University of Southern California Dissertations and Theses

A unified geodatabase design for sinkhole inventories in the United States

(USC Thesis Other)

A unified geodatabase design for sinkhole inventories in the United States

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

A Unified Geodatabase Design for Sinkhole Inventories in the United States

by

Ebrahim “Tony” Khan

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)

May 2016

Copyright© 2016 by Ebrahim “Tony” Khan

iii
Table of Contents
List of Figures ................................................................................................................................. v
List of Tables ................................................................................................................................. vi
Acknowledgements ....................................................................................................................... vii
List of Abbreviations ................................................................................................................... viii
Abstract .......................................................................................................................................... ix
Chapter 1 Introduction .................................................................................................................... 1
1.1 What are Sinkholes? ............................................................................................................1
1.2 Sinkholes and Society ..........................................................................................................2
1.3 Sinkholes in the State of Florida ..........................................................................................4
1.4 Research Objectives .............................................................................................................7
1.5 Structure of this document ...................................................................................................8
Chapter 2 Existing Sinkhole Inventories and Related Research ..................................................... 9
2.1 National and International Sinkhole Studies........................................................................9
2.1.1. National Studies .........................................................................................................9
2.1.2. International Studies ................................................................................................12
2.2 SGS Sinkhole Inventories ..................................................................................................12
2.3 FGS and KGS Databases as Models for This Study ..........................................................15
2.4 The Database Design Process ............................................................................................16
2.4.1. Database Design Evaluation ....................................................................................17
2.5 The National Address Database – a Unified Database Design Example...........................19
2.6 Summary ............................................................................................................................20
Chapter 3 A Unified Sinkhole Geodatabase Design ..................................................................... 22
3.1 User Communities .............................................................................................................22
3.1.1. Expert GIS Users: Geologists ..................................................................................22
3.1.2. Intermediate GIS Users: Insurance Fraud Investigators ..........................................23
3.1.3. Novice GIS Users: General Public ..........................................................................23
3.2 Design Principles ...............................................................................................................24
3.2.1. Insight from Sinkhole Studies ..................................................................................24
3.2.2. Key Sinkhole Database Models ...............................................................................24
3.2.3. Distillation of State Sinkhole Database Models ......................................................27
3.2.4. Identification of Feature Layers ...............................................................................29
3.2.5. Why Use an ArcGIS Geodatabase? .........................................................................29
3.3 Unified Geodatabase Design ..............................................................................................30
3.3.1. Database Structure ...................................................................................................30
3.3.2. Attributes..................................................................................................................32

iv
3.3.3. Geodatabase structure ..............................................................................................35
3.4 Geodatabase Design Data Integrity ...................................................................................37
3.4.1. Integration of Both Sinkholes and Sinkhole Areas Features ...................................37
3.4.2. Cross-Referencing VGI Data ...................................................................................38
3.4.3. Primary and Foreign Keys .......................................................................................39
3.5 Summary of Database Design Steps Completed ...............................................................40
Chapter 4 Evaluation of the Unified Geodatabase Design ........................................................... 41
4.1 Study Area for Implementation Tests ................................................................................41
4.2 Construction of the Prototype Unified Geodatabase .........................................................42
4.2.1. Sinkhole Areas Polygon Feature Class ....................................................................43
4.2.2. Sinkhole Point Feature Class ...................................................................................44
4.2.3. VGI Tables ...............................................................................................................46
4.2.4. Primary and Foreign Keys .......................................................................................46
4.3 Testing the Prototype with User Community Sample Queries ..........................................47
4.3.1. Geologist Use Case Queries .....................................................................................48
4.3.2. Insurance Fraud Investigators Use Case Queries .....................................................50
4.3.3. General Public Use Case Queries ............................................................................52
4.4 Additional Evaluation Perspectives ...................................................................................54
4.4.1. Level of Completeness .............................................................................................54
4.4.2. Database Design Quality..........................................................................................55
Chapter 5 Conclusions and Future Work ...................................................................................... 56
5.1 Suggestions to Improve VGI Collection and Implementation ...........................................56
5.1.1. A GeoJSON Interface for VGI Data Collection ......................................................57
5.1.2. OpenStreetMap VGI Data Collection ......................................................................58
5.2 A Web Map Interface .........................................................................................................59
5.3 National Implementation ...................................................................................................59
5.4 Conclusion .........................................................................................................................60
REFERENCES ............................................................................................................................. 61
Appendix A: Attributes Emulated from the FGS Database .......................................................... 64

v
List of Figures
Figure 1 Karst topography .............................................................................................................. 2
Figure 2 Carbonate outcrops globally, excluding Evaporites ........................................................ 2
Figure 3 Existing karst topography in the Contiguous United States ............................................. 4
Figure 4 Aerial view of the Winter Park Sinkhole ......................................................................... 5
Figure 5 Three types of sinkholes, mechanisms and resulting features .......................................... 6
Figure 6 Database quality dimensions .......................................................................................... 19
Figure 7 Entity-Relationship Diagram ......................................................................................... 31
Figure 8 Entity-Attribute-Relationships Diagram ....................................................................... 36
Figure 9 The State of Florida and study area ................................................................................ 42
Figure 10 Sinkhole area locations in study area ........................................................................... 43
Figure 11 Sinkholes in study area ................................................................................................. 45
Figure 12 Python script for calculating key values ....................................................................... 47
Figure 13 A sample GeoJSON interface ....................................................................................... 58

vi
List of Tables
Table 1 SGS sinkhole database sources ........................................................................................ 14
Table 2 Comparison between SGS databases and international studies ....................................... 15
Table 3 Database design steps ...................................................................................................... 17
Table 4 SGS databases contributing to the UG design ................................................................. 26
Table 5 Sources and common attribute themes in SGS inventories ............................................. 27
Table 6 UG attributes emulated from Florida database ................................................................ 28
Table 7 SHA polygon feature class attributes ............................................................................... 32
Table 8 Sinkhole point feature class attributes ............................................................................. 33
Table 9 VGI Source table attributes.............................................................................................. 34
Table 10 VGI Duplicate Data table attributes ............................................................................... 35
Table11 Attribute-domain associations ........................................................................................ 37
Table 12 Database design steps summary .................................................................................... 40
Table 13 Geologist use case sample query 1 ................................................................................ 48
Table 14 Geologist use case sample query 2 ................................................................................ 49
Table 15 Geologist use case sample query 3 ................................................................................ 50
Table 16 Insurance fraud investigator use case sample query 1 ................................................... 51
Table 17: Insurance fraud investigator use case sample query 2 .................................................. 51
Table 18 Insurance fraud investigator use case sample query 3 ................................................... 52
Table 19 General public use case sample query 1 ........................................................................ 53
Table 20 General public use case sample query 2 ........................................................................ 53
Table 21 General public use case sample query 3 ........................................................................ 53

vii
Acknowledgements
I want to thank my thesis advisor, Dr. Kemp, for her guidance, insight, patience and keeping me
on track and Dr. Swift and Dr. Vos for their input on possible thesis directions. I would like to
thank my supervisor, Doug McKay, who was extremely accommodating when I needed to
modify my work schedule throughout the GIST Program at University of Southern California.
Dr. Hastings suggestion on a potential approach for the thesis was invaluable. Finally, I am
grateful for the support my Mom gave me, without whom it would have been impossible to
finish the thesis.

viii
List of Abbreviations
EARD Entity-Attribute-Relationship Diagram
ERD Entity-Relationship Diagram
FGS Florida Geological Survey
KGS Kentucky Geological Survey
GIS Geographic Information System
SGS State Geological Survey
SHA Sinkhole Area
UG Unified Geodatabase
USGS United States Geological Survey
VGI Volunteered Geographic Information

ix
Abstract
Sinkholes are naturally occurring geologic phenomena which form when karst erosion causes the
surface to collapse. Karst formations can be found globally as a result of water eroding soluble
bedrock which creates features such as fissures, caves, and sinkholes. In the United States, every
state except Rhode Island has the presence of karst terrain and, therefore, the potential of
developing sinkholes. Sinkhole formation can negatively impact society, manifesting mostly as
property damage, and in some tragic cases, causing a loss of life. There is a lack of protocols for
tracking and recording sinkhole events data nationally. The sinkhole inventories that are
available do not include all sinkhole activity and are primarily found among different State
Geological Surveys (SGS) databases.
The objective of this thesis was to create a single unified geodatabase (UG) schema based
on existing SGS sinkhole databases. The majority of SGS sinkhole data is in the public domain
and is of an authoritative source while only two states are utilizing Volunteered Geographic
Information (VGI). Two states, in particular, Florida and Kentucky, influenced the geodatabase
design because of their developed structure and relative completeness respectively. The proposed
UG combines authoritative and VGI elements from multiple databases. It is composed of two
feature classes and three tables that are joined by primary and foreign keys. Additional design
elements stem from database design theory and sinkhole research studies. The geodatabase
design was tested by implementing a prototype database for a portion of Florida. The design was
evaluated against the needs of three potential user communities: geologists, insurance fraud
investigators, and the general public. Based on these fundamentals, a single UG template was
created that can be implemented at the SGS level, and lay the foundations for a national
geodatabase in the future.

1

Chapter 1 Introduction
Sinkholes are naturally occurring features in karst landscapes. The unpredictable nature of the
ground collapsing suddenly or subsiding over time can result in property destruction, injuries,
and in some cases fatalities if the collapse occurs where people reside. Since karst is present
throughout the U.S., there is a need for recording sinkhole locations in a comprehensive unified
geodatabase as population growth spreads into susceptible karst landscapes. Database models for
the recording of sinkholes that have been implemented by multiple state geological surveys in
the US all vary significantly. Thus, this thesis presents a unified geodatabase template that can be
used to unify existing sinkhole databases within a single standard.
1.1 What are Sinkholes?
Sinkholes are the most recognizable features of karst topography. Karst topography is the
result of natural geologic processes that cause soluble bedrock such as Carbonates and
Evaporites to dissolve (Fleury, Carson, and Brinkmann 2008). As Figure 1 illustrates, this
geologic solution results in caves, fissures, and tubes underground. Often subsidence occurs on
the surface, and sometimes the surface collapses resulting in sinkholes.
While karst formations are a global phenomenon indicated by the red areas representing
carbonate rock outcrops in Figure 2, the scope of this study is sinkhole activity in the U.S. where
every state has karst terrain except Rhode Island (Tobin and Weary 2004). Although sinkholes
are a type of geologic hazard, there is no regional or national database tracking such events in the
U.S., nor is there a set of standards to record such information. This lack of a central database or
common database structure can result in inconsistent data recording of sinkhole formation within
different states, developing inefficiencies that affect society’s safety and functionality.

2

Figure 1 Karst topography (Modified from University of Texas at Austin 2015)

Figure 2 Carbonate outcrops globally, excluding Evaporites (Tichy 2010)
1.2 Sinkholes and Society
Sinkhole events occur frequently throughout the U.S., but not all are catastrophic in
nature. However, the following cases within the last two years from around the contiguous U.S.
gained national media attention, emphasizing the significant threats for society. On June 5, 2015

3

in Colorado, a police officer was driving on a road when his vehicle was swallowed by a
sinkhole. Fortunately, in this case, no deaths occurred (Whitehead and Shiff 2015). On May 23,
2015, a large sinkhole appeared on a golf course in Missouri after heavy rains (Associated Press
2015). Three significant events from 2013 that occurred in Florida illustrate the true danger of
sinkholes. The first event in February ended with the death of 37-year-old Jeff Bush whose home
partially collapsed into a 20-foot wide sinkhole (Malone 2013). Sadly, his body was never
recovered (NOVA 2015). In August, a second incident involved a three-story building of a resort
close to Walt Disney World becoming engulfed by a 100-foot wide sinkhole. Luckily, there were
no fatalities (Liston 2013). The last event in November 2013 destroyed two homes when a 90-
foot wide and 50-foot deep sinkhole collapsed (Malone 2013).
Such dramatic instances are actually rare among the many sinkhole incidents that impact
people in their daily lives. The Citizens Property Insurance Company in Florida reported that
prior to 2010 less than one percent of the sinkhole claims submitted to Citizens Property were for
‘‘catastrophic’’ ground collapse of the surface below or near the building resulting in significant
damage to the building (Zisman 2013). Other sinkhole collapses do not result in astonishing
circumstances and, therefore, do not garner media attention, but the danger is ever present.
Considering such unsettling sinkhole incidents, it seems unusual that there is no
consolidated database that attempts to track this information at a regional or national level.
Having one database would assure many kinds of end users that they have the most complete
data for meeting their business or personal needs. In the absence of a single database, a large
number of variable state-level collections makes it is very difficult to know the extent of sinkhole
activity and its impacts on society in the U.S. Some questions that could be answered with a
unified geodatabase include: how much property damage is attributed to sinkholes? What is the

4

number of people that have been injured or killed by sinkholes? What are the most active areas
of sinkhole activity? Is there a specific activity that is correlating with the most active sinkhole
areas? The goal of this study is to design a unified sinkhole geodatabase so that such inquires and
countless others can be answered easily and uniformly across the U.S.
1.3 Sinkholes in the State of Florida
Karst terrain can be found throughout the U.S., with the exception of Rhode Island. In 16
states, at least 25 percent of their terrain is composed of karst topography. This national extent is
shown by the orange areas in Figure 3. Florida is one of the most affected sinkhole regions in the
world (Galve et al. 2011) and with approximately 81 percent of the terrain containing karst
topography has the highest proportion in the nation (Tobin and Weary 2004). It is also the third
most populated state (Census Bureau 2014) thus giving it a high probability to have negative
societal consequences from karst terrain, mostly in terms of property damage. As of 2005, it was
estimated that sinkholes cost Floridians between 22 to 65 million U.S. dollars annually (Galve et
al. 2011).

Figure 3 Existing karst topography in the Contiguous United States (Tobin and Weary 2004)

5

A significant Florida sinkhole incident is the Winter Park Sinkhole. In 1981, a massive
sinkhole formed in Winter Park, Florida. It was 320 feet wide and 90 feet deep (Figure 4) and
was estimated to cost up to four million U.S. dollars in damages. The Winter Park Sinkhole
prompted the State to create the Florida Sinkhole Research Institute (FSRI) to gather information
on sinkholes, although the Florida Geological Survey (FGS) had been accumulating data since
1907 (Zisman 2013). The FGS database is still in use today and was the impetus for this study.

Figure 4 Aerial view of the Winter Park Sinkhole (Orlando Sentinel 1981)
While sinkhole data collection began in the early 20
th
century, it was not until 1959 that
Florida’s Legislature introduced guidelines for insurance claims relating to sinkholes which were
arising from a lack of clear descriptions (Zisman 2013). The Florida Sinkhole Statute attempted
to clarify the definition of a sinkhole and the types of damage that can be attributed to them as
illustrated in Figure 5. The Statute has had several revisions spanning into the mid-2000s to
include topics such as sinkhole insurance claim definitions, testing standards for sinkholes and
alternative procedures for resolution of disputed sinkhole claims (Zisman 2013).

6

Figure 5 Three types of sinkholes, mechanisms and resulting features (Zisman 2013)
However, because there are no strict procedures to evaluate properties for sinkhole risk,
claimants can attribute property damage to sinkholes without proper evidence. Even though
damage may have nothing to do with sinkhole activity – but rather poor building construction,
other natural causes or neglect – it is usually cheaper for insurance companies to pay the claim
instead of disputing it in court (Zisman 2013). It is important to address that after the extensive
damage of Hurricane Andrew in 1992, there was a lack of insurance availability. The State of
Florida responded by creating the state-run Citizens Property Insurance Corporation. According
to Zisman (2013), this entity in 2009 released data showing that it paid out $97 million U.S.
dollars in sinkhole losses while collecting only $19.6 million U.S. dollars in sinkhole coverage
premiums.
Zisman (2013) also states that claimants of sinkhole damage may not even use the funds
towards any kind of repair of the associated property, and if they sell the property, the next
owner's safety may be in jeopardy if the current owner does not disclose the sinkhole damage in

7

fear of losing property value. Florida law requires all property owners to purchase sinkhole
insurance, which contributes to the higher cost of living even though there may not be significant
sinkhole activity in a given area (Fleury, Carson, and Brinkmann 2008). Given all of these
situations, it is clear that having access to consistently structured and reliable sinkhole
information would be a great asset in Florida and nationwide.
1.4 Research Objectives
In his study examining the legislation for sinkholes, Zisman (2013) concisely captures the
need for a central database to alleviate insurance costs for the State of Florida. His statement
forms the basis of this thesis’s objectives:
Data on the occurrence and nature of sinkholes in sinkhole-prone areas will be
beneficial in the planning for insurance and development in these areas. Much of
this information currently exists but is not shared or is difficult to obtain…. [I]t is
recommended that the Florida Geologic Survey be tasked with the responsibility
of developing a database base [sic] of sinkhole information [obtained] from
insurance companies and other sources. This information should be made
available to the public…. Such a database would improve knowledge of sinkhole
occurrence and lead to reduced insurance costs through a better understanding of
the occurrence of karst features. (Zisman 2013, 101)
This project builds on these recommendations, realizing that such a model could be
applied to other states with similar sinkhole related challenges, benefitting society at much larger
scale. Present sinkhole inventory methodologies lack common data collection and synthesis
methods which would be necessary to create a regional or national database. A single data source
would eliminate redundant data efforts aimed at reaching similar ends. The goal of this research
is to design a unified geodatabase that can incorporate existing state-level sinkhole inventories
into a single standard structure making access to and use of this information more efficient.
This goal was achieved by locating and evaluating sinkhole inventories developed in
seventeen U.S. states. The needs of three end user groups – geologists, insurance fraud

8

investigators and the public – were integrated into the study. The first two groups would find
such a resource to be useful for their business needs while the public would benefit from a
heightened awareness of their surroundings. Based on existing inventories and the needs of these
user groups, a unified sinkhole geodatabase was designed. This geodatabase design was
implemented and tested in a section of Hillsborough County, Florida, an area with a very high
number of sinkholes. Given the extensive amount of sinkhole-related research previously
undertaken in Florida and the related repercussions society has to contend with there, Florida
provides an excellent case study area in which to test the unified geodatabase design.
1.5 Structure of this document
This document is divided into four additional chapters. Chapter Two describes existing
state-level sinkhole databases and sinkhole related studies to gain the domain knowledge
necessary to inform the design of the unified geodatabase for inventorying sinkholes. In Chapter
Three, potential end user needs are summarized, the structure of the FGS database used to guide
the attribute design of the geodatabase is outlined, and the Kentucky Geological Survey’s
methodology for the authoritative collection of sinkhole data, which was used to populate the
implemented test geodatabase, is explained. Chapter Four assesses the success and limitations of
the geodatabase design by implementing a prototype database and testing it with a series of
sample queries from the three potential user communities. Finally, Chapter Five discusses the
future directions this project could go beyond the original objectives of this study.

9

Chapter 2 Existing Sinkhole Inventories and Related Research
The question this thesis seeks to answer is: What combination of design factors for the unified
geodatabase (UG) could be integrated that would allow existing state-level sinkhole inventories
to be more efficient at the state level? This chapter is divided into three sections. It begins with a
brief overview of research about sinkholes carried out within the U.S. and internationally. That is
followed with an overview of existing State Geological Survey (SGS) sinkhole inventories and
databases, focusing especially on those by the Florida Geological Survey (FGS) and Kentucky
Geological Survey (KGS) which are most relevant to this project because of their well-developed
status. Next an overview of the database design process and an evaluation framework is
provided. Finally, an example of another unified database that is in the process of being
implemented concludes this chapter.
2.1 National and International Sinkhole Studies
This section discusses how national and international research on sinkholes aligns with
this research. The national studies are very similar; therefore, discussing one in detail is
sufficient, allowing focus to shift to the international studies, whose methodologies are more
relevant to the UG design. Sinkhole research has primarily focused on delineating sinkhole
features or creating models to predict future sinkhole activity. There is no research evidence of a
successful long-term development of a database to record sinkhole events in a consistent manner
in these studies, something this thesis attempts to resolve.
2.1.1. National Studies
Sinkhole studies in the U.S. have been conducted in several states including Florida,
Illinois, Kentucky, Minnesota, and Texas. The Texas study verified only 20 sinkholes (Harbert

10

2014) and is therefore excluded from this study due to its small size. Gao et al. (2006) created a
Karst Features Database for Minnesota using data from the state geological survey (SGS) and
topographic maps focused on organizing and making the sinkhole data accessible publicly in
Esri’s ArcView along with Microsoft Access tabular formats. According to the authors, the main
advantage of using ArcView was the ability to display features in a spatial context rather than
only using tables. One important point mentioned was the need for different database
permissions where the database administrator has full access while some individuals can only
edit attributes and the public can only view the data. This tiered access structure is an important
consideration in the design of the UG to ensure data quality by reducing human error.
The other studies were interested in identifying and mapping sinkholes during a limited
research period. This approach was understandable since the Areas of Interest (AOI) were large,
for example, Monroe County in Illinois is almost 400 square miles and had almost 3,000
sinkholes inventoried (Angel et al. 2004). A survey of the State of Kentucky by the Kentucky
Geologic Survey (KGS) from 1999 to 2003 resulted in an inventory of over 100,000 sinkhole
areas, depressions where sinkholes or other karst features could occur (Lee 2005). Fleury,
Carson, and Brinkmann (2008) recorded about 1,500 sinkholes in four counties on the west coast
of Florida.
A large database compiled by an academic researcher in Tennessee reportedly has over
54,000 sinkholes, but only 18,081 of those with a depth of at least three meters or more are
available as a table (Dunigan 2015). This restricts the full evaluation of the dataset because the
remaining sinkholes were provided only as web maps without a download option. From the
subset of data available, the attributes concentrate on the sinkhole size and location –
coordinates, perimeter, area, depth, volume, and elevation – attributes that are similar to those

11

included in the authoritative databases described in the next section. Based on the date of the
table file’s creation, it appears that this data was compiled in 2013 and has not been updated
since that time. Since Dunigan’s work cannot be evaluated fully and what could be found did not
yield any new information, it was decided that further consideration of this database as input to
the design was unnecessary.
The national research studies generally collected the locations of sinkholes using USGS
topographic maps and on-screen digitizing in ArcGIS. Angel et al. (2004) mapped sinkholes in
Monroe County in Southern Illinois, an area that is defined by 42 Public Land Survey System
(PLSS) sections. Using 1:24,000 USGS topographic maps to delineate sinkholes, the goal was to
identify sinkholes with hardcopy map manual counting and compare the accuracy of the
concurrent GIS count. Researchers classified sinkholes into four categories: simple, complex,
compound, and ponded. The latter two were derived in GIS with algorithms. The initial GIS
setup took longer to create but was as reliable as hand counting: 2,823 sinkholes to a hand count
of 2,830. This shows that topographic map interpretation is an effective method to identify
sinkholes, the method which most of the SGS use to create inventories.
The main strength of Angel et al.’s study was the confirmation that using digital
topographic maps with a five-foot contour interval allowed the identification of small sinkholes
that are the most common. The weakness in this study, as with the others, was it did not consider
a long-term database design. If they had done so, the collected data could have been stored and
built upon by others in the future. As the focus of that study and the others, with the exception of
the KGS inventory, was to test methods for recording sinkholes, consideration of the future use
of this data was not recognized as a goal. It seems that an opportunity was missed by not making
the preliminary sinkhole data collected available to build upon by creating a standard database

12

template that others could utilize. While they do not focus on databases structures, these studies
do demonstrate the type of data sources that can be used to populate a UG. Thus incorporating
Esri’s ArcGIS tools and geodatabase into the design of a UG can allow a smooth transition as
considerable sinkhole data already exists in shapefiles and scanned maps can be easily digitized.
2.1.2. International Studies
In contrast, international studies that were conducted in Spain, Belgium, Italy, the Island
of Crete, and the Dead Sea area between Israel and Jordan had smaller AOIs and added a couple
of other methods to identify and generate sinkhole susceptibility maps using Digital Elevation
Models (DEMs) and Ground Penetrating Radar (GPR). Three of these studies, those conducted
in Belgium (Eeckhaut et al. 2007), Spain (Galve et al. 2009) and Malaysia (Al-Kouri et al. 2013),
provide key foundations for the design of the UG. These have been selected because they clearly
state what variables contribute to sinkhole formation and thus ones that should be considered for
inclusion in the UG attribute design. Since some of these variables coincide well with existing
SGS sinkhole inventory attributes which are discussed in the next section, these studies are
further examined there.
2.2 SGS Sinkhole Inventories
This section weighs the SGS sinkhole database advantages and disadvantages and how
they contribute towards the thesis goals. The FGS and KGS databases are discussed in detail as
they form the foundations for the UG template. The research conducted for this thesis was
through online scholarly databases and internet web searches to locate enough examples of
sinkhole databases to form a consensus for a unified template. Therefore, it is not necessarily an
exhaustive list as that was not the objective of this thesis.

13

The 16 U.S. states with 25 percent or more of their terrain in karst topography are
included in this study. Although Colorado only has 14 percent of its area in karst terrain, the state
tracks and offers sinkhole data publicly, and is therefore included here, making a total of 17
states examined for their handling of sinkhole information.
Each of these states was placed into one of four categories based on their method of
distributing their inventories of sinkhole occurrences. First, some SGS such as Alabama, Indiana,
Iowa and Kentucky have publicly available GIS layers. Second, others like Missouri and Ohio
have kept their GIS data internal but published static electronic or web maps. Third, Colorado,
Florida and Pennsylvania distribute a combination of GIS layers and static maps. Lastly,
Georgia, Illinois, New Mexico, Tennessee, Texas, Vermont and Wisconsin, after extensive
searching turned up no SGS sinkhole inventories, do not appear to have any publicly available
sinkhole inventory data.
Table 1 summarizes techniques used by the various SGS to generate databases or maps
along with other relevant information about the states. These databases and studies include
digitized karst features from sources such as United States Geological Survey (USGS)
topographic maps, LiDAR data, field site photographs and Volunteered Geographic Information
(VGI). The number of features is the number of records found in the databases. The percent land
surface as karst is derived from Tobin and Weary’s (2004) research, which is depicted in
Figure 3 in Chapter 1.

14

Table 1 SGS sinkhole database sources.
Category State
% of Land
Surface as
Karst
# of Features
in the
database
Sources Format
GIS
Layers
Alabama 29.4 6,460 USGS Topo Map Point Shapefile
Iowa 47.4 37,548
SSURGO spot data,
LiDAR, Photos
Point and Polygon
Shapefile
Indiana 43.0 1,201 Field Sketch Maps Polygon Shapefile
Kentucky 42.3 101,176 USGS Topo Map Polygon Shapefile
Maps

Florida 81.2
3,580; 43
verified
Mostly VGI
Web map (Points)
and Point Shapefile
Missouri 58.1 15,981 USGS Topo Map State map JPEG
Ohio 31.7 975 LiDAR, Photos County maps PDF
Pennsylvania 27.0 141,274
USGS Topo Map,
Aerial Imagery
Web map (Points)
GIS
Layers
and Maps
Colorado 14.1 1,557
Hardcopy records,
National Agriculture
Imagery Program
Point Shapefile and
PDF Map

South
Carolina
27.3 28
SGS Investigation,
Maps
Point and Polygon
Shapefile,
State map – PDF
No Public
Digital
Data
Georgia 56.4 -
- -
Illinois 26.0 - Hardcopy Maps
Static County Maps,
Polygon Feature
Class
New Mexico 32.0 -
- -
Tennessee 50.8 -
- -
Texas 28.1 -
- -
Vermont 29.3 -
- -
Wisconsin 26.1 - Mostly VGI -

Returning to the discussion in the previous section about attributes included in national
research studies on sinkholes, Table 2 shows attribute categories (“Variable”) listed in some of
the SGS databases that match those in three key national research studies (Eeckhaut et al. 2007,
Galve et al. 2009, and Al-Kouri et al. 2013). It can be seen here that Florida’s database has the
most compatibility with the susceptibility studies variables.

15

Table 2 Comparison between SGS databases and international studies

For the UG to be successful, the existing databases used as design models had to be as
complete as possible in both the number of sinkholes included and the range of attributes used.
Building upon sinkhole databases designed by various SGS should allow the UG to be replicated
within different states easily and create a smoother transition to a national universal geodatabase
in the future.
2.3 FGS and KGS Databases as Models for This Study
For various reasons, the sinkhole databases created by Florida and Kentucky became key
examples for the design and assessment of the UG developed in this project. Florida has the
highest amount of karst topography in the nation at 81% of its terrain covered and Kentucky has
42% (Tobin and Weary 2004).
The FGS data is composed largely of volunteered geographic information (VGI). VGI’s
advantage is access to a large, voluntary labor force capable of collecting geographic data that
would otherwise take official agencies using traditional methods long periods of time to perform
(Goodchild 2007). While this means that a large amount of data can be quickly collected, VGI
can have considerable data quality problems. The major issues with VGI arise from the
collection of data by private citizens with no formal qualifications or rigorous quality assurance
mechanisms in place to ensure result accuracy. The FGS database suffers from these problems
Variable Geological Survey Eeckhaut Galve Al-Kouri
Sinkhole Dimensions Florida, Iowa, Kentucky   
Time Florida   
Human Impact/ Land Use Florida   
Topography Florida   
Hydrology – Water Bodies, Precipitation Florida   
Geology – Soil Type, Bedrock Colorado, Florida   
Hydro Chemistry    

16

but also exists because of the unique advantages of VGI that allow rapid data collection through
many volunteers. It provides an important design key for the UG and a data source for a test
implementation of the UG.
The most significant concern of the FGS’s database is how the VGI data is collected.
There is a two page PDF form used to report a sinkhole event found on the FGS Website
(http://www.dep.state.fl.us/geology/default.htm). As will be seen in the next chapter, the form is
far too complex for a private citizen to fill out completely and accurately without specialized
geological knowledge and access to GIS layers. That is likely why many of the attribute fields in
the database are empty. Also, on many occasions, individuals want to remain anonymous, and so
the contact information attributes are often also blank. Improving their reporting form is beyond
the scope of this study but is discussed in Chapter 5.
The KGS database covers the entire State of Kentucky and has over 100,000 sinkhole
areas recorded making it the most complete sinkhole areas inventory in the U.S. In contrast to the
Florida VGI inventory, the value of this inventory for the purposes of this project is the data
collection method. Here the complete survey of the state was conducted by digitizing as
polygons from USGS topographic maps all depressions indicating areas of potential sinkhole
activity. As discussed in Chapter 4, this methodology provides a means to validate the FGS VGI
data used in the case study.
2.4 The Database Design Process
While the type of database used for the UG is a spatial database, there is a common set of
steps used to create any database. These are described in detail by multiple authors. Microsoft
(2015), Sekstrin (2015) and sources as far back as the University of California, Berkeley (1997)
provide similar useful frameworks. Based on these sources, Table 3 summarizes the ten steps in

17

database design that were applied to the UG design process in this thesis. The completion of each
of these steps is described in the next two chapters, with steps one to nine outlined in Chapter 3
and Chapter 4 dedicated to testing the geodatabase as noted in step ten.
Table 3 Database design steps
Step Description
1 Determine the purpose of your database
2 Find and organize the required data
3 Create a simplified E – R Diagram
4 Divide the data into entities and attributes
5 Decide which entities and attributes you want to store in each layer
6 Specify primary and foreign keys
7 List cardinal relationships between layers
8 Create detailed ER Diagram with keys and cardinal relationships
9 Convert detailed ER Diagram into geodatabase format
10 Test geodatabase and refine design as needed
Source: Adapted from Microsoft (2015) and Sekstrin (2015)
2.4.1. Database Design Evaluation
The methodology for the evaluation of a database design lacks a set of agreed upon
standards (Lukyanenko and Parsons 2012). Hoxmeier (1998) outlined one possible framework
that continues to be widely used (see for example Cherfi, Akoka and Comyn-Wattiau 2011 and
Singh et al. 2011). He suggests there are four dimensions — process, data, model and behavior –
that can be used to assess database quality for a specific problem domain. Each of these
encompasses various aspects of database design (Figure 6):
 The process dimension is where the domain knowledge used to build the database
design occurs both conceptually and physically. This is also the step where database
implementation and performance evaluation occurs, thus ensuring the database is not
being hindered by poor design features and maintenance operations.

18

 The database data quality dimension addresses the data itself that makes up the
database. These are factors ranging from accuracy, source data, accounting for
temporal changes and security safeguards that any standalone data or database must
account for to ensure integrity.
 The database model dimension is concerned with how well users and the database
model interact. Factors such as representation, scope, and consistency help answer
questions such as is it easy to use, understandable, and consistent and how much of
the problem domain does it respond to?
 Finally, the database behavior dimension assesses how effectively the database
responds to the problem domain. The effectiveness of the solution can vary based on
multiple factors such as: the experience of the database designer, scope of the
problem domain and the three previous dimensions of database design quality.

19

Figure 6 Database quality dimensions (Hoxmeier 1998)
As discussed later in Chapter 4, these database design quality evaluation aspects provide
a useful template by which to assess the final design of the UG.
2.5 The National Address Database – a Unified Database Design Example
The UG design undertaken in this research is not unique in its attempt to unify
information compiled by multiple entities into a single database. A current example of an
opportunity for a unified database is being addressed by the U.S. Department of Transportation
(Otto 2015). There is a widely-recognized need to have a single national database with every
address in it. Groups or activities that would benefit include emergency responders, Census
tracking, income tax collection, and natural disaster planning. These are just a few of the
applications that a centralized address database would support (McKinney 2015). Similar to the

20

case of the various SGS sinkhole databases having their own designs and varying levels of data
collection, address data collection relies on local and state entities to gather this information. As
of 2013, only 22 states had some type of address data collection plan implemented and of those
22, only eight had data for the entire state (Otto 2015).
A National Address Database Summit was held at the Maritime Institute in Linthicum,
MD on April 8-9, 2015 where 58 participants attended from all levels of government as well as
private and non-profit organizations. Four main points were agreed upon at the summit, which
included: (1) local authorities would be responsible for data collection; (2) state agencies would
aggregate the local level datasets; (3) Tribal Nations, US Territories and the District of Columbia
will have a role in the NAD; and (4) federal leadership and support is required for a national
implementation of NAD. Although this summit did not provide specific information about
database design, it did recognize the need for standardization of data, taking into account
logistics when so many agencies are involved and deciding on using data collection methods that
would be implementable by agencies with limited resources (Applied Geographics 2015). These
final three objectives were taken into consideration for the UG because they are applicable to the
work performed in this thesis.
2.6 Summary
After reviewing the database design process, international and national sinkhole research
studies, as well as the SGS databases, a strategy for developing the design of the UG emerged.
While the national studies and three SGS examples provided context as to why sinkhole areas
should also be included as a separate feature layer, the international studies argued the need to
include some of the sinkhole development factors as attributes in the UG layers.

21

The scale of the AOIs present in the studies led to evaluation of the geodatabase design
by doing a prototype implementation in only a portion of Hillsborough County, Florida, a county
that had the most reported sinkholes in the state’s database. Testing and refining the UG design
at this scale should reveal problems relevant at the state scale of use that is the initial goal for the
UG. As this thesis focuses on a standard geodatabase design incorporating data from various
sources, it is essential to include attributes common to various database models and consider the
differences that exist to ensure that most people benefit when using the UG. As explained next in
Chapter 3, using the most comprehensive existing SGS databases as templates provided the
foundation on which to proceed through the database design steps.

22

Chapter 3 A Unified Sinkhole Geodatabase Design
The primary objective of this thesis is to design a unified geodatabase template to consistently
manage sinkhole data collections. Adapted from existing SGS sinkhole database models, the
unified template is designed to incorporate the key attributes and functions exhibited by them.
The design also considers the needs of anticipated user communities, specifically geologists,
insurance fraud investigators and the general public who have a range of uses that the UG should
be able to resolve. This chapter is divided into three sections: a description of the key user
groups, an overview and summary of the characteristics of the various SGS database models, and
a description of the steps and decisions leading to the final design of the UG template.
3.1 User Communities
To ensure the design of the UG is useful, three user communities were identified as
potential consumers of the UG. Each user community has a different level of GIS expertise;
geologists are considered to be experts, insurance fraud investigators are intermediate, and the
general public are novices. In this section, an overview of what each user community’s needs are
and how the UG should fulfill them is provided. Later in Chapter 4, possible queries for each
user community are tested as one step in the evaluation of the design.
3.1.1. Expert GIS Users: Geologists
Geologists would use the UG as an additional resource in combination with their own
GIS data for research similar to the sinkhole studies occurring around the world and for hazard
mapping. This group would be doing advanced spatial analysis, aerial interpretation, and
independent sinkhole mapping studies, thereby members of this group would not be limited by
the contents of the UG nor lack of GIS abilities to meet their objectives. Realizing that such

23

advanced users would be one of the communities using the UG, the decision not to include
attributes such as “slope,” “land use,” and “trigger mechanism” is valid because they can easily
locate and add layers with this information to their analysis. Specifically, attributes such as
“trigger mechanism” cannot be easily acquired without expert knowledge and would require a
field check. In addition, if the final database were to be populated from VGI reporting (i.e. non-
authoritative sources), such attributes may be incorrectly reported.
3.1.2. Intermediate GIS Users: Insurance Fraud Investigators
Insurance fraud investigators would find the UG useful for querying specific information
and combining it with some of their own data to meet their business needs. In comparison to the
geologists, insurance fraud investigators would be relying on this geodatabase to complement
their own data and make it more complete. There would be less abstract and analytical use of this
data and more practical applications such as tracking how many properties are damaged by
sinkhole events. From an insurance fraud investigator’s perspective, attributes concerning if a
reported sinkhole is a true sinkhole, whether property damage occurred and if repairs are planned
are among the most pressing information needed for them to perform their business transactions.
The UG should allow them to cross-reference with their own information to reduce fraudulent
and duplicate claims.
3.1.3. Novice GIS Users: General Public
The general public would have the simplest GIS queries and visualization needs for the
UG. Curiosity and safety concerns are the main reasons the public would be viewing this data
and thereby raising awareness of their surroundings. People would be curious to know how many
sinkholes have been reported in the city where they live, which could then lead to awareness of a
safety concern as to where they are. However, their queries and explorations of the data may not

24

have a specific goal like research in a geologist’s case or a part of their occupation as is the case
for insurance fraud investigators. Therefore, the information included in the UG for the public is
for awareness purposes only and does not have a precise use when compared to the other user
communities.
3.2 Design Principles
The UG was designed by integrating three components: potential needs of three user
communities, insight from previous research on sinkhole phenomena, and existing state sinkhole
database models. The potential user communities were discussed in the previous section. This
section summarizes the choices made about the fundamental principles that guided the database
design.
3.2.1. Insight from Sinkhole Studies
Previous sinkhole research studies, such as those discussed in Chapter 2 which explore
risks and causes of sinkholes, suggest that given the multitude of factors that may be
incorporated into such studies, there is no need to include in a sinkhole database a great deal of
additional information that geologists and others might have or use from other sources. Adding
this information into the main sinkhole database tables would greatly and unnecessarily expand
their size. Therefore, layers such as land use, precipitation, slope, karst formations, and many
more are not included in the UG template. Such data is easily found and can be related spatially
as needed.
3.2.2. Key Sinkhole Database Models
From the initial 17 states included in this study, 12 had some type of published sinkhole
data. These 12 karst-rich states with published data were categorized in Table 1 by those who

25

provide: (1) public GIS data about sinkholes, (2) published maps showing sinkholes, either web
based or static, and (3) those using a combination of the first two categories. As Table 1 shows,
while they did have published data, Missouri and Ohio only had maps of the point locations of
sinkholes and besides the spatial display did not provide additional information that would be
useful here to inform UG attribute choices. This reduced the set of states with sinkhole databases
of interest in the design phase to ten.
Table 4 shows the ten databases were organized mostly using ArcGIS shapefiles and, in a
couple of cases, ArcGIS feature classes. This lends support for the use of an ArcGIS geodatabase
as the technology for the database implementation. Most of the GIS data used a point format to
represent a single sinkhole event. In a few instances, in Iowa, Kentucky, Pennsylvania and South
Carolina, polygons were utilized to capture sinkhole areas and not individual sinkholes. As
explained earlier, sinkhole areas are land depressions where sinkholes or other karst features
could occur. After examining these two approaches that the state databases were using for
recording features, it was determined that the UG would need to have both a sinkhole point
feature class and a sinkhole areas polygon feature class, a multi-feature model used in Iowa.

26

Table 4 SGS databases contributing to the UG design. * Indicates no data found.
State Data Format
Spatial
Display
Data Source
Alabama Point Shapefile Point USGS Topo Map
Colorado Point Shapefile and PDF Map Point
Hardcopy Records and USDA
Aerial Imagery
Florida Point Shapefile and Web Map Point
Mostly VGI based on available
data
Illinois* Polygon Feature Class Polygon Metadata - Hardcopy Maps
Indiana Polygon Shapefile Polygon Hardcopy Maps
Iowa Point and Polygon Shapefile
Polygon and
Point
SSURGO spot data, LiDAR, Field
Site Photographs
Kentucky Polygon Shapefile Polygon USGS Topo Map
Pennsylvania
Web Map with shapefile/geodatabase
and raster download options
Polygon and
Point
USGS Topo Map and USDA
Aerial Imagery
South
Carolina
Point and Polygon Shapefile,
Coastal area map - PDF
Polygon and
Point
SGS Field Studies
Wisconsin* Unknown Unknown
Mostly VGI based on online
reporting form

Table 4 also shows that all state geological survey sinkhole databases except Florida and
Wisconsin were compiled authoritatively, relying primarily on expert map interpretation. Florida
and Wisconsin were the only two states that relied on Volunteered Geographic Information as a
primary source of information.
Table 5 summarizes the attribute categories that the ten states with detailed databases had
developed, showing which ones they had in common. It is worth pointing out that the attribute
themes of Florida and Wisconsin are very similar because they both use VGI, while among the
authoritative sources, Alabama and Pennsylvania have several in common with the VGI source
states. To accommodate both VGI and authoritative sources, it was decided that the UG should
emulate the attributes used in Florida and include a combination of attributes from the
authoritative databases as well. Florida’s database was chosen as the foundation of the UG
because existing databases kept VGI and authoritative methodologies separate which would
complement each other if merged. It is important to note that Florida’s set of attributes were

27

chosen over Wisconsin’s since, although their sinkhole online reporting form was located and
potential attributes could be surmised, no actual data could be accessed.
Table 5 Sources and common attribute themes in SGS inventories

Alabama
Colorado
Florida
Illinois
Indiana
Iowa
Kentucky
Pennsylvania
South
Carolina
Wisconsin
Data Source

Topo Map                 
VGI         
Field Site Photograph         
Imagery             
Attributes
Lat/Long coordinates             
Sinkhole/SHA Size              
County Name              
Geologic Comments                
General Comments             
Occurrence Date              
3.2.3. Distillation of State Sinkhole Database Models
The Florida Geological Survey database table served as the primary attribute model for
the unified geodatabase. Modifications made for the UG included removing redundant attribute
fields for location, such as county and zip code which can easily be referenced with other GIS
layers and including features such as a cave, fissure or general subsidence even though they are
not strictly sinkholes. To illustrate how state database models were incorporated into the design,
Table 6 shows the attributes emulated from the Florida database into the UG database. The
complete list of attributes from the Florida database is shown in Appendix A.

28

Table 6 UG attributes emulated from Florida database
Attribute Name from Florida
Database
Florida Description Attribute description in UG
TRUE_SINK Verified sinkhole
Separates sinkholes from
other karst features
DATE_REV Date Revised Record Update
COUNTY County Location reference attribute
RPT_SOURCE Source of report Who reported it
RPT_PHONE Report phone number Contact information
RPT_NAME Report name Contact information
EVENT_DATE Date of reported event Date of event
SINSHAPE Shape Sinkhole Shape
SINLNGTH Length Dimension
SINWIDTH Width Dimension
SINDEPTH Depth Dimension
PROPDAM Property damage Was there damage?
REPAIRED Feature repaired Sinkhole repaired – Y/N
PLANNED Repairs planned Repairs planned – Y/N
COMMENTS Other comments Comments about event
COMMENTS2 Additional comments Comments about event
WITNAM Witness Name Name of person who saw it
WITPHONE Witness Phone Contact witness about event

The 12 SGS with some type of data collections had relied on either an authoritative data
collection method or VGI for compiling sinkhole locations. These distinct approaches were
merged in the unified geodatabase template in order to allow the public to be involved in
reporting potential sinkholes that an SGS would not be able to collect with its own staff in the
same amount of time. This would allow SGS staff to prioritize verifying VGI sinkhole reports in
urban settings where the risk is greater and free them to document sinkholes in more rural and
remote areas where VGI reporting may be less frequent.
In the UG, a separate VGI Source table contains information about the person reporting
the sinkhole. This was put into a different, linked table to separate contact information from the
primary point feature class. This not only allows that information to be kept confidential for
some users, but it also keeps the main sinkhole table more compact when compared with the
bulky FGS database table.

29

3.2.4. Identification of Feature Layers
The UG has two separate layers for the geographic features. Replicating many of the SGS
examples, the most important feature layer in the database is the sinkhole point feature layer.
However, two states using authoritative methods, Kentucky, and Pennsylvania, did not compile
individual sinkhole locations but rather they recorded areas of sinkhole activity. These areas
were derived from USGS topographic maps by identifying the upper-most closed contours
around depressions. These sinkhole areas (SHAs) represent places where sinkholes are or could
be occurring. Therefore, it was decided to include such SHAs as a layer in the unified
geodatabase to show possible sinkhole prone locations in addition to individual sinkholes. As
described later in Chapter 4, this method of identifying SHAs was used as a means of populating
the prototype database.
3.2.5. Why Use an ArcGIS Geodatabase?
An ArcGIS geodatabase was chosen as the technology in which to implement the design.
A file geodatabase was chosen primarily because it can store spatial and tabular information and
for three additional reasons: customization options, file organization, and storage capacity. The
geodatabase offers a range of options to customize attributes and to organize different data types.
The use of domains and subtypes for attributes helps eliminate human error when entering data
and provides a fixed number of responses to choose from, thus encouraging record completion.
Importantly, all data is stored in a single file that is a huge advantage because multiple
data types are all in one location. With standalone shapefiles, which are composed of a collection
of files with the same name and different file extensions, component files can get lost or deleted
by accident as some users may not use ArcCatalog to gather them all in one folder, relying on
Windows Explorer. The projection file, attribute table or index file could be misplaced resulting

30

in data not loading and, depending on the user’s GIS ability, could make this task difficult to
resolve. Finally, a file geodatabase has no data storage limit while a personal geodatabase is
limited to two Gigabytes (GB). This leaves one fewer issue to worry about as a national
geodatabase in the future combined with other layers could easily exceed two GB.
3.3 Unified Geodatabase Design
This section reviews the overall structure of the geodatabase design. It begins with a
summary of the basic database structure, followed by a description of the attributes included in
each database component. It concludes with a description of the full geodatabase with
relationships and domains outlined.
3.3.1. Database Structure
The database is composed of two tables and two feature classes. The database structure is
represented by the basic Entity-Relationship Diagram (ERD) shown in Figure 7. The polygon
feature class is the digitized depressions that show SHAs while the point feature class contains
the sinkholes. The two tables are for any sinkhole events that came from a VGI source. The ERD
also shows the primary and foreign keys and the types of relationships between the different
datasets.

31

Figure 7 Entity-Relationship Diagram of the UG structure
Data to populate the sinkhole point feature class may come from both authoritative
sources and VGI sources. Authoritative data will populate only the point feature class. If the data
comes from a VGI source, the sinkhole point in the feature class will have additional information
related to it in the two VGI tables.
The VGI Duplicate Data table is important because when a sinkhole event occurs, there is
a chance that multiple reports could be describing the same feature, and this data needs to be
validated by the SGS as one record in the point feature class. The multiple representations are
deleted from the point feature class after validation occurs. At the same time, it is important to
save the multiple reports of the same feature to the VGI Duplicate Data table. They can be useful
for end users to validate entries themselves because the SGS may have not yet verified if it is a
true sinkhole or a single event. This type of a situation could be very helpful for insurance fraud
investigators who would want to accurately track sinkhole claims.
The relationships between the feature classes and tables are as follows: a sinkhole area
can have one or more sinkholes intersecting it, a sinkhole may not intersect a sinkhole area at all;
a sinkhole point record can have multiple VGI duplicate data records linked to it, a VGI

32

duplicate record may not have a link to any sinkhole point; a VGI source record can have links to
multiple VGI duplicate data records and finally, a VGI duplicate record must have a link to a
VGI source record. Sinkhole points do not always coincide with sinkhole areas because they are
derived from different data sources and sinkhole areas in most cases represent significantly
larger features than the majority of sinkholes.
3.3.2. Attributes
The following tables (Tables 7 to 10) show the attributes that are included in each
component of the UG along with a description of what they represent. For each of the
components, the primary and foreign keys are combined with the U.S. Census Bureau state and
county FIPS codes. These are unique numbers assigned to every state and county in the U.S.
making them ideal for identifying each record uniquely at the state level and providing a way to
organize records at the national level in the future. The source column indicates which SGS
contributed to the particular attribute design, if the designation is “Multiple” then three or more
SGS databases had the same or very similar attribute in their designs. Finally, if the designation
is “New”, then it was created for this design.
Table 7 SHA polygon feature class attributes
Attribute
Name
Definition Source Type
Character
Length
ID*
Primary Key using State FIPS Code and
number of up to 999,999. To link Point and
Polygon Layers
Iowa Long Integer 8
Area Area of SHA in acres Kentucky Double 4
D_Type
SHA Depression Type
(Dry, Lake, Marsh or Water)
New Text 20
Quad_Name USGS Topographic Map name Alabama Text 20
Year_Pub Year USGS Map was published Alabama Short 4
Year_Rev Year USGS Map was revised Alabama Short 4
Comments Additional comments about the SHA New Text 50

33

Table 8 Sinkhole point feature class attributes
Attribute
Name
Definition Source Type
Character
Length
ID*
Foreign Key using State FIPS Code and
number of up to 999,999. To link Point and
Polygon Layers
Iowa Long Integer 8
GEOID*
Primary Key using State FIPS Code, County
FIPS Code and a number up to 999,999. To
link Point layer to VGI Source Table and
VGI Duplicate Data Table
Iowa Long Integer 11
Feature_Type Type of feature - sinkhole, fissure, cave, etc. Illinois Text 20
Sink_Shp
Sinkhole Shape –circular, elongated,
unknown
Florida Text 20
Sink_Verified
SGS verification of event– Y, N or P
(pending)
Florida Text 1
Source Data Source – VGI, Topo, NAIP
Alabama,
Florida
Text 15
Date Date of occurrence Florida Date n/a
Date_Revised Date record updated Florida Date n/a
Address Address associated with sinkhole Florida Text 50
City City name of event Multiple Text 20
State State in which event occurred Multiple Text 20
Length Feature length (ft) Florida Short Integer 3
Width Feature width (ft) Florida Short Integer 3
Depth Feature Depth (ft) Florida Short Integer 3
Prop_Dmg Did property damage occur – Y,N, Unk Florida Text 3
Repairs
Were features and property repairs planned –
both Y, Both N, Feature Y & Prop N, vice-
versa, Unk
Florida Text 10
Drainage
Any drainage in or around feature – Y, N or
Unk
Illinois Text 3
Comments Additional comments about feature Multiple Text 250

34

Table 9 VGI Source table attributes
Attribute
Name
Definition Source Type
Character
Length
GEOID*
Primary Key using State FIPS Code,
County FIPS Code and a number up to
999,999. To link Point layer to VGI table
Iowa Long Integer 11
XID*
Foreign Key using an “X” then State
FIPS Code, County FIPS Code and a
number up to 999,999. To link VGI
Duplicate Data Table to VGI Source
Table
Iowa Text 12
Contact
Y/N answer if the person(s)/ organization
want their information to be public
knowledge and are open to being
contacted regarding event(s)
Florida Text 1
Name First and Last Florida Text 30
Organization
If individual not available perhaps an
organization is provided
Florida Text 40
Address Mailing address for contact Florida Text 50
Email Email for contact Florida Text 30
Phone Phone number for contact Florida Text 20
Property_Owner Is report source, property owner –Y/N Florida Text 1

35

Table 10 VGI Duplicate Data table attributes
Attribute
Name
Definition Source Type
Character
Length
XID*
Primary Key using an “X” then State FIPS
Code, County FIPS Code and a number up to
999,999. To link VGI Duplicate Data Table to
VGI Source Table
Iowa Text 12
GEOID*
Foreign Key using State FIPS Code, County
FIPS Code and a number up to 999,999. To
link Point layer to VGI Duplicate Data Table
Iowa Long Integer 11
Feature_Type Type of Feature – sinkhole, fissure, cave, etc. Florida Text 20
Sink_Shp Sinkhole Shape –circular, elongated, unknown Florida Text 20
Sink_Verified
SGS verification of event Dropdown menu –
Y, N or P (pending)
Florida Text 1
Date Date of occurrence Florida Date n/a
Date_Revised Date record updated Florida Date n/a
Address Address associated with sinkhole Florida Text 50
City City name of event Multiple Text 20
State State in which event occurred Multiple Text 20
Length Feature length (ft) Florida Short Integer 3
Width Feature width (ft) Florida Short Integer 3
Depth Feature Depth (ft) Florida Short Integer 3
Prop_Dmg Did property damage occur – Y,N, Unk Florida Text 3
Repairs
Were features and property repairs planned –
both Y, Both N, Feature Y & Prop N, vice-
versa, Unk
Florida Text 10
Drainage
Any drainage in or around feature – Y, N or
Unk
Illinois Text 3
Comments Additional comments about feature Multiple Text 250

3.3.3. Geodatabase structure
Having defined the included attributes, Figure 8 shows an expanded Entity-Attribute-
Relationship Diagram (EARD) of the two feature classes and two tables that make up the UG
along with their corresponding attributes and associated domain values. The attributes to which
the domains, shown in the lower portion of Figure 8, are applied are listed in Table 11. Domains
denoted with a “*” or “**” are shared by more than one attribute while all other domains apply
to an attribute of the same name. Finally, all the keys and relationships between the components
are displayed in Figure 8 encompassing the complete structure of the UG.

36

Figure 8 Entity-Attribute-Relationships Diagram of the UG structure

37

Table11 Attribute-domain associations
Domain Name Component Associated Attributes
Y/N VGI Source Contact, Property_Owner
Y/N/Unk. Point and VGI Duplicate Data Drainage, Property_Dmg
All others
Point, Polygon, and VGI
Duplicate Data
Domain and Attribute
have the same name

3.4 Geodatabase Design Data Integrity
There are at least three ways that the design of the unified geodatabase manages to
increase data integrity: (1) the use of the SGS methodologies that generated the sinkhole areas
layer; (2) cross-referencing several VGI records to accurately as possible identify possible
sinkhole feature(s); and (3) having primary and foreign keys between components is an essential
feature to reduce data entry errors and duplicate records.
3.4.1. Integration of Both Sinkholes and Sinkhole Areas Features
The majority of the SGS databases were populated using some type of terrain analysis. In
some cases, they used USGS topographic maps to identify closed contours, LiDAR to generate
DEMs and identify depressions, or other remote sensing techniques such examining NAIP
Imagery to identify individual sinkhole features or SHAs. The use of such techniques is most
helpful in mapping SHAs as these are features with areas in acres that are larger than most
individual sinkholes. In turn, these locations provide an indication of where sinkholes could be
occurring. As demonstrated later in the implementation tests (Chapter 4), of the 211 VGI
sinkholes from the Florida Database located in the portion of Hillsborough County, 195 were
within half a mile of an SHA and 210 were within a mile. In fact, flagging sinkhole features
located in areas lacking SHAs might be used as an indication of errors, or possibly areas that
should be early targets for a field check by SGS staff.

38

3.4.2. Cross-Referencing VGI Data
The separation of data collected as VGI from that collected by authoritative techniques in
the SGS databases sometimes weakened data integrity of existing designs since the components
could not be easily integrated. That is why it was decided to combine both methods of data
collection in the UG. VGI accounts can cover large areas in a fraction of the time that SGS staff
can survey them. VGI also identifies features that the SGS may want to investigate further.
Perhaps VGI’s most essential quality is having multiple witness accounts about the same
feature. Having more than one witness account provides a wealth of intelligence about a single
reported feature from different people that the SGS staff can filter through before finalizing the
record in the sinkhole feature class in the UG.
For example, consider a scenario where three people have reported through a VGI
reporting interface details about the same sinkhole. It is likely that all reported locations will
differ slightly, so their duplication will not be immediately apparent. When an SGS staff member
examines the reports and deduces that they represent the same sinkhole feature, only one spatial
representation will be determined to be the most accurate and kept in the point feature class.
However, just as the three spatial locations can vary, so can the attributes and the spatial point
chosen as the most accurate location wise may not have the most accurate and complete
attributes. In this situation, the attributes from another record are copied to the one with the best
spatial accuracy. Finally, all the records are kept in the VGI Duplicate Data table with a foreign
key relating them to the spatial record in the point feature class.
When appropriate, multiple accounts will become associated with a single sinkhole point
feature while the multiple accounts continue to be preserved in the VGI Duplicate Table,
whether they have or have not been verified by the SGS. This would allow interested parties to

39

pursue their own investigations of these reports, possibly uncovering additional information or
verifying if a feature is a sinkhole or not, and informing SGS staff for their attention.
However, multiple reports of the same feature create a potential problem resulting in
duplicate records. This is addressed by the use primary and foreign keys which provide a way to
identify related records. The design of these keys is discussed in the next section.
3.4.3. Primary and Foreign Keys
Multiple VGI witness accounts can easily create duplicate records and other errors in data
entry if there is no way to identify them separately. That is why the UG uses a simple structure to
organize VGI accounts in separate tables to show how VGI source information was derived for
the sinkhole point feature class. Using a set of primary and foreign keys to identify records
uniquely and allow them to be related between components is an important means to ensure data
quality. This is valuable in the situations when multiple VGI records exist for the same feature or
when the SGS have their own data and choose to combine it with some VGI to update an
existing record.

40

3.5 Summary of Database Design Steps Completed
To summarize the design process accomplished in this project and described in this
chapter, Table 12 matches each of the database design steps outlined in Chapter 2 with the task
completed in this project that best fits the description.
Table 12 Database design steps summary
Step Description Task Completed
Relevant Section
or Table
1
Determine the purpose of
your database
Acquired domain knowledge Section 1.1, 1.2
2
Find and organize the
required data
Investigated SGS databases
Section 2.2, 2.3,
Table 1
3
Create a simplified ER
Diagram
Determined database model Figure 7
4
Divide the data into entities
and attributes
Decided which entities and
attributes are essential to design
Table 5
5
Decide which entities and
attributes you want to store in
each layer
Decided which entities and
attributes are not redundant
Section 3.3.2
6
Specify primary and foreign
keys
Created keys based on U.S.
Census Bureau FIPS codes
Section 3.3.2
7
List cardinal relationships
between layers
Determined how the four layers
relate to each other
Section 3.3.1
8
Create detailed ER Diagram
with keys and cardinal
relationships
Insured consistency of process
steps 4 – 7
Section 3.3, 3.4,
Figure 8
9
Convert detailed ER Diagram
into geodatabase format
Created geodatabase in ArcGIS
10.3.1
Section 4.2
10
Test geodatabase and refine
design as needed
Discussed in next chapter Section 4.3, 4.4

Having described the design of the UG, the next chapter describes how a prototype of the
geodatabase was implemented and populated as a means of evaluating the success of the design.

41

Chapter 4 Evaluation of the Unified Geodatabase Design
In order to evaluate the unified geodatabase design, it was necessary create, populate and test a
prototype implementation. This chapter discusses four related topics: (1) the study area used for
the test implementation of the geodatabase design, (2) the construction and population of the
prototype geodatabase, (3) the results of a sample of queries for each of the user communities
that can be answered using the prototype sinkhole geodatabase, and (4) additional database
design qualities.
4.1 Study Area for Implementation Tests
The study area chosen is a portion of Hillsborough County in Florida. This County had
556 reported sinkholes as of the August 31, 2015, release of the Florida database, the most of any
Florida County. It was, therefore, considered a good testing location for the geodatabase design.
In Figure 9, the map on the left shows the State of Florida and Hillsborough County. The map on
the right shows the study area highlighted by a black rectangle that was defined by four 1: 24,000
scale USGS topographic quad maps: Citrus Park, Gandy Bridge, Sulphur Springs, and Tampa.
The points represent reported sinkholes from the Florida Geological Survey sinkhole database.
There are two main clusters of sinkholes in Hillsborough County: the one that was chosen as the
study area is within the major metropolitan area of Tampa and had 211 sinkholes recorded in the
existing database.

42

Figure 9 The State of Florida and study area
4.2 Construction of the Prototype Unified Geodatabase
An empty prototype geodatabase was created using the schema shown in Figure 8. It
consists of four components: a polygon feature class, a point feature class, and two tables to
house the VGI information. The projected coordinate system was set to the appropriate Florida
State Plane system. The data from the Florida VGI database was then loaded into the point
feature class and tables while the polygon features were created using on-screen digitizing from
raster topographic map images. Finally, primary and foreign keys were created using a python
script. The population of each component with data and the creation of the keys are discussed in
separate subsections below.

43

4.2.1. Sinkhole Areas Polygon Feature Class
Emulating the Kentucky Geological Survey sinkhole area data collection method, first a
polygon feature class was created and then the uppermost closed contour depressions visible in
raster images of the four USGS topographic maps encompassing the study area were captured
using on-screen digitizing. Figure 10 shows the resulting 1,083 sinkhole areas that were digitized
in this study area.

Figure 10 Sinkhole area locations in study area
These areas were categorized into four types of depressions based on what was found on
the topographic maps: dry, lake, marsh, and water. The “D_Type” or “depression type” attribute

44

was created to take account of these different designations. The difference between lake and
water designations is that lakes are named, and water depressions have no names on the
topographic maps used to digitize them. While Kentucky’s sinkhole areas data structure does not
include such a type attribute, this enhancement was inspired by the Indiana Geological Survey.
In the attribute table of their sinkhole layer, Indiana differentiated depressions that had
hydrological features such as sinking streams from sinkhole areas that were dry. Since such
information is available on the topographic maps, it was decided to capture this information
during the on-screen digitizing process. Thus, sinkhole area depression features were digitized
and this information was added to the “D_Type” attribute. In this step, the iterative nature of the
geodatabase design process is illustrated since the process of populating the polygon feature
class informed this aspect of the database design.
4.2.2. Sinkhole Point Feature Class
The spatial locations and the majority of the attribute data for the sinkhole feature class in
the prototype geodatabase were acquired from the downloaded Florida sinkhole shapefile
(http://www.dep.state.fl.us/gis/datadir.htm). An empty point feature class was first created in
ArcCatalog with attributes of similar names but the same data type as the original shapefile
because there were a few cases where the names in the UG point layer were modified to make
them less cryptic. The next step, while still in ArcCatalog, was to load the spatial data from the
original shapefile and match the attributes correctly before transferring. This process was
accomplished by right-clicking on the point feature class and selecting the “load data” option.
This is how the attributes with modified names were matched correctly to one another.
Once all the data from the Florida source were loaded, a final step was to select only the
sinkholes within the study area. First, a polygon boundary was created from the four topo maps

45

that represent the study area. Then, the “Select by Location” function was used to select
sinkholes within the polygon study area boundary. This yielded 211 sinkholes making up the
prototype UG sinkhole point feature class (Figure 11). Any sinkholes located outside of the study
area were deleted.

Figure 11 Sinkholes in study area

46

4.2.3. VGI Tables
The original Florida shapefile contained VGI related attribute fields and sinkhole
characteristic attribute fields in a single table containing 51 attributes. Each record in the Florida
shapefile was considered VGI data unless otherwise noted by the “True_Sink” attribute. If the
value was “Y” or “N” that indicated that the Florida SGS staff had verified this record and
determined if it was a true sinkhole or not, making this authoritative. In the final database, these
records are included as “Sink_Verified” = Y. If the record had a “U” value in the “True_Sink”
attribute designating the feature as unknown or if the field was blank indicating that it has not
been authoritatively verified by the SGS and is of VGI origin, the data is deemed unreliable as it
has not been checked yet (thus, “Sink_Verified” = N). The table was exported from the shapefile
in ArcMap and copied: one copy to be used as the source for the VGI Source table and a second
copy for the VGI Duplicate Data table. The attributes that were not incorporated in the prototype
UG were deleted, including from the point feature class. New ones that were going to be the
primary and foreign keys were added. Finally, both tables were imported into the prototype
geodatabase.
4.2.4. Primary and Foreign Keys
Multiple primary and foreign keys were needed to link the two feature classes and VGI
tables. Before any keys were calculated, the original primary key in the Florida database
“REF_NUM” attribute column was loaded initially into the sinkhole feature class and retained in
the two tables. This ensured all records could be matched before migrating to the new primary
and foreign keys.
The keys were then calculated by using a python script modified from Duggan (2013)
that utilizes the FIPS codes in the calculation as demonstrated in Figure 12 for the SHA polygon

47

feature class in ArcMap. The majority of the script was inserted in the Pre-Logic Script Code
box to calculate each record and for each of the two feature classes and two tables. The final
script code: “autoIncrement()” was inserted in the smaller box below the Pre-Logic Script Code
one. In the case of the SHA polygon feature class, the “ID” attribute was designated as the
primary key and was calculated. The “pStart” value is made up of the State FIPS code for Florida
which is twelve and a six digit number. The “pStart” value is set to begin at 12,000,000 because
it is a combination of the State FIPS code and a six digit number. The “pInterval” is set to 1, to
incrementally increase each record by 1. A similar approach was used to calculate the primary
and foreign key values in the remaining point feature class and two tables individually using this
same script but with different values. The final step was to remove the join and delete the
“REF_NUM” attribute column.

Figure 12 Python script demonstration for calculating key values for SHA feature class
(after Duggan 2013)
4.3 Testing the Prototype with User Community Sample Queries
The usability of the UG design was tested by running several sample queries from the
three user communities previously identified. For each of the user communities, three sample

48

queries were explored. Each set of queries consists of at least one spatial and one attribute based
scenario. Below, results of the queries are reported based on the prototype geodatabase
constructed.
To keep this evaluation of the design uncomplicated, for these queries, it is assumed each
user has downloaded the geodatabase and has access to and is capable of using ArcMap. In the
future, it is anticipated that the UG would be implemented in a web application through which
the database could be queried directly, and the selected data downloaded as shapefiles, tables or
displayed on a web map. This future work is discussed further in Chapter 5.
4.3.1. Geologist Use Case Queries
Tables 13 to 15 provide sample queries that geologists may perform on the UG to extract
data to be used in ArcMap. Additional data needed for processing these queries is assumed to
have been obtained from locally stored collections or public domain data portals.
Table 13 Geologist use case sample query 1
Query
List the sinkholes that are pending confirmation, that are within a quarter
mile of an SHA and that caused property damage.
Objective
Prioritize investigating sinkhole reports that are pending based on other
factors such as if they are within a quarter mile of an SHA and caused
property damage.
Additional
data needed
No additional data needed.
Procedure
In ArcMap,
1. Buffer features in the SHA feature class by ½ mile.
2. Select by Location to select features from the sinkhole feature class that
intersect the SHA buffers.
Result 47 sinkholes selected

49

Table 14 Geologist use case sample query 2
Query
Is there any relationship between the type of soil and aquifers present for
sinkholes of a certain shape and a depth of ten feet or more?
Objective Investigate some of the geologic properties around sinkholes.
Additional
data needed
Soil and Aquifer layers.
Procedure
In ArcMap,
1. Intersect Sinkhole layer with soil and aquifer layers.
2. Select by Attributes to select features from the sinkhole layer that
meet the following SQL statement: “Depth >= 10.”
3. Summarize the number of sinkholes within each type of soil and
aquifer.
Sample
Result
Layer Classification
Soil 1 sinkholes in Limestone
27 sinkholes in Medium Fine Sand and Silt
Aquifers 8 sinkholes in Carbonate Rock
20 sinkholes Other Rocks

It is interesting to see that more sinkholes did not occur in the carbonate rock type aquifer
or limestone type soil. This could indicate VGI reporting errors in the original FGS database that
was imported or errors in the boundaries on the other layers. It could certainly lead the geologist
to further investigations.

50

Table 15 Geologist use case sample query 3
Query Create a risk assessment map of sinkhole occurrence.
Objective
Implement a ranking system in which sinkholes are given a higher value the
closer they are to sinkhole areas. Ranks are defined as:
1. Highest risk, sinkholes within SHAs.
2. Sinkholes more than 0 and less than ¼ mile from SHAs.
3. Sinkholes more than ¼ mile and less than ½ miles from SHAs.
4. Sinkholes more than ½ miles from SHAs.
Additional
data needed
No additional data needed.
Procedure
In ArcMap,
1. Open the “Generate Near Table” tool in ArcToolbox.
2. The ‘Input Feature’ is the ‘Sinkhole’ feature class and the ‘Near
Feature’ is the ‘SHA’ feature class.
3. Open the resulting table and create an attribute column ‘Miles.’
4. Open the Field Calculator for the Miles column and enter the
following: ‘NEAR_DIST/5,280’ to convert feet to miles.
5. Create an attribute column ‘Rank.’
6. While still in the table, click on the ‘select by attributes’ option
7. To calculate Rank, select for each class sequentially and enter the
appropriate value in the Rank column as follows:
Rank 1: ‘Mile = 0’
Rank 2: ‘Mile >0 AND Mile<= .25’
Rank 3: ‘Mile > .25 AND Mile <= .5’
Rank 4: ‘Mile > .5’
Result
16 sinkholes as rank 1 or the highest risk of sinkhole occurrence, 194 as rank
2 (zero-quarter mile from SHA), 1 as rank 3 (quarter-half mile) and 0 as rank
4 (half-one mile) or minimal risk of sinkhole occurrence.
4.3.2. Insurance Fraud Investigators Use Case Queries
Tables 16 to 18 provide sample queries that insurance fraud investigators may generate
on the UG. Again, it is assumed that the geodatabase has been downloaded and accessed through
ArcMap. Any additional data is assumed to have been acquired by the user from their internal
collection or public domain sources.

51

Table 16 Insurance fraud investigators use case sample query 1
Query
Compare the sinkholes in the state database that have not been reported to
have caused property damage with those for which the insurance firm has
paid compensation.
Objective Reduce fraudulent claims.
Additional
data needed
Insurance firm’s data table that lists all sinkholes on which they have paid
claims. Geographic reference is by UG geodatabase “GEOID” attribute.
Procedure
In ArcMap,
1. Import the company table and join to the sinkhole point feature class
by “GEOID” attribute.
2. Select sinkholes for which the company data shows compensation
paid and the state database shows “Property_Dmg = N.”
3. Export the resulting attribute table for further review.
Sample
Result
62 sinkholes had caused property damage according to the state database
while 51 had not. The remaining results are dependent on the insurance
firm’s data to analyze which were paid though are not considered to have
caused damage in the state database.
Table 17: Insurance fraud investigators use case sample query 2
Query List zip code areas that contain ten or more sinkhole occurrences.
Objective Calculate monthly premiums on perceived sinkhole risk.
Additional
data needed
Zip Code layer found on U.S. Census Bureau Data Portal.
Procedure
1. Intersect Zip Code and Sinkhole layers.
2. Use Summarize to list by zip code the number of sinkhole
occurrences.
Result
There were 25 zip codes in the study area, and six had greater than ten
sinkholes: 33618 – 58, 33613 – 28, 33612 – 22, 33624 – 15, 33617 – 13, and
33614 – 11.

52

Table 18 Insurance fraud investigators use case sample query 3
Query
Locate sinkhole VGI records with a phone contact related to the sinkhole
claim.
Objective
An insurance fraud investigator wants to conduct interviews regarding a
sinkhole claim and needs to gather additional details about this particular
feature.
Additional
data needed
Insurance firm’s data table of pending sinkhole claims. Locations are
indicated with lat/long coordinates.
Procedure
In ArcMap,
1. Right-click the “VGI Duplicate Data” table and choose the ‘Relate’
option.
2. Choose the “GEOID” attribute from the first drop down menu and
then select the “sinkhole” feature class to relate in the second drop
down menu and finally select the “GEOID” attribute in the third
dropdown menu.
3. Load insurance firm’s data table and make a point feature class from
X, Y coordinates.
4. Using the “Buffer” tool in ArcToolbox, make a ½ mile buffer for the
sinkhole feature of interest.
5. Intersect the buffer result with the UG sinkhole feature class points.
6. Click the ‘Relate’ function in the point feature class to see which VGI
records in the VGI Duplicate Data table have adequate contact
information.
Sample
Result
2 sinkhole VGI reports are selected, but only one has contact information for
conducting an interview.

4.3.3. General Public Use Case Queries
Tables 19 to 21 provide sample queries that the general public may generate using the
sinkhole geodatabase in ArcMap. This group has the simplest queries and so rather than actually
doing three individual queries, the first query is intended to represent many of the most common
questions that could be asked. This might include questions such as how many sinkholes have
occurred in a county or how many sinkholes are within a certain distance from a location. These
can be spatial or attribute queries. Any additional data is assumed to be acquired and processed
by the user. The third query (Table 20) represents one that might be made by an individual who

53

has some GIS experience and can perform some additional steps beyond those needed for the
first two query examples.
Table 19 General public use case sample query 1
Query List number of sinkholes in a specified area.
Objective Discover how many sinkholes are in the city that an individual resides in.
Additional
data needed
No additional data needed.
Procedure
In ArcMap,
1. Select by Attributes where “City = Tampa.”
Result 110 sinkholes were selected.
Table 20 General public use case sample query 2
Query
Where are any sinkholes that caused property damage, were at least ten feet
deep, and occurred over the last decade?
Objective Curiosity to see how many sinkholes meet this criteria and where they are.
Additional
data needed
No additional data needed.
Procedure
In ArcMap,
1. Select by Attributes where “Property_Dmg = Y.”
2. Select from Current Selection , “Depth >= 10”
3. Select from Current Selection , “Date > date 2005-01-01 00:00:00”
Result Two sinkholes were selected.
Table 21 General public use case sample query 3
Query List number of sinkholes occurring in forest or wetland areas.
Objective What is the magnitude of sinkhole incidence in non-urban areas?
Additional
data needed
National Land Cover Data (NLCD)
Procedure
In ArcMap,
1. Open the “Extract Values to Points” in ArcToolbox.
2. Select the ‘Input Point Features’ as the ‘sinkhole’ feature class and
select the ‘Input Raster’ to NLCD.
3. Open the generated point layer from step 2 and scroll to the
“RASTERVALU” attribute and right-click it to ‘Summarize.’
4. Open the Summary table where the “Count_ RASTERVALU”
indicates the number of sinkholes per value, refer to the NLCD key
that shows what each value represents.
Result
For this query, Forest is represented by values of 41-43 and Wetlands as 90
and 95. 35 sinkholes were selected; 9 have a value of 42 or Evergreen Forest,

54

25 have a value of 90 or Woody Wetlands, and 1 had a value of 95 or
Emergent Herbaceous Wetlands.

After conducting possible test queries for the three user communities, it can be concluded
that the UG is able to deliver on the demands of real world needs. The geodatabase can be used
in standalone queries or in conjunction with additional datasets that are available in the public
domain, such as the USGS and U.S. Census Bureau data portals. A geodatabase created from the
UG template was capable of being used to perform tabular, spatial or a combination of both
kinds of queries to meet an objective for all three user communities.
4.4 Additional Evaluation Perspectives
In addition to the usability evaluation discussed in the previous section, there are other
perspectives by which the UG design can be evaluated. These include assessments of the
database level of completeness and the database design quality.
4.4.1. Level of Completeness
When it comes to any sinkhole inventory database, it will never be complete because
sinkholes continue to occur. It is rather an ongoing exercise to be as complete as possible at any
point in time. The decision to merge data collected by authoritative sources and data from VGI
collection techniques for the UG is one way of making sinkhole inventories more complete.
Authoritative data, for example, may be published on an annual basis when satellite imagery is
used for collection or there may be several years between publications if collection depends on
new topographic maps becoming available. In the time period between data publications, many
sinkholes may go recorded.
In addition, collecting possible sinkhole features from volunteered public reports
enhances the traditional data sources. VGI allows for monthly if not weekly updates to the

55

database. Of course the SGS may not to be able to verify immediately if each VGI report is a true
sinkhole. However, the UG structure allows such data to be included and information about the
VGI sources are also listed allowing users to contact the sources and perform their own
investigations. This dual source approach provides an improvement over the authoritative-only
approach which would have much longer periods between updates.
4.4.2. Database Design Quality
Finally, it is useful revisit the database design quality evaluation aspects outlined in
Figure 6. These provide a template by which to assess the overall UG design. The design process
utilized the four dimensions of database design, focusing especially on the process, data, and
model aspects. From the process quality, gaining the domain knowledge about sinkholes through
existing research and finding databases with the SGS provided a solid platform of organizing the
data. In regards to the data aspect, designing the UG in a way to represent authoritative and VGI
collection methods with attribute accountability was important. Finally, the model aspect of
choosing to house the data in a geodatabase in an easily understandable format as most of the
SGS data are using shapefiles allows a smoother transition of information to the new UG
platform.
Having demonstrated through various perspectives that the UG design is sound and
appropriate, this report now concludes in Chapter 5 with a consideration of the possibilities for
future development and use of the UG including support for a web map viewing and download
platform.

56

Chapter 5 Conclusions and Future Work
By beginning with an effort to understand the existing state of SGS sinkhole databases and then
finding a way to bring VGI and authoritative methods together, this thesis builds the case that the
unified sinkhole geodatabase designed here provides an effective database template for
implementation at the state level. This chapter discusses ways in which the UG may be further
enhanced in the future. Building upon the possibility of implementing the UG at the state level,
three additional directions that could be the next steps in the evolution of the UG are:
(1) improvements in methods for collection of VGI data about sinkholes, (2) development of a
web map interface to display and download the sinkhole inventory data, and (3) implementation
of the UG at the national level.
5.1 Suggestions to Improve VGI Collection and Implementation
Only two SGS are using VGI as part of their data collection methods. Florida has a two-
page PDF form (http://www.dep.state.fl.us/geology/geologictopics/sinkhole.htm) while
Wisconsin has a two-page web based form (http://www.tfaforms.com/209523). A
recommendation to improve VGI data collection in general is to simplify the VGI reporting
form, a revision that would be particularly helpful in the case of Florida’s sinkhole reporting.
Also, both states rely on tabular reporting that could be greatly enhanced by adding functionality
to record spatial features. Rather than collecting the location information and terrain
characteristics by data entry from the keyboard, location details could be easily captured through
a web interface. Two methods for this are proposed below.

57

5.1.1. A GeoJSON Interface for VGI Data Collection
One simple mechanism for collecting the spatial data in conjunction with the simplified
reporting forms discussed above could be through GeoJSON. The SGS website where the
reporting form is could also have a link and quick tutorial of how GeoJSON functions. The
simple GeoJSON interface could allow a contributor to easily digitize features using available
imagery as a reference. Figure 13 shows an example of this interface. The digitized features
could then be exported as a shapefile and sent to the SGS along with the reporting form. This is
one way to increase the accuracy of the sinkhole database should it rely on a VGI collection
methodology and also as a way to have a stream of updates that the SGS could verify more
efficiently. Although this approach may appear to create a bottleneck in updates given the ease
with which they can be submitted, it would ensure more data integrity when compared to a
method where many users can edit the same features simultaneously or sequentially, such as the
OpenStreetMap project.

58

Figure 13 A sample GeoJSON interface (http://geojson.io/#map=2/20.0/0.0)
5.1.2. OpenStreetMap VGI Data Collection
The OpenStreetMap (OSM) project is a successful example of VGI contributors around
the world digitizing and updating features on a web map interactively. The users contributing to
OSM have personal knowledge about locations allowing them to correct features more
effectively than the traditional mapping agency that would not be as aware of localities in other
countries or remote locations.
The states could use the OSM model and have contributors digitize sinkhole features
online which that could then be verified by geological survey staff. Such a system seems like a
more efficient way to handle sinkhole inventories compared to the GeoJSON approach given the
existing OSM infrastructure, but there are some drawbacks. Having all the data available
instantly online before all the features are verified is an important issue. In addition, digital
vandalism where people make deliberately erroneous features may be difficult to detect by

59

contributors or geological staff compromising data integrity an issue recognized by Gao et al.
(2006) who argued that a tiered structure of database access permissions would aid in combatting
such occurrences. The OSM approach would work well if the geological surveys could dedicate
staff to monitor the online dataset otherwise it does not make sinkhole inventories more efficient.
5.2 A Web Map Interface
It has been mentioned that a web map may be the most effective way to display and
possibly distribute the sinkhole data. Florida and Pennsylvania are examples of states that have
sinkhole web maps, and Pennsylvania includes an option allowing user communities to
download the data. Being able to view, query and download the data would make it accessible
for all the different user communities, including those without access to ArcGIS. Using a web
map in conjunction with an updated VGI collection method would allow updates to become
available faster and seamlessly compared to publishing a new dataset every few months.
5.3 National Implementation
As demonstrated by the National Address Database project, when a unified database
implementation is needed, it requires a complex set of tasks to be undertaken involving many
interest groups. In order for the effort to succeed, these groups must be coordinated, agreeing on
the database structure, data collection procedures and clear deliverable milestones. This would
apply to a national sinkhole database because of the multiple SGS involved who will have to
agree upon similar standards for their state databases while simultaneously laying the foundation
for a national database to be implemented in the future. As the geological surveys are the
primary data collectors and compliers for sinkholes, it would make sense to have the
coordination for a national database to be managed by the USGS at the federal level since they

60

are in a position to provide a feedback loop with the states, enabling a common database design
template to be reached.
There are several aspects that would need resolution in a national version. Decisions
would need to be made regarding the inclusion of database fields that are applicable in only some
states. Definitions would need to be debated. For example, the definition of a sinkhole may vary
among geologists. What other karst features should be included or defined differently for
inclusion in the database would need discussion. Does the VGI contact information get omitted
at the national level because it is only relevant at the state level? Clearly, there are many
important choices to make before a template that can be used at the national level is determined.
Once a consensus is reached for the database design, existing collections can be
transferred to the new platform from the state level. It would then be the responsibility of the
multiple SGS to engage local government agencies, private institutions, universities and the
general public to upload sinkhole related information using a common collection method. This
approach is similar to the U.S. Census Bureau’s Community TIGER portal discussed by Otto
(2015) for the National Address Database.
5.4 Conclusion
The UG proposed by the author of this thesis is only one possible design for a standard
template for sinkhole inventory purposes. It was devised in the context of a wide range of
different approaches to a common problem. Researching existing databases yielded design
improvements that are included in the proposed universal sinkhole geodatabase template outlined
here. It is a possible first step to creating a national sinkhole database, but realizing that objective
remains a task for the institutions responsible for managing sinkhole data who will be able to
build upon the framework presented herein.

61

REFERENCES
Angel, Julie C., Daniel O. Nelson, and Samuel V. Panno. 2004. "Comparison of a new GIS-
based technique and a manual method for determining sinkhole density: An example
from Illinois' sinkhole plain." Journal of Cave and Karst Studies 66, no. 1: 9-17.
Accessed: November 14, 2013.
Al-Kouri, Omar, A. Al-Fugara, Samih Al-Rawashdeh, Balqies Sadoun, and Biswajeet Pradhan.
2013. "Geospatial Modeling for Sinkholes Hazard Map Based on GIS & RS Data."
Journal of Geographic Information System 5 (6): 584-592.
http://search.proquest.com/docview/1490579576?accountid=14749.
Applied Geographics. 2015. “National Address Database Summit Report.” Accessed: January, 2
2016. https://www.nsgic.org/public_resources/NAD_Summit_Final_Report_070815.pdf.
Associated Press 2015. “Huge sinkhole damages golf course in Missouri.” ABC13, May 23.
Accessed June 16, 2015. http://abc13.com/news/huge-sinkhole-damages-golf-course-in-
missouri/737917/.
Cherfi, Si-saïd Samira, Jacky Akoka and Isabelle Comyn-Wattiau. “Federating Information
System Quality Frameworks using a Common Ontology.” Paper presented at the
Proceedings of the 16th International Conference on Information Quality, pp. 160-173.
Adelaide, Australia, November 18 - 20, 2011.
http://www.scopus.com/inward/record.url?eid=2-s2.0-
84871555845&partnerID=40&md5=c564fd5479e97b9a40fb02be12286e82.
Chicago Tribune. N.D. "Pictures: Winter Park sinkhole." Orlando Sentinel, 1981. Accessed:
December 6, 2013. http://www.chicagotribune.com/news/nationworld/os-fla360-pictures-
winter-park-sinkhole-20121113,0,5366877.photogallery.
Duggan, Nicholas. 2013. The Spatial Blog. “GIS Tips – ArcGIS, Create Sequential Numbers in
the Attribute table (without using = FID!!).” Aug 15. Accessed November 17, 2015.
https://dragons8mycat.wordpress.com/2013/08/15/gis-tips-arcgis-create-sequential-
numbers-in-the-attribute-table-without-using-fid/comment-page-1/.
Dunigan, Tom. 2013. “Tennessee Sinkholes.” Accessed: June 10, 2015.
http://tnlandforms.us/landforms/sinks.php.
Eeckhaut, M. Van Den, J. Poesen, M. Dusar, V. Martens and Ph. Duchateau. 2007. “Sinkhole
formation above underground limestone quarries: A case study in South Limburg
(Belgium).” Geomorphology 91, no. 1–2:19-37. Accessed: November 13, 2013.
http://dx.doi.org/10.1016/j.geomorph.2007.01.016.
Fleury, E. Spencer, Steve Carson and Robert Brinkmann. 2008."Testing Reporting Bias in the
Florida Sinkhole Database: An Analysis of Sinkhole Occurrences in the Tampa
Metropolitan Statistical Area." Southeastern Geographer 48, no. 1: 38-52. Accessed:
November 15, 2013.

62

Galve, Jorge P., Juan Remondo and Francisco Gutiérrez, 2011. “Improving sinkhole hazard
models incorporating magnitude–frequency relationships and nearest neighbor analysis.”
Geomorphology 134, no.1–2: 157-170. Accessed: November 14,
2013.http://dx.doi.org/10.1016/j.geomorph.2011.05.020.
Galve, Jorge P., F. Gutiérrez, J. Remondo, J. Bonachea, P. Lucha and A. Cendrero. 2009.
“Evaluating and comparing methods of sinkhole susceptibility mapping in the Ebro
Valley evaporite karst (NE Spain).” Geomorphology 111, no. 3–4: 160-172. Accessed:
November 14, 2013. http://dx.doi.org/10.1016/j.geomorph.2009.04.017.
Gao, Yongli, Robert G. Tipping and E. Calvin Alexander Jr. 2006."Applications of GIS and
database technologies to manage a karst feature database." Journal of Cave and Karst
Studies 68, no. 3: 144-152. Accessed: November 24, 2013.
Harbert, Artemis. 2013. “Texas Sinkholes.” Accessed: June 10, 2015.
http://artemisharbert.com/gis-projects/texas-sinkholes/.
Hoxmeier, John A. 1998. “Typology of database quality factors.” Software Quality Journal 7,
no. 3: 179-193. Accessed December 11, 2015.
http://dx.doi.org/10.1023/A:1008923120973.
Liston, Barbra. 2013. “UPDATE 3-Guests saved as Florida resort building falls into sinkhole.”
Reuters, August 12. Accessed: November 13, 2013.
http://www.reuters.com/article/2013/08/12/usa-florida-sinkhole-
idUSL2N0GD0HO20130812.
Lukyanenko, Roman, and Jeffrey Parsons. 2012. "Impact of Conceptual Modeling Approaches
on Information Quality: Theory and Empirical Evidence." 17th International Conference
on Information Quality, p. 5. Accessed December 10, 2015.
http://dx.doi.org/10.13140/2.1.3942.9442.
Malone, Shyann. 2013. WTSP-TV, Tampa-St. Petersburg, Fla. “2 houses likely lost because of
Florida sinkhole.” USA Today, November 14. Accessed: November 14, 2013.
http://www.usatoday.com/story/news/nation/2013/11/14/florida-sinkhole-dunedin-
homes/3524877/.
McKinney, Richard. 2015. “DOT presses forward on National Address Database.” July 17.
Accessed November 27, 2015. https://www.transportation.gov/fastlane/pressing-forward-
on-national-address-database.
Microsoft. 2015. “Database design basics.” Accessed December 10, 2015.
https://support.office.com/en-US/article/Database-design-basics-EB2159CF-1E30-401A-
8084-BD4F9C9CA1F5.
NOVA. 2015. “Sinkholes—Buried Alive.” by Larry Klein. Aired January 28, 2015 on PBS.
Accessed June 16, 2015. http://www.pbs.org/wgbh/nova/earth/sinkholes.html.

63

Otto, Greg. 2015. Fed Scoop. “Coming Soon: An open database of every U.S. address.”
September 28. Accessed November 27, 2015. http://fedscoop.com/national-address-
database.
Sekstrin Limited. 2015. “The Sekstrin Project.” Accessed December 10, 2015.
http://sekstrin.org/theproject.html.
Singh, A. J., Balvir Singh Thakur, Rajesh Chauhan, and Vikas Sharma. 2011. "Comprehensive
Framework for quality Database Applications." Accessed December 10, 2015.
http://ijoes.vidyapublications.com/paper/SI2/52.pdf.
Tichy, Ulrich. 2010. Wikimedia Commons. “Global distribution of major outcrops of carbonate
rocks.” February 6. Accessed June 19, 2015.
https://commons.wikimedia.org/wiki/File:Carbonate-outcrops_world.jpg.
Tobin, Bret D. and David J. Weary. 2004. Digital Engineering Aspects of Karst Map: A GIS
version of Davies, W.E., Simpson, J.H., Ohlmacher, G.C., Kirk, W.S., and Newton, E.G.
1984. "Engineering aspects of karst: U.S. Geological Survey, National Atlas of the
United States of America." Accessed: April 12, 2015. http://pubs.usgs.gov/of/2004/1352/.
United States Census Bureau. 2014. "Florida Passes New York to Become the Nation’s Third
Most Populous State, Census Bureau Reports." U.S. Department of Commerce. Last
Modified: December 23, 2014. Accessed: June 16, 2015.
http://www.census.gov/newsroom/press-releases/2014/cb14-232.html.
University of California, Berkeley. School of Information Management and Systems. 1997.
“Files and Databases.” August 28. Accessed December 10, 2015.
http://courses.ischool.berkeley.edu/i202/f97/Lecture6/sld001.htm.
University of Texas at Austin. N.D. “What is Karst?” Accessed May 4, 2015.
http://www.esi.utexas.edu/outreach/caves/karst.php.
Whitehead, Darren, and Blair Shiff. 2015. “Police SUV swallowed by sinkhole in Colorado.”
USA Today, June 5. Accessed June 16, 2015.
http://www.usatoday.com/story/news/nation/2015/06/05/colorado-police-suv-swallowed-
sinkhole/28531325/.
Zisman, E.D. 2013. “The Florida sinkhole statute: its evolution, impacts and needed
improvements.” Carbonates and Evaporites 28, no. 1-2: 95-102. Accessed: November
14, 2013. http://dx.doi.org/10.1007/s13146-013-0144-7.

64

Appendix A: Attributes Emulated from the FGS Database
Attribute Name Description Emulated Why or Why Not Emulated
REF_NUM FGS Assigned Reference Number *  *Temporary unique ID
ADDED_BY Person who added the record Redundant
TRUE_SINK Verified sinkhole  
Separates sinkholes from other
karst features
DATE_ADD Date Added  Redundant
DATE_REV Date Revised   Record Update
LONG_DD Longitude Degrees Redundant
LAT_DD Latitude Degrees Redundant
COUNTY County  Require a reference attribute
TWNSHP PLSS Township Redundant
RANGE PLSS Range Redundant
SECTION PLSS Section Redundant
QTRSECT1 Quarter section Redundant
QTRSECT2 Quarter section Redundant
RPT_SOURCE Source of report   Who reported it
RPT_PHONE Report phone number   Contact information
RPT_NAME Report name   Contact information
EVENT_DATE Date of reported event   Date of event
OSTREET Owner's address Redundant
OCITY Owner's city Redundant
OZIP Owner's zip code Redundant
EVT_ADDR Event address Redundant
SIZDIM Dimensions Redundant
SINSHAPE Shape   Sinkhole Shape
SINLNGTH Length   Dimension
SINWIDTH Width   Dimension
SINDEPTH Depth   Dimension
SLOPE Slope of Sides Complex for VGI Report Source
WATSIN Water visible Complex for VGI Report Source
WATBLS Water below land surface (FT) Complex for VGI Report Source
LIMVIS Limestone visible Complex for VGI Report Source
CAVVIS Cave visible Complex for VGI Report Source
SUBRATE Subsidence rate Complex for VGI Report Source
TRIGGERS Triggering mechanisms Complex for VGI Report Source
COL_CODE Pre-collapse indicators Complex for VGI Report Source
PROPDAM Property damage   Was there damage?

65

Attribute Name Description Emulated Why or Why Not Emulated
REPAIRED Feature repaired   Sinkhole repaired – Y/N
PLANNED Repairs planned  Repairs planned – Y/N
DRAINSTR Drainage structures present  Complex for VGI Report Source
LUCODE Land use code Complex for VGI Report Source
SOILTYPE Soil type Complex for VGI Report Source
COMMENTS Other comments   Comments about event
COMMENTS2 Additional comments   Comments about event
ACCESS Access to sink Complex for VGI Report Source
WITNAM Witness Name   Name of person who saw it
WITADDRE Witness Address  Redundant
WITCZIP Witness Zip Code  Redundant
WITPHONE Witness Phone   Contact witness about event
EM_Hard_Co Unknown n/a
ACCURACY Unknown n/a

Source: Florida Geological Survey (2015)

Abstract (if available)

Abstract Sinkholes are naturally occurring geologic phenomena which form when karst erosion causes the surface to collapse. Karst formations can be found globally as a result of water eroding soluble bedrock which creates features such as fissures, caves, and sinkholes. In the United States, every state except Rhode Island has the presence of karst terrain and, therefore, the potential of developing sinkholes. Sinkhole formation can negatively impact society, manifesting mostly as property damage, and in some tragic cases, causing a loss of life. There is a lack of protocols for tracking and recording sinkhole events data nationally. The sinkhole inventories that are available do not include all sinkhole activity and are primarily found among different State Geological Surveys (SGS) databases. ❧ The objective of this thesis was to create a single unified geodatabase (UG) schema based on existing SGS sinkhole databases. The majority of SGS sinkhole data is in the public domain and is of an authoritative source while only two states are utilizing Volunteered Geographic Information (VGI). Two states, in particular, Florida and Kentucky, influenced the geodatabase design because of their developed structure and relative completeness respectively. The proposed UG combines authoritative and VGI elements from multiple databases. It is composed of two feature classes and three tables that are joined by primary and foreign keys. Additional design elements stem from database design theory and sinkhole research studies. The geodatabase design was tested by implementing a prototype database for a portion of Florida. The design was evaluated against the needs of three potential user communities: geologists, insurance fraud investigators, and the general public. Based on these fundamentals, a single UG template was created that can be implemented at the SGS level, and lay the foundations for a national geodatabase in the future.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

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

PDF

Web-based relational spatiotemporal geodatabase of glaciers in California

PDF

Building a geodatabase design for American Pika presence and absence data

PDF

Generating trail conditions using user contributed data through a web application

PDF

Finding environmental opportunities for early sea crossings: an agent-based model of Middle to Late Pleistocene Mediterranean coastal migration

PDF

GIS data curation and Web map application for La Brea Tar Pits fossil occurrences in Los Angeles, California

PDF

Developing an archaeological specific geodatabase to chronicle historical perspectives at Bethsaida, Israel

PDF

Cartographic design and interaction: An integrated user-centered agile software development framework for Web GIS applications

PDF

Validation of volunteered geographic information quality components for incidents of law enforcement use of force

PDF

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

PDF

Design and implementation of an enterprise spatial raster management system

PDF

Preparing for earthquakes in Dallas-Fort Worth: applying HAZUS and network analysis to assess shelter accessibility

PDF

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

PDF

Building a spatial database of biochar research and practice with Web-GIS

PDF

A spatial narrative of alternative fueled vehicles in California: a GIS story map

PDF

Alaska Hike Search: designing a mapping application for Alaskan trails and user contributed hazards

PDF

Assessing the reliability of the 1760 British geographical survey of the St. Lawrence River Valley

PDF

A Web GIS application for airport pavement management

PDF

Cal ToxTrack: a full stack Web GIS for mapping pollution in California

Asset Metadata

Creator Khan, Ebrahim (Tony) (author)

Core Title A unified geodatabase design for sinkhole inventories in the United States

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 02/08/2016

Defense Date 01/11/2016

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

Tag database design,Florida,geodatabase,GIS,karst,OAI-PMH Harvest,sinkhole,unified

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Kemp, Karen (committee chair), Swift, Jennifer (committee member), Vos, Robert (committee member)

Creator Email ebrahimk@usc.edu,tonykhanek@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-205890

Unique identifier UC11277339

Identifier etd-KhanEbrahi-4086.pdf (filename),usctheses-c40-205890 (legacy record id)

Legacy Identifier etd-KhanEbrahi-4086.pdf

Dmrecord 205890

Document Type Thesis

Format application/pdf (imt)

Rights Khan, Ebrahim (Tony)

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