51% due to suboptimal scan
sampling distribution that hindered tree crown delineation (Tiede, Hochleitner, and Blaschke
2005). Zhang and Qiu (2012) developed a neural network-based approach to identify urban tree
species at the individual tree level from LiDAR and hyperspectral imagery. Their method was
able to detect individual trees, estimate their tree metrics, and also identified species types with
an accuracy of 96% in detecting individual urban trees and 68% in tree species identification.
22

Shrestha and Wynne (2012) developed prediction models to estimate biophysical parameters
such as height, crown area and biomass for over 2,000 individual trees in central Oklahoma with
the use of a multiple linear regression model. Using LiDAR, they were able to achieve a high
level of accuracy for estimating individual tree height (R
2
= 0.89), crown diameter (R
2
= 0.90),
and biomass (R
2
= 0.67).
2.4.3 Studying Urban Forests from a Socio-ecological or Urban Landscape Ecology Perspective
Despite the recent surge in interest in studying the sociological components of ecological
phenomena occurring in urban environments, urban social science and landscape ecology
research related to urban forests remains very limited in scope: Zhang et al. (2007) studied the
attitudes surrounding urban forests among Montgomery, AL residents and found that higher
income, more educated individuals were more likely to financially support tree planting
programs. They also found that characteristics such as race, gender, and residence were not
statistically significant factors in explaining attitudes toward urban forestry programs. Troy et al.
(2007) used geocoded point crime data and high resolution UTC data with ordinary least squares
and spatially adjusted regression, and found a strong and inverse relationship between tree
canopy and crime: a 10% increase in tree canopy was associated with roughly a 12% decrease in
crime. Iverson and Cook (2000) found a strong correlation between household income and
population density and UTC in Chicago, IL. Flocks et al. (2011) applied a random sampling
approach to study urban forest cover structural inequities and found that white areas had greater
tree density, cover, and species diversity while African American areas had the lowest tree
density and LAI (Leaf Area Index). Interestingly, Hispanic areas were found to have more trees
per hectare than African Americans, the greatest individual tree LAI, and more trees in excellent
condition than either of the two aforementioned groups.
23

CHAPTER THREE: METHODS
To explore the relationship between Hispanics and urban forest structure in the City of West
Covina, data was extracted with image analysis software and subsequently analyzed together
with census and cadastral data within a GIS environment. Eight explanatory variables were
derived at the census block group level including: Average Parcel Size, Average Building Age,
and Average Property Value. In addition, other demographic and socioeconomic variables
including Percent Hispanic, Percent Asian and Median Household income were also analyzed.
This chapter provides a description of the study area chosen, the data and the data sources
employed, the OBIA techniques implemented to extract urban forest structure information, and
the geostatistical methods used to explore possible connections between culture and privately
owned trees.
3.1 Study Area
The city of West Covina is located approximately 20 miles northeast of downtown Los Angeles
(Figures 6 and 7), and had a population of 106,000 as of the 2010 Census. With 54% of the
residents being Hispanic, the City was ranked 25
th
by the US Census Bureau among cities with
the highest percentage of Hispanics in 2010. In addition, 34% of West Covina’s residents
reported that they were foreign born during the last decennial census. The second largest
minority group, Asians, accounted for approximately 26% of its population and was therefore
included in the analysis. The city was chosen as a study area because of the varying demographic
nature of its different neighborhoods, its large and growing Hispanic population, and because in
an earlier study, it was found to have higher than usual UTC losses despite having some of the
largest average residential lot sizes in the county (Lee 2012).

24

Figure 6: Map of Los Angeles County, Hispanic
populations and West Covina study area location

25

Figure 7: Map of the City of West Covina and major points of interest

3.2 Data and Sources
All of the data used in the image analysis and feature extraction originated from the LAC
Department of Regional Planning and the Los Angeles Region Imagery Acquisition Consortium
(LARIAC) Program. The LARIAC Program involves multiple organizations, county
departments and various public agencies in a multi-beneficiary, cost-sharing spatial data
acquisition program. The program launched its first acquisition project in 2006 and
accomplished its fourth (LARIAC 4) in 2014. Most of the LARIAC data employed in this study
originated from the original 2006 acquisition, executed by Infotech Enterprises America using a
single-pass DMC (Digital Mapping Camera) system. The RGB and CIR data were delivered as
26

two separate products with a 1.5 ft horizontal accuracy at a 95% confidence level. For this study,
a 2006 leaf-off 12 in RGB orthogonal image originally acquired at a 4 in resolution was used in
the extraction, as well as a readily available 5-foot NDVI (Normalized Difference Vegetation
Index) raster layer derived from the 2006 RGB and Color Infrared (CIR) orthophotos. A 5-foot
Normalized Digital Surface Model (nDSM) derived from LARIAC’s 2006 LiDAR data was also
used in the feature extraction as height information. Typically, an nDSM is derived by
normalizing a Digital Surface Model (DSM) with a gridded Digital Terrain Model (DTM):
𝑛𝐷𝑆𝑀 = 𝐷𝑆𝑀 − 𝐷𝑇𝑀 (1)
However, the DSM used for this work was normalized by subtracting a Digital Elevation
Model (DEM) from the DSM instead. Although many of these terms are often used
interchangeably, a DEM represents the bare earth surface, whereas a DTM usually attempts to
incorporate more geographic elements and natural features such as rivers and other break lines
(Hashemi 2008; Li et al. 2010). In an urban setting, a DTM is usually based on height points of
streets (Hashemi 2008; Li et al. 2010). Because the DTM was not available in the necessary
gridded format, the readily available LAC “Height” file or DEM-normalized nDSM was utilized
for height information. Vertical accuracy for the LiDAR used to derive all of the aforementioned
terrain datasets was tested with 0.82 ft accuracy at a 95% confidence level (LARIAC Product
Guide 2006). Lastly for the high-resolution feature extraction, a “Building Outlines” vector layer
captured from stereo imagery during the 2008 LARIAC 2 acquisition was employed as a mask to
aid in the classification and to reduce processing times. For the GIS analysis, the aforementioned
building layer, an LAC Assessor’s 2013 cadastral layer and a 2010 US Census Bureau census
block group data layer and corresponding Topologically Integrated Geographic Encoding and
Referencing (TIGER) vector reference layer downloaded from the US Census Bureau website
27

were also incorporated into the analysis. All of the data was preprocessed to match a single
geographic projection (NAD 1983, State Plane, Zone V) and clipped to the extent of the city
boundary of West Covina.
3.3 Remote Sensing Methods
An OBIA approach was used to extract both the general UTC area as well as individual trees and
tree heights. Trimble
®
eCognition Developer was chosen as the image analysis software. The
first of its kind, the software was founded by Nobel laureate and OBIA pioneer Gerd Binning
and his team of researchers and was first launched in 2000 by Munich, Germany-based Company
Definiens
®
(and acquired by Trimble
®
in 2010) (Nassbaum 2008; Esch et al. 2008). Arguably,
much of the software’s success can be attributed to their software implementation of multi-
resolution segmentation for the purposes of remote sensing, allowing for multi-source, multi-
region, multi-method and multi-scale image analysis, that is well suited to landscape ecology and
other urban forest research. Multi-resolution segmentation was the main segmentation algorithm
used in this study and Esch et al. (2008, p. 2) describes the segmentation in the following
manner: the segmentation is controlled by heterogeneity criteria color and shape. Color ℎ
𝑐 is
calculated:
ℎ
𝑐 = ∑ 𝑤 𝑏 𝑏 ∗ 𝜎 𝑏 (2)
where b = the band, w b = the weight, and σb = the standard deviation of the band. The shape is
composed of both smoothness ℎ
𝑠𝑠
and compactness ℎ
𝑠𝑐
which are calculated as follows:
ℎ
𝑠𝑠
= 1/𝑘 (3)

ℎ
𝑠𝑐
= 𝑙 /√𝑛 (4)
28

where 𝑙 = the length of the object’s outline, 𝑘 = the shortest length of the bounding box, and 𝑛 =
the number of pixels of the object. After each merge process, the change of heterogeneity, which
flows into the fusion value 𝑆 𝑓 , was calculated in the following manner:
𝑆 𝑓 = 𝑤 𝑠 ℎ
𝑐 + (1 − 𝑤 𝑠 )ℎ
𝑠 (5)
where, 𝑤 𝑠 represents a user-defined weighting factor of the shape criterion.
The fusion value was then compared with a user-defined scale parameter which defines
the maximum allowable heterogeneity of the image objects. By varying the scale parameter,
arbitrary object levels with scale-specific segment sizes can be generated. An example of multi-
resolution segmentation can be seen in Figure 8.

Figure 8 Example of multi-resolution segmentation

3.3.1 Extracting UTC
To extract the general UTC area, a multi-resolution segmentation was first applied to the RGB
bands of the 1 ft orthoimage, taking advantage of the highest resolution of the dataset in this first
segmentation to create the first primitive or sub-objects. Because the red band is known to
29

provide the most vegetation information of the three bands, a higher weight could have
potentially been assigned to it; however, weighting each of the three bands equally for the
segmentation provided the best results for this particular extraction process. The smallest
possible scale parameter or highest possible spatial resolution was chosen, allowing for the edges
of even small trees to be properly segmented. The shape and compactness were set to 0.9 and
0.1, respectively. Setting the highest shape value possible (0.9) meant that less weight was given
to color, providing good discrimination between grass and tree canopy cover. The compactness
value (0.1) was chosen by testing several parameter combinations. Higher compactness values
would have resulted in more bounded objects. Although some research has focused on the
optimization of these parameters, this process still remains largely an iterative one for
researchers, highly dependent on the dataset as well as the information being extracted.
In the next step, classification, the NDVI layer was used to differentiate between
vegetation and non-vegetation sub-objects. Using a threshold condition of ≥ -0.045, lower than
the more common > 0.1 NDVI threshold used to classify vegetation, leaf-off and sparsely leaved
trees were clearly identified as such. Figure 9 shows a leaf-off tree in a parking lot (marked with
an X) on the left, and is identified as a tree object in green on the right along with other green
canopy. Within residential lots, lowering this threshold resulted in small misclassifications
mainly over portions of light colored roofs, walkways, and driveways. These misclassifications
were largely addressed during subsequent steps in the analysis. More prominent errors occurred
over commercial areas where light colored surfaces tend to be more common. Here, larger areas
of parking lots and commercial building roofs were confused with vegetation. These larger
errors, however, were not relevant to the scope of this study and were later filtered out when
30

Figure 9 Example of leaf-off tree being identified as a tree in image analysis

single-family home residential lots were isolated for the analysis. For the purposes of this study,
the benefit of more accurate leaf-off tree detection greatly outweighed the remaining degree of
error.
Next, the LiDAR was used to discriminate between vegetation of different heights.
Common thresholds for tree heights range between 5 and 8 ft, largely dependent on the tree type.
For this study, a height threshold of 5 ft was chosen to classify trees, allowing for more
flexibility in the data post-extraction where lower height values could easily be filtered out. More
importantly, due to the coarseness of the NDVI and nDSM as well as the natural structure of
most tree tops, tree heights tended to be underestimated. Setting this minimum height allowed for
better identification of smaller trees and also of the edges of trees. This step also largely
addressed any misclassifications over walkways and driveways. Finally, the vector building
outline from LAC Assessor’s 2013 cadastral data was used to mask out buildings, largely
addressing the misclassifications over roofs. A merging operation was applied to tree canopy
31

objects before extracting UTC objects and exporting them as vector polygon data. Figure 10
summarizes the aforementioned workflow.

Figure 10 UTC extraction workflow
The resulting UTC extraction can be seen in Figures 11 and 12. Because of the
coarseness in the LiDAR data, the UTC was slightly overestimated. This can clearly be seen in
Figure 11, where the UTC is overlaid on a 4 inch RGB orthoimage. Figure 13 illustrates this
problem more clearly. The coarseness of the nDSM in violet juxtaposed with the 1 ft orthoimage
seen in green. The lightest violet colors show the highest points in the image. Figure 12 shows
the resulting UTC extraction in vector format with and without a base image layer.
32

Figure 11 Resulting UTC extraction

Figure 12 Resulting UTC vector layer

33

Figure 13 LiDAR data overlaid over a 1 ft orthoimage
3.3.2 Individual Tree Identification
In order to identify individual tree tops, the NDVI and nDSM layers were again utilized. A
multi-resolution segmentation was applied with the lowest possible scale parameter, with a
weight of 2 for the nDSM height layer and a weight of 1 for the NDVI layer. A lower emphasis
on shape (0.3) and a higher compactness value (0.7) that would create more bounded objects was
chosen because of the coarseness of the data (Figure 9). Using only these two layers of
information provided better individual tree detection than when a sharper image was
incorporated into the analysis. The same classification criteria for UTC was then applied, ≥ 0.045
for the NDVI and ≥ 5 ft for the nDSM. Here, a chessboard segmentation that divides the image
into a predefined grid would commonly be applied to the areas classified as tree canopy before
applying a local maxima algorithm to identify potential individual tree tops. Instead, a spectral
difference segmentation that merges neighboring objects according to their mean layer intensity
values was first applied, using both the nDSM and NDVI layers as equally valued weights. This
34

method provided better results in this particular case with the particular set of parameters. The
assigned search range value in the find local maxima algorithm is predefined normally
depending on the forest type. The spectral difference segmentation provided more meaningful
objects for the search range. In this highly heterogeneous urban forest, this method provided
more accurate local maxima designation. After some clean-up operations, the resulting local
maxima points or individual tree crown points were exported as vector data with height and
NDVI attribute information. Figure 14 below describes the workflow and Figure 15 shows the
individual tree identification results.

Figure 14 Individual tree extraction workflow
35

Figure 15 Extracted individual tree locations
3.4 GIS Methods
In traditional statistics, observations are assumed to exist in a world that is abstract and void of
any external influences, in spatial statistics, the observations are real world phenomena with a
location, and almost always exhibit regional and local trends. Post feature extraction, the
remotely sensed data was imported into ArcGIS 10.2 for geostatistical analysis. An Ordinary
Least Squared (OLS) global regression was applied to model the relationship between forest
structure and several socioeconomic, demographic and cadastral variables. A Global Morans I
was then applied to the residuals to check for spatial autocorrelation that would indicate spatial
clustering of residuals and misspecification of the model due to some spatial variables not
accounted for in the model but present in the relationship(s).
3.4.1 Dependent Variables
Four dependent variables were derived for the analysis, all with the use of the extracted urban
forest data. Two aggregation levels were used to derive these variables: the first at the parcel
level, and the second at the census tract level (the coarsest level of granularity and the level used
36

for the regression analysis). Dependent variables at the census block group level included:
Average Tree Density per Parcel (TDP); average Percent UTC Canopy Cover (PUTCC) per
parcel; Trees per Person (TPP); and Average Tree Height (ATH) per parcel. Table 1 describes
each of these variables in more detail.
Table 1 List of dependent variables at the US Census block group level
Variable Abbreviation Explanation Problems
Average Tree
Density per Parcel
(TDP)
Avg_TDP Number of individual trees per sq. ft of
available planting area within each
residential parcel, aggregated to the
census block group level
Available planting area
does not exclude pools or
other hardscape and
further image processing
would be necessary to
address; error in feature
data extraction
Average Percent
UTC Cover per
Parcel (PUTCC_p)
Avg_PUTCC_p Average percent urban tree canopy cover
of available planting area within each
residential parcel, aggregated to the
census block group level.
Available planting area
does not exclude pools or
other hardscape; does not
include UTC cover over
buildings; error in feature
data extraction
Average Tree
Height (ATH)
ATH The average tree height in census block
group
Error in feature data
extraction
Trees per Person
(TPP)
TPP Total number of trees in block
normalized by the total census block
group population
Error in the data
extraction; US Census
Bureau estimates
3.4.2 Derivation of Variables
To calculate dependent and explanatory variables, the LAC Assessor’s cadastral data was first
used to isolate single-family home parcels (SFHP). All other parcels including commercial and
multiple family dwellings parcels were filtered out. Next, an erase overlay operation was applied
to mask out buildings in SFHP parcels, leaving a theoretical Parcel Potential Planting Area
(PPAP) within SFHP Parcels:
PPAP = Single Family Home Parcel (SFHP) – Building Footprint (6)
PPAP did not exclude pools or other impermeable surfaces, and further land-use classification
methods or data layers would have to be added to identify them. Next, an overlay operation
37

intersecting parcel features and the extracted individual trees was applied. This assigned a parcel
identification number to each individual tree. This information was then summarized to a table
by parcel number and was subsequently joined back to the parcel layer. The Tree Density per
Parcel (TDP) was then calculated by dividing the total number of trees in each parcel t, by the
potential planting area:
𝑇𝐷𝑃 =
𝑡 𝑃𝑃𝐴
(7)
Next, to calculate the average PUTCC per parcel (the percentage of UTC within the
potential planting area of each SFHP), the extracted UTC features were intersected with the
SFHP layer, assigning a parcel identifier to each feature. Again a field summary operation was
applied to the parcel identifier. A dissolve operation was used, aggregating UTC polygons to the
parcel level, then calculating the area for each. Then the total UTC area u of each parcel was
calculated and then divided by the PPA:
𝑃𝑈𝑇𝐶𝐶 =
𝑢 𝑃𝑃𝐴
(8)
Another overlay intersect operation was applied to aggregate data to a second level. The
average TDP and PUTCC per parcels for each census block was then calculated. To calculate the
average PUTCC in each census block group, the UTC areas in each parcel were summed by
census block group to a table, joined back to the analysis and then divided by total PPA in each
census block group. The Average Tree Height (ATH) for each census block group was
calculated and included as a dependent variable. Trees per Person (TPP) was also calculated by
normalizing the total number of trees in each census block group by the total census block group
population. Because these variables were measured at different scales, the possibility of
38

introducing issues of MAUP (modifiable areal unit problem) and/or ecological fallacy into the
models exists.
3.4.2 Explanatory Variables
Eight explanatory variables were derived at the census block group level including: Average
Parcel Size, Average Building Age, and Average Property Value. In addition, single-level census
block group level data calculations were performed to obtain: Population Density, Percent
Hispanic, and Percent Asian. Two existing Census data fields, Median Age and Median
Household Income, were also included as explanatory variables (Table 2). Figure 16 shows
variables and data sources used to derive each variable.

Table 2 List of explanatory variables aggregated to the US Census block group level
Variable Abbreviation Explanation Problems/Potential Error
Population
Density
Pop_Den People per square mile US Census Bureau estimates
Percent Hispanic Perc_Hisp Percentage of Hispanics living in the
census block group
US Census Bureau estimates (historically
thought to be underestimated)

Percent Asian Perc_Asian Percentage of Asians living in the census
block group
US Census Bureau estimates (historically
thought to be underestimated)

Median Age Med_Age Median age of census tract population US Census Bureau estimates
Median Income Med_Incm Median income of census tract
population
US Census Bureau estimates
Average Parcel
Size
Avg_PrclSz Average parcel size in census block
group, sq. ft
Error in cadastral data
Average
Building Age
Avg_BldgAg Average Age of home in the census
block group
Error in cadastral data
Average
Property Value
Avg_PrpVal Average single home property value
within study are and study parcels
Error in cadastral data

39

Figure 16 Data layers used to produce each variable

3.4.3 Exploratory Regression and OLS
Next, an exploratory regression was applied to each of the four dependent variables with the
eight explanatory variables. The explanatory regression was used to explore variable significance
when choosing explanatory variables for OLS models. All variables with no significance were
removed. The suggested regression model(s) with the largest number of variables and consisting
of only statistically significant p-values < .10 and highest R
2
values, were used as the starting
point for fitting the OLS models for each of the dependent variables. A series of OLS models
were tested for each variable. Results and models are discussed in the next chapter.
Residential
UTC
Average Parcel Size
Average Building Age
Average Property Value
Average Tree Density per
Parcel (TDP)
Average Percent UTC Cover
per Parcel (Avg_PUTCC_p)
Trees per Person (TPP)
Average Tree Height (ATH)

Population Density
Percent Hispanic
Percent Asian
Median Age
LAC Parcels
Individual
Tree Points
Building
Outlines
2010 US
Census
Block Data
40

CHAPTER FOUR: RESULTS
Six of the 79 census block groups (CBGs) spanning the City of West Covina in 2010 were
excluded from the analysis because they did not contain parcels classified as single-home
residential, leaving 73 remaining census block groups in the analysis. With most of its land
contained in the north, the results portrayed two distinct sides of West Covina: the northeast and
the northwest. The northeast had higher Asian populations, higher incomes, larger lot sizes and
newer homes. The northwest had higher concentrations of Hispanic populations and portrayed a
more checkered and complex picture. While some links were found between urban forest
structure and Hispanics, the more significant relationships were found between forest structure
variables and Asian populations.
4.1 Spatial Patterns
Spatial patterns were detected in the demographic, socioeconomic, housing stock and urban
forest structure data used in the analysis. The denser northwest was found to have larger numbers
of CBGs with higher population densities (Figure 17a). Other CBGs with high population
densities are located sporadically over the remainder of the city. Many of the least densely
populated groups tended to be the ones with higher median ages (Figure 17b). Many census
block groups did not follow this pattern, however. Higher concentrations of Hispanics tended to
be located in the northeast portion of the City, while Asian populations tended to concentrate in
the southwest (Figures 18a and 18b). The least densely populated CBGs tended to have some of
the highest incomes with central West Covina having among the highest. Many census block
groups with higher median incomes also occurred in the eastern portion of the city (Figure 19a).
Some higher income neighborhoods were found in the northwest part of the city where Hispanic
populations also tended to concentrate (Figure 19a), despite the higher property values
41

Figure 17 Maps of Population Density (a) and Median Age (b) by census block group
Figure 18 Maps of Percent Hispanic (a) and Percent Asian (b) by census block group
(a) (b)
(a)
(b)
42

Figure 19 Maps of Median Household Income (a) and Average Parcel Size (b) by CBG
Figure 20 Maps of Average Building Age (a) and Average Parcel Size (b) by census block group
(a)
(a)
(b)
(b)
43

and larger lot sizes found in the eastern areas of the city (Figures 19b and 20b). Some high
Hispanic, high income and low property value census block groups were located in the
northwestern quadrant. Older homes also tend to be located in the northwest, with most homes in
these areas having been built over 50 years ago (Figure 20a).
The urban forest structure data that was derived did not show such clear patterns at the
CBG level, with the exception of TDP. The average tree density per parcel (TDP) was found to
be higher in the south and eastern portions of the city, as well as in some CBGs located in the
northwest part (Figure 21a). A similar but less pronounced pattern can be seen for UTC cover.
The average percent UTC cover in these areas seemed to be higher, where higher property
values, more people of Asian descent and higher incomes were also found. Some CBGs in the
northeast part of the city also showed higher tree densities and percentages of UTC cover. The
average height of trees (ATH) and trees per person (TPP) did not show strong patterns (Figure
22), however there were some clusters of CBGs with higher ATH in the eastern, western, and
central areas of the city.
4.2 Summary of Regression Results
Three tables are included for each of the modeled relationships: an exploratory summary of
variable significance, a table showing OLS regression results including coefficients and p-values,
and a table showing other OLS statistical tests. None of the variables in the models were found to
have VIF (Variance Inflation Factor) values greater than 7.5 (> 7.5). Values greater than 7.5
would have indicated high multi-collinearity between variables (O’Brian 2007; Esri 2014). With
the exception of Average Tree Density in Parcel, all variables used in the models were used were
at the > 90% significance level for either probability or robust probability values. The Koenker

44

Figure 21 Maps of average TDP (a) and average PUTCC (b) per parcel

Figure 22 Maps of ATH (a) and TPP (b) by census block groupue
(a)
(a)
(b)
(b)
45

Breusch-Pagan test was not statistically significant (p < 0.01) for any of the models. A
statistically significant Koenker (BP) value would have indicated the relationships modeled were
not consistent due to non-stationarity or heteroscedasticity (Cai et al. 1998; Esri 2014). Jarque-
Bera, one of the most commonly applied tests to evaluate model fit was not found to be
statistically significant (p < 0.01) for any of the models. (Jarque and Bera 1987; Esri 2014). This
indicated that the residuals were not found to be normally distributed and that no significant bias
was found in the predictions due to non-linear relationships, data outliers or other problems. (Esri
2014).
Tables 3 through 5 show the results for the average tree density (TDP) per parcel model.
A statistically positive relationship was found between the percentage of Asians living in the area
and tree density in residential parcels. In addition, parcels with older homes tended to have lower
tree densities irrespective of parcel size.

Table 3 Exploratory regression results after applying OLS regression to all possible
combinations of explanatory variables: average Tree Density per Parcel (TDP)

Variable Percent Significant Percent Negative Percent Positive
PERC_ASIAN 100.00 0.00 100.00
AVG_BLDGAG 100.00 100.00 0.00
AVG_PRPVAL 50.00 50.00 50.00
PERC_HISP 25.00 64.06 35.94
MED_AGE 23.44 99.22 0.78
AVG_PRCLSZ 18.75 48.44 51.56
POP_DEN 0.00 82.03 17.97
MED_INCM 0.00 15.62 84.38
46

Table 4 OLS regression coefficients and significance levels: average TDP

Variable Coefficient P-Value Robust P-Value VIF
Intercept 73.536757 0.000000*** 0.000000*** --------
PERC_HISP 0.061253 0.199206 0.135847 1.800252
PERC_ASIAN 0.151828 0.007432*** 0.013586** 1.920284
AVG_BLDAG -0.501719 0.000000*** 0.000000*** 1.557480
*Significant at 90% level **Significant at 95% level ***Significant at 99% level

Table 5 OLS regression diagnostics: average TDP
n 73
Multiple R
2
0.616370
Adjusted R
2
0.599690
Koenker (BP) Statistic

12.571826
Jarque-Bera Statistic

16.699200
Global Morans I 1.604524

Tables 6 through 8 show the results for the percentage of UTC over residential lots OLS model.
A positive relationship was also found between higher percentages of UTC on residential lots
and Asian populations. In addition, median age and property value were also found to be
statistically significant but negatively correlated with UTC.
Table 6 Exploratory regression results after applying OLS regression to all possible
combinations of explanatory variables: average Percent UTC Cover (PUTCC) per parcel

Variable Percent Significant Percent Negative Percent Positive
AVG_PRPVAL 97.66 100.00 0.00
PERC_ASIAN 53.12 0.00 100.00
MED_AGE 45.31 100.00 0.00
AVG_PRCLSZ 10.16 50.00 50.00
AVG_BLDGAG 9.38 68.75 31.25
MED_INCM 7.81 100.00 0.00
POP_DEN 0.00 96.88 3.12
PERC_HISP 0.00 39.84 60.16
47

Table 7 OLS regression coefficients and significance levels: average PUTCC per parcel
Variable Coefficient P-Value Robust P-Value VIF
Intercept 37.853496 0.000000*** 0.000000*** --------
PERC_ASIAN 0.087237 0.025322** 0.027469** 1.143105
MED_AGE -0.178216 0.030954** 0.045515** 1.076361
AVG_PRPVAL -0.000014 0.006765*** 0.029085** 1.099178
*Significant at 90% level **Significant at 95% level ***Significant at 99% level

Table 8 OLS regression diagnostics: average PUTCC per parcel

n 73
Multiple R
2
0.178507
Adjuster R
2
0.142789
Koenker (BP) Statistic

5.891550
Jarque-Bera Statistic

1.300332
Global Morans I, z-Score 0.708727

A statistically significant negative relationship was found between the number of
Hispanics in each census block and the average height of trees (ATH) within SFHP parcels in the
study area (Tables 9-11). Higher numbers of Hispanic residents were associated with lower tree
heights. In contrast, the relationship between Asian populations and tree heights was positive.
Taller trees tended to be correlated with higher Asian population numbers. Median age was
negatively correlated: a younger demographic profile in the census block group was associated
with taller trees. Higher incomes, higher property values and larger parcel sizes were also
associated with taller trees. Properties with older homes tended to have shorter trees in addition
to lower concentrations and lower percentages of canopy cover.

48

Table 9 Exploratory regression results after applying OLS regression to all possible
combinations of explanatory variables: Average Tree Height (ATH)
per census block group

Table 10 OLS regression coefficients and significance levels: ATH per census block group
Variable Coefficient P-Value Robust P-Value VIF
Intercept 33.266574 0.000000* 7.608690 --------
PERC_HISP -0.063328 0.005185*** 0.010868*** 1.660652
MED_AGE -0.212528 0.000058*** 0.000057*** 1.386687
MED_INCM 0.000034 0.035142** 0.016711*** 1.203189
AVG_PRCLSZ 0.000344 0.000010*** 0.000002*** 3.362716
AVG_BLDGAG -0.107541 0.010924*** 0.041534** 2.288284
AVG_PRPVAL 0.000021 0.000597*** 0.002280*** 4.931695
*Significant at 90% level **Significant at 95% level ***Significant at 99% level
Table 11 OLS regression diagnostics: ATH per census block group
n 73
Multiple R
2
0.446353

Adjuster R
2
0.396022
Koenker (BP) Statistic

5.000781
Jarque-Bera Statistic

1.520840
Global Morans I, z-Score
0.319906

Variable Percent Significant Percent Negative Percent Positive
AVG_PRCLSZ 100.00 0.00 100.00
MED_AGE 96.88 100.00 0.00
AVG_BLDGAG 57.81 100.00 0.00
MED_INCM 50.00 0.00 100.00
PERC_HISP 46.88 100.00 0.00
AVG_PRPVAL 43.75 50.00 50.00
PERC_ASIAN 29.69 0.00 100.00
POP_DEN 0.00 100.00 0.00
49

The TPP (Trees per Person) dependent variable model found that more trees per person
were associated with higher incomes, higher property values and lower median ages (Tables 12-
14). Areas with higher percentages of Hispanics tended to have fewer trees per person. Higher
Asian populations pointed to more robust urban forest plots, with more trees per person being
associated with higher concentrations of Asian residents.

Table 12 Exploratory regression results after applying OLS regression to every possible
combination of explanatory variables for Trees per Person (TPP) per census block group

Table 13 OLS regression coefficients and significance levels: Trees per Person (TPP) per
census block group
Variable Coefficient P-Value Robust P-Value VIF
Intercept -0.790664 0.391922 0.446162 --------
PERC_HISP -0.011530 0.112201 0.050013** 1.330502
MED_AGE 0.027891 0.109947 0.098454* 1.265537
MED_INCM 0.000011 0.066781* 0.003808*** 1.197055
AVG_PRPVAL 0.000004 0.000335*** 0.066007* 1.233927
*Significant at 90% level **Significant at 95% level ***Significant at 99% level

Variable Percent Significant Percent Negative Percent Positive
MED_INCM 100.00 0.00 100.00
PERC_HISP 89.84 100.00 0.00
AVG_PRPVAL 87.50 0.00 100.00
MED_AGE 74.22 0.00 100.00
PERC_ASIAN 29.69 9.38 90.62
AVG_BLDGAG 10.94 48.44 51.56
AVG_PRCLSZ 7.81 50.00 50.00
POP_DEN 0.00 21.09 78.91
50

Table 14 OLS regression diagnostics: Trees per Person (TPP)
n 73
Multiple R
2
0.413886
Adjuster R
2
0.379408
Koenker (BP) Statistic

25.757151
Jarque-Bera Statistic

73.106057
Global Morans I, z-Score
0.577187

4.3 Research Questions
The goal of course, was to use the aforementioned results to answer the four research questions
posed in Chapter 1. The following bullets summarize what was learnt about socio-cultural and
ecological processes and outcomes in the City of West Covina.
1. Are there quantifiable differences in urban forest structure within
predominantly Hispanic and non-Hispanic neighborhoods in LAC?
An inverse relationship between higher tree densities and the height of trees and
Hispanic populations was found, however, although the p-values were significant,
the coefficients were small. The highest UTC was associated with CBGs with
large Asian populations and no significant relationship was found between UTC
and percent Hispanic.
2. If so, what do these structural differences suggest about urban tree
management practices taking place in Hispanic neighborhoods?
The results support the notion that Hispanic neighborhoods tend to have a higher
number of smaller trees with no significant increase or decrease in UTC cover.
However, although the relationship was significant it was not found to be
51

particularly strong and extending the study to a larger area may yield clearer and
more significant results.
3. How does urban forest structure in Hispanic neighborhoods compare with
that in neighborhoods of a different makeup?
Asian populations were found to have positive associations between all forest
structure data. A statistically significant and positive relationship was found
between large Asian populations, tree density, tree height and UTC cover.
4. Are there other socioeconomic, demographic or physical influences affecting
urban forest structure in LAC?
A negative, statistically significant relationship was found between building age
and tree density. Older homes tended to have lower tree densities. UTC and
median age also had a negative relationship. More UTC was associated with
CBGs with younger populations and lower property values. Taller trees were
associated with CBGs with younger populations, higher incomes, larger parcels,
newer homes and higher property values.

52

CHAPTER 5: DISCUSSION AND CONCLUSIONS
Are residential green spaces in Hispanic neighborhoods more rural in nature? If there are more
urban agricultural practices taking place in certain areas what does that mean for the landscape?
Are there more fertilizers and pesticides being used that may be leaching into the soil? What are
the implications given the water shortages in California? These are larger questions that should
be addressed.
This study set out to analyze the relationship between culture and urban ecosystems by
studying the relationship between Hispanic populations and urban forests in Los Angeles.
Because the largest percentage of Hispanic immigrants in Los Angeles have historically come
from rural, often agriculturally fertile areas in Mexico, urban forest structure was studied to
identify possible differences in the management practices of privately owned residential trees in
Hispanic neighborhoods. Using remote sensing techniques to extract urban forest structure data
and geostatistical tools to map and analyze these relationships, three variables describing urban
forest structure within single home residential parcels for the City of West Covina were derived
and analyzed: the average density of trees within each parcel, the percentage of UTC cover over
each residential lot, and the average height of trees in lots spanning 73 census block groups in the
study area. When controlling for several factors including parcel size, property value and income
levels, a statistically significant relationship at the 90% confidence level was found between
Hispanic and/or Asian populations and all three dependent variables.
The lower tree heights found in Hispanic neighborhoods support the idea that Hispanic
immigrants may have larger numbers of fruit trees that tend to be kept at shorter heights. This
could also suggest larger percentages of younger trees. Overall, one of the most significant
(positive) relationships identified was between Asian populations and tree density within each
parcel: the percentage of Asians and average building age in the census block alone explained
53

over 60% of the variation in tree density. There were other positive links between Asians and
urban forests: neighborhoods with higher percentages of Asian residents were found to have
taller trees as well as higher percentages of UTC cover. Obtaining individual tree point data
provided more samples and therefore richer data extraction at multiple level. Delineation of each
individual tree crown for measurement, which was not achieved in this study, would have
provided additional information on urban forest structure. Finer-scale inputs with individual tree
points, refining the study area and applying two levels of aggregation (parcel and parcel census
block group) were methods applied in an attempt to yield finer-scale results. In future studies,
ownership rates as well as other ethnic and racial backgrounds should be incorporated. In
addition to a change detection analysis, extremely useful in studying these relationships, an
improved feature extraction, achieving individual tree crown delineation and more accurate
delineation of potential planting area per parcel (PPAP) could be useful. Utilizing more cognitive
variables during the feature extraction would have improved the image classification and feature
data extraction, increasing accuracy. In addition, being able to differentiate between tree species
could provide valuable information and clearer answers (see Figure 18 for example).
The interactions studied here, however, remain highly complex, and more work, layered
on top the results shown here and using other methods such as random surveys of residents,
would be needed to accept or reject the hypothesis that Hispanics and/or Asians have a unique
approach to managing urban trees. The self-sustainability practices as well as the cultural,
political, economic, and ecological dynamics experienced by an immigrant both in their place of
origin as well as their new environment all have the potential to influence the value placed on
and the management practices of privately owned urban residential green spaces. The degree of
need for assimilation in the host country also plays a role. The term ‘‘ecology of prestige’’, for
54

example, suggests that a household’s land management decisions are influenced by its desire to
uphold the prestige of its community and outwardly express its membership in a given lifestyle
group (Troy et al. 2007). Studying these spatial relationships can shed insight into these
dynamics and their significance (Figure 23).

Figure 23 Photo of the front yard of a home in a West Covina Hispanic neighborhood showing a
newer avocado tree and full grown orange tree.
Photo by Kathy M. Ulloa

Understanding these relationships can inform us in improving our urban forests as efforts
need to reach beyond government agencies. Outreach campaigns can be targeted towards
maintaining established trees and not just planting new ones. More education on the importance
and care of trees may prove useful in Hispanic neighborhoods. As forest structure spatial data
becomes increasingly common in urban centers, exploiting the data to analyze these complex
relationships can serve as groundwork for other studies.
55

Differing management practices can have unexpected implications. Because shorter trees
take up more space near the ground, for example, it can mean less open space available for
children to play. This could potentially also mean closer and more frequent contact with
fertilizers and pesticides used for fruit trees. Although the importance of urban forests has been
well studied, understanding the ecological processes taking place and how factors like culture is
influential remains underexplored. It is important to understand these dynamics both to mitigate
losses and improve urban forest health. As the world increasingly becomes more urban, and
complex urban ecosystems grow in size and importance, the spatial sciences will increasingly
take a lead role in their study, seizing the opportunity for improvements through better
understanding of processes and what they offer.

56

REFERENCES
Alberti, Marina, John Marzluff, Eric Sulenberger, Gordon Bradley, Clare Ryan, and Craig
Zumbrunnen. 2003. “Integrating Humans into Ecology: Opportunities and Challenges for
Studying Urban Ecosystems." BioScience 53(12): 1169-1179.
Anon. 1928. “Mexican Immigration.” The Baltimore Sun, March 5.
Bacher, Uwe and Helmut Mayer. 2000. "Automatic Extraction of Trees in Urban Areas from
Aerial Imagery." International Archives of Photogrammetry and Remote
Sensing 33(B3): 51-57.
Blaschke, Thomas. 2010. "Object Based Image Analysis for Remote Sensing." ISPRS Journal of
Photogrammetry and Remote Sensing 65(1): 2-16.
Borjas, George. J. 2007. Mexican Immigration to the United States. Chicago, IL: University of
Chicago Press.
Cai, Zong W., Clifford M. Hurvich, and Chih-Ling Tsai. 1998. "Score tests for heteroscedasticity
in wavelet regression." Biometrika 85(1): 229-233.
Chubey, Michael S., Steven E. Franklin, and Michael A. Wulder. 2006. "Object-based Analysis
of Ikonos-2 Imagery for Extraction of Forest Inventory Parameters." Photogrammetric
Engineering and Remote Sensing 72(4): 383-385.
Dagodag, W. Tim. 1975. "Source Regions and Composition of Illegal Mexican Immigration to
California." International Migration Review 9(4): 499-511.
Dwyer, Mark C., and Robert W. Miller. 1999. "Using GIS to Assess Urban Tree Canopy
Benefits and Surrounding Greenspace Distributions." Journal of Arboriculture
25:102-107.
Encyclopedia Britannica. 2015 “Mesa Central”: Encyclopedia Britannica Online. Accessed
February 10, 2015. http://www.britannica.com/EBchecked/topic/376531/Mesa-Central.
57

Esch, T., M. Thiel, M. Bock, A. Roth, and S. Dech. 2008. "Improvement of Image Segmentation
Accuracy Based on Multiscale Optimization Procedure." Geoscience and Remote Sensing
Letters, IEEE 5(3): 463-467.
Esri. 2014. “Interpreting OLS Results”. ArcGIS Desktop Help 10.1. Accessed August 31, 2015.
http://resources.arcgis.com/en/help/main/10.1/index.html#//005p00000030000000
Evans, Erv. 2014. “Trees of Strength”. North Carolina State University Department of
Horticulture. Accessed April 4, 2014.
http://www.ncsu.edu/project/treesofstrength/benefits.htm.
Flocks, Joan, Francisco Escobedo, Jeff Wade, Sebastian Varela, and Claudia Wald
"Environmental Justice Implications of Urban Tree Cover in Miami-Dade County,
Florida. 2011" Environmental Justice 4(2)
(2011): 125-134.
Fussell, Elizabeth 2004 "Sources of Mexico's Migration Stream: Rural, Urban, and Border
Migrants to the United States." Social Forces 82(3): 937-967.
Gao, Yan and Jean Francois Mas. 2008. "A comparison of the Performance of Pixel Based
and Object Based Classifications over Images with Various Spatial Resolutions." Online
Journal of Earth Sciences 2(1): 27-35.
Given, Susan, Linwood H. Pendleton, and Alexandra B. Boehm. "Regional Public Health Cost
Estimates of Contaminated Coastal Waters: A Case Study of Gastroenteritis at Southern
California Beaches." Environmental Science and Technology 40(16): 4851-4858.
Gronemeyer, Pam. 2013. “Tools and Methods for Effective Rangeland Management: Object
Based Classification.” From The Landscape Tool Box. Accessed May 20, 2014.
http://wiki.landscapetoolbox.org/doku.php/remote_sensing_methods:object-
58

based_classification.
Grove, J. Morgan, Austin R. Troy, Jarlath M. O’Neil-Dunne, William R. Burch, Mary L.
Cadenasso, and Steward Pickett. 2006. "Characterization of Households and its
Implications for the Vegetation of Urban
Ecosystems." Ecosystems 9(4): 578-597.
Hashemi, Sam. 2008. “Automatic Peaks Extraction from Normalized Digital Surface
Model (NDSM). The International Archives of the Photogrammetry, Remote Sensing and
Spatial Information Sciences. 37(B3).
Irani, Frederick M. and Michael F. Galvin. 2007 “Strategic urban forests assessment: Baltimore,
Maryland.” Maryland Department of Natural Resources.
Iverson, Louis R. and Elizabeth A. Cook. 2000. "Urban forest cover of the Chicago region and
its relation to Household Density and Income." Urban Ecosystems 4(2): 105-124.
Jarque, Carlos M. and Anil K. Bera. 1987. "A test for normality of observations and regression
residuals." International Statistical Review/Revue Internationale de Statistique 55(2):
163-172.
Lee, Su J. 2012. "Effects of Building Modifications and Municipal Policies on Urban Green
Cover.” PhD diss., University of Southern California.
Lehrbass, Brad, and Jinfei Wang. 2010. "Techniques for Object-based Classification of Urban
Tree Cover from High-resolution Multispectral Imagery." Canadian Journal of Remote
Sensing 36(2): S287-S297.
Lewis, Reta J. 2013. “Global Connectivity for Urban Leaders.” Speech presented at the German
Marshall Fund (GMF) Headquarters, Washington, DC, May 21. Accessed February 19,
2014. http://www.state.gov/s/srgia/2013/209839.htm.
59

Li, Jing, and David Wong. 2015. "Effects of DEM sources on Hydrologic Applications."
Computers, Environment and Urban Systems 34 (3): 251-261.
Library of Congress. 2015 “Mexican Immigration”
http://www.loc.gov/teachers/classroommaterials/presentationsandactivities/presentations/
immigration/mexican2.html.
Los Angeles Department of Water and Power, City Plants Program 2015. City Plants. Accessed
February 20, 2015. http://www.cityplants.org.
Los Angeles Region Imagery Acquisition Consortium. 2006. “LARIAC Product Guide 2006-
07”. http://egis3.lacounty.gov/dataportal/wp-
content/uploads/2012/05/ProductGuideFinal.pdf
Mitchell, Andy. 2008. ESRI Guide to GIS Analysis Volume 2: Spatial Measurement & Statistics.
Redlands, CA: ESRI Press.
McGee, John A., Susan D. Day, Randolph H. Wynne, and M. Barbara White (2012). “Using
Geospatial Tools to Assess the Urban Tree Canopy: Decision Support for Local
Governments.” Journal of Forestry 110(5): 275-286.
McIntyre, Sue, and Richard Hobbs. 1999. "A Framework for Conceptualizing Human Effects on
Landscapes and its Relevance to Management and Research Models." Conservation
Biology 13(6): 1282-1292.
McPherson, Gregory E., James R. Simpson, Xiao Qingfu and C Wu. 2011. "Million Trees Los
Angeles Canopy Cover and Benefit Assessment." Landscape and
Urban Planning 99: 40-50.
Moskal, L. Monika, and Guang Zeng. 2011. “Retrieving Forest Inventory Variables with
Terrestial Laser Scannng in Urban Heterogeneous Forest” Remote Sensing 4:1-20.
60

Nassauer, Joan I. 1997. Placing Nature: Culture and Landscape Ecology. Washington, D.C.:
Island Press.
Nussbaum, Sven, and Gunter Menz. 2008. “Object-based Image Analysis and Treaty
Verification: New Approaches in Remote Sensing-Applied to Nuclear Facilities in Iran.”
Springer Science & Business Media.
Nowak, David J., Miki Kuroda, and Daniel E. Crane. "Tree Mortality Rates and Tree Population
Projections in Baltimore, Maryland, USA." Urban Forestry & Urban Greening 2(3):139-
147.
O’Brien, Robert M. 2007. "A Caution Regarding Rules of Thumb for Variance Inflation
Factors." Quality & Quantity 41 (5): 673-690.
Shashua-Bar, Limor and Michael E. Hoffman. 2000. "Vegetation as a Climatic Component in the
Design of an Urban Street: An Empirical Model for Predicting the Cooling Effect of
Urban Green Areas with Trees." Energy and Buildings 31(3): 221-235.
Shrestha, Rupesh, and Randoph H Wynne. 2012. “Estimating Biophysical Parameters of
Individual Trees in an Urban Environment Using Small Footprint Discrete-return
Imaging Lidar.” Remote Sensing 4(2):484-508.
Telles, Edward M. and Vilma Ortiz. 2008. Generations of Exclusion: Mexican-Americans,
Assimilation, and Race. New York: Russell Sage Foundation.
Tiede, Dirk, Gregor Hochleitner and Thomas Blaschke. 2005. "A Full GIS-based Workflow for
Tree Identification and Tree Crown Delineation Using Laser Scanning." In ISPRS
Workshop CMRT 5: 29-30.
Troy, Austin, J. Morgan Grove, and Jarlath M. O’Neil-Dunne. 2012. "The Relationship Between
Tree Canopy and Crime Rates Across an Urban–rural Gradient in the Greater Baltimore
61

Region." Landscape and Urban Planning 106(3): 262-270.
Troy, Austin, J. Morgan Grove, Jarlath PM O’Neil-Dunne, Steward Pickett, and Mary L.
Cadenasso. 2007. "Predicting Opportunities for Greening and Patterns of Vegetation on
Private Urban Lands." Environmental Management 40(3): 394-412.
Turner, Monica.G. 2001. Landscape Ecology in Theory and Practice: Pattern and Process. New
York: Springer.
US Census Bureau. 2010. “2010 Census Briefs: The Hispanic Population 2010”
Accessed February 10, 2015. http://www.census.gov/prod/cen2010/briefs/
c2010br-04.pdf.
Valdez, Zulema. 2005. “Segmented Assimilation among Mexicans in the Southwest.” The
Sociological Quarterly 47(3): 397-424.
Vermont Monitoring Cooperative. “Jarlath O'Neil-Dunne - Mapping Vermont From Above:
Turning Data into Information.” YouTube video. 2009.
https://www.youtube.com/watch?v=vQ5ubZ_kpT0
Jeffrey T. Walton, David J. Nowak, and Eric J. Greenfield. 2008. "Assessing Urban Forest
Canopy Cover Using Airborne or Satellite Imagery." Arboriculture and Urban Forestry
34(6) 334-340.
Wu, Jianguo 2008. "Making the Case for Landscape Ecology an Effective Approach to Urban
Sustainability." Landscape Journal 27(1): 41-50.
Wu, Jianguo 2010. "Landscape of Culture and Culture of Landscape: Does Landscape Ecology
Need Culture?" Landscape Ecology 25(8): 1147-1150.
Zhang, Caiyn and Fang Qiu. 2012. "Mapping Individual Tree Species in an Urban Forest Using
Airborne Lidar Data and Hyperspectral Imagery." Photogrammetric Engineering &
62

Remote Sensing 78(10): 1079-1087.
Zhang, Yaogi, Anwar Hussain, Jinyang Deng and Neil Letson. "Public Attitudes toward Urban
Trees and Supporting Urban Tree Programs." Environment and Behavior 39(6): 797-814.

Abstract (if available)

Abstract The relationship between culture and urban forests is explored by analyzing residential urban trees within the privately owned residential lots of City of West Covina residents in Los Angeles County, CA. Because the largest percentage of Hispanic immigrants in Los Angeles have historically come from rural, often agriculturally fertile areas in Mexico, urban forest structure was studied to identify possible differences in the management practices of privately owned residential trees in Hispanic neighborhoods

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Urban green space accessibility and environmental justice: a GIS-based analysis in the city of Phoenix, Arizona

PDF

Geographic object based image analysis for utility scale photovoltaic site suitability studies

PDF

Identification and analysis of future land-use conflict in Mecklenburg County, North Carolina

PDF

A comparison of pixel-based, object-based, deep learning, and data fusion image classification approaches in delineating urban tree canopy in City Terrace

PDF

Spatial narrative of the invasive lionfish in the Western Atlantic and Caribbean oceans: a GIS story map

PDF

Classifying the alpines: developing a methodology to track environmental changes in the alpines utilizing remote sensing and ecozone vegetation patterns

PDF

The wildland-urban interface in Lassen County, California: a change analysis 2000–2015

PDF

Peoples of Washington historical geographic information system: geocoding culture using archival standards

PDF

Out-of-school suspensions by home neighborhood: a spatial analysis of student suspensions in the San Bernardino City Unified School District

PDF

Precipitation triggered landslide risk assessment and relative risk modeling using cached and real-time data

PDF

Detection and accuracy assessment of mountain pine beetle infestations using Landsat 8 OLI and WorldView02 satellite imagery: Lake Tahoe Basin-Nevada and California

PDF

Water quality in the Los Angeles River: a remote sensing based analysis

PDF

Social media canvassing using Twitter and Web GIS to aid in solving crime

PDF

Light rail expansion in Houston using viable path corridors and least cost path: alternatives for the failed University and Uptown lines

PDF

Risk analysis and assessment of non‐ductile concrete buildings in Los Angeles County using HAZUS‐MH

PDF

Creating a geodatabase and web-GIS map to visualize drone legislation in the state of Maryland

PDF

A contributory Web-based application for documenting historic resources: French colonial era art deco architecture in Hanoi

PDF

Using Landsat and a Bayesian hard classifier to study forest change in the Salmon Creek Watershed area from 1972–2013

PDF

Exploring commercial catch: creating a responsive Florida fisheries web GIS using ASP.NET, the Esri JavaScript API 4.x, and Calcite Maps

PDF

Exploring land use changes in the city of Irvine’s master plan

Asset Metadata

Creator Ulloa, Kathy Maria (author)

Core Title Exploring the relationship between urban forests and ethnicity in West Covina using GIS and object-based image analysis

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/21/2015

Defense Date 08/21/2015

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

Tag GIS,Hispanic,Los Angeles,OAI-PMH Harvest,object-based image analysis,object-oriented image analysis,regression analysis,remote sensing,urban ecology,urban forestry,West Covina

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wilson, John P. (committee chair), Swift, Jennifer (committee member), Vos, Robert O. (committee member)

Creator Email katmariaulloa@outlook.com,kmandrad@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-187217

Unique identifier UC11274584

Identifier etd-UlloaKathy-3943.pdf (filename),usctheses-c40-187217 (legacy record id)

Legacy Identifier etd-UlloaKathy-3943.pdf

Dmrecord 187217

Document Type Thesis

Format application/pdf (imt)

Rights Ulloa, Kathy Maria

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