analyzing the effect of political and natural...
TRANSCRIPT
ANALYZING THE EFFECT OF POLITICAL AND NATURAL CRISES ON CONTRIBUTION PATTERNS TO CROWD-SOURCED DATA PLATFORMS AND
SOCIAL MEDIA ACTIVITIES
By
AHMED AHMOUDA
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
2018
© 2018 Ahmed Ahmouda
To my parents, my wife, and my children
4
ACKNOWLEDGMENTS
I would like to thank all the people who have helped me in carrying out this work.
I would like to offer my spatial thanks to Dr. Hartwig Hochmair, my advisor, for his
patient guidance, useful critiques, and the continuous support of my Ph.D. study and
research. I could not have imagined a better advisor and mentor for my Ph.D. study.
Besides my advisor, I would like to thank the rest of my thesis committee for their
encouragement and insightful comments.
My grateful thanks are also extended to Adam Benjamin, Levente Juhász, Sreten
Cvetojevic, and Majid Alivand for all the help and support they provided during my
study.
Special thanks should be given to Mohamed Salem, my employer who passed
away last year, for his support and encouragement to complete my study. I will be ever
grateful for his assistance and support.
I would also like to send my thanks to all my family, friends, and workmates in
Libya.
I would like to express my deepest gratitude and love to my wife for her support,
quiet patience, and for making it possible. My deepest love is also extended to my
children, Suhaib, Romaissa, and Shaimaa, for their sweet smiles.
Finally, and most important, I wish to give special thanks to my parents for their
sacrifices, support, and encouragement throughout my study.
5
TABLE OF CONTENTS page
ACKNOWLEDGMENTS .................................................................................................. 4
LIST OF TABLES ............................................................................................................ 7
LIST OF FIGURES .......................................................................................................... 8
LIST OF ABBREVIATIONS ........................................................................................... 10
ABSTRACT ................................................................................................................... 11
CHAPTER
1 INTRODUCTION .................................................................................................... 13
Objectives ............................................................................................................... 16
Dissertation Outline ................................................................................................ 17
2 USING VOLUNTEERED GEOGRAPHIC INFORMATION TO MEASURE NAME CHANGES OF ARTIFICIAL GEOGRAPHICAL FEATURES AS A RESULT OF POLITICAL CHANGES: A LIBYA CASE STUDY ................................................... 18
Background and Literature Review ......................................................................... 22
Methodology ........................................................................................................... 27
Study Sites ....................................................................................................... 27 Data Sources and Data Collection Methods ..................................................... 28 Analysis ............................................................................................................ 33
Results .................................................................................................................... 35 Detection of Name Changes ............................................................................ 36
Local and External Contributors ....................................................................... 39 Reliability of Name Changes ............................................................................ 39
Concluding Discussion............................................................................................ 40
3 ANALYZING THE EFFECT OF EARTHQUAKES ON OPENSTREETMAP CONTRIBUTION PATTERNS AND TWEETING ACTIVITIES ................................ 52
Previous Work ........................................................................................................ 53
Collaborative Mapping and Crisis Events ......................................................... 53 Social Media Activity and Crisis Events ............................................................ 55 Human Dynamics and Crises ........................................................................... 56
Data and Methods .................................................................................................. 57 Study Areas ...................................................................................................... 57
Data Collection and Preparation ....................................................................... 58 Analysis Methods ............................................................................................. 60
OSM ........................................................................................................... 60
6
Twitter ........................................................................................................ 63
Analysis of Contribution and Activity Patterns ......................................................... 65
OSM Data Contributions................................................................................... 65 Nepal earthquake ....................................................................................... 65 Italy earthquake ......................................................................................... 68
Tweet Activity Patterns ..................................................................................... 70 Nepal earthquake ....................................................................................... 70
Italy earthquake ......................................................................................... 71 Human Dynamics ................................................................................................... 72 Concluding Discussion............................................................................................ 74
4 USING TWITTER TO ANALYZE THE EFFECT OF HURRICANES ON HUMAN MOBILITY PATTERNS ........................................................................................... 87
Data and Methods .................................................................................................. 92
Study Areas ...................................................................................................... 92 Hurricane Harvey ....................................................................................... 92 Hurricane Matthew ..................................................................................... 92
Data Collection and Preparation ....................................................................... 93 Data Analysis ................................................................................................... 93
Results .................................................................................................................... 95 User Mobility ..................................................................................................... 95 Hashtags Use Changes.................................................................................... 98
Concluding Discussion............................................................................................ 99
5 CONCLUSIONS ................................................................................................... 104
LIST OF REFERENCES ............................................................................................. 111
BIOGRAPHICAL SKETCH .......................................................................................... 122
7
LIST OF TABLES
Table page 2-1 Number of geographical polygon features in different labeling categories for
Tripoli .................................................................................................................... 42
2-2 Number of geographic polygon features in different labeling categories for Benghazi ............................................................................................................... 42
2-3 Number of streets in different labeling categories in Tripoli ................................... 43
2-4 Number of streets in different labeling categories in Benghazi ............................. 43
3-1 Monthly activities of local and external contributors who also mapped during the Nepal earthquake ............................................................................................ 77
4-1 Mean and median (km) displacements before, during, and after the hurricanes with significance levels for differences between distance medians ..................... 100
4-2 Frequency of Twitter hashtags used before, during, and after the analyzed hurricanes ........................................................................................................... 100
8
LIST OF FIGURES
Figure page 2-1 Editing history in Wikimapia ............................................................................... 44
2-2 Editing history in Google Map Maker ................................................................ 44
2-3 New names of artificial geographic features from VGI and people.................... 45
2-4 Labeling of geographic features in Tripoli ......................................................... 46
2-5 Labeling of geographic features in Benghazi .................................................... 47
2-6 Labeling of streets in Tripoli .............................................................................. 48
2-7 Labeling of streets in Benghazi ......................................................................... 49
2-8 Type of contributors providing old place names ................................................ 50
2-9 Type of contributors providing new place names .............................................. 50
2-10 Number of old and new feature names with no local knowledge reference shared on different numbers of VGI sources ..................................................... 51
3-1 Affected and control areas of the analyzed earthquakes in Nepal/India and Italy ................................................................................................................... 78
3-2 Edit and user counts for OSM ways in the affected Nepal area and number of local and external contributors including HOT members ............................... 79
3-3 Monthly numbers of first OSM changesets submitted by local and external contributors who mapped in the affected Nepal area during the earthquake .... 80
3-4 Most prominent OSM node and way features mapped in the affected Nepal area ................................................................................................................... 80
3-5 Edit and user counts for OSM ways in the affected Italy area and number of local and external contributors including HOT members ................................... 81
3-6 Monthly numbers of first OSM changesets submitted by local and external contributors who mapped in the affected Italy area during the earthquake ....... 82
3-7 Most prominent OSM node and way features mapped in the affected Italy area ................................................................................................................... 82
3-8 Average monthly number of tweets in the affected Nepal area and the New Delhi reference area and daily number of tweets in both areas ........................ 83
9
3-9 Average monthly number of tweets in the affected Italy area and the Sulmona reference area and daily number of tweets in both areas .................. 84
3-10 Distribution of reported OSM sources by local and external contributors in the affected Nepal area ..................................................................................... 85
3-11 Nepal earthquake: OSM contributors and Twitter users by country, overlayed by user flow lines derived from both sources .................................... 86
4-1 Affected areas of Hurricane Harvey in Houston and Hurricane Matthew in Miami-Dade County and North and South Carolina ........................................ 101
4-2 Distribution of trip counts for different distance bands within the Houston affected area before, during, and after Hurricane Harvey ............................... 102
4-3 Power-law and truncated power-law distribution models for displacement data within the Houston affected area before, during, and after Hurricane Harvey ............................................................................................................. 102
4-4 Distribution of trip counts for different distance bands within the Miami-Dade affected area before, during, and after Hurricane Matthew ............................. 103
4-5 Distribution of trip counts for different distance bands within North and South Carolina affected areas before, during, and after Hurricane Matthew ............. 103
10
LIST OF ABBREVIATIONS
API
FEV
GIS
GPS
HOT
JSON
LDA
OECD
Application Programming Interface
Flood Event Viewer
Geographic Information System
Global Positioning System
Humanitarian OpenStreetMap Team
JavaScript Object Notation
Latent Dirichlet Allocation
Organisation for Economic Co-operation and Development
OSM
SQL
USGS
OpenStreetMap
Structured Query Language
United States Geological Survey
VGI Volunteered Geographic Information
11
Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy
ANALYZING THE EFFECT OF POLITICAL AND NATURAL CRISES ON
CONTRIBUTION PATTERNS TO CROWD-SOURCED DATA PLATFORMS AND SOCIAL MEDIA ACTIVITIES
By
Ahmed Ahmouda
December 2018
Chair: Hartwig H. Hochmair Major: Forest Resources and Conservation
Political crises, such as revolutions, and natural crises, such as earthquakes or
floods, lead to an increase in topical information shared in social media and crowd-
sourced data platforms. These types of crises, especially natural crises, also influence
human mobility. Using several case studies of political and natural crises from around
the world this research investigates how crisis events affect the contribution patterns to
crowd-sourced platforms and social media activities, and how they change human
mobility patterns.
The first case study takes place in a part of Libya and, using five Web 2.0
platforms (OpenStreetMap, Wikimapia, Google Map Maker, Panoramio, and Flickr),
analyzes if and how Volunteered Geographic Information (VGI) reflects name changes
of geographical features as a result of the revolution in 2011. Results vary strongly
between VGI data sources.
The second case study analyzes short-term (weeks) and longer-term (half year)
changes in OpenStreetMap (OSM) mapping behavior and tweet activities as a
consequence of earthquakes in Nepal and Central Italy as case studies. Results show a
12
significant increase of OSM mapping and tweeting activities during the events. It was
also found that only a small percentage of OSM external mappers travels to the affected
regions to map, whereas most of them rely on desktop mapping instead. Additionally,
the spatio-temporal information of posted tweets together with keyword filters helps to
identify a subset of users who most likely traveled to the affected regions for support
and rescue operations.
The final part of the dissertation analyzes the effect of two hurricanes in the
United States on local human mobility patterns, more specifically trip distance
(displacement), using Twitter data. The study examines three geographical areas
affected by hurricanes Harvey (2017) and Matthew (2016). The results show that
hurricanes limit the mobility of residents at the local scale, possibly due to winds, floods,
or road closures.
This research shows political and natural crisis events do have an impact on the
activities in crowd-sourced and social media platforms as well as on human mobility
patterns.
13
CHAPTER 1 INTRODUCTION
A crisis is “a specific, unexpected and non-routine event or series of events that
create high levels of uncertainty and threaten or are perceived to threaten an
organization’s high priority goals” (Seeger, Sellnow, & Ulmer, 1998). In times of crises
and disasters, the Internet is an effective mean in helping communities (Lev-On, 2012).
Especially the change in the paradigm of Web use towards data sharing and
crowdsourcing in the Web 2.0, away from one-directed data consumption, provides new
opportunities to get insight into a local community’s perception of crisis events. The
Web 2.0 allows the public to use various channels (blogs, social media) to share
information in the form of photos, videos, comments, or maps on crises and disasters
(Ostermann & Spinsanti, 2011).
The Web 2.0 also facilitates the sharing of crowdsourced spatial data, often
referred to as Volunteered Geographic Information (VGI) (Goodchild, 2007a), which can
help to better understand and manage crises. The free access and the use of
crowdsourced mapping platforms, such as OSM, Wikimapia, and formerly also Google
Map Maker, enable and motivate users, including non-geographers, to participate in
mapping activities by adding spatial data to online maps (Plantin, 2015). As of the end
of 2017, the OSM community comprised more than 4 million registered users, and the
OSM database contained over four billion tracking points and 400 million ways1. The
participation of the user community has had a significant impact on the quality and
timeliness of community-based map products, especially in response to disasters and
1 https://wiki.openstreetmap.org/wiki/Stats
14
crises of different kinds. A prominent example is the worldwide web mapping response
to the 2010 Haiti earthquake on the OSM platform (Zook, Graham, Shelton, & Gorman,
2010). Haiti severely lacked spatial information of its infrastructure, which complicated
the relief efforts during the early days after the earthquake. To mitigate this lack of
information, numerous volunteers and organizations from around the world contributed
to the OSM database by uploading geometries of different artificial features (e.g.,
buildings and streets) which they digitized from satellite images. These voluntary
contributions played an important role in rescue and relief operations on site. Another
example is the mapping involvement of the Humanitarian OSM Team (HOT) in
response to Typhoon Yolanda that hit the Philippines in 2013 (Palen, Soden, Anderson,
& Barrenechea, 2015). Within a month after the event, 3,648,537 node edits were made
and 3,294,981 new nodes were added by 1,574 OSM contributors. This corresponds to
an average of 2,292 edits per mapper. This mapping effort provided geographic data
that helped agencies and local responders in recovery and emergency response to this
incident.
Social media facilitates the sharing of news via online communication networks,
such as Facebook, Twitter, and LinkedIn, which oftentimes comes with spatial
information. The development of modern means of communications through mobile
platforms has reformed the way we communicate with each other (Veil, Buehner, &
Palenchar, 2011). The free and easy use of these media platforms provides more
opportunities and new ways for communication and data sharing in crises. As of 2017
53% of adults across emerging countries use social media sites (Poushter, Bishop, &
Chwe, 2018). Daily social networking usage of worldwide internet users amounted to
15
135 minutes per day in 20172. According to data published by the Pew New Media
Index3, more than 20 million tweets about Hurricane Sandy were posted from October
27 through November 1, 2012. The data revealed that more than 50% of the Twitter
conversations and posts were related to news, information, photos, and videos about
the storm. Besides people supplying information, government officials in New Jersey
and New York cities used Twitter during the storm to provide updates to the citizens
about warnings, evacuation, and water use.
VGI and social media have been widely used to manage relief efforts of natural
disasters and to analyze human movements as a result of natural crises. Whereas the
short-term effects of such crises on the pattern of social media activity and VGI mapping
have already been examined in previous studies (Goolsby, 2010; Zook et al., 2010), the
effects on long-term VGI contribution patterns, social media usage, and human mobility
around affected regions are underexplored and not yet well understood. Also, the types
of edits (e.g. change of a feature name) and the types of new objects (e.g. roads,
buildings, water wells, etc.) provided by the OSM community during their mapping
activities in response to natural disasters have not yet been explored, although such
information could be helpful in identifying damaged local infrastructure in need for repair
and maintenance.
This research attempts to provide a more comprehensive description of how
different types of crisis events (e.g. hurricanes and earthquakes) impact the contribution
patterns to crowd-sourced data platforms and social media activities as well as human
2 https://www.statista.com/statistics/433871/daily-social-media-usage-worldwide/
3 http://www.journalism.org/2012/11/06/hurricane-sandy-and-twitter/
16
mobility patterns. This analysis of crisis effects adds to the body of knowledge on the
evolution of crowd-sourced mapping and social media platforms, and the growth and
retention of their user base. It can help governments to better understand how to use
community-based information during such events. It also provides a good
understanding of human movement patterns, especially with respect to natural
disasters. The analysis clearly demonstrates that such crises attract new users to VGI
mapping platforms and also increases activities on social media platforms. The analysis
also distinguishes between local and external contributors and identifies their
nationalities. It determines from which countries they travel to such crisis events for
mapping and support activities. Hence, this research contributes to a better
understanding of global travel patterns in times of natural disasters.
Objectives
This dissertation uses recent crisis events in different parts of the world as case
studies: the Libyan revolution (2011), an earthquake in Nepal (2015), another
earthquake in Italy (2016), Hurricane Matthew (2016), and Hurricane Harvey (2017).
The dissertation aims to expand our knowledge on the change in VGI contribution
patterns, social media usage, and human movement patterns as a consequence of
crises by addressing important questions, such as: Do crises change the patterns of
contributions to VGI and social media platforms? If so, when does the change take
place (e.g. before, during, or after the crisis)? How does the change vary between
different VGI data sources? Do crises influence human mobility patterns in response to
such events? Are these changes stronger for some sub-groups of users than for
others? The main goal of the dissertation is accomplished by pursuing a set of
objectives that are summarized as follows:
17
1. Measure and compare name changes of artificial geographic features as a result of political changes among different VGI data sources
2. Discover the change in contribution patterns of VGI mapping and social media before, during, and after earthquakes on short-term and longer-term periods using OSM and Twitter, as well as travel patterns to these events from different countries
3. Discover the effect of hurricanes on human mobility patterns at the local level using tweets
Dissertation Outline
Chapter 2 and 3 are self-contained journal articles. Chapter 2 was published in
GeoJournal (Ahmouda & Hochmair, 2017). Chapter 3 was published in Geospatial
Information Science (Ahmouda, Hochmair, & Cvetojevic, 2018). The first objective of the
dissertation was accomplished by evaluating five different VGI platforms
(OpenStreetMap, Wikimapia, Google Map Maker, Panoramio, and Flickr) using the case
of the Libyan revolution in chapter 2. The second objective of the dissertation was
accomplished by analyzing the effects of two natural disasters (Nepal and Italy
earthquakes) on the OSM and Twitter platforms in chapter 3. The third objective of the
dissertation was accomplished, as described in Chapter 4, by analyzing the effects of
two major hurricanes (Harvey 2017 and Matthew 2016) within three affected areas in
the United States (Houston, TX; Miami-Dade County, FL; and North and South
Carolina) using Twitter data. Finally, a discussion of this dissertation and plans for future
work are provided in Chapter 5.
18
CHAPTER 2 USING VOLUNTEERED GEOGRAPHIC INFORMATION TO MEASURE NAME CHANGES OF ARTIFICIAL GEOGRAPHICAL FEATURES AS A RESULT OF
POLITICAL CHANGES: A LIBYA CASE STUDY
Libya, a country in North Africa, suffered 42 years of a dictatorial regime which
seized power in a coup in 1969. In order to perpetuate its takeover, its ideals, and the
names of its leaders, this regime changed many names of artificial geographical
features, such as roads, schools, and plazas. These new names were used to represent
the ideas of this regime and to blur the identity of the former regime in the country.
Therefore, most of these new names were associated with the names of the coup, its
date, its leaders, and its ideals. The regime fell in 2011 during the Libyan revolution,
which officially lasted from February 15 to October 23, 2011 and was part of the Arab
Spring revolutionary wave. During and after the revolution, yet another change of
names of artificial geographical features occurred in Libya, now reflecting names
associated with the 2011 revolution, and undoing some name changes enacted in the
previous regime. This new phase of name changes was also noticed by the geospatial
Web community which in its own way disseminated the word about the revolution. It did
so by adding updated feature names in a variety of Web-based geo-portals that allow
the sharing of crowd-sourced geospatial information. This study analyzes, using the
Libyan revolution as a showcase, the usability of crowd-sourced information for the
identification of name changes of artificial geographical features as a consequence of
changes in a political system. The case study is based on two districts in Libya (Tripoli
and Benghazi). It compares name changes between five crowd-sourced datasets and
uses alternative data sources, such as local knowledge of residents in the analyzed
areas, as a reference dataset for comparison when available.
19
Place names are a prominent research topic among geographers. Some of the
many important characteristics about place names are that they reflect the identity of
people, embody ideologies, and provide a common means to toponymic
commemoration by creating a “geography of memory” (Alderman, 1996, p. 51). The
latter means that the history of the past can be transferred to the present through
commemorative place naming, such as commemorative street names that define the
historical memory of a nation (Rose-Redwood, 2008). In cultural geography, place
names have played a significant role in establishing new identities of nations. New
identities could be achieved by renaming geographical places to deface the ideas of the
former regimes (Alderman & Inwood, 2013). Place names are claimed to be essential
for a better understanding of the political landscape (Zelinsky, 1983) since place naming
is considered an essential part of the political process (Cohen & Kliot, 1992). Starting
with the French Revolution, renaming streets commonly occurred after political changes
when a new group comes to power (Azaryahu, 1997). This renaming is an act of
diminishing the impact of the previous regime and establishing the authority of the new
ruling group. Studies that analyze name changes of places due to political changes rely
on a variety of sources to do so. Examples of sources include: newspaper articles,
maps and reports generated by naming commissions for analyzing street name
changes in East Berlin in the context of German reunification (Azaryahu, 1997);
Websites of dedicated geographic name councils for analyzing name changes of
municipalities in South Africa in the context of apartheid (Guyot & Seethal, 2007); or city
government (Bucharest Primărie) documents for analyzing changes of street names in
Bucharest in the context of post-socialism (Light, 2004). In geographic regions (e.g.
20
countries, cities) where these types of official data sources (newspaper articles,
governmental databases or maps) about feature names and/or their changes are not
available crowd-sourced spatial information, collected and shared on Web 2.0
platforms, could provide a viable alternative source of information for the detection of
name changes. Analyzing to which extent such crowd-sourced geodata reflect names
changes of artificial features is, therefore, the overarching goal of this study.
Names of artificial geographic features, including parks, schools, and streets, are
typically assigned by governments at various different hierarchical administrative levels
(Alderman & Inwood, 2013). However, in cases of weak public governance (e.g. due to
limited resources), changed place names, may not (yet) have been officially declared
but be formed by the local community within a toponymic process. The focus of this
study is to analyze the change of official feature names (i.e. relating to the Gaddafi
regime) to community driven names, through the Libyan revolution in 2011. Examples
for non-governmental, community-based place names, can be found on several
Facebook websites that were posted by local teachers in the study area. These
websites contain directories of local schools with their names before and after the
revolution, whereas official governmental resources for new school names do not exist.
These Facebook websites allow parents to identify new names for schools of their
children. The schools listed in these directories were also used during local election
periods as election centers. These facts mean that those new school names, though not
contained in governmental gazetteers or maps, were generally accepted as valid by the
community. These community-based names are somewhat related to vernacular place
names, some of which are also community driven (e.g. vernacular regions in cities).
21
However, vernacular place names typically refer to larger areas than individual artificial
features. They are vague descriptions of locations for which no official boundary exists
due to their fuzzy object boundaries, or for which their commonly perceived boundaries
differ from official ones (Jones, Purves, Clough, & Joho, 2008). Names of vernacular
places are not the subject of this presented study.
This study revolves around crowd-sourced spatial information as a potential data
source to document changes in place names where governmental reports, historic
maps, or other official documents are not available. The introduction of the Web 2.0 and
the integration of GPS in various mobile devices, such as smartphones and cameras,
facilitated the emergence of crowd-sourcing which generates spatial information through
the Web community. The Web 2.0 significantly changed Web user behavior, from
primarily one-directional data consumption to a bi-directional collaboration in which
users are able to interact with and provide information to central sites. Crowd-sourced
spatial information is oftentimes referred to as Volunteered Geographic Information,
short VGI (Goodchild, 2007a) or neogeography (Turner, 2006). Its technologies are
used in emergency response, spatial decision making, participatory planning, and
citizen science (Haklay, 2013). VGI comes in many forms, ranging from informal,
emotional, and unstructured annotations of places, like in Wikimapia, to geometrically
accurate representations of physical features, like in OpenStreetMap (OSM) (Bittner,
2016).
(Glasze & Perkins, 2015) state that crowd-sourcing offers a radical alternative to
conventional ways of map making, challenging the hegemony of official and commercial
cartographies. They point out that crowd-sourcing offers a forum for different voices,
22
elaborating on the fact that maps, and geoinformation in general, can never be neutral
or objective, but instead are always embedded in specific social contexts of production
and use. (Fraser Taylor & Caquard, 2006) suggest that cartography’s notion of a ‘map
user’ no longer implies a discrete singular consumer of cartographic communication.
Instead, online applications, developed by the open-source community, enable a user to
integrate individual data to create maps in collaboration with others. The so-called
democratization of GIS is facilitated by tools and applications that incorporate increased
levels of interactivity and data manipulation made available to the Web community
(Couclelis, 2003), and by integrating local geographic knowledge through the
involvement of new communities (Dunn, 2007). A review of archived messages on the
Humanitarian OpenStreetMap Team (HOT), Crisis Mappers, and Crisis Commons
listserves shows that individuals and grassroots groups regularly discuss how to
represent and share their knowledge and how to integrate new technologies in disaster
management (Burns, 2014). (Elwood, 2008, 2010) points out that citizens and
grassroots groups begin to generate spatial data that is popular among government
officials, altering the roles of petitioner and provider. (Johnson & Sieber, 2013) find that
governments have long been providing online services to citizens, and that nowadays
governments (primarily Western-style) foster citizen engagement through Web 2.0. This
approach facilitates transparency and effectiveness of government services (Brewer,
Neubauer, & Geiselhart, 2006).
Background and Literature Review
Place names have become a subject of interest for numerous studies in
Geography due to their important role in building the identity of nations and because
they are frequently rewritten in the event of political and ideological changes (Alderman,
23
2016). These studies (Azaryahu, 1997; Azaryahu & Golan, 2001; Azaryahu & Kook,
2002; Cohen & Kliot, 1992) broadly illustrate how place names, especially
commemorative street names, are being manipulated by the ruling elite in countries like
Russia, Germany, Israel, Romania, and former Yugoslavia to reform the national and
the historical identity of the nation. Scholars have also explored the role of place names
for ethnicities finding and expressing their identities. As a case study for the southern
United States, (Alderman, 1996; Alderman & Inwood, 2013) elaborate on how the
renaming of streets in honor of Martin Luther King, Jr created a new geography of
memory, commemorating the historical experiences, struggles, and achievements of
African Americans in a region where its landscape was controlled by white people.
Renaming of streets may indicate as a positive symbol of southern black communities
given that name changes occur more often when black people control the government
or mobilize to petition their local governments. A book by (Monmonier, 2006) discusses,
among others, the politics of map names in conflict zones like Cyprus and Israel, and
international disputes about name features of Antarctica, the ocean floor, and the moon.
It specifically elaborates on two types of applied toponyms, i.e., names of geographical
features and names of settled places, omitting street names and names of constructed
features. The notion that maps can never be neutral or objective but are embedded in a
social context (Glasze & Perkins, 2015) becomes evident in VGI platforms, where
volunteers provide their local knowledge by contributing through the GeoWeb
(Stephens, 2013). One prominent example of this phenomenon is an OSM “tagging
war” with disagreement over the use of Greek or Turkish names in the Turkish-
controlled area of Cyprus (Perkins, 2014). Another example is continued disagreement
24
over name forms of contested territories such as Jerusalem or in Kosovo. (Mooney &
Corcoran, 2012a, 2012b) analyzed in detail the phenomenon of “tag flip-flopping”. They
found that especially the values of the “name” and “highway” attributes, which are
attached to mapped features in OSM, such as roads, are subject to frequent change. As
for the name attribute, the authors suspect that objects with two name value
assignments could be a result of place name spelling errors, incorrect naming, or
splitting of a way or a polygon feature into two or more new objects, leading to objects
with different names. The OSM highway attribute more closely specifies the type of
mapped road, street or path, such as primary or residential. Flip-flopping of this attribute
can be attributed to uncertainty among contributors regarding the designation for a
given highway object. This is demonstrated for a road feature whose highway value has
been flipped 88 times between “Trunk” and “Construction” by different users (Mooney &
Corcoran, 2012b).
VGI can facilitate the collection of spatial information for community mapping
programs and provides a promising way to include formerly excluded people in geodata
creation (Tulloch, 2008). However, VGI reflects community-based biases for different
VGI platforms, such as OSM and Wikimapia (Bittner, 2016). Analysis of such biases
provides a more refined view of VGI usage than the often stated digital divide
(Goodchild, 2007a; Heipke, 2010), which suggests different levels of access to
information and technology among population groups with varying levels of
socioeconomic and demographic composition (Chow, 2012). (Tulloch, 2008) describes
a case study where the New Jersey State Department of Environmental Protection
collected and mapped data on vernal pools across the state based on the participation
25
of volunteers. Similarly, a U.S. Geological Survey (USGS) project encouraged
volunteers to contribute data to The National Map using online editing tools (Poore,
Wolf, Korris, Walter, & Matthews, 2012). Freely available governmental data has been
commonly imported into various VGI data collections, such as OSM, providing a
comprehensive database for the community to add on their own collected data (Zielstra,
Hochmair, & Neis, 2013). (Coleman, Georgiadou, & Labonte, 2009) identify different
contexts in which individuals voluntarily contribute spatial information in support of a
given purpose. These contexts include mapping and navigation, social networks,
civic/governmental, and emergency reporting. The paper proposes a classification of
data contributors into five overlapping categories, ranging between “neophyte”
(someone with no formal background in a subject) and “expert authority” (someone who
possesses an established record of providing high-quality products). The study
consolidates and summarizes also the various motivators behind contributions to free or
open-source software and Wikipedia, which include altruism, professional or personal
interest, intellectual stimulation, social reward, and pride of place. The latter is a driving
factor for individual mappers to update road centerlines in OSM or Google Map Maker
in their hometown. Constructive motivational factors for contributions to VGI can be
grouped into intrinsic (e.g. altruism, fun/recreation, and learning) and extrinsic (e.g.
social rewards, career, and personal reputation) (Neis & Zielstra, 2014). Casual OSM
mappers are primarily motivated by general principles of free availability of mapping
data, i.e. help others by providing free digital maps, whereas serious mappers were
more motivated by learning, gaining local knowledge, and to some extent by career
motivations (Budhathoki & Haythornthwaite, 2013). (Zook et al., 2010) review how the
26
information production in different Web-based services was used during the Haiti relief
effort. The authors point out duplicated efforts between OSM and Google Map Maker for
people who utilize both sources. Since Google retains the intellectual property of all
information created with Map Maker, data is not portable between OSM and Map
Maker, leading to different spatial coverage between both data sources. Not all VGI
contributors may be interested in providing objective or reliable information. This could
be motivated by mischief, a hidden agenda of special interest groups, or malice and/or
criminal intent in hope of personal gain, leading to vandalism. Vandalism is an individual
or group attack on active sites for the purpose of data corruption (Neis, Goetz, & Zipf,
2012). For example, in 2015 Google experienced attacks from a user of the Map Maker
platform who created a large-scale prank on the map. Therefore, Map Maker has since
then been limited to selected countries (Map Maker, 2015) before it is scheduled to be
retired entirely in 2017 (Map Maker, 2016). Several VGI platforms do include automated
vandalism detections tools. Existing methods to counteract data vandalism in OSM,
based on historical user edits, are reviewed in (Neis et al., 2012).
As opposed to VGI mapping platforms, such as OSM, Wikimapia, and Google
Map Maker, photo-sharing platforms focus on point events. These platforms allow a
Web user to upload geotagged images to a server and to share them with the
community. Geotagging is a common method to georeference user-generated content
online and to turn photographs into geographic information (Elwood, Goodchild, & Sui,
2012). Shared photos can be annotated with a variety of metadata, including textual
tags, title, geographic position, and capture time. (Ames & Naaman, 2007) developed a
taxonomy of motivations to annotate Flickr photos. The first dimension, “sociality,”
27
relates to whether the tag’s intended usage is by the individual photographer him or
herself, or by others. The second dimension, “function,” refers to a tag’s intended uses,
which can be either to facilitate later organization and photo retrieval or to communicate
some additional context to photo viewers. (Hollenstein & Purves, 2010) found that 70%
of all analyzed georeferenced Flickr images include specific place name tags, where
place names at the granularity of city names were by far the most common ones.
Comparing geotagged position and image content a study by (Zielstra & Hochmair,
2013) revealed that Panoramio photos have a better positional accuracy than Flickr
images and that positional accuracy varies by world region for both data sources.
Regarding the localness of user generated content, (Hecht & Gergle, 2010) found that
53 percent of Flickr users contribute, on average, content that is 100 km or less from
their specified home location, whereas the corresponding number drops to 23% for the
English Wikipedia. (Neis, Zielstra, & Zipf, 2013) showed for OSM that for some world
regions a large percentage of contributions (e.g. Istanbul with slightly over 50%) comes
from members whose home region is separated by more than 1000 km from the
analyzed area. (Zielstra, Hochmair, Neis, & Tonini, 2014) found that OSM users
contribute a more diverse set of features in their home region than in external regions.
Methodology
Study Sites
This research uses two districts, Tripoli and Benghazi, as study sites. Tripoli is
one of 22 districts of Libya and located in the northwestern part of the country on the
Mediterranean Sea. Its population is about 1,067,000 and its area is about 835 km² (322
mi²). Benghazi is another district of Libya and located in the northeastern part of the
country on the Mediterranean Sea with a population of about 667,000 and an area of
28
about 11,372 km² (4390 mi²). By comparing five VGI data sources, this study reviews
name changes of 49 artificial areal geographic features in the Tripoli district and 44 in
the Benghazi district in or after 2011. The analysis is restricted to those features for
which a name change was either reflected in any of the five VGI platforms or in one of
the alternative data sources (e.g. Facebook Websites) we had access to. Areal features
analyzed include universities, institutes, schools, kindergartens, mosques, clinics,
stadiums, towers, neighborhoods, squares, and markets. In addition, the study analyzes
name changes for five major roads in Tripoli and for four major roads in Benghazi.
Data Sources and Data Collection Methods
The analysis in this study is limited to artificial geographic features and excludes
natural geographic features, such as lakes and rivers. This is because natural features
in the study area did not undergo name changes as a result of political changes.
Information about name changes of artificial geographical features comes from two
types of sources which are (1) VGI datasets (OSM, Wikimapia, Google Map Maker,
Panoramio, and Flickr) and (2) Facebook sites listing local school directories as well as
local knowledge and personal communication with local residents. Regarding VGI as a
data source for name changes, only features satisfying at least one of the following
conditions were retained for further analysis: (a) At least one old VGI name among all
VGI sources was associated with the previous regime; (b) At least one new VGI name
was associated with the revolution.
OSM is an open source mapping platform that provides online map tools to
create, store, and edit geographic features (e.g., roads, land use, or buildings) on a
worldwide map. For the analysis, the full history planet dump file, which stores edits of
each feature ever created in OSM, was used. The file was downloaded in March 2015
29
in pbf format and the history data of Libya was extracted using the OSM-history-splitter
tool. This splitter tool extracts a data subset from the full history planet dump based on
geographic boundaries. The data was then imported into a PostgreSQL database, and
SQL queries were used to identify dates of name changes for all features in the
database. Only features falling inside the boundaries of the two analyzed districts in
Libya were retained for further analysis.
Wikimapia, another mapping platform analyzed for this study, allows users to
draw rectangles and polygons at places of interest, to add an object description, and to
modify place entries (both location and description) that were created by other users
(Mummidi & Krumm, 2008). The Wikimapia application programming interface (API)
facilitates searching, updating, and downloading data from Wikimapia maps. However,
the history of feature edits must be reviewed manually on the website itself. Besides
using the typical search terms associated with the former regime and the Libyan
revolution in the free text search field, features from relevant classes, such as schools,
hospitals, and universities, were selected through the “Categories” menu and manually
reviewed for name changes. Figure 2-1 illustrates the identification of a feature name
change as an example. It shows that the Al-Fateh University (a name associated with
the Gadhafi dictatorial regime) was renamed to the more neutral name Tripoli University
(see upper text box). Besides changing a place name, Wikimapia allows users also to
edit the description of a feature, as shown in the lower text box in Figure 2-1.
Google Map Maker is an online application that provides users with tools to add,
edit, and update map information of geographic features, such as roads and buildings.
Feature changes will, after a review process by Google, be reflected on Google Maps.
30
Google Map Maker has been available since 2008 and is to be retired in March 2017
(Map Maker, 2016), with its editing functions being gradually migrated to Google Maps.
For our study, the free text search field was used to search for features whose names
were associated with the former regime and the revolution. The editing history was
manually reviewed for identified objects in March 2015. As opposed to Wikimapia,
Google Map Mapper provides exact dates of edits independent of the time that has
passed since the edit. Figure 2-2 shows the editing of the same university feature as
before in Google Map Maker, which was performed on August 26, 2011. As can be
seen, the old and new names match between Wikimapia and Google Map Maker in
Arabic and English except for small differences in the order of words in English,
revealing general data consistency in this example.
The photo-sharing platforms Panoramio and Flickr are the two remaining VGI
data sources in the presented analysis. Both platforms provide an API, which was used
to download geotagged photos for Tripoli and Benghazi in April 2015. Place names,
which can be attached to the photo by the user, were obtained from photo titles and
other tags. Place names in analyzed geotagged photos stem exclusively from the
knowledge of local users who manually geotagged photos. This is because automated
geotagging of photos through mobile devices requires detailed base maps with feature
names. Such base maps and underlying detailed gazetteers do not exist for the
analyzed areas in this study. Since the Panoramio and Flickr APIs do not provide a
photo editing history, a name change of the artificial geographic feature of interest could
only be identified by comparing titles and tags between nearby photos that were posted
before and after the revolution date, respectively. The geographic focus for this
31
comparison was on locations where the other three map based VGI sources, or
alternative data sources, indicated a name change during or after 2011.
The second source of information on feature name changes was knowledge from
local residents and local institutions that was not mapped on a VGI platform but
available in some other form. We reached out to the Urban Planning Department of the
Libyan government to obtain documentation about changes of place names after the
revolution, which was, however, not available from the department. Instead, the website
called “The Electronic Gate for Schools” (http://smsm.ly/maps/), which was created by
the Department of Technologies and Maintenance of Educational Facilities, provided
changed names of schools. The Website allows the public to add new schools to the
map, to submit a report about name changes of existing schools, and to describe and
categorize schools. Facebook, which is popular in Libya, provided further information on
this topic. Several Facebook pages were used by local teachers to introduce the new
names of local schools whose names were changed after the revolution. Parents use
these pages as directories to identify new names given to the schools of their children.
Also, for elections taking place after the revolution, government agencies used the
schools listed in these directories as election centers. Another resource for changed
place names was local residents in both districts. Some of these residents were
contacted and asked to share their local knowledge about name changes of any
artificial geographic features.
The author’s local knowledge about naming conventions of places in Libya
before and after the 2011 revolution was crucial to identify feature names associated
with the former regime and with the revolution. Typical terms used in names associated
32
with the former regime include “First of September”, “Al-Jamahiriya”, and “Al Fateh
Revolution”, whereas terms associated with the Libyan revolution in 2011 include, for
example, “17th February”, “independence”, and “martyrs”. Especially for OSM,
Wikimapia, and Google Map Maker searching for these core terms written in Arabic
language was the key to find features with names relating to the period before and after
the revolution. VGI platforms allow typing in tag values as free text and in different
languages, and spelling errors may occur (Longueville, Luraschi, Smits, Peedell, &
Groeve, 2010). For the analyzed features, Arabic and English were the only two
languages that users used to add or update place information on the five used VGI
sources. Therefore, before-after name comparison of a feature may involve these two
languages. Use of Arabic, which is the official language in that region, points towards a
local user, as opposed to tourists who more likely add place related information in
English. Several spelling errors were detected in VGI based place names. However,
spelling errors were ignored for the purpose of name change detection in our analysis if
the name portraits the correct meaning or the correct description.
Although most of the old feature names were associated with the names of the
Gadhafi dictatorial regime, not all of them were changed to be associated with names of
the revolution. Instead, some feature names were changed to be associated with their
city or region (e.g. University of Tripoli), or they represented more general names (e.g.
New Libya School). Similarly, some of the old features names were not associated with
the Gadhafi dictatorial regime, but they were nevertheless changed to be associated
with the names of the revolution. For example, the Court Square in Benghazi was
renamed to Freedom Square when the revolution began.
33
Figure 2-3 maps polygon and road features in both districts that were analyzed in
this study. With a few exceptions, all of these features have new names that are
associated with the Libyan revolution in at least one of the used reference sources.
Features in the map are classified by the source that provides the new feature name,
i.e. VGI, other local knowledge (labeled “people”), or both. Figure 2-3 shows that VGI
and alternative information sources mostly complement each other since for only a few
features new names were mentioned in both types of data sources (orange square
symbol). Although the location with new names obtained from VGI overlaps only little
with local residents’ knowledge about actual name changes, the author’s local
knowledge was used to check VGI name changes for plausibility, besides applying VGI
internal data checks. Information on new names for road features was either provided
through VGI or local knowledge, but not through both.
Analysis
Not all objects (polygons and streets) were present in each of the five VGI
sources. Therefore, the first step was to identify all locations that had either (1) an
artificial object with a name change in at least one VGI data source or (2) an artificial
object mentioned in at least one of the alternative data sources (e.g. Facebook page).
This process resulted in 49 polygon and five road features for Tripoli, and 44 polygon
and four road features for Benghazi. Next, all of these locations were classified into five
labeling categories for each VGI source (see Table 2-1 through Table 2-4). The labeling
categories are sorted by the amount of information their features provide regarding
names and name changes. A feature was assigned to the first labeling category if it was
completely missing in a VGI source, i.e. not mapped, but if it existed in another data
source. For Panoramio and Flickr, the existence of a feature was determined by
34
whether it was shown in at least one photo nearby the location or not (independent of
title or tag information). A feature was assigned to the second labeling category if it was
mapped in the VGI dataset but lacked any name information.
An object was assigned to the third labeling category (“old name”) if it had only a
name that was used before the revolution but not updated. The “new name” category
holds features that contain only names used after the revolution but no names from
before the revolution. Finally, the fifth labeling category contains features with both old
and new names, i.e. reflecting a name change during or after the revolution. An
example for the last class was provided in Figure 2-1 and Figure 2-2. Changes were
only analyzed up to the first occurrence during or after the revolution for the analysis.
Since the posed research question evolves around the 2011 revolution, subsequent
name changes were not considered.
Another aspect of the conducted analysis was the distinction between local and
external VGI users who contributed to name changes of the artificial features in a
specific VGI data source. For the three-map based VGI sources (OSM, Wikimapia, and
Google Map Maker), the distinction between local and external user was determined
from the number of contributions within and outside Libya. A contributor is considered
local if the total number of changes made by the contributor within Libya exceeds or
equals that of the number of changes made outside Libya. A contributor is considered
external if that is not the case. In Wikimapia and Google Map Maker, there are some
contributors that cannot be identified as local or external because they contributed to
these VGI sources without usernames, making it impossible to see their editing history.
These contributors were classified as unknown users in the analysis. For Panoramio
35
and Flickr, local and external contributors were distinguished by the number of days
between first and last photo contribution of a user in Libya (Hauthal & Burghardt, 2016).
More specifically, the contributor is considered external if this date difference is 30 days
or less, and local if the difference is more than 30 days. A contributor with just one
photo uploaded was classified as an unknown user.
The second source of information on feature name changes was local
knowledge. For the few features where new feature names can both from VGI and local
knowledge, the latter was used as a reference data source to check the former
regarding the correctness of the changed names. That is, if local knowledge was
available for a feature, only those VGI features where the old or new name matched the
local knowledge were retained.
When no alternative local data source was available the most likely correct
feature name was based on the majority of the used VGI data sources that shared the
same feature name. Besides a few exceptions, all features analyzed in this study that
had an old or new feature name in several VGI data sources shared similar names that
matched and were therefore assumed to be correct. If an old and new feature name
appeared in only one VGI data source that feature was still retained (and assumed
correct). This was to increase the sample size of the study in order to be able to derive
more informative VGI editing patterns as a result of political changes.
Results
The section begins with presenting results relating to the primary research
question of this study, namely to what extent which VGI data is able to indicate a name
change. This is followed by characterizing the user base that caused name changes in
36
the analyzed VGI data sources in terms of local vs. external users. The last part reports
results about the reliability of name changes.
Detection of Name Changes
Table 2-1 lists for Tripoli and the five VGI sources the classification of polygon
features into the five labeling categories. Results in the first category show that Flickr
has the highest number of missing features among the five VGI sources, whereas
Wikimapia has no missing features. Looking at the last three categories, it can be seen
that Wikimapia provides the highest number of objects with old and new names (21),
which means that it is the most complete individual data source to consult for name
changes. The next best data source in this aspect are Google Map Maker and
Panoramio with seven features containing old and new names. OSM provides the least
amount of relevant information since only one feature was found to reflect name
changes. The shared photo portal Flickr ranges somewhere in-between. The bottom
row in Table 2-1 provides for each VGI data source the number of different contributors
that added new feature names for those features falling into the fourth and fifth
category. It shows that each data source has a diverse pool of data contributors and
that updates were not provided by a single person only. Wikimapia, with the largest
number of mapped features, exhibits also the largest number of different mappers.
Figure 2-4 maps the classification results. The maps reveal different spatial patterns of
mapped and missing features between the five used VGI sources. Wikimapia maps all
analyzed features both in the center of Tripoli and its suburbs. For OSM and Flickr,
mapped features are primarily concentrated in the city center, whereas Google Map
Maker and Panoramio demonstrate about the same level of completeness in the city
center and in the suburbs. Based on data from Table 2-1, a chi-square test of
37
independence was performed to examine the relation between labeling category and
VGI platform. The relation between these variables is significant (𝑥2 (16, 𝑁 = 245) =
131.46, 𝑝 < 0.00001). This means that the proportion of features falling into the different
labeling categories varies by VGI data source.
Table 2-2 shows the same type of count data for Benghazi. Some differences
compared to the Tripoli count data (Table 2-1) can be observed. First, Panoramio has a
much higher proportion of missing features than for Tripoli. A reason could be that this
district is less frequently visited by tourists who are the main group of contributors to
Panoramio. Second, the number of features reflecting a name change (last category) is
for all VGI categories equal to or lower than the numbers found for Tripoli. Wikimapia is
the only VGI data source providing name updates for more than one feature. Although
the total number of mapped OSM features is higher for Benghazi than for Tripoli, for
only one feature the name has been updated as well. In general, the low combined
numbers from the last two categories indicate that the major source for updated names
in this district comes from sources other than VGI, e.g. Facebook or personal
communication, revealing that the VGI community is not as established in this district as
in Tripoli. Figure 2-5 maps results from Table 2-2 for Benghazi. As for Tripoli, Wikimapia
maps all the features both within the center and the suburbs of the city. OSM and
Google Map Maker show similar relative densities in mapped features between the city
center and suburbs. Both Panoramio and Flickr map the same two features, one in the
city center, and one far out in the suburbs. Based on data from Table 2-2, a chi-square
test of independence was performed to examine the relation between labeling category
and VGI platform. The relation between these variables is significant (𝑥2 (16, 𝑁 =
38
220) = 157.10, 𝑝 < 0.00001). This means that the proportion of features falling into the
different labeling categories varies by VGI data source.
Table 2-3 lists for Tripoli and the five VGI sources the classification of road
features into the five labeling categories. In general, few roads with name changes were
identified from VGI and non-VGI sources both for Tripoli and for Benghazi, meaning that
apparently fewer roads than areal features were renamed after the revolution. All five
data sources mapped at least some streets, and all data sources except for Flickr
reflected also name changes for some streets. OSM, Wikimapia, and Google Map
Maker provide, relatively speaking, the most comprehensive information about road
name updates, whereas shared images provide the least comprehensive information.
This pattern differs somewhat from Table 2-1, where shared photo services performed
better than OSM in the fifth category. This indicates that the OSM community focuses
generally more on roads, whereas shared photos are mostly taken from areal objects of
limited spatial extent, such as buildings or plazas. Figure 2-6 maps these results, where
black and green for Flickr, indicate that this data source does not provide any
information on street names.
Table 2-4 provides corresponding numbers for Benghazi. As was already
observed for polygon features, the data is less complete than for Tripoli with respect to
name update information. Benghazi streets are the only dataset where Wikimapia does
not provide any name updates, whereas Google Map Mapper reveals the highest
number of features that contain old and new names. None of the two photo sharing
services provides any information about updated names. Results for Benghazi streets
39
are mapped in Figure 2-7. OSM, Wikimapia, and Google Map Maker map all the used
streets of the city, whereas Panoramio and Flickr have just one mapped street.
Local and External Contributors
Figure 2-8 plots for each data source the number of local, external, and unknown
contributors who contributed at least one old feature name. For Flickr, Panoramio, and
OSM old feature names come primarily from local contributors. Similarly, in Wikimapia
most contributors who could be classified based on their editing history were local.
However, more than a half of user could not be classified. Google Map Maker reveals
very little information in this aspect, due to numerous anonymous edits.
Figure 2-9 shows a similar pattern for users contributing new place names. It
clearly demonstrates that the local contributors are primarily responsible for providing
new name information in the various VGI data sources. The percentage of unknown
users in Google Map Maker and Wikimapia declined compared to contributions of old
feature names.
Reliability of Name Changes
For features where a local reference information (e.g. Facebook page entry) was
not available to verify the new name, the most likely old and new feature names were
determined based on the majority vote among the five VGI data sources. Figure 2-10
shows how many VGI data sources shared an old or new name on features that had no
local knowledge reference. In general, the old feature names appear more often on
more than one different VGI platform than the new names, meaning that old names are
somewhat more reliable than new ones. Features that have their name shown in only
one VGI data source (left most group of bars) cannot be checked through VGI data
sources alone, but were still retained for further analysis.
40
Sometimes, since different mappers were involved in reporting name changes for
the same feature on different VGI websites, updated feature names varied slightly
between VGI data sources. Differences can be partially explained by variations in the
translation from Arabic to English. An example is the new name of the Al-Fateh
University (a name associated with the Gadhafi dictatorial regime). The name was
updated to Tripoli University in Wikimapia and to University of Tripoli in Google Map
Maker (compare Figure 2-1 and Figure 2-2). However, for some updated names,
differences between VGI data sources went beyond the typical nuances caused by
translations or spelling errors. In total, two names in OSM, two names in Google Map
Maker and Panoramio, and six names in Wikimapia were updated differently than the
most likely correct new names. For such a feature, the new (incorrect) name for the
affected VGI source was ignored (i.e. treated like absent) when it came to assigning the
VGI data source to one of the five labeling categories (Table 2-1, Table 2-2). Such a
feature could then only be assigned either to the first or the second label category for
that VGI data source.
Concluding Discussion
For Tripoli, more data reflecting name changes could be obtained from VGI data
sources than from local knowledge sources, illustrating the effectiveness of a system
that uses citizens as voluntary sensors (Goodchild, 2007b) in an organized way. The
comparison of VGI data completeness indicates that Wikimapia, at least in this region of
the world, provides generally the most complete data source among all analyzed VGI
data sources and is one of the preferred VGI platforms by local citizens. In general, VGI
data provided more information in Tripoli than in Benghazi, reflecting the varying degree
of data quality between different locations (Hochmair & Zielstra, 2015; Neis et al., 2013).
41
The presented approach can be transferred to other places around the world and
be applied to man-made events (e.g. political) or natural events (e.g. earthquakes) that
have effects on geographic features mapped in several VGI sources. The challenge of
finding named features in sparsely mapped regions, like in Tripoli and Benghazi, and
determining the characteristics of toponymy before and after the event, will remain.
42
Table 2-1. Number of geographical polygon features in different labeling categories for Tripoli
Labeling category OSM Wikimapia Google Map Maker
Panoramio Flickr
Feature does not exist 34 0 20 20 43 Feature without name 9 2 3 8 2 Feature with old name only 3 22 9 11 0 Feature with new name only 2 4 10 3 1 Feature with old and new name
1 21 7 7 3
Total 49 49 49 49 49
No. of different users updating
3 21 13 4 4
Table 2-2. Number of geographic polygon features in different labeling categories for
Benghazi
Labeling category OSM Wikimapia Google Map Maker
Panoramio Flickr
Feature does not exist 26 0 31 42 42 Feature without name 5 1 1 0 1 Feature with old name only 9 38 6 1 0 Feature with new name only 3 1 5 1 1 Feature with old and new name
1 4 1 0 0
Total 44 44 44 44 44
No. of different users updating
2 7 5 1 1
43
Table 2-3. Number of streets in different labeling categories in Tripoli
Labeling category OSM Wikimapia Google Map Maker
Panoramio Flickr
Feature does not exist 0 0 0 1 2 Feature without name 0 0 0 1 3 Feature with old name only 1 1 1 1 0 Feature with new name only 0 0 1 0 0 Feature with old and new name
4 4 3 2 0
Total 5 5 5 5 5
No. of different users updating
4 3 2 1 0
Table 2-4. Number of streets in different labeling categories in Benghazi
Labeling category OSM Wikimapia Google Map Maker
Panoramio Flickr
Feature does not exist 0 0 0 3 3 Feature without name 1 0 0 0 0 Feature with old name only 0 4 2 1 1 Feature with new name only 2 0 0 0 0 Feature with old and new name
1 0 2 0 0
Total 4 4 4 4 4
No. of different users updating
2 0 2 0 0
44
Figure 2-1. Editing history in Wikimapia
Figure 2-2. Editing history in Google Map Maker
45
Figure 2-3. New names of artificial geographic features from VGI and people
46
Figure 2-4. Labeling of geographic features in Tripoli
47
Figure 2-5. Labeling of geographic features in Benghazi
48
Figure 2-6. Labeling of streets in Tripoli
49
Figure 2-7. Labeling of streets in Benghazi
50
Figure 2-8. Type of contributors providing old place names
Figure 2-9. Type of contributors providing new place names
0 10 20 30 40 50 60 70 80 90 100
OSM
Wikimapia
Google Map Maker
Panoramio
Flickr
Number of users
local external unknown
0 10 20 30 40
OSM
Wikimapia
Google Map Maker
Panoramio
Flickr
Number of users
local external unknown
51
Figure 2-10. Number of old and new feature names with no local knowledge reference
shared on different numbers of VGI sources
0
10
20
30
40
1 source 2 sources 3 sources 4 sources 5 sources
Nu
mb
er o
f fe
atu
res
old name new name
52
CHAPTER 3 ANALYZING THE EFFECT OF EARTHQUAKES ON OPENSTREETMAP
CONTRIBUTION PATTERNS AND TWEETING ACTIVITIES
Crowd-sourced data, such as volunteered geographic information (VGI)
(Goodchild, 2007a) and social media posts, have been used to manage relief efforts
around natural disasters (Haworth & Bruce, 2015; Zook et al., 2010), to study the
formation of user mapping communities after such disasters (Budhathoki &
Haythornthwaite, 2013), and to analyze human dynamics as a result of such events
(Wang & Taylor, 2014a) . While short-term effects of crisis events on Twitter activity
patterns (Goolsby, 2010) and OSM mapping patterns (Zook et al., 2010) in response to
natural crises have already been analyzed in previous studies, their effects on longer
term VGI contribution patterns and social media usage, as well as on human travel
patterns (human dynamics) toward affected regions are less understood. Also,
information about the types of edits and the categories of new features mapped by the
OSM community in response to natural disasters could reflect how the degree of
completeness of the different feature categories is affected (i.e. increased) by such
mapping efforts. The objectives of this study are to determine:
• how earthquakes, as examples of natural disasters, change both short-term and longer-term activity behavior in OSM and Twitter
• how the proportion of local and external data contributors changes throughout such events
• how contributions of organized mapping communities, such as HOT, differ from those of other mappers
• patterns of human dynamics in response to such events
• feature types primarily mapped in OSM in response to such events
53
The analysis provides new insights into OSM data growth patterns (Neis & Zipf,
2012) and user loyalty (Napolitano & Mooney, 2012), adding on to previously analyzed
factors that affect VGI contribution patterns, such as land use type (Alivand & Hochmair,
2017; Arsanjani, Mooney, Helbich, & Zipf, 2015), socio-economic factors (Heipke,
2010), organized mapping events and campaigns (Dittus, Quattrone, & Capra, 2017),
and natural disasters, such as earthquakes (Poiani, dos Santos Rocha, Degrossi, & de
Albuquerque, 2016). It also contributes to the large research topic of human movement
pattern analysis from crowd-sourced and social media data (Valle et al., 2017),
especially in response to natural disasters (Goodchild & Glennon, 2010). This
knowledge will improve planning abilities for crisis management in the future, since it
reveals which type of OSM data will get mapped and continuously updated after such
an event, and to which extent it contains local contributions, allowing to draw
conclusions about its data quality (Zielstra et al., 2014). The case study revolves around
two crisis events, which are a 2015 earthquake in Nepal and a 2016 earthquake in
central Italy.
Previous Work
Collaborative Mapping and Crisis Events
Over the past decade, volunteer mappers have used a growing number of tools
and platforms to contribute geospatial information to crowd-sourced data repositories in
response to natural and political crises (Ahmouda & Hochmair, 2017; Ziemke, 2012;
Zook et al., 2010). An early example of collaborative mapping efforts could be observed
in the aftermath of a 2010 earthquake in Haiti, where a Haiti Ushahidi site and the OSM
platform provided important street maps and infrastructure information for relief efforts
(Goolsby, 2010). The Humanitarian OSM Team (HOT), which supports the generation
54
of free up-to-date maps for relief efforts through collaborative mapping, also contributed
to the 2010 Haiti earthquake as one of the many tasks completed in response to various
crises1, including the Typhoon Yolanda in the Philippines 2013 (Palen et al., 2015).
Several studies have analyzed the processes that are involved in collaborative
mapping during crisis events, as well as the short-term changes in contribution behavior
to crowd-sourcing platforms. For example, Poiani et al. (2016) review the effect of an
earthquake which hit Nepal on April 25, 2015, on OSM contributions between April 15
and May 15. Results reveal a sharp rise in the number of active contributors around the
event, which can be partially explained by mapping parties organized in Bangalore
(India) and in Barcelona (Spain) shortly after the earthquake. The vast majority (about
99%) of nodes and ways during that 1-month observation period were edited or added
after the earthquake, showing that the event triggered a large wave of user participation.
Soden and Palen (2016) conducted 36 semi-structured interviews with GIS practitioners
and information managers who were involved in the emergency response to that
earthquake. The study identified an increased level of collaboration and spatial data
sharing between map producers and the Nepal government compared to previous
crises, reflected by 8,000 OSM volunteer mappers contributing after the Nepal
earthquake, compared to the much smaller number of 500 contributors mapping around
the Haiti (2010) earthquake over the same time period (Soden & Palen, 2014). That
higher number of participants is, however, also driven by the general increase in the
number of registered OSM users from 200,000 in 2010 to around 2.2 mio. in 20152.
1 http://wiki.openstreetmap.org/wiki/Humanitarian_OSM_Team
2 https://wiki.openstreetmap.org/wiki/Stats
55
Previous papers that examined OSM contribution patterns in the context of natural
disasters discussed the number of edits or edited objects in response to events and the
proportion of new versus more experienced mappers (Palen et al., 2015; Poiani et al.,
2016), but did not elaborate on longer term changes in the contribution patterns over
time. Other studies describe recruitment efforts and emerging collaborations in disaster
relief efforts (Dittus et al., 2017), and the interaction between the mapping community
and stakeholders in crowd-based disaster mapping (Soden & Palen, 2016).
Social Media Activity and Crisis Events
While VGI platforms are primarily used for mapping tasks, social media
platforms, such as Twitter, YouTube, and Facebook, provide platforms to share
information and opinions about crisis events, such as refugee movements (Rettberg &
Gajjala, 2016), terrorist attacks (Cassa, Chunara, Mandl, & Brownstein, 2013),
earthquakes (Crooks, Croitoru, Stefanidis, & Radzikowski, 2013), tsunamis (Acar &
Muraki, 2011), or flood events (De Albuquerque, Herfort, Brenning, & Zipf, 2015). The
activity level in social media networks increases during crisis events (Austin, Fisher Liu,
& Jin, 2012) to share up-to-date information about the events and to provide emotional
support (Cvetojevic & Hochmair, 2018).
Tweet analyses use different pieces of information from tweets, including tweet
message, location, timestamp, or linked information such as images (Cvetojevic,
Juhasz, & Hochmair, 2016; Steiger, Albuquerque, & Zipf, 2015). Middleton, Middleton,
and Modafferi (2014) introduce a social media crisis mapping platform for natural
disasters that, based on real-time Twitter data, maps affected regions. Resch, Usländer,
and Havas (2018) use tweets to assess the footprint of and damages caused by natural
disasters through a combination of Latent Dirichlet Allocation (LDA) for semantic
56
information extraction and local spatial autocorrelation for hot-spot detection. The
granularity of geographic references in tweet texts that were used to specify the spatial
extent of two earthquakes in Italy and Myanmar are analyzed in (Zahra, Ostermann, &
Purves, 2017), showing that a larger number of toponyms are used for Myanmar than
for Italy, although most of the toponyms for the Myanmar earthquake relate to areas in
countries surrounding Myanmar, such as Bangladesh and India, but not Myanmar itself.
Human Dynamics and Crises
Previous work has shown that crisis events can affect movement patterns of
people at different geographic scales. Li, Airriess, Chen, Leong, and Keith (2010)
assess the effect of Hurricane Katrina on the migration and return process of New
Orleans residents, which includes an analysis of evacuation return rates, experiences,
motivations to return or stay, and evacuation routes. The study used data from surveys,
phone and in-person interviews, and focus groups, but no crowd-sourced mapping or
social media data. Another paper uses geo-tagged tweets to study the influence of three
tropical storms (Hurricane Sandy, Typhoon Wipha, and Typhoon Haiyan) on human
mobility patterns, revealing that during the events the frequency of longer trips
decreased significantly (Wang & Taylor, 2014a). Geotagged data from about 500,000
Twitter users between 2011 and 2013 were used to infer trends of out-migration rates
for various countries around the world, including Mexico, Spain, Greece, and Ireland,
due to economic conditions (Zagheni, Garimella, & Weber, 2014).
Besides this research, travel dynamics toward natural events (e.g. for aid and
support teams) at the worldwide scale have not yet been addressed in previous work.
The studies and methods presented in this chapter will add to the body of knowledge
that connects travel patterns to natural crises using VGI and social media data.
57
Data and Methods
Study Areas
The first earthquake analyzed had a 7.8-moment magnitude scale (M) and
occurred on April 25, 2015, at an approximate depth of 8.2 km and 80 km northwest of
Kathmandu, the capital of Nepal (Figure 3-1a). Over 9,000 people were killed and more
than 23,000 were injured. On May 12, 2015, a major aftershock occurred near the
Chinese border between Kathmandu and the Mount Everest with a magnitude of 7.3 M.
This aftershock killed more than 200 people and injured more than 2,500 people. This
earthquake is considered the worst natural disaster that has hit Nepal since 1934.
The second earthquake analyzed hit Central Italy on August 24, 2016, with a
magnitude of 6.2 M (Figure 3-1b). It caused the death of 297 people and injury of at
least 365 people. Severe damage was reported in some towns, including Amatrice,
Accumoli, and Pescara del Tronto. The estimated economic loss ranged between $1
billion and $11 billion3.
OSM edits and Twitter activity data were extracted for the study areas affected
by earthquakes (red areas in Figure 3-1), and for nearby control areas (blue squares in
Figure 3-1). The extent of the affected areas was determined from earthquake maps of
the United States Geological Survey (USGS)4. The control areas were chosen in a way
to be close to, but not directly affected by the earthquakes in question, and to reflect
similar population characteristics (rural vs. urban) as the affected areas. For the Nepal
earthquake, New Delhi is the closest nearby large city to Kathmandu. For the Italy
3 http://www.royalgazette.com/re-insurance/article/20160901/reinsurers-face-up-to-166m-italy-quake-loss
4 https://earthquake.usgs.gov/earthquakes/eventpage/us20002926/executive; https://earthquake.usgs.gov/earthquakes/eventpage/us10006g7d/executive
58
earthquake which took place in the Umbria region in the central mountain region, the
area around Sulmona, a small city in the mountains south of the affected area was
chosen as a control area.
Data Collection and Preparation
Data for this study come from OSM and the Twitter streaming API. Only
geocoded tweets were used, which make around 1% of all tweets (Graham, Hale, &
Gaffney, 2014).
For OSM, the full planet history dump file, which contains feature edits, was
downloaded. Selected OSM changeset files, which contain all changes made to OSM
features by any mapper in a geographic area, were downloaded as well. Next, the
OSM-history-splitter tool was applied to extract data for the areas of interest from the
history dump file. This data was imported into a PostgreSQL database using the OSM-
history-importer. Next, SQL queries were run to identify features that were newly
mapped or edited during the analysis time period of this study. As a further step in OSM
data preparation, the following two criteria were considered for the removal of OSM data
edits that were most likely generated by automated scripts, so called bots (Zielstra et
al., 2014):
• Remove edits from changesets with a tag comment "mechanical=yes" or "bot=yes"
• Remove edits on nodes and ways in changesets with more than 4000 edits
The bots were first identified in OSM changesets and corresponding edits were
subsequently removed from features in the areas of interest.
Geo-tagged Twitter data were collected 6 months before and 6 months after the
earthquakes from the affected and nearby control regions and downloaded in
59
JavaScript Object Notation (JSON) format which contains tweet id, time created, tweet
content, hashtags, location, and user profile information. These data were subsequently
stored in a PostgreSQL database. When posting a tweet, a user has two options to
provide geolocation information. The first, less precise option, is through a place tag,
which is suggested by the Twitter app based on the current user location, resulting in a
bounding box in the JSON field. The second option is exact coordinates, resulting in a
latitude/longitude pair of the current user position in the JSON field. Before May 2015,
users could opt to have all their tweets geocoded with exact coordinates in Twitter apps
(e.g. Twitter for Android). However, since May 2015, due to a change in the app user
interface functionality, the user needs to request the use of exact coordinates for each
individual tweet, which significantly reduces the number of tweets posted with exact
coordinates.
As part of Twitter data preparation, tweets from automated Twitter accounts
(bots) were excluded, where travel speed, tweet frequency, and tweet source were
considered. A speed threshold of 200 kilometers per hour between subsequent tweets
of the same user and a frequency threshold of more than 150 tweets per day were
applied (Azmandian, Singh, Gelsey, Chang, & Maheswaran, 2012; Zhang & Paxson,
2011). Also, only tweets from mobile applications or computer applications that people
typically use to tweet (e.g. Twitter for Android, Twitter for iOS, or twitter.com) were kept
for further analysis. Tweets posted from sources, such as Google, Big Planet Earth, and
TweetMyJOBS, were excluded. Also, only original tweets were used for further analysis,
and retweets were excluded.
60
Analysis Methods
OSM contribution patterns and Twitter activity patterns were analyzed over a
one-year period for each event, which extends between 6 months before and 6 months
around the crisis event under consideration. The next few paragraphs describe the
extraction of user patterns from OSM and Twitter data sources in more detail.
OSM
OSM editing patterns were extracted from the OSM history dump file. By
comparing two consecutive rows, one can identify the edits that occurred to a feature
between two versions, and when they occurred. Three basic operations can be applied
to each OSM feature, which are create, modify, and delete (Ramm & Topf, 2010; Rehrl,
2013; Zielstra et al., 2014). Nodes and ways are the basic features of OSM where the
nodes represent points and ways represent linear or areal features consisting of a list of
nodes that define a polyline. Nodes with tags represent point features, such as points of
interest, and differ therefore from nodes on ways which usually have no tags but provide
solely geometry information to define the shape of ways. In this study, nodes were
filtered by their tags to extract standalone point features, and hence nodes without tags
(as found in ways) were not considered for node statistics. The following operations on
OSM nodes and ways were taken into account for the analysis of contribution patterns:
• Operations on nodes:
• Create geometry (new node)
• Remove geometry (deleted node)
• Change position of geometry
• Add/remove/update a tag
• Operations on ways:
• Create geometry (new way)
• Remove geometry (delete way)
• Change position of geometry (move a node on way)
• Add node to way
61
• Remove node from way
• Add/remove/update tag Besides monthly frequency plots of operations, statistical tests, such as the chi-
squared test of independence were performed to determine differences in OSM
contribution patterns between an affected area and its reference area. The number and
type of operations were also analyzed for different OSM feature types. OSM offers 23
primary feature types, including amenity, highway, shop, or building. The primary
feature type of an OSM object can be extracted through the key tag in the key-value
pair of the analyzed OSM object.
Various methods can be applied to determine the home region of an OSM
mapper (Neis & Zipf, 2012; Zielstra et al., 2014). In this study, the analysis of a user’s
home region was built upon three criteria, which are the first changeset created, the
number of changesets made, and the number of changes made by the user. More
specifically, if the location of the first point created and the centroid of the first
changeset created are within the affected area, and the affected area is within the
country in which a user contributed most operations, the user is called local to the
region. This region becomes then the user’s home region. The same is also true if either
most changeset centroids or most feature operations fall within the affected area. In
case of a tie in count numbers between the affected area and other areas, the user is
considered local to the affected area. This approach was conducted locally on
downloaded data, and no online OSM tools were used to determine the user’s home
region.
A user is called external to the affected area, if his or her home region does not
overlap with the affected area. In addition, if the user’s home region is located in a
62
different country than the affected area, the user is considered to be an international
user. The defining criteria for a local mapper provided above correctly identify a mapper
as an external (or international) mapper even if his or her first point is mapped in the
crisis region (and not in the mapper’s true home region) but the other criteria are not
satisfied. Such a case is common for international HOT members who start contributing
to OSM as part of a disaster mapping task, but then move on to other countries for later
tasks. In principle, an external user could map remotely within the affected area just at
the time of the event and thus be incorrectly classified as local based on the above
criteria. To evaluate this type of error, mapping operations after the event were
analyzed in more detail for all users. Results show that for the Nepal earthquake 43.6%
of local users made OSM changes after the earthquake (i.e., between 7 days and 6
months after the event) exclusively within the affected area, whereas only 4.8% of them
made changes outside the affected area. For the Italy earthquake, the corresponding
numbers are 33.3% and 4.4%, respectively. Although this method does not provide a
ground truth evaluation, it indicates that the applied method to distinguish between local
and external mappers works reasonably well. This is, because most mappers that are
classified as local contributors focus indeed almost exclusively on the affected area.
For human dynamics related to the mapping of crisis events, it is necessary to
understand where an OSM mapper was physically located before, during, and after a
crisis event, respectively. The “source” tag of a changeset, if populated by the user, can
help to determine whether a mapper was physically present in a region while mapping
(namely, if using a local mapping source), or whether the user conducted desktop
mapping from satellite imagery instead. Although individual OSM features also have a
63
source tag, these are sparsely populated (Juhász & Hochmair, 2016), so that this study
relies solely on source tag information in changesets. An OSM contributor’s simplified
travel trajectory can then be derived by connecting two regions of local mapping
activities of that contributor, namely the crisis region near the earthquake, and the
mapper’s prior mapping region away from the earthquake, such as the mapper’s home
region. When aggregating the identified travel trajectories from all OSM contributors a
flow map can be generated that reflects the general event-related travel pattern among
OSM users traveling to the crisis event. Physically present users in a crisis region could
theoretically use remotely sensed sources only (e.g. “Bing” satellite images) for OSM
data edits so that they would not be discovered through our approach. However, it is
quite unlikely that such a local mapper uses only a remote source in the crisis region
after traveling there.
An alternative method to distinguish between local and remote mapping (which is
not applied in our study though) is supervised machine learning approaches, e.g.
Decision Tree, Support Vector Machine, or Logistic Regression on OSM edits (Basiri,
Amirian, & Mooney, 2016). This would require a substantial manually annotated training
set of edits based on known local mapping tasks and remote mapping tasks, and a
selection of measurable OSM editing variables (e.g. proportion and variety of primary
keys or values in edited features, density of edited nodes on polyline features) for the
prediction of the mapping type.
The average number of geo-referenced tweets per day, week, and month was
used to analyze differences in tweet activities between different time periods and
locations, and to assess the effect of the earthquakes on tweeting behavior. In addition,
64
the proportion of event-related tweets among all tweets was computed for different
points in time, using event-related hashtags, such as #nepalearthquake, #terremotiitalia,
#prayfornepal, or #terremoto2016, and event-related keywords, such as earthquake,
sisma, nepalquake, nepalshakes, devastation, disaster, and collapse. The lists of
keywords and hashtags related to the events were constructed manually in an iterative
process e.g. by parsing tweet content and identifying hashtags and keywords in English
(and some in Italian) that were common and frequently used during the events. As
shown in (Hong, Convertino, & Chi, 2011), more than half of posted tweets are in
English, which, together with the nine other top languages, cover 95.6% of tweets.
Italian and Nepali were not among them, which is why these languages were not used
in our analysis, with the exception of a few hashtags in Italian that could be easily
identified. Furthermore, in order to determine if a user was local or external to a region
of interest, and to determine the user’s home country, the tweeting history of each user
who posted from the earthquake region during the event time was obtained from the
Twitter API, which allows downloading the latest 3,000 tweets of a user. A user was
considered local if the majority of the user’s geotagged tweets fell within the affected
area.
To identify an actual trip of a Twitter user to one of the analyzed crises regions
the user’s locations of tweet activities before, during, and after the crisis events were
compared. More specifically, a user who posted tweets from the affected area during
but not before or after the event and who also posted tweets before or after the event
from an outside area, was considered to be traveling to the crisis area. From among this
identified group of traveling Twitter users, those who were likely traveling because of the
65
event were identified through event based keywords and hashtags found in their tweet
posts. Keywords related to rescue operations, such as rescue, relief, support, and
emergency and hashtags related to rescue operations, such as #rapidresponseteam,
#earthquakeresponse, #searchandrescue, and #rapidrescueteam, were used for this
purpose. The movement of a user was considered to be related to the crisis event if at
least one of his or her tweets contained such as a keyword or hashtag and was posted
during the time of the event in the affected area. In the generated flow maps, all
observed travel movements were aggregated for each user, so that one unit on a flow
arrow corresponds to travel of one user.
Analysis of Contribution and Activity Patterns
The section describes observed OSM and Twitter contribution patterns, and
begins with the OSM data analysis for both earthquakes, followed by Twitter analysis.
OSM Data Contributions
Nepal earthquake
Figure 3-2a shows the monthly counts of created OSM way features, changes of
way tags and way geometries, and users contributing to ways before, during, and after
the months of the Nepal earthquake events (April 25, 2015, and May 12, 2015). A spike
in April and May in all four charts reveals a significant impact on the mapping
contributions in OSM in the affected area during the event period and up to 2 months
later. A similar pattern was observed for nodes in the Nepal area. As opposed to this,
the Nepal control area (New Delhi) does not reflect an increase in mapping activities
during the event periods (chart not shown for brevity).
Monthly user plots show a significant increase both in the number of local and
external contributors around the time of the earthquake (Figure 3-2b), where the relative
66
increase is higher for local than for external contributors. This suggests the presence of
an active local mapping community during the crisis. The figure shows also a
permanent contributor base of HOT members among external mappers before, during,
and after the event, whereas local activities only spike during the event, also for local
HOT users.
Next, the contribution history of users is analyzed in more detail. In Table 3-1,
values in the center row (boldface) refer to the subset of users who mapped during the
earthquake in the crisis region. Rows above and below indicate which proportion of
these users mapped also before or after the event, respectively. The rightmost four
columns subdivide these user counts for local and external users. The second column
shows that few users mapped in the region before and during the event (up to 0.26%),
whereas the percentage of contributors who continue to map after the event remains
somewhat higher for at least 3 months (0.50% or more). External contributors make a
significantly higher share than local mappers both before and after the event. As
opposed to this, local users constitute the larger group during the crisis with 54.7%.
Figure 3-3 plots on a timeline from 2005 to 2015 the month in which users who
mapped during the earthquake in the affected Nepal area submitted their first OSM
changeset. A large percentage of local users (84.6%) (Figure 3-3a) and external users
(39.5%) (Figure 3-3b) signed up for OSM during the weeks of the earthquake, which is
true both for HOT members and other mappers. This illustrates the power of such
events to attract new OSM contributors.
Figure 3-4 shows for the most frequently added OSM feature types the number
of new nodes (i.e. point features) (Figure 3-4a) and ways (Figure 3-4b) mapped during
67
the analyzed one-year period. The x-axis represents the month, the y-axis the OSM
feature type, and the z-axis the number of contributors mapping a feature type. The
color value indicates the number of nodes or ways added (according to the legend).
Both for nodes and ways, buildings are most frequently mapped during the earthquake,
suggesting that buildings receive most attention during earthquake-related events. For
ways, highways are the second most mapped feature type. This dominance of mapping
highway and building features reflects the expected HOT-OSM mapping behavior in
disaster response, which is apparently also followed by non-HOT members. While for
created ways, buildings are the most frequently mapped features throughout the year
(Figure 3-4b), for nodes this is only the case during the earthquake (Figure 3-4a),
revealing a change in the mix of features that are mapped in the affected region as a
result of the earthquake.
A chi-squared test of independence was conducted to check for a statistical
association between the change in the proportion of mapped feature types between
two-time periods (e.g. before and during the earthquake) and mapping region. A first
test compared the change in the proportion of node features added before and during
the earthquake between the Nepal earthquake area and the New Delhi reference area,
which was found to be significantly different between both regions (𝑥2 (5, 𝑁 = 11975) =
1401.4, 𝑝 < 2.2𝑒−16). This means that the earthquake has a significant effect on the
types of features being mapped when compared to a non-affected area. The same test
procedure was conducted to compare the change in contribution behavior before and
after the earthquake between the affected and the reference area (𝑥2 (5, 𝑁 = 1556) =
42.085, 𝑝 < 0.0001) suggesting that the earthquake affects also the composition of
68
feature types being mapped after the earthquake. Similar test results were obtained for
way features.
Italy earthquake
During the month of the Italy earthquake event, which took place on August 24,
2015, OSM mapping activities peaked significantly (Figure 3-5a), whereas the Sulmona
reference area showed no increase in OSM mapping activities whatsoever (plots not
shown for brevity).
The bar chart in Figure 3-5b indicates an increase both in the number of local
and external contributors around the time of the earthquake, where the relative increase
is higher for local than for external users. An Italian task manager project was opened
during the earthquake and managed by an Italian OSM team5. Figure 3-5b shows that
both external and local contributors are part of the Italian OSM team. While an increase
in local OSM team member activities (left chart) lasts until 2 months after the
earthquake, external OSM team members showed increased levels of activities only
during the event (right chart).
Similar to the Nepal earthquake case, among contributors who mapped during
the earthquake in the Italy crisis region, it is primarily the external contributors who
continue their mapping activities in the region after the event and who already mapped
before the event. The difference to Nepal is that local mappers are the minority of users
(13.1%) also during the earthquake (compared to 54.7% for Nepal), meaning that the
earthquake triggered only a moderate increase in mapping activities within the local
OSM community. This can also be seen by the relatively small number of users in the
5 http://osmit-tm.wmflabs.org/project/13
69
left chart of Figure 3-5b for August 2016, compared to much larger user numbers in the
right chart for that same month.
The timelines in Figure 3-6 show that a large percentage of local contributors
(42.2%) who mapped during the Italy earthquake began their first contributions around
that time (Figure 3-6a), whereas a larger portion of external members (87.2%) had
mapped before the first day of the earthquake (Figure 3-6b). Therefore, the Italy
earthquake attracted a somewhat different, i.e. more experienced, OSM mapper crowd
than this was the case for Nepal.
The proportion of created OSM node and way features of different types varied
throughout the analyzed year (Figure 3-7a and Figure 3-7b). Specifically, buildings
(ways) and amenities (nodes) were more frequently mapped during and after the
earthquake than before the event, suggesting that the mapping community considered
these feature types particularly relevant to be mapped. As opposed to the Nepal case,
only a few buildings were mapped as nodes, but instead mostly as ways. This areal
digitization approach requires somewhat more experience. The largely external
mapping community present in the affected Italy area with its multi-year mapping
experience likely brings the necessary skills, which explains this somewhat different
mapping pattern compared to Nepal.
A chi-squared test revealed a statistical association between the change in the
proportion of mapped node features (𝑥2 (5, 𝑁 = 217) = 53.346, 𝑝 < 0.0001) and way
features (𝑥2 (5, 𝑁 = 17207) = 2087, 𝑝 < 2.2𝑒−16) before and during the earthquake and
the mapping region (i.e. affected area vs. reference area). A similar significant impact of
70
the region was observed for a before and after comparison of mapped OSM feature
types for node and way features.
Tweet Activity Patterns
Nepal earthquake
The average monthly numbers of geo-referenced tweets with exact coordinates
posted during the analyzed one-year period around the Nepal earthquake (April 25,
2015) and the aftershock (May 12, 2015) are shown in Figure 3-8a (left). The same data
for the New Delhi reference area as a control is shown in Figure 3-8a (right). The Nepal
chart reveals a peak during the first month of the earthquake (April 2015), followed by a
sharp decline in tweet numbers in May. The latter is caused by a change in the Twitter
app interface which requires more steps from the user to post geo-tagged tweets with
exact coordinates. The reference area in New Delhi shows a similar tweet pattern,
however, with a less prominent peak during the 2 months of the earthquake compared
to March. In both plots, the right axis of the chart (ranging between 0 and 1) shows the
magnitude of contributions in the different months relative to the peak months,
suggesting a drop of tweets numbers by over 50 percent due to changes in the
application user interface. The reason we analyzed the Nepal earthquake and not
another earthquake that did not undergo this decline in tweets due to technical changes
is the magnitude of the Nepal earthquake, which allows to observe pronounced
changes in Twitter activities. For example, a Myanmar earthquake occurred on the
same day as the Italy earthquake (Zahra et al., 2017). However, it was less powerful
than the one in Italy, causing fewer casualties (four people died compared to 297 in
Italy) and less damage. Figure 3-8a also shows the monthly local and external user
tweeting activities. A small increase in the proportion of tweets from external tweeting
71
persons can be observed during the 2 months of the earthquake for Nepal, but not for
New Delhi. This could be the result of international aid organizations operating in the
affected area during these 2 months.
A refined analysis of daily tweet numbers for April and May reveals a similar peak
on the day of the first earthquake (April 25) in the affected area (Figure 3-8b left) and
the reference area (Figure 3-8b right). Additional hashtag and keyword analysis shows
that a significant proportion of tweets in both regions is thematically related to the Nepal
earthquake (darker bars in the figure). As opposed to this, hashtags and keywords
related to the aftershocks are only found in the Nepal area. These results indicate that
the Nepal earthquake was initially of interest to the tweeting community across an
extended, international region, but that the interest in this topic faded after the first
earthquake.
Italy earthquake
A spike in the average number of tweets during the month of the earthquake
(August 2016) both within the affected area and the reference area (Sulmona) is shown
in Figure 3-9a.
A closer look at daily tweet numbers, however, reveals that only the affected area
(Figure 3-9b left) experiences a peak on the day of the earthquake (August 24),
whereas the reference area (Figure 3-9b right) experiences no peak on that day.
Additional hashtag and keyword analysis reveals that a significant proportion of tweets
is thematically related to the earthquake only in the affected area (darker bars in Figure
3-9b left), but not in the reference area. This shows that the effect of the earthquake on
Twitter behavior is limited to a local area in the Italy case.
72
Human Dynamics
This section derives human travel patterns from OSM contributions and tweet
activities for both analyzed earthquake regions. For OSM, the source tag of a changeset
can among others indicates a local data collection activity (e.g. “GPS” or “survey”) or the
use of remotely sensed online sources which do not require a physical presence at the
mapped site (e.g. “satellite imagery” or “Bing Maps”). Oftentimes, however, the source
field is left empty. Figure 3-10 shows for local (a) and external (b) OSM mappers the
distribution of data sources during the 2 months around the Nepal earthquake event
(April 1 through May 31). To identify an external mapper who physically traveled to the
crisis area to contribute to local mapping activities, the user needs to demonstrate local
mapping activities in an external area before the crisis followed by local mapping
activities in the affected area during the crisis (April 25 to May 7).
For the Nepal earthquake, review of OSM data identified more than 1390
international contributors (i.e. from outside Nepal) who mapped during that event in the
affected area. These contributors were from 113 different countries (Figure 3-11a).
Sixteen of these contributors traveled to Nepal to participate in local mapping activities
related to the Nepal earthquake event, based on the source tag analysis described
before. Most of these contributors were from Europe. The flow lines in the figure reflect
human dynamics in relation to this crisis event.
Location analysis of geotagged tweets under consideration of event-related
hashtags and keywords identified more than 150 international users from 44 different
countries who tweeted during the Nepal earthquake event from within the affected area,
but not all of them tweeted from other areas before the event. Among the 150 users, 23
traveled to Nepal to participate in aid and rescue operations, based on identified
73
previous tweet activities outside Nepal and analyzed keywords (Figure 3-11b). Most of
these users were found to travel from the United States. In addition to this, 75 of the 98
external Nepal users (i.e. users who posted primarily from outside the affected area but
from within Nepal) were identified to travel from other areas of Nepal to the earthquake
affected area for the purpose of aid and rescue operations, based on keyword analysis.
Additionally, the analysis discovered another 65 movements to the Nepal crisis region
from within and outside Nepal that matched the expected spatial movement pattern but
did not have a rescue operation related keyword. The choropleth map in Figure 3-11b
shows the number of international users from different home countries around the world
who tweeted during the event in the affected area (but who did not necessarily travel
there in response to the event to participate in rescue operations).
OSM data analysis for the Italy earthquake identified 50 international contributors
from 25 different countries who mapped remotely during that event in the affected area.
Three users traveled from other regions in Italy to the area affected by the earthquake,
but no OSM user from outside Italy visited the affected region for crisis mapping, based
on source tag analysis.
Location analysis of geotagged tweets identified 15 international users from 6
different countries who were tweeting during the Italy earthquake from the affected area,
however none of these posts contained keywords related to rescue operations. Also, 17
of the external Twitter users came from other regions of Italy to the area affected by the
earthquake and used keywords related to rescue operations.
For both OSM and Twitter, the number of contributors per country and travelers
to the earthquake region is sufficiently high for Nepal to generate flow and choropleth
74
maps for visualization, but it is too small to run more detailed statistical analyses (e.g.
proportion of travelers among the country population, the influence of travel distance on
travel activities). For OSM, the countries with most data contributions to the Nepal
earthquake affected area correspond to those countries with a generally active OSM
mapping community (Neis & Zipf, 2012), including Germany, the U.S., and some other
European countries. Also India, which is bordering Nepal to the north, reflects a high
OSM contribution activity. Figure 3-11b shows that the US, which is one of the countries
with the highest Twitter penetration rate in the world (Hawelka et al., 2014), has the
highest number of users tweeting from the event in Nepal besides neighboring India.
Since the sample size of event related tweets is smaller for the Italy earthquake, it was
not possible to draw clear conclusions about human dynamics and international tweets
patterns for that event, which is why no corresponding map is shown for the Italy
earthquake. The travel analysis in this study reflected the level of internationality
associated with the different crisis events. The Nepal affected area receives more
international mapping activities and registers more international travel activities than
Italy, which could indicate that developing countries need (and receive more) assistance
from abroad than developed countries for such events, based our case study.
Concluding Discussion
For both analyzed study areas the number of local and external OSM
contributors increased significantly during the time of the crises, where external OSM
contributors kept a more active level of mapping in the crisis region after the event than
local contributors. A series of chi-squared tests showed that an earthquake has a
significant effect on the OSM feature types being mapped in both cases of Nepal and
Italy, reflecting a disproportionally strong emphasis on the mapping of building and
75
highway features after the earthquake, as can be expected from the mission of HOT
mapping tasks. The analysis results show more OSM feature edits and users
contributing to event-related mapping efforts in Nepal than in Italy (compared Figure 3-2
and Figure 3-5). This could be explained by the lack of detailed OSM maps in the Nepal
region before the earthquake, as a consequence of a poorer economy, less developed
communication infrastructure, or other social and environmental factors (Gröchenig,
Brunauer, & Rehrl, 2014; Heipke, 2010).
Twitter data analysis, based on keyword selection, showed that news about
earthquakes are posted both in affected areas (whether they are rural or urban) and to
some extent in relatively nearby urban areas (i.e. New Delhi), but not in nearby rural
areas (Sulmona control area for Italy). This suggests that this type of disaster is treated
as a local event in the Twitter information landscape in rural areas, but reaches beyond
the affected area in, typically better connected, metropolitan areas with a wider range of
information channels about current events.
The study demonstrates that especially OSM can provide relatively up-to-date in-
situ information about areas affected by natural disasters which is necessary for
effective disaster management (Poser & Dransch, 2010). It shows that part of the OSM
community is engaged in elevated mapping activities up to a month or two after such an
event, where especially the continued engagement of the (small) local mapping
community can contribute to the accurate mapping of features that may have been
affected by the earthquake (e.g. closed roads). The two showcases of this study, which
captured OSM contribution behavior after earthquakes in remote (Nepal) and rural
(Italy) areas, showed that external users played a major role in OSM mapping. Although
76
large, metropolitan areas around the world with a strong OSM presence have a
considerable share of external mapper contributions (Neis et al., 2013), we would
nevertheless expect that, in case of a natural disaster in an urbanized area with strong
OSM presence, the mapping activities in the aftermaths of such an event would be
primarily conducted by the local community on site due to its mapping experience and
local knowledge. We would also expect that the need for physical travel to the site,
which was found to be extensive in the Nepal case, decreases if such an event
occurred in an urban area with a strong local OSM mapping community already present.
77
Table 3-1. Monthly activities of local and external contributors who also mapped during the Nepal earthquake
Date Number of users
% of users Number of local users
% of local users
Number of external users
% of external users
Nov 2014 8 0.14% 1 12.50% 7 87.50%
Dec 2014 9 0.16% 0 0.00% 9 100.00%
Jan 2015 9 0.16% 0 0.00% 9 100.00%
Feb 2015 12 0.21% 0 0.00% 12 100.00%
Mar 2015 15 0.26% 3 20.00% 12 80.00%
25 Apr 2015 – 18 May 2015
5761 100.00% 3152 54.71% 2608 45.27%
Jun 2015 80 1.39% 10 12.50% 70 87.50%
Jul 2015 34 0.59% 5 14.71% 29 85.29%
Aug 2015 29 0.50% 5 17.24% 24 82.76%
Sep 2015 11 0.19% 0 0.00% 11 100.00%
Oct 2015 9 0.16% 0 0.00% 9 100.00%
Nov 2015 8 0.14% 1 12.50% 7 87.50%
78
Figure 3-1. Affected and control areas of the analyzed earthquakes in Nepal/India (a)
and Italy (b)
79
Figure 3-2. Edit and user counts for OSM ways in the affected Nepal area (a) and
number of local and external contributors including HOT members (b)
80
Figure 3-3. Monthly numbers of first OSM changesets submitted by local (a) and
external (b) contributors who mapped in the affected Nepal area during the earthquake
Figure 3-4. Most prominent OSM node and way features mapped in the affected Nepal
area
81
Figure 3-5. Edit and user counts for OSM ways in the affected Italy area (a) and number
of local and external contributors including HOT members (b)
82
Figure 3-6. Monthly numbers of first OSM changesets submitted by local (a) and
external (b) contributors who mapped in the affected Italy area during the earthquake
Figure 3-7. Most prominent OSM node and way features mapped in the affected Italy
area
83
Figure 3-8. Average monthly number of tweets in the affected Nepal area and the New Delhi reference area (a) and daily number of tweets in both areas (b)
(a)
(b)
84
Figure 3-9. Average monthly number of tweets in the affected Italy area and the
Sulmona reference area (a) and daily number of tweets in both areas (b)
85
Figure 3-10. Distribution of reported OSM sources by local (a) and external (b)
contributors in the affected Nepal area
86
Figure 3-11. Nepal earthquake: OSM contributors (a) and Twitter users (b) by country,
overlayed by user flow lines derived from both sources
87
CHAPTER 4 USING TWITTER TO ANALYZE THE EFFECT OF HURRICANES ON HUMAN
MOBILITY PATTERNS
Natural crises, such as earthquakes, hurricanes, and floods affect human
mobility in the form of evacuation, migration, and return processes (Li et al., 2010). For
example, in 2005, Hurricane Katrina affected around 1.7 million people who had to be
evacuated in the state of Louisiana (Coussens & Goldman, 2007). Using interviews,
focus group discussions, and door-to-door surveys, that study examined the spatial
morphology of routes, the volumes of evacuees as well as their return rates and
experience. Hurricane Matthew of 2016 led to the evacuation of more than one million
people across Cuba and 1.5 million people in the State of Florida1. In 2017, the floods
from Hurricane Harvey around Houston, Texas displaced more than 30,000 people and
inundated hundreds of thousands of homes2. Understanding human mobility in times of
such events is important for support and rescue operations. Various sources have been
used to identify human mobility patterns during such events, including social media and
online mapping platforms. One study used tweets to analyze the short-term
perturbations on three other weather events, which are Hurricane Sandy, Typhoon
Wipha, and Typhoon Haiyan, concluding that tropical cyclones can significantly disturb
travel frequencies and displacement probabilities (Wang & Taylor, 2014a). The study
also showed that in most cases displacement data fits the power-law, and hence the
Lévy Walk during and after the event, better than both the lognormal distribution and
exponential distribution. Using the same data from Hurricane Sandy, a follow-up study
1 https://en.wikipedia.org/wiki/Hurricane_Matthew
2 https://en.wikipedia.org/wiki/Hurricane_Harvey
88
revealed that the center of mass and the radius of gyration of each individual's
movements were highly correlated during perturbation and steady states (Wang &
Taylor, 2014a; Wang & Taylor, 2014b), suggesting that human mobility data (center of
mass and the radius of gyration) obtained in steady states, might be used to predict the
perturbation state. Analysis of tweets posted before, during, and after fifteen destructive
cases across five types of natural disasters (typhoons, winter storms, earthquakes,
wildfires, rainstorms) further confirmed that the power-law can describe human mobility
(Wang & Taylor, 2016). Using a set of anonymized mobile phone billing records (for
voice and text services) one study analyzed perturbations in communication and
mobility patterns in response to eight emergencies (e.g. earthquake, bombing) and eight
non-emergency events (e.g. concert) (Bagrow, Wang, & Barabasi, 2011). Results
indicate that strong propagation of information at the global scale occurs only during
real emergency situations, whereas non-emergency information is more spatially
constrained. Phone record data were also used to track population movements in
response to the Haiti 2010, earthquake, determining a net outflow of 20% of the Port-
au-Prince pre-earthquake population within 19 days of the earthquake (Bengtsson, Lu,
Thorson, Garfield, & Von Schreeb, 2011). Analysis of cell phone data at the national
level showed that travel patterns of individual users may be approximated by a Levy law
model up to a distance characterized by the radius of gyration, but not longer
(Gonzalez, Hidalgo, & Barabasi, 2008). Random walk models have been extended with
the observed traits that the tendency to explore additional locations decreases with
time, and that there is a large probability to return to a location visited before (Song,
89
Koren, Wang, & Barabási, 2010; D. Wang, Pedreschi, Song, Giannotti, & Barabasi,
2011).
Distance, an individual move in a certain period of time, is the main factor in
modeling human mobility (Barbosa et al., 2018). Different terms (e.g., jump length, flight
length, displacement, or trip) are common terms to describe the distance covered
between two stops. Mobility models, such as random way point (RWP) or walk models
(e.g., Brownian motion (BM)) are commonly used in computer networking research for
more accurate understanding of mobility patterns (Rhee et al., 2011). Analyses on
mobility patterns (Shin, Hong, Lee, & Chong, 2008) show that human mobility appears
to be similar to what physicists have called Levy Walks, a model first used to describe
the diffusion of atypical particle not obtained by BM. The mean of flight lengths and
mean pause time between flights are the characteristics used to describe the diffusion
of tiny particles in the BM model. When flight lengths have a scale-free distribution, their
second moment (variance) is not finite, and the particles make Levy walks subject to
atypical diffusion. Hence, the mean squared displacement (MSD), the average squared
distance from the origin after a time 𝑡, of particles making Levy walks is proportional to
𝑡𝛾 where 𝛾 < 1, describing super-diffusion. This super-diffusion comes from the heavy-
tail distribution of the constituent flights (short flights and occasionally long flights) of
Levy walks (Rhee et al., 2011). Oftentimes, power law pause times are accompanied
with Levy walks (known as Levy walk with trapping). Flight and pause time distributions
of Levy walks closely follow (truncated) power-law distributions (Shin et al., 2008).
However, distributions of power-law pause times are not required to identify Levy walks.
90
Tweets and the history of OpenStreetMap edits have been used to analyze
worldwide travel patterns to regions affected by two earthquakes for the purpose of on-
site mapping and assistance with rescue operations (Ahmouda et al., 2018). Another
study analyzed international migration between OECD countries based on tweets, for
example, out-migration rates from Mexico and Southern European countries due to
economic reasons (Zagheni et al., 2014). Tweets were also used to identify refugee
migration patterns from the Middle East and Northern Africa to Europe during the initial
surge of refugees aiming for Europe in 2015 (Hübl, Cvetojevic, Hochmair, & Paulus,
2017), illustrating the limitations of tweets for this purpose due to data scarcity in many
countries. Visual analytics tools have been developed for the visual comparison of
movement patterns from tweets for the purpose of event detection and helping cope
with disaster situations (Krueger, Sun, Beck, Liang, & Ertl, 2016).
Twitter hashtags (keywords with the (#) symbol) draw attention to a particular
topic or event and are actively used in data retrieval. Twitter users can create new
hashtags or use existing ones. Hashtags have been widely used in tracking topical
information during and after crisis situations. For example, Twitter hashtags related to
the Paris attacks in November 2015 helped to model the propagation of information
shared around the world through Twitter (Cvetojevic & Hochmair, 2018), demonstrating
the role of hashtag language, profession of the user, and distance from the event on
information propagation. Spatial and temporal characteristics of the Twitter feed were
also used as a sensor network to assess the extent and impact of earthquakes, for
example in a 2011 earthquake in Christchurch (New Zealand) (Bruns & Burgess, 2012),
a 2013 earthquake south-west of Washington D.C. (Crooks et al., 2013), or the Napa
91
(CA, USA) earthquake in 2014 (Resch et al., 2018). The latter study used a combination
of Latent Dirichlet Allocation (LDA) for semantic information extraction and local spatial
autocorrelation for hotspot detection. A review of techniques for event detection in
Twitter is provided in (Atefeh & Khreich, 2015), where the distinction is made between
event type (specified or unspecified), detection task (retrospective or new event
detection), and detection method (supervised or unsupervised), many of which use
hashtags as Twitter specific features. Using machine-learning methods (Support Vector
Machine, supervised Latent Dirichlet Allocation) in combination with several public
disaster-related datasets it was shown that a classification model can be trained that
automatically collects hashtags highly related to the event and specific only to it
(Murzintcev & Cheng, 2017).
Since most previous studies focused on determining the effects of crises events
on large and regional scale movements, such as evacuations, less is known about the
effect on local mobility patterns, e.g. for the population that stays in a region during such
an event. As a step in this direction, the goal of this study is to investigate the pattern of
trip distances (displacements) of human movements before, during and after two major
hurricanes in three study areas in the United States, based on human data obtained
from Twitter. It also examines the change in Twitter hashtags usage during hurricanes,
which demonstrates a close relationship between user perception of the event and
mobility patterns.
92
Data and Methods
Study Areas
This research examines three different urban areas in the United States, which
are Houston, TX (Figure 4-1a), Miami-Dade County, FL (Figure 4-1b), and North and
South Carolina (Figure 4-1c) that were affected by two major hurricane events. The
extent of the three affected areas (red areas in Figure 4-1) was obtained from Flood
Event Viewer (FEV) of the United States Geological Survey (USGS) for Hurricane
Harvey in August 2017 event3 and Hurricane Matthew in October 20164.
Hurricane Harvey
Hurricane Harvey was a category 4 major hurricane that formed on 17 August
2017 and dissipated on 2 September 2017. It caused severe damage and destruction to
hundreds of thousands of homes and the death of 103 people throughout Texas. The
hurricane dumped a significant amount of rainfall of up to 30 in (760 mm), causing
extensive flooding in many areas including the Houston metropolitan area. This
hurricane is considered to be the wettest tropical cyclone in the history of Texas and the
United States.
Hurricane Matthew
Hurricane Matthew was a Category 5 hurricane that formed on 28 September
2016 and dissipated on 10 October 2016. The hurricane also caused damage of 1.9
billion US $ and far-reaching destruction in Haiti and the southeastern United States.
3 https://www.usgs.gov/special-topic/hurricane-harvey
4 https://water.usgs.gov/floods/events/2016/matthew
93
Over 500 people were killed and more than 400 were injured. Many areas within Florida
and North and South Carolina were exposed to major flooding for several days.
Data Collection and Preparation
The Twitter streaming API was used to collect geocoded tweets in the analyzed
regions, where only tweets which came with exact coordinates were used for further
analysis. Data were downloaded in JavaScript object notation (JSON) format for the
Houston affected area between 1 August 2017 and 30 September 2017 and for the
Miami-Dade County and North and South Carolina study areas between 1 September
2016 and 31 October 2016 and stored into a PostgreSQL database. SQL queries were
used to extract tweet metadata including user id, time created, and geometry location of
the obtained tweets in addition to the profile information of users, such as username.
Tweets from automated Twitter accounts (bots) were excluded before further
analysis, using the Botometer (formerly known as BotOrNot) Web service that evaluates
Twitter accounts to distinguish between human and bot accounts (Davis, Varol, Ferrara,
Flammini, & Menczer, 2016). This service uses a Twitter screen name to examine the
activity of a Twitter account based on which a score is assigned. Higher scores indicate
a higher probability of a bot account. Python was used to execute this service.
Data Analysis
Data analysis included the examination of trip distances during, two weeks
before, and two weeks after the hurricane landfalls from users who posted tweets from
within the affected areas before, during, and after the events. This ensured that the
same sample of Twitter users was considered for trip distance comparison between the
different time periods. For Hurricane Harvey (Houston area), the observation time
during the event was set to two days, which were Monday, August 28 and Tuesday,
94
August 29, 2017. The same weekdays two weeks before (August 14 and 15, 2017) and
two weeks after (September 18 and 19, 2017) the event were chosen for comparison.
For Hurricane Matthew (Miami-Dade County), Thursday, October 6, 2016, was set as
the analyzed event date, and September 22 and October 20, 2016, were assigned for
comparison dates before and after the hurricane, respectively. For North and South
Carolina, Saturday, October 8 and Sunday, October 9 were the observed event dates,
with 25 through 26 September 2016 and 22 through 23 October 2016 used for
comparison before and after the hurricane.
The displacements for each user were calculated in kilometers between each two
consecutive tweet locations using the Haversine formula (Chopde & Nichat, 2013). The
average and the median of displacements from all users before, during, and after the
events were computed. Significance levels of differences in median distances before,
during, and after the event were then determined through the Mann-Whitney U test.
Furthermore, observed displacements were grouped into six distance bands (<= 5, >
5 <= 10, > 10 <= 25, > 25 <= 50 , > 50 <= 100, 𝑎𝑛𝑑 > 100 𝑘𝑚) and the number of
displacements was counted for each band before, during, and after the event. A Chi-
square test of independence was then performed to determine the statistical association
between distance band and observation period.
To determine the distribution of obtained displacements they were fit to a power-
law model (Clauset, Shalizi, & Newman, 2009):
𝑝(𝑥) ∝ 𝑥(−𝛽) (4-1)
and a truncated power-law model (Gonzalez et al., 2008):
𝑝(𝑥) ∝ 𝑥(−𝛽) 𝑒𝑥𝑝(−𝑥 𝑘⁄ ) (4-2)
95
where 𝑥 is the variable of interest (displacement), 𝛽 is the power low exponent, and 𝑘 is
the cutoff value. Also, a Likelihood Ratio test was performed to compare the goodness
of fit between the two applied models and to determine which model provides a better fit
to observed displacement data. The 𝑅 packages movr and lmtest were used for
implementing the power-law distribution models and for conducting the Likelihood Ratio
test, respectively.
Besides analyzing changes in trip distances, a frequency count of Twitter
hashtags in the affected areas was conducted to identify an eventual change in Twitter
topic focus before, during, and after the events. This would give insight into the
magnitude of the effect the hurricanes have on the Twitter community, and could,
therefore, tie changed mobility behavior to the events
Results
User Mobility
Figure 4-2 shows the number of observed trips in the different distance ranges
for the Houston area before, during, and after Hurricane Harvey. The proportion of short
trips (<= 5𝑘𝑚) is highest during the hurricane period (red columns) with 50% compared
to 27% before and 30% after the event. That is, longer trips above 5km occurred less
frequent during the hurricane, and no trips over 50km were observed in the study area
during the hurricane. A chi-square test revealed a significant association between trip
distribution and time period before/during the hurricane (𝜒2 = 11.169, 𝑑𝑓 = 1, 𝑝 <
0.001) and during/after the hurricane (𝜒2 = 8.600, 𝑑𝑓 = 1, 𝑝 = 0.003). This means that
the hurricane causes significant perturbations on travel behavior within the localized
affected area, with trips becoming shorter, and overall also less frequent during the
event.
96
The mean and median offsets between any two subsequent tweet locations were
calculated for each study area before, during, and after the events and are reported in
Table 4-1. In line with the chi-square test conducted above, results for the Houston area
confirm that the mean and median displacement during Hurricane Harvey (𝑀 = 11𝑘𝑚,
𝑀𝑑𝑛 = 5𝑘𝑚) are smaller than the corresponding values before (𝑀 = 16𝑘𝑚, 𝑀𝑑𝑛 =
10𝑘𝑚) and after (𝑀 = 17𝑘𝑚, 𝑀𝑑𝑛 = 9𝑘𝑚) the hurricane, supporting the finding of
limited mobility during the hurricane event. A Mann-Whitney U test showed that the
differences in distance medians between before and during the hurricane (𝑈 = 3892.5,
𝑝 = 0.005, 𝑟 = 0.190) and during and after the hurricane (𝑈 = 4981, 𝑝 = 0.025, 𝑟 =
0.143) are significant.
The distribution of displacements for the Houston area is fit to power-law and
truncated power-law models for the time periods before (Figure 4-3a), during (Figure 4-
3b), and after (Figure 4-3c) Hurricane Harvey. Both models reveal high R-squared
values, although R-squared values are higher for the truncated power-law throughout
(0.82, 0.92, 𝑎𝑛𝑑 0.85) compared to the power law (0.80, 0.87, 𝑎𝑛𝑑 0.83). A Likelihood
Ratio test revealed that the truncated power-law provides a significantly better model fit
to observed displacements than the power-law (𝛥𝜒2(1) = 4.7922 , 𝑝 = 0.028) during the
hurricane, but not before and after the hurricane.
The number of observed trips in the different distance ranges for the Miami-Dade
County affected area before, during, and after Hurricane Matthew are shown in Figure
4-4. Also here, the proportion of short trips (<= 5𝑘𝑚) is highest with 66% during the
hurricane, compared to 52% before and 43% after the event. Also, no trips over 50km
were observed during the hurricane. A chi-square test revealed a significant association
97
between trip distribution and time period before/during the hurricane (𝜒2 = 5.652, 𝑑𝑓 =
1, 𝑝 = 0.017) and during/after the hurricane (𝜒2 = 12.508, 𝑑𝑓 = 1, 𝑝 < 0.001).
Therefore, Hurricane Matthew causes significant perturbations on local mobility
behavior within the Miami-Dade County affected area.
As shown in Table 4-1, the mean and median values of displacements are
smaller during Hurricane Matthew (𝑀 = 5𝑘𝑚, 𝑀𝑑𝑛 = 2𝑘𝑚) than before (𝑀 = 7𝑘𝑚,
𝑀𝑑𝑛 = 4𝑘𝑚) and after (M = 9km, Mdn = 6km) the hurricane in Miami-Dade County. A
Mann-Whitney U test showed that the differences in distance medians between before
and during the hurricane (𝑈 = 7729.5, 𝑝 = 0.019, 𝑟 = 0.141) and during and after the
hurricane (𝑈 = 4461.5, 𝑝 < 0.001, 𝑟 = 0.266) are significant.
The truncated power-law (R-squared = 0.96) fits the distribution of trips better
than the power-law (R-squared = 0.93) before Hurricane Matthew (Miami-Dade County),
although this difference was not statistically significant. Both models reveal an equal fit
to the observed offsets during the hurricane (R-squared = 0.92) and after the hurricane
(R-squared = 0.77).
Also, for North and South Carolina, the proportion of short trips was highest
during Hurricane Matthew (68%), compared to before (53%) and after (49%) (see
Figure 4-5). A chi-square test revealed a significant association between trip distribution
and time period before/during the hurricane (𝜒2 = 3.935, 𝑑𝑓 = 1, 𝑝 = 0.047) and
during/after the hurricane (𝜒2 = 5.935, 𝑑𝑓 = 1, 𝑝 = 0.014). This indicates also a
significant impact of the hurricane event on trip distances.
For the affected areas in North and South Carolina the mean and median values
of displacements are smaller during Hurricane Matthew (𝑀 = 11𝑘𝑚, 𝑀𝑑𝑛 = 3𝑘𝑚) than
98
before (𝑀 = 20𝑘𝑚, 𝑀𝑑𝑛 = 4𝑘𝑚) and after (𝑀 = 24𝑘𝑚, 𝑀𝑑𝑛 = 5𝑘𝑚) the hurricane (see
Table 4-1). A Mann-Whitney U test showed that the difference in distance medians
between during/after the hurricane is significant (𝑈 = 2891, 𝑝 < 0.021, 𝑟 = 0.175).
The truncated power-law shows a much better model fit with observed
displacements (R-squared: 0.91, 0.89, and 0.78) than the power-law (R-squared: 0.70,
0.72, and 0.54) before, during, and after Hurricane Matthew for the North and South
Carolina area, respectively. The Likelihood Ratio test revealed that these differences in
model fit were significant before (𝛥𝜒2(1) = 17.022, 𝑝 < 0.001), during (𝛥𝜒2(1) = 8.929,
𝑝 = 0.002), and after the hurricane (𝛥𝜒2(1) = 11.479, 𝑝 < 0.001).
Hashtags Use Changes
For each hurricane and analysis area, the ten most frequently used hashtags
before, during, and after the event were extracted. They are listed in descending order
in Table 4-2 for the three areas, where hurricane-related hashtags are underlined. For
Hurricane Harvey (top section), hashtags related to the event constitute the majority of
the top ten hashtags during the hurricane (second column), including #hurricaneharvey
and #prayforhouston. No event-related hashtags occurred in the top ten list before and
after the event. This clearly suggests that the hurricane affects the topics tweeted about
during the event.
For Hurricane Matthew, two event-related hashtags (#hurricanematthew, and
#matthew) were found among the top ten hashtags during the hurricane time in Miami-
Dade County, but none before and after. These two hashtags and another event-related
hashtag (#HurricaneMatthew) also occurred in the most ten used hashtags during the
hurricane in North and South Carolina.
99
Concluding Discussion
Results show that displacements become shorter during hurricanes, and that, as
expected, hurricanes limit mobility during these events. This result supports findings
from previous research analyzed for other events (Wang & Taylor, 2014a). The results
also demonstrate that the distribution of displacements can be well approximated by the
power-low models, where the truncated power-law sometimes outperforms the power-
law. This shows that the Levy walk model closely resembles human mobility patterns
even during hurricane events. Twitter data, which were used in this study, do not allow
to extract pause times at locations, which are often accompanied with Levy walk
analysis. Nevertheless, the displacement distribution obtained from Twitter data
conforms with the characteristics of Levy walks (Shin et al., 2008). The high R squared
values across analyzed time periods also shows the resilience of people’s movement
during hurricanes in the sense that there are some invariant characteristics in travel that
remain undisturbed during hurricane events. This study, therefore, contributes to a
better understanding of people’s changed mobility behavior during emergency events.
Twitter data analysis, based on used hashtags, shows that the hurricane events affect
the use of hashtags on Twitter where Twitter users increase the frequency of event-
related hashtags during the hurricanes.
100
Table 4-1. Mean and median (km) displacements before, during, and after the hurricanes with significance levels for differences between distance medians
Period Houston Miami-Dade
North and South Carolina
Mean Median P-value
Mean Median P-value
Mean Median P-value
Before 16 10 0.005 7 4 0.019 20 4 0.275 During 11 5 - 5 2 - 11 3 - After
17 9 0.025 9 6 < 0.001
24 5 0.021
Table 4-2. Frequency of Twitter hashtags used before, during, and after the analyzed
hurricanes
Hurricane/ Study area
Before During After
Hashtag Freq. Hashtag Freq. Hashtag Freq.
Harv
ey/H
ou
sto
n
ireallybreakmusic Repost Mp3waxx GloryFitness houston realtor Houston fitnessmodel jessegreene fitnessmotivation
16 16 14 6 6 4 4 4 4 4
Repost hurricaneharvey prayforhouston houston houstonstrong texas hurricane HurricaneHarvey Hurricaneharvey prayfortexas
129 23 18 9 8 7 7 5 4 4
GloryFitness ireallybreakmusic Repost fitnessmotivation MP3Waxx fitnessmodel breakfast FollowTheSmell ClientMelissaE Houston
28 24 21 19 19 18 10 6 6 5
Ma
tth
ew
/
Mia
mi-D
ad
e C
ou
nty
southbeach miamibeach beachlife southbeachlocal usa tropical beachvolleyball palmtrees Miami fitness
9 9 9 9 8 8 8 6 6 6
miamibeach miami beachlife southbeachlocal southbeach beachvolleyball tropical nature_perfection hurricanematthew matthew
14 11 10 10 10 8 7 6 6 6
Take1TakeOver miami BarackObama TagsForLikes ingodwetrust weedvision daytrader ironteam fuck925 StrongerTogether
33 8 5 5 5 5 5 5 5 5
Ma
tth
ew
/
Nort
h a
nd
So
uth
Ca
rolin
a
northcarolina 3 hurricanematthew 21 fccstudentministry 4 sunset 3 HurricaneMatthew 5 tunarun2016 3 uncw 2 surfcity 4 wilmingtonnc 3 CharlieWilson 2 northcarolina 4 bleachmyfilm 3 surfcitync 2 matthew 4 kurebeach 2 surfcity 2 surfcitync 3 TheMatrimony 2 shotoniphone 2 OnePiece 2 wilmington 2 iphonography 2 charleston 2 outerbanks 2 wilmingtonnc 2 h2ography 2 Chicago 2 knottsisland 2 topsailisland 2 PoeticJustice 2
101
Figure 4-1. Affected areas of Hurricane Harvey in Houston (a) and Hurricane Matthew in
Miami-Dade County (b) and North and South Carolina (c)
102
Figure 4-2. Distribution of trip counts for different distance bands within the Houston
affected area before, during, and after Hurricane Harvey
Figure 4-3. Power-law and truncated power-law distribution models for displacement
data within the Houston affected area before (a), during (b), and after (c) Hurricane Harvey
103
Figure 4-4. Distribution of trip counts for different distance bands within the Miami-Dade
affected area before, during, and after Hurricane Matthew
Figure 4-5. Distribution of trip counts for different distance bands within North and South
Carolina affected areas before, during, and after Hurricane Matthew
104
CHAPTER 5 CONCLUSIONS
The importance of VGI and social media in spreading spatial information has
become crucial over recent years (Asur & Huberman, 2010; Konečný & Reinhardt,
2010; Thackeray, Neiger, Hanson, & McKenzie, 2008). The increase of spatial
information available on online sites makes tracking and analysis of people’s opinions
and their travel patterns possible (Ceron, Curini, Iacus, & Porro, 2014; Lenormand,
Gonçalves, Tugores, & Ramasco, 2015). This research extends previous knowledge by
providing an analysis of the contribution patterns in VGI mapping and social media
activities in response to crises. Various studies were provided in this dissertation to
achieve this goal.
As a first study, the research analyzed the effects of political changes on the
names of artificial geographical features. Many of the names, especially those that were
associated with previous regimes, were changed to become associated with the
revolutions, their dates, their leaders, or their martyrs. Political changes, which occurred
as a result of the revolution in Libya in 2011, had a clear impact on names of artificial
geographical features, such as streets, schools, and hospitals, which were originally
associated with the former regime. Some information about these changes was
reflected in data disseminated by the geospatial Web community. This is particularly
valuable because the Libyan government has no websites or maps that individuals can
consult to obtain current name information. This case study demonstrates an example
of the democratization of GIS, where citizens and grassroots groups share their
knowledge to generate spatial data that are not provided by government officials (Dunn,
2007; Elwood, 2008). In addition to this, VGI provides, especially in the absence of
105
official governmental data, information in a special social context (that of the aftermath
of the Libyan revolution), and may therefore not be completely objective (Glasze &
Perkins, 2015). VGI platforms offer a forum for different voices, reflecting the state of
perception of a local environment at a given time. Nevertheless, VGI provides a useful
supplement to limited governmental resources to better understand how names of
artificial geographical features reflect changes in political systems.
The research also conducted another study to analyze the contribution patterns
to OSM and tweeting activities as a result of earthquake events. The study expands
previous research on event-based VGI contribution behavior (Neis et al., 2013; Zielstra
et al., 2014) in various aspects. It analyzes the proportion of local and external
contributors in mapping and tweeting activities before, during, and after the analyzed
earthquakes, the change in OSM feature types mapped, the proportion of newbies and
experienced OSM users mapping in response to these events, the effect of the events
on tweet frequency in affected and outside reference regions, as well as worldwide
travel patterns to disaster events. The study presented a novel approach to derive an
OSM mapper’s physical presence in an event area during mapping, using the source
tag from changesets. There are limitations to this method, such as data scarcity or
potential incorrect tagging, but the approach is a first attempt to produce a general map
about travel dynamics, which could possibly be refined in future steps.
As a result, an increase of international activities in OSM during the events can
be attributed to focused mapping campaigns of the Humanitarian OSM Team (HOT) (for
Nepal) and the Italian OSM community participating in an organized mapping task (for
Central Italy). However, a closer look at mapping activities shows that from among
106
external mappers only a small portion actually travels to affected regions, but primarily
relies on desktop (armchair mapping) instead, as the analysis of OSM data sources
reveals. Despite this, members of these HOT and other local mapping task communities
retain elevated levels of mapping activities in the affected areas even months after the
events, as opposed to other external and local OSM contributors. A significant change
in the proportion of OSM feature classes being modified during and after events
suggests that big data provide insight into the type of infrastructure mostly affected by
such disasters. Twitter data reveals a spike in the proportion of external visitors in
affected areas, indicating the presence of international rescue teams, news reporters, or
redevelopment assistance teams. Despite these findings, parts of the analysis are
limited by data scarcity. This is especially true for movement analysis based on OSM
source information. In order to get more detailed information about OSM mappers
traveling, contributors would need to be encouraged to more completely provide source
information during their tagging process. Also, while the percentage of geo-tagged
tweets has been low (about 1%) since the beginning of Twitter, the change in the
Twitter app functionalities further reduced the percentage of tweets that are geotagged
with exact coordinates. This effect was demonstrated in connection with the Nepal
Twitter dataset. Exact position information is, however, necessary for the analysis of
local events, where country and even city level geocoding is too coarse. In addition to
this, in some countries, such as Nepal, the geotagging information cannot be obtained
at the city place level. This posed some limitations on the spatial analysis of users’
tweeting history, which was used to distinguish between external and local users. In
107
conclusion, it can be stated that natural crises have a clear impact on OSM contribution
patterns and tweet activities as well as on human dynamics.
The steady growth of information available on social media portals provides new
opportunities for tracking and analyzing people’s mobility (Lenormand et al., 2015). This
research also conducted another study that assessed the effect of hurricanes on human
mobility patterns based on Twitter data. It examined three different urban areas in the
United States (Houston, TX; Miami-Dade County, FL; and North and South Carolina)
that were recently affected by two major hurricanes. To obtain a cleaner assessment of
the hurricane effect on user mobility tweets were only considered from users who
tweeted before, during, and after the hurricane. To mitigate any effects that might
influence people’s movement closer before the event (e.g., preparing for a hurricane)
and after the crises (e.g., flooding, road damage, and blockage) the analysis periods
were spread to two weeks before and after the events. The study also demonstrated the
change in tweeted topics during the hurricanes through frequency counts of hashtags
before, during, and after hurricanes.
Using different statistical tests and methods, differences in human movement
patterns before, during, and after each event were identified, showing that the
proportion of shorter trips becomes significantly higher during the hurricane compared
to two weeks before and after. Moreover, the analysis also shows a significant change
in the mobility range of people, where the means and medians of people’s
displacements during the time of the events are smaller than before and after the
events. The results also found that the displacement distribution is well approximated by
a power-law with high R-squared values, which are sometimes higher for the truncated
108
power-law. These results show a better fit of truncated power-law than power-law, most
likely due to the finite size of the study areas considered (Barbosa et al., 2018).
Furthermore, the results from displacement data fitting of truncated power-law
distribution show that the Levy walk model describes human mobility in the analyzed
events. These results support the findings from studies on human mobility (Gonzalez et
al., 2008; Shin et al., 2008), where mobile phone and GPS displacements follow
truncated power-law distribution. The study also provides a better understanding of how
hurricanes influence people to change their use of hashtags on Twitter. More
specifically, Twitter users increase the use of hurricane-related hashtags during the
emergency events. In general, this study adds to the body of knowledge that connects
human mobility to natural crises.
The recent proliferation of spatial information on the Internet through novel
interactive platforms allows the user community to share data with others and engage in
interaction. Governments began to use Internet technologies to communicate with
citizens (Johnson & Sieber, 2013). Government services can gain credibility and
efficiency through this kind of interaction between governments and citizens. Also, this
relationship encourages citizens to more actively participate in decision-making
processes.
Johnson and Sieber (2013) identify two reasons for the use of user-generated
data sources by government agencies. First, the possibility for citizens, whether they
reside inside or outside a particular jurisdiction, to serve as their environment sensors.
This can help a government to build and maintain its spatial data infrastructure to be
used by decision-makers. Tulloch (2008) provides a case study based on the citizen
109
verification of data that have been collected by the official governmental on vernal
pools. This case creates an example of how government can save money and time by
utilizing the knowledge of citizens in order to support decision-making. Second, VGI can
be a valuable resource for governments as a form of participation of citizens. Contrary
to the opinion of the citizens as sensors, this process treats the use of VGI as an
opportunity for citizens as partners to participate in the production of social, economic,
and environmental goals with the task of strengthening civil society. This strengthening
forms authority basis for the government employee initiatives to support official policies.
Consequently, understanding contribution patterns and activities on online VGI
platforms becomes essential. In this research, the impact of crises on the contributions
to VGI and social media platforms became clear. It revealed more details about which
groups of people contribute and what kind of information they provide. This is
particularly valuable since such results reflect the involvement of citizens and different
web communities. This research also expands the analysis of the editing history of
mapped objects to other VGI mapping platforms, which has so far been mostly limited to
the OSM platform. That is, analyzing the editing history of Wikimapia and Google Map
Maker objects, as explained in Chapter 2, allowed to identify events and name changes.
Also, this study is one of the first to compare the completeness of the editing history
between different VGI mapping platforms. Results showed that especially Wikimapia
was one of the preferred VGI mapping platforms by local citizens in the Libya study
area. Furthermore, this research shows a new way of using Flickr and Panoramio, as
examples of VGI photo platforms, in detecting spatio-temporal changes of information
caused by an event. Moreover, a new approach to discover human dynamics was
110
carried out by this research which led to produce flow maps about travel dynamics
related to crisis events, based on OSM and Twitter data.
For future work of this dissertation, the analyses could be expanded to other
regions of the Arab world and beyond, where could the political events cause a
substantial number of changes in feature names, too. Alternative data sources, such as
tweets or images from social media, regarding their usefulness of detecting name
changes of man-made features could be also reviewed. Moreover, the analyses could
be extended to other events, such as wildfires or floods. Another potential aspect of
future is to determine if individual users contribute to one or several crowd-sourcing
platforms in the case of crisis events and whether related information is cross-tagged
between different data sources (Juhász & Hochmair, 2016).
111
LIST OF REFERENCES
Acar, A., & Muraki, Y. (2011). Twitter for crisis communication: lessons learned from Japan's tsunami disaster. International Journal of Web Based Communities, 7(3), 392-402.
Ahmouda, A., & Hochmair, H. H. (2017). Using Volunteered Geographic Information to measure name changes of artificial geographical features as a result of political changes: a Libya case study. GeoJournal, 83(2), 237-255.
Ahmouda, A., Hochmair, H. H., & Cvetojevic, S. (2018). Analyzing the effect of earthquakes on OpenStreetMap contribution patterns and tweeting activities. Geo-spatial Information Science, 21(3), 195-212.
Alderman, D. H. (1996). Creating a New Geography of Memory in the South:(Re) naming of Streets in Honor of Martin Luther King, Jr. Southeastern Geographer, 36(1), 51-69.
Alderman, D. H. (2016). Place, naming and the interpretation of cultural landscapes. Heritage and Identity, edited by Brian Graham and Peter Howard, 195-213.
Alderman, D. H., & Inwood, J. (2013). Street naming and the politics of belonging: spatial injustices in the toponymic commemoration of Martin Luther King Jr. Social & Cultural Geography, 14(2), 211-233.
Alivand, M., & Hochmair, H. H. (2017). Spatiotemporal analysis of photo contribution patterns to Panoramio and Flickr. Cartography and Geographic Information Science, 44(2), 170-184.
Ames, M., & Naaman, M. (2007). Why We Tag: Motivations for Annotation in Mobile and Online Media. In Proceedings of the SIGCHI conference on Human factors in computing systems (pp. 971-980). New York: ACM.
Arsanjani, J. J., Mooney, P., Helbich, M., & Zipf, A. (2015). An exploration of future patterns of the contributions to OpenStreetMap and development of a Contribution Index. Transactions in GIS, 19(6), 896-914.
Asur, S., & Huberman, B. A. (2010). Predicting the future with social media. Paper presented at the Web Intelligence and Intelligent Agent Technology (WI-IAT), 2010 IEEE/WIC/ACM International Conference on, Toronto, ON.
Atefeh, F., & Khreich, W. (2015). A survey of techniques for event detection in twitter. Computational Intelligence, 31(1), 132-164.
Austin, L., Fisher Liu, B., & Jin, Y. (2012). How audiences seek out crisis information: Exploring the social-mediated crisis communication model. Journal of Applied Communication Research, 40(2), 188-207.
112
Azaryahu, M. (1997). German reunification and the politics of street names: The case of East Berlin. Political Geography, 16(6), 479-493.
Azaryahu, M., & Golan, A. (2001). (Re) naming the landscape: The formation of the Hebrew map of Israel 1949–1960. Journal of Historical Geography, 27(2), 178-195.
Azaryahu, M., & Kook, R. (2002). Mapping the nation: street names and Arab‐Palestinian identity: three case studies. Nations and nationalism, 8(2), 195-213.
Azmandian, M., Singh, K., Gelsey, B., Chang, Y.-H., & Maheswaran, R. (2012). Following human mobility using tweets. In International Workshop on Agents and Data Mining Interaction (pp. 139-149). Berlin Heidelberg: Springer.
Bagrow, J. P., Wang, D., & Barabasi, A.-L. (2011). Collective response of human populations to large-scale emergencies. PloS one, 6(3), e17680.
Barbosa, H., Barthelemy, M., Ghoshal, G., James, C. R., Lenormand, M., Louail, T., . . . Tomasini, M. (2018). Human mobility: Models and applications. Physics Reports.
Basiri, A., Amirian, P., & Mooney, P. (2016). Using crowdsourced trajectories for automated OSM data entry approach. Sensors, 16(9), 1510.
Bengtsson, L., Lu, X., Thorson, A., Garfield, R., & Von Schreeb, J. (2011). Improved response to disasters and outbreaks by tracking population movements with mobile phone network data: a post-earthquake geospatial study in Haiti. PLoS medicine, 8(8), e1001083.
Bittner, C. (2016). Diversity in volunteered geographic information: comparing OpenStreetMap and Wikimapia in Jerusalem. GeoJournal, 82(5), 887-906.
Brewer, G. A., Neubauer, B. J., & Geiselhart, K. (2006). Designing and Implementing E-Government Systems Critical Implications for Public Administration and Democracy. Administration & Society, 38(4), 472-499.
Bruns, A., & Burgess, J. E. (2012). Local and global responses to disaster:# eqnz and the Christchurch earthquake. Paper presented at the Disaster and emergency management conference, conference proceedings.
Budhathoki, N. R., & Haythornthwaite, C. (2013). Motivation for Open Collaboration: Crowd and Community Models and the Case of OpenStreetMap. American Behavioral Scientist, 57(5), 548-575.
Burns, R. (2014). Moments of closure in the knowledge politics of digital humanitarianism. Geoforum, 53, 51-62.
113
Cassa, C. A., Chunara, R., Mandl, K., & Brownstein, J. S. (2013). Twitter as a sentinel in emergency situations: lessons from the Boston marathon explosions. PLOS currents, 5.
Ceron, A., Curini, L., Iacus, S. M., & Porro, G. (2014). Every tweet counts? How sentiment analysis of social media can improve our knowledge of citizens’ political preferences with an application to Italy and France. New Media & Society, 16(2), 340-358.
Chopde, N. R., & Nichat, M. (2013). Landmark based shortest path detection by using A* and Haversine formula. International Journal of Innovative Research in Computer and Communication Engineering, 1(2), 298-302.
Chow, T. E. (2012). “We Know Who You Are and We Know Where You Live”: A Research Agenda for Web Demographics. In D. Sui, S. Elwood, & M. Goodchild (Eds.), Crowdsourcing Geographic Knowledge: Volunteered Geographic Information (VGI) in Theory and Practice (pp. 265-285). Berlin: Springer.
Clauset, A., Shalizi, C. R., & Newman, M. E. (2009). Power-law distributions in empirical data. SIAM review, 51(4), 661-703.
Cohen, S. B., & Kliot, N. (1992). Place-Names in Israel's Ideological Struggle over the Administered Territories. Annals of the association of American geographers, 82(4), 653-680.
Coleman, D. J., Georgiadou, Y., & Labonte, J. (2009). Volunteered Geographic Information: The Nature and Motivation of Produsers. International Journal of Spatial Data Infrastructures Research, 4(1), 332-358.
Couclelis, H. (2003). The Certainty of Uncertainty: GIS and the Limits of Geographic Knowledge. Transactions in GIS, 7(2), 165-175.
Coussens, C., & Goldman, L. (2007). Environmental Public Health Impacts of Disasters: Hurricane Katrina: Workshop Summary: National Academies Press.
Crooks, A., Croitoru, A., Stefanidis, A., & Radzikowski, J. (2013). # Earthquake: Twitter as a distributed sensor system. Transactions in GIS, 17(1), 124-147.
Cvetojevic, S., & Hochmair, H. H. (2018). Analyzing the spread of tweets in response to Paris attacks. Computers, environment and urban systems, 71, 14-26.
Cvetojevic, S., Juhasz, L., & Hochmair, H. H. (2016). Positional Accuracy of Twitter and Instagram Images in Urban Environments. GI_Forum, 1, 191-203.
Davis, C. A., Varol, O., Ferrara, E., Flammini, A., & Menczer, F. (2016). Botornot: A system to evaluate social bots. Paper presented at the Proceedings of the 25th International Conference Companion on World Wide Web.
114
De Albuquerque, J. P., Herfort, B., Brenning, A., & Zipf, A. (2015). A geographic approach for combining social media and authoritative data towards identifying useful information for disaster management. International Journal of Geographical Information Science, 29(4), 667-689.
Dittus, M., Quattrone, G., & Capra, L. (2017). Mass participation during emergency response: event-centric crowd-sourcing in humanitarian mapping. Portland: Association for Computing Machinery (ACM).
Dunn, C. E. (2007). Participatory GIS — a people's GIS? Progress in Human Geography, 31(5), 616-637.
Elwood, S. (2008). Grassroots groups as stakeholders in spatial data infrastructures: challenges and opportunities for local data development and sharing. International Journal of Geographical Information Science, 22(1), 71-90.
Elwood, S. (2010). Geographic information science: emerging research on the societal implications of the geospatial web. Progress in Human Geography, 34(3), 349-357.
Elwood, S., Goodchild, M. F., & Sui, D. Z. (2012). Researching Volunteered Geographic Information: Spatial Data, Geographic Research, and New Social Practice. Annals of the association of American geographers, 102(3), 571-590.
Fraser Taylor, D., & Caquard, S. (2006). Cybercartography: Maps and Mapping in the Information Era. Cartographica: The International Journal for Geographic Information and Geovisualization, 41(1), 1-6.
Glasze, G., & Perkins, C. (2015). Social and Political Dimensions of the OpenStreetMap Project: Towards a Critical Geographical Research Agenda. In J. Jokar Arsanjani, A. Zipf, P. Mooney, & M. Helbich (Eds.), OpenStreetMap in GIScience (pp. 143-166). Cham: Springer International Publishing.
Gonzalez, M. C., Hidalgo, C. A., & Barabasi, A.-L. (2008). Understanding individual human mobility patterns. nature, 453(7196), 779.
Goodchild, M. F. (2007a). Citizens as sensors: the world of volunteered geography. GeoJournal, 69(4), 211-221.
Goodchild, M. F. (2007b). Citizens as Voluntary Sensors: Spatial Data Infrastructure in the World of Web 2.0 (Editorial). International Journal of Spatial Data Infrastructures Research (IJSDIR), 2, 24-32.
Goodchild, M. F., & Glennon, J. A. (2010). Crowdsourcing geographic information for disaster response: a research frontier. International Journal of Digital Earth, 3(3), 231-241.
115
Goolsby, R. (2010). Social media as crisis platform: The future of community maps/crisis maps. ACM Transactions on Intelligent Systems and Technology (TIST), 1(1), 7.
Graham, M., Hale, S. A., & Gaffney, D. (2014). Where in the world are you? Geolocation and language identification in Twitter. The professional geographer, 66(4), 568-578.
Gröchenig, S., Brunauer, R., & Rehrl, K. (2014). Digging into the history of VGI data-sets: results from a worldwide study on OpenStreetMap mapping activity. Journal of Location Based Services, 8(3), 198-210.
Guyot, S., & Seethal, C. (2007). Identity of place, places of identities: change of place names in post-apartheid South Africa. South African Geographical Journal, 89(1), 55-63.
Haklay, M. (2013). Citizen Science and Volunteered Geographic Information: Overview and Typology of Participation. In D. Sui, S. Elwood, & M. Goodchild (Eds.), Crowdsourcing Geographic Knowledge (pp. 105-122). Dordrecht: Springer Netherlands.
Hauthal, E., & Burghardt, D. (2016). Using VGI for analyzing activities and emotions of locals and tourists. Paper presented at the Link-VGI workshop in connection with the AGILE 2016, Helsinki.
Hawelka, B., Sitko, I., Beinat, E., Sobolevsky, S., Kazakopoulos, P., & Ratti, C. (2014). Geo-located Twitter as proxy for global mobility patterns. Cartography and Geographic Information Science, 41(3), 260-271.
Haworth, B., & Bruce, E. (2015). A review of volunteered geographic information for disaster management. Geography Compass, 9(5), 237-250.
Hecht, B. J., & Gergle, D. (2010). On the “Localness” of User-Generated Content. In Proceedings of the 2010 ACM conference on Computer supported cooperative work (pp. 229-232). New York: ACM.
Heipke, C. (2010). Crowdsourcing geospatial data. ISPRS Journal of Photogrammetry and Remote Sensing, 65(6), 550-557.
Hochmair, H. H., & Zielstra, D. (2015). Analysing user contribution patterns of drone pictures to the dronestagram photo sharing portal. Journal of Spatial Science, 60(1), 79-98.
Hollenstein, L., & Purves, R. (2010). Exploring place through user-generated content: Using Flickr tags to describe city cores. Journal of Spatial Information Science(1), 21-48.
116
Hong, L., Convertino, G., & Chi, E. H. (2011). Language Matters In Twitter: A Large Scale Study. Proceedings of the fifth international AAAI conference on weblogs and social media (pp. 518–521).
Hübl, F., Cvetojevic, S., Hochmair, H., & Paulus, G. (2017). Analyzing refugee migration patterns using geo-tagged tweets. ISPRS International Journal of Geo-Information, 6(10), 302.
Johnson, P. A., & Sieber, R. E. (2013). Situating the adoption of VGI by government. In D. Sui, S. Elwood, & M. Goodchild (Eds.), Crowdsourcing geographic knowledge (pp. 65-81). Netherlands: Springer.
Jones, C. B., Purves, R. S., Clough, P. D., & Joho, H. (2008). Modelling vague places with knowledge from the Web. International Journal of Geographical Information Science, 22(10), 1045-1065.
Juhász, L., & Hochmair, H. H. (2016). Cross-Linkage Between Mapillary Street Level Photos and OSM Edits. In T. Sarjakoski, M. Y. Santos, & T. Sarjakoski (Eds.), Geospatial Data in a Changing World: Selected papers of the 19th AGILE Conference on Geographic Information Science: (Lecture Notes in Geoinformation and Cartography, pp. 141-156). Berlin: Springer.
Konečný, M., & Reinhardt, W. (2010). Early warning and disaster management: the importance of geographic information (Part A). International Journal of Digital Earth, 3(3), 217-220.
Krueger, R., Sun, G., Beck, F., Liang, R., & Ertl, T. (2016). TravelDiff: Visual comparison analytics for massive movement patterns derived from Twitter. Paper presented at the Pacific Visualization Symposium (PacificVis), 2016 IEEE.
Lenormand, M., Gonçalves, B., Tugores, A., & Ramasco, J. J. (2015). Human diffusion and city influence. Journal of The Royal Society Interface, 12(109), 20150473.
Lev-On, A. (2012). Communication, community, crisis: Mapping uses and gratifications in the contemporary media environment. New Media & Society, 14(1), 98-116.
Li, W., Airriess, C. A., Chen, A. C.-C., Leong, K. J., & Keith, V. (2010). Katrina and migration: Evacuation and return by African Americans and Vietnamese Americans in an eastern New Orleans suburb. The professional geographer, 62(1), 103-118.
Light, D. (2004). Street names in Bucharest, 1990–1997: exploring the modern historical geographies of post-socialist change. Journal of Historical Geography, 30(1), 154-172.
Longueville, B., Luraschi, G., Smits, P., Peedell, S., & Groeve, T. (2010). Citizens as Sensors for Natural Hazards: A VGI integration Workflow. Geomatica, 64, 41-59.
117
Map Maker. (2015). Countries being mapped. Retrieved from https://support.google.com/mapmaker/answer/155415?topic=1094318&hl=en. Accessed 21 November 2016.
Map Maker. (2016). Google Map Maker graduates to Google Maps. Retrieved from https://support.google.com/mapmaker/answer/7195127. Accessed 21 November 2016.
Middleton, S. E., Middleton, L., & Modafferi, S. (2014). Real-time crisis mapping of natural disasters using social media. IEEE Intelligent Systems, 29(2), 9-17.
Monmonier, M. (2006). From Squaw Tit to Whorehouse Meadow: How Maps Name, Claim, and Inflame. Chicago: University of Chicago Press.
Mooney, P., & Corcoran, P. (2012a). The Annotation Process in OpenStreetMap. Transactions in GIS, 16(4), 561-579.
Mooney, P., & Corcoran, P. (2012b). Characteristics of heavily edited objects in OpenStreetMap. Future Internet, 4(1), 285-305.
Mummidi, L. N., & Krumm, J. (2008). Discovering points of interest from users’ map annotations. GeoJournal, 72(3-4), 215-227.
Murzintcev, N., & Cheng, C. (2017). Disaster Hashtags in Social Media. ISPRS International Journal of Geo-Information, 6(7), 204.
Napolitano, M., & Mooney, P. (2012). MVP OSM: A tool to identify areas of high quality contributor activity in OpenStreetMap. The Bulletin of the Society of Cartographers, 45(1), 10-18.
Neis, P., Goetz, M., & Zipf, A. (2012). Towards Automatic Vandalism Detection in OpenStreetMap. ISPRS International Journal of Geo-Information, 1(3), 315-332.
Neis, P., & Zielstra, D. (2014). Recent Developments and Future Trends in Volunteered Geographic Information Research: The Case of OpenStreetMap. Future Internet, 6(1), 76-106.
Neis, P., Zielstra, D., & Zipf, A. (2013). Comparison of Volunteered Geographic Information Data Contributions and Community Development for Selected World Regions. Future Internet, 5(2), 282-300.
Neis, P., & Zipf, A. (2012). Analyzing the contributor activity of a volunteered geographic information project—The case of OpenStreetMap. ISPRS International Journal of Geo-Information, 1(2), 146-165.
Ostermann, F. O., & Spinsanti, L. (2011). A conceptual workflow for automatically assessing the quality of volunteered geographic information for crisis management. In Proceedings of AGILE (Vol. 2011, pp. 1-6).
118
Palen, L., Soden, R., Anderson, T. J., & Barrenechea, M. (2015). Success & scale in a data-producing organization: The socio-technical evolution of OpenStreetMap in response to humanitarian events. In Proceedings of the 33rd annual ACM conference on human factors in computing systems (pp. 4113-4122). New York: ACM.
Perkins, C. (2014). Plotting practices and politics: (im) mutable narratives in OpenStreetMap. Transactions of the Institute of British Geographers, 39(2), 304-317.
Plantin, J.-C. (2015). The politics of mapping platforms: participatory radiation mapping after the Fukushima Daiichi disaster. Media, Culture & Society, 37(6), 904-921.
Poiani, T. H., dos Santos Rocha, R., Degrossi, L. C., & de Albuquerque, J. P. (2016). Potential of collaborative mapping for disaster relief: A case study of OpenStreetMap in the Nepal earthquake 2015. In System Science (HICSS), 2016 49th Hawaii International Conference on (pp. 188-197). Koloa: IEEE.
Poore, B. S., Wolf, E. B., Korris, E. M., Walter, J. L., & Matthews, G. D. (2012). Structures data collection for the National Map using volunteered geographic information. US Geol. Surv. Open-File Rep, 1209, 34.
Poser, K., & Dransch, D. (2010). Volunteered geographic information for disaster management with application to rapid flood damage estimation. Geomatica, 64(1), 89-98.
Poushter, J., Bishop, C., & Chwe, H. (2018). Social Media Use Continues to Rise in Developing Countries but Plateaus across Developed Ones. Pew Research Center, 22.
Ramm, F., & Topf, J. (2010). OpenStreetMap (3rd ed.). Berlin: Lehmanns Media.
Rehrl, K., Gröchenig, S., Hochmair, H.H., Leitinger, S., Steinmann, R., and Wagner, A. (2013). A conceptual model for analyzing contribution patterns in the context of VGI. In Progress in Location Based Services (Lecture Notes in Geoinformation and Cartography, pp. 373-389). Berlin: Springer.
Resch, B., Usländer, F., & Havas, C. (2018). Combining machine-learning topic models and spatiotemporal analysis of social media data for disaster footprint and damage assessment. Cartography and Geographic Information Science, 45(4), 362-376.
Rettberg, J. W., & Gajjala, R. (2016). Terrorists or cowards: negative portrayals of male Syrian refugees in social media. Feminist Media Studies, 16(1), 178-181.
Rhee, I., Shin, M., Hong, S., Lee, K., Kim, S. J., & Chong, S. (2011). On the levy-walk nature of human mobility. IEEE/ACM transactions on networking (TON), 19(3), 630-643.
119
Rose-Redwood, R. S. (2008). From number to name: symbolic capital, places of memory and the politics of street renaming in New York City. Social & Cultural Geography, 9(4), 431-452.
Seeger, M. W., Sellnow, T. L., & Ulmer, R. R. (1998). Communication, organization, and crisis. Annals of the International Communication Association, 21(1), 231-276.
Shin, R., Hong, S., Lee, K., & Chong, S. (2008). On the Levy-walk nature of human mobility: Do humans walk like monkeys? Paper presented at the Proc. IEEE INFOCOM.
Soden, R., & Palen, L. (2014). From crowdsourced mapping to community mapping: The post-earthquake work of OpenStreetMap Haiti. In COOP 2014-Proceedings of the 11th International Conference on the Design of Cooperative Systems, 27-30 May 2014, Nice (France) (pp. 311-326). Cham: Springer.
Soden, R., & Palen, L. (2016). Infrastructure in the Wild: What Mapping in Post-Earthquake Nepal Reveals About Infrastructural Emergence. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (pp. 2796-2807). New York: ACM.
Song, C., Koren, T., Wang, P., & Barabási, A.-L. (2010). Modelling the scaling properties of human mobility. Nature Physics, 6(10), 818.
Steiger, E., Albuquerque, J. P., & Zipf, A. (2015). An advanced systematic literature review on spatiotemporal analyses of Twitter data. Transactions in GIS, 19(6), 809-834.
Stephens, M. (2013). Gender and the GeoWeb: divisions in the production of user-generated cartographic information. GeoJournal, 78(6), 981-996.
Thackeray, R., Neiger, B. L., Hanson, C. L., & McKenzie, J. F. (2008). Enhancing promotional strategies within social marketing programs: use of Web 2.0 social media. Health promotion practice, 9(4), 338-343.
Tulloch, D. L. (2008). Is VGI participation? From vernal pools to video games. GeoJournal, 72(3-4), 161-171.
Turner, A. (2006). Introduction to Neogeography. Sebastopol, CA: O'Reilly Media.
Valle, D., Cvetojevic, S., Robertson, E. P., Reichert, B. E., Hochmair, H. H., & Fletcher, R. J. (2017). Individual Movement Strategies Revealed through Novel Clustering of Emergent Movement Patterns. Scientific Reports, 7, 44052.
Veil, S. R., Buehner, T., & Palenchar, M. J. (2011). A work‐in‐process literature review: Incorporating social media in risk and crisis communication. Journal of contingencies and crisis management, 19(2), 110-122.
120
Wang, D., Pedreschi, D., Song, C., Giannotti, F., & Barabasi, A.-L. (2011). Human mobility, social ties, and link prediction. Paper presented at the Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining.
Wang, Q., & Taylor, J. E. (2014a). Quantifying and Comparing Human Mobility Perturbation during Hurricane Sandy, Typhoon Wipha and Typhoon Haiyan. Procedia Economics and Finance, 18, 33-38.
Wang, Q., & Taylor, J. E. (2014b). Quantifying human mobility perturbation and resilience in Hurricane Sandy. PloS one, 9(11), e112608.
Wang, Q., & Taylor, J. E. (2016). Patterns and limitations of urban human mobility resilience under the influence of multiple types of natural disaster. PloS one, 11(1), e0147299.
Zagheni, E., Garimella, V. R. K., & Weber, I. (2014). Inferring international and internal migration patterns from twitter data. In Proceedings of the 23rd International Conference on World Wide Web (pp. 439-444). New York: ACM.
Zahra, K., Ostermann, F. O., & Purves, R. S. (2017). Geographic variability of Twitter usage characteristics during disaster events. Geo-spatial Information Science, 20 (3), 231-240.
Zelinsky, W. (1983). Nationalism in the American Place-Name Cover. Names, 31(1), 1-28.
Zhang, C. M., & Paxson, V. (2011). Detecting and analyzing automated activity on twitter. In N. Spring & G. F. Riley (Eds.), Passive and Active Measurement. PAM 2011 Berlin: (LNCS 6579, pp. 102–111). Berlin: Springer.
Zielstra, D., & Hochmair, H. H. (2013). Positional accuracy analysis of Flickr and Panoramio images for selected world regions. Journal of Spatial Science, 58(2), 251-273.
Zielstra, D., Hochmair, H. H., & Neis, P. (2013). Assessing the Effect of Data Imports on the Completeness of OpenStreetMap – A United States Case Study. Transactions in GIS, 17(3), 315-334.
Zielstra, D., Hochmair, H. H., Neis, P., & Tonini, F. (2014). Areal Delineation of Home Regions from Contribution and Editing Patterns in OpenStreetMap. ISPRS International Journal of Geo-Information, 3(4), 1211-1233.
Ziemke, J. (2012). Crisis mapping: The construction of a new interdisciplinary field? Journal of Map & Geography Libraries, 8(2), 101-117.
121
Zook, M., Graham, M., Shelton, T., & Gorman, S. (2010). Volunteered Geographic Information and Crowdsourcing Disaster Relief: A Case Study of the Haitian Earthquake. World Medical & Health Policy, 2(2), 7-33.
122
BIOGRAPHICAL SKETCH
Ahmed Ahmouda was born in Samno, Libya. In 2004, he graduated with a
Bachelor of Science in civil engineering from Engineering Academy Tajoura, Tajoura,
Libya. In 2014, he graduated with a Master of Science in geographic information
systems from the University of Redland, California, US. He was accepted as a Ph.D.
student and graduate assistant at the University of Florida in 2015 and graduated with a
Ph.D. in the School of Forest Resources and Conservation with a concentration in
geomatics in 2018.