Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tjss20

Download by: [University of Connecticut] Date: 28 January 2016, At: 11:44

Journal of Spatial Science

ISSN: 1449-8596 (Print) 1836-5655 (Online) Journal homepage: http://www.tandfonline.com/loi/tjss20

Towards an interoperable online volunteeredgeographic information system for disasterresponse

Chuanrong Zhang, Tian Zhao & Weidong Li

To cite this article: Chuanrong Zhang, Tian Zhao & Weidong Li (2015) Towards an interoperableonline volunteered geographic information system for disaster response, Journal of SpatialScience, 60:2, 257-275, DOI: 10.1080/14498596.2015.972996

To link to this article: http://dx.doi.org/10.1080/14498596.2015.972996

Published online: 24 Nov 2014.

Submit your article to this journal

Article views: 141

View related articles

View Crossmark data

Towards an interoperable online volunteered geographic information systemfor disaster response

Chuanrong Zhanga*, Tian Zhaob and Weidong Lia

aDepartment of Geography & Center of Environmental Sciences and Engineering, University of Connecticut,Storrs, CT, USA; bDepartment of Computer Science, University of Wisconsin-Milwaukee, Milwaukee, WI,

USA

Although volunteered geographic information (VGI) (voluntarily provided crowd-sourcingspatial data) applications have proven to be useful for gathering information about a crisis, theseapplications have provided only limited utilities for response coordination. One major problemof existing implemented VGI systems is heterogeneous data semantics. To provide more utilitiesto disaster response, a VGI system that can encode data semantics is needed. The objective ofthis research is to propose such a VGI system based on the state-of-the-art Geospatial SemanticWeb technologies. A prototype for disaster response in Connecticut was implemented. Althoughthere are still issues waiting for further studies, the implemented prototype shows great promise.

Keywords: volunteered geographic information (VGI); geospatial semantic web; disasterresponse

1. Introduction

Emergency response and disaster management

have greatly benefited from recent advances in

information technologies, especially in GIS

and remote sensing (Cova 1999). However,

findings have shown that the use of GIS for

disaster management can readily fail due to the

inability to rapidly acquire spatial data and the

lack of interoperable GIS (Zerger & Smith

2003). According to a critical report released

by the U.S. House Select Bipartisan Commit-

tee (U.S. House Select Bipartisan Committee

2006), the federal government’s ineffective

response to Hurricane Katrina was partly a

failure of sharing and accessing information

in a timely manner (White house 2006).

The experiences suggest that the real barriers

to emergency response and disaster manage-

ment are in most cases difficulties in sharing

and integrating the heterogeneous data rather

than the lack of data (Donkervoort et al. 2008).

Data sharing facilitated by the advances of

network technologies is hampered by the

incompatibility of a variety of data models and

semantics used at different sites.

To facilitate exchange and sharing of spatial

data built on initial expenditures, Spatial Data

Infrastructures (SDIs), which are data infra-

structures including interactively connected

spatial data, metadata, and tools, have been

developed in many countries in the past two

decades (Rajabifard et al. 2006). Although the

fast development of SDIs and OGC (Open

Geospatial Consortium) web service technol-

ogies has undoubtedly improved sharing and

synchronisation of geospatial information

q 2014 Mapping Sciences Institute, Australia and Surveying and Spatial Sciences Institute

*Corresponding author. Email: chuanrong.zhang@uconn.edu

Journal of Spatial Science, 201

Vol. 60, No. , 257–275, http://dx.doi.org/10.1080/14498596.2015.9729962

5

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

across the diverse sources, literature in SDIs

shows that there are limitations in the current

SDI implementation and it might still be

difficult to find data sources from the currently

implemented SDIs. The implemented SDIs only

emphasise technical data interoperability via

web services and standard interfaces, and

cannot resolve semantic heterogeneity problems

in spatial data sharing. However, difference in

semantics used in diverse data sources is one of

the major problems in spatial data sharing and

data interoperability (Bishr 1998).

In addition, it is difficult to obtain up-to-

date spatial information from the existing data

of SDIs. The informational databases that were

built up over many years through GISs require

trained GIS professionals to operate and

maintain. However, disasters can significantly

affect the local geography, making the existing

spatial data and maps outdated instantly. The

dynamic nature of a disaster situation requires

timely updating of a variety of data/infor-

mation from various sources at federal, state,

and local levels. However, it is challenging to

enable disaster responders to obtain updated

data/information rapidly based on the cur-

rently implemented SDIs. Although existing

data in the currently implemented SDIs can

provide disaster responders useful information

such as political boundaries, population, social

and economic siutations, availability of relief

and services, locations of shelters and

hospitals, and biophysical variables, they lack

up-to-date information such as accessibility to

roads and damaged areas, and number and

locations of injured people. For example,

shelters and hospitals may be closed because

of the outage of power, thus they will no longer

be useful for helping the people affected.

Therefore, timely up-to-date information is

critical for disaster response. The effectiveness

of disaster response not only depends on fast

access to a multitude of distributed existing

data but also relies on obtaining up-to-date

information to save lives.

Remotely sensed images can provide the

timely information that is particularly useful in

disaster response. For example, after the 2010

Haiti earthquake, groups of volunteers

remapped Haiti using satellite imagery. How-

ever, despite recent advances in both in situ and

remote sensing systems, there are still some

phenomena that cannot be sufficiently

measured, such as storms and the closing

down of hospitals. In addition, measurements

from sensor systems may not be available due

to communication interruptions or destruction

of sensors; for example, water gauges may be

destroyed by very severe floods. Moreover,

sensors may not be able to take measurements

at critical moments, as often happens in severe

weather conditions when images from optical

remote sensing systems are obstructed by

clouds. Also for remote sensing, a time delay

occurs due to data acquisition and processing.

Currently, satellite scheduling can take up to

25 hours. This means that images cannot be

acquired on the same day when a disaster

occurs. Some of these gaps may be filled by

Volunteered Geographic Information (VGI),

which is the crowd-sourcing geographic data

provided voluntarily by individuals and a

special case of the larger Web phenomenon

known as user-generated content (Goodchild

2007a, b; Jones et al. 2012).

Advances in geospatial positioning (GPS),

Web mapping, cellular communications (smart

phones), and wiki-based collaboration tech-

nologies have now outpaced the original

visions of the architects of SDIs (e.g. Craglia

et al. 2008). Collaborative web-based efforts

such as Open Street Map, the Ushahidi project,

Sahana2, Google Earth/Maps, Wikimapia, and

Flickr have been used to assist in disaster

response situations (e.g. Laituri & Kodrich

2008; Goodchild & Glennon 2010; Zook et al.

2010; Kawasaki et al. 2013; Stefanidis et al.

2013). Several rapid and successful VGI

deployments employing the aforementioned

collaborative web-based software helped to

coordinate disaster responses recently, such as

the earthquake that struck Haiti in January

2010, the Gulf of Mexico Oil Spill in April

2010, and the series of damaging wildfires that

C. Zhang et al.2258

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

affected Santa Barbara between 2007 and 2009

(Goodchild & Glennon 2010). The launch of

the Ushahidi platform in Haiti demonstrated

the potential of using mobile technology for

information gathering and communication

during disaster response and management.

Several case studies have shown the added

value of using VGI in various types of crisis

events (De Longueville et al. 2010a) such as

earthquakes (De Rubeis et al. 2009), forest

fires (De Longueville et al. 2009), political

crises (Bahree 2008), hurricanes (Hughes &

Palen 2009), floods (De Longueville et al.

2010b), and terrorist attacks (Palen et al. 2009).

Although these VGI (or crowd-sourcing)

applications have proven useful for gathering

information about a crisis, these applications

only provided limited utilities for response

coordination (Gao et al. 2011). This is mainly

caused by lack of compatibility within

different software packages and heterogeneity

among different data sources (Poser & Dransch

2010; Zook et al. 2010; Morrow et al. 2011).

One major problem of VGI is the existence of

the heterogeneous semantic problem. The

challenges of semantic interoperability have

been recognised in the literature (e.g. Zhang

et al. 2010c). Problems arise when VGI is to be

used as decision-support information because

the user-generated annotations are often

unstructured and their contents are not

machine-readable. It is difficult to use the

user-generated data to support the spatial

analysis required in decision-making processes

because computers do not understand the

contents of VGI. For example, with the

currently deployed VGI systems, it is not

easy to get answers to the questions:

Which roads have been blocked (thus

cannot be accessed) by the hurricane Irene

disaster? And which is the best evacuation

route?

The information required to find the

answers cannot be readily retrieved because

the information needs to be extracted from

multiple data sources and the extraction

process is time-consuming. It may take several

hours or days for a professional GIS person to

collect the needed information, update the

existing spatial database, and perform the

spatial analysis to find out the best evacuation

route.

Therefore, a semantic-empowered analysis

tool, which is able to automatically integrate

and analyse a broad variety of spatial data

contents, is needed to provide more utilities for

response coordination. The objective of this

research is to propose such a system for the

public and emergency responders, who usually

do not have many GIS skills, to update the

existing databases and automatically search

for the needed information. We propose such

an interoperable online VGI system based on

state-of-the-art Geospatial Semantic Web

technologies to overcome the limitations of

the currently implemented SDI and VGI

systems.

2. A framework of an interoperable onlineVGI system

The framework of an interoperable online VGI

system based on Geospatial Semantic Web

technologies for disaster response is proposed

as shown in Figure 1. The overall goal of this

framework is to overcome the semantic

heterogeneity limitation of the currently

implemented SDI and VGI systems and make

use of unstructured citizen-generated contents

for decision-making in disaster response.

The tools in the proposed VGI system should

be available and useful to many stakeholders,

especially local authorities, disaster respon-

ders, and local residents.

The framework is based on a distributed

local-responsible web service architecture.

Heterogeneous local geospatial databases

such as shapefiles and geodatabases in

ArcGIS, remotely sensed satellite images in

GeoTIFF format, and Geomedia databases are

published using OGC web services such as

Web Feature Services (WFSs), Web Map

Services (WMSs), andWeb Coverage Services

Journal of Spatial Science 3259

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

(WCSs). Real-time observation and spatial

data mined from social media network

(e.g., Twitter, Facebook, Chat, Skype, blogs,

and emails) can be used for adding, deleting,

and updating spatial feature data such as

accessibility to roads and damaged areas,

availability of relief and services, and numbers

and locations of injured people. Ontology,

which formally represents knowledge as a set

of concepts and the relationships among those

concepts within a domain and supports

reasoning about concepts, is used to add

computer processable meaning (semantics) to

the online VGI system over the World Wide

Web. The OGC web services are connected to

local ontologies through local data source

adapters. Heterogeneous local ontologies are

integrated into an ontology server for web

service discovery and integration. Users such

as disaster responders, authorities, and local

residents, who usually do not have much GIS

knowledge or skills, can update the existing

heterogeneous GIS databases from a variety of

disparate sources through user-friendly

graphic interfaces based on their real-time

observations or other knowledge, such as

mining information from social media network

(e.g., Twitter, Facebook, Chat, Skype, blogs,

and emails).

The major advantage of the proposed

framework is to allow emergency responders

or local residents to remotely update (add,

delete, change) semantically heterogeneous

GIS databases, and automatically query and

integrate data from a variety of different

sources based on data content. In such an

infrastructure each local ontology server offers

a lookup for local geospatial concepts within

its spatial scope. The local ontology server

should be maintained by the local community

running the service. Thus, the stored ontology

can be accurate and always the most updated.

Instead of introducing all of the technologies

applied in the framework, in the following

sections we only introduce the main advanced

technologies applied in the framework such as

Figure 1. A framework of the proposed interoperable online VGI system.

C. Zhang et al.4260

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

remotely updating spatial data, matching

geospatial features to the predefined geospatial

ontologies, and heterogeneous geospatial

ontology integration.

Remotely updating spatial data

To allow users such as disaster responders,

authorities, and local residents to remotely

update the outdated existing spatial data

distributed in different sources in real time,

the proposed frameworkmakes full use of OGC

Web Feature Services (WFSs). The WFSs

manipulate spatial data at feature level based on

OGC’s simple features (e.g. points, lines, and

polygons) (Zhang et al. 2003; Peng & Zhang

2004; Zhang & Li 2005). They are written in

XML (Extensible Markup Language) and use

an open-source standard GML (Geography

Markup Language) to represent features. Data

in GML are stored in text format, which is a

vendor-neutral universal format. BecauseGML

is not locked into a proprietary binary format, it

is easy to integrate GML data into other data

across a variety of platforms and devices. The

proprietary systems’ support of GML inWFS is

through DataStores. The DataStores can

transform a proprietary data format such as

ESRI’s Shapefiles into the GML feature

representation.

To support remotely updating spatial data

in the DataStores, five operations are defined

in the OGC WFS: GetCapabilities, Describe-

FeatureType, GetFeature, Transaction, and

LockFeature. GetCapabilities describes the

capabilities of a WFS server, such as which

feature types it can serve and what operations

are supported on each feature type; Describe-

FeatureType informs the structure of any

feature type upon a request; GetFeature

retrieves feature instances; LockFeature pro-

cesses a lock request on one or more instances

of a feature type for the duration of a

transaction; Transaction provides transaction

requests for such operations on features as

create, update, and delete. The capability of

creating, deleting, and updating features over

theWeb in real time allows users to manipulate

geospatial data remotely at feature level, which

may provide the most updated data for

conducting further spatial analysis, modelling,

and other operations. For example, local

residents can instantly edit a road feature to

flood status in its remote databases by using the

update capability of a WFS over the Web and

the first disaster responders can add a new

flood affected location to their remote data-

bases by using the create capability of a WFS

over the Web.

Although OGC WFSs provide capabilities

of remotely updating geospatial data, the

currently implemented OGC WFSs have

limitations: They only emphasise technical

data interoperability via standard interfaces

and cannot resolve semantic heterogeneity

problems in spatial data sharing (Zhang et al.

2010c). Thus to achieve the goal of remotely

updating spatial data at semantic level over the

Internet, in the proposed framework we extend

the existing OGC WFSs with geospatial

semantic web technologies. We map OGC

WFS descriptions to OWL ontologies to

provide a semantically based view of the web

services, which spans from abstract descrip-

tions of capabilities of the services to the actual

feature data contents that exchange with other

services. We focus on specifying semantic

descriptions for three operations: GetCapabil-

ities, DescribeFeatureType, and GetFeature.

We specify semantics, the meaning of

languages (e.g., words, phrases), for each

FeatureType and FeatureProperty in defined

GetCapabilities, DescribeFeatureType, and

GetFeature operations. The semantics of

FeatureTypes and FeatureProperties in WFSs

are mapped to disjunctions or conjunctions of

(possibly negated) concepts in OWL ontolo-

gies. Because the definitions of the semantic

concepts in the enhanced semantic WFSs are

available at referenced uniform resource

identifier (URI) ontology databases on the

Web, the WFS providers and the clients have a

means of sharing terms. The results are that, by

taking an OWL description of a WFS, a WFS

Journal of Spatial Science 5261

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

client can distinguish and properly interpret all

FeatureTypes and FeatureProperties in Get-

Capabilities, DescribeFeatureType, and Get-

Feature operations.

Matching geospatial features to thepredefined geospatial ontologies

To realise updating and discovery of geospa-

tial feature data at the semantic level, one

important challenge is to match geospatial

features to the predefined geospatial ontologies

in order to connect local ontology servers to

the corresponding geospatial web servers.

In our previous studies we manually matched

geospatial features to the predefined geospatial

ontologies (Zhao et al. 2008; Zhang et al.

2010a, b, c). In this research, we apply the

supervised machine learning techniques to

automate matching of the geospatial features

to the predefined geospatial ontologies. The

idea is to develop a matching tool with

machine learning components to identify

potential matching candidates. Using the

developed tool users should be able to make

final adjustments through an interface to

eliminate erroneous or unnecessary matching

candidates and output the correct ones. Note

that this matching step has been done offline to

set up the local ontology servers, and manual

improvement can always be made if the

automatic matching has failed to generate the

satisfactory results.

The basic steps of the supervised machine

learning method are provided as follows:

1. Define domain ontology concepts using

knowledge in literature.

2. Create a training data set by labelling

the selected geospatial schemas with

the correct ontology concepts.

3. Determine the feature representation of

the learned functions which are used to

map ontology concepts to geospatial

schemas. The features may include

geometry type, property name, values,

and data-types of the geospatial sche-

mas. Other features such as bounding

box, density, and relations of geometries

may also be considered.

4. Determine the structure of the learned

functions and the corresponding learn-

ing algorithms. We apply decision tree

learning (Berikov and Litvinenko 2003)

for this step since it is able to handle

categorical data (e.g., geometries of

points, lines, polygons).

5. Apply learning algorithms on the

training set to obtain parameters,

which will be adjusted to maximise

performance on the test data sets.

For example, let’s consider a scenario

where a hurricane has struck the town of New

Haven in Connecticut. To take immediate

rescue action, the emergency responders need

to find evacuation routes. Suppose we have a

domain ontology for a transit system and we

also have a collection of web features of transit

data such as streets, bus routes, bus stops,

patterns, time-points, trips, schedules, and

facilities. These web features are derived from

several regions with some differences and

similarities. As a set of training data, we need to

select a subset of web features and manually

map them to ontological concepts. After that,

we need to run a machine learning algorithm to

derive a matching function f using the training

data. Then we will apply the matching function

to the rest of the web features. To generate f, we

need to decide how to match ontology classes,

how to match ontology properties, and how

to generate ontology individuals. For example,

to decide how to match web feature types to

ontology classes, we can use feature geometry

(type and density), feature type name, feature

property list, and property values as indepen-

dent variables. The ontology class to which a

web feature should be matched is considered as

a dependent variable.

We use a decision tree learning algorithm

to classify web features into each ontology

class. Figure 2 shows an example of a decision

tree for classifying web features into the

C. Zhang et al.6262

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

ontology classes of Street, Route, Stop, and

Facility. In this example, we use two indepen-

dent variables – geometry type and density.

We start with four web features for each class at

the top node of the decision tree. Through the

learning algorithm, we find out that the sub-tree

of point geometry includes all of the web

features for Stop and Facility classes while the

sub-tree of line geometry includes all of the

features of Street and Route classes. There is no

feature for the sub-tree of polygon as expected.

Since there are many more bus stops than

facilities, the predictive attributes of geometry

density are able to separate features of Stop

from those of Facility. Similarly, we can

separate features of Street from those of

Route because Street class contains features

with a higher density than Route class does.

We can also use a related strategy to determine

how to match web feature properties with

ontology properties. For web feature proper-

ties, we consider web feature type, data-type,

property value, and property name.

The advantage of the machine learning

approach is that it can identify the hidden

semantic links between the existing ontology

concepts and the geospatial schemas of the

legacy data. Though itmay not resolve all of the

semantic ambiguities, supervised machine

learning can potentially extract the maximum

amount of information from the training data,

which has been annotated with the help of

domain experts. We can also use relevance

feedback, which grades the relevance of the

matching results as determined by experts, to

improve the precision of the learning algorithm.

Heterogeneous geospatial ontologyintegration

In our prior work, we have used a single-domain

ontology to ensure semantic interoperability

(Zhang et al. 2007; Zhao et al. 2008). However,

Figure 2. An example of a decision tree for classifying web features into several ontology classes.The dependent variable is the number of web features in each ontology class including Route, Stop, Street, andFacility. The independent variables are geometry type and density.

Journal of Spatial Science 7263

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

it is impractical to develop a global ontology for

all disaster management applications that

support the tasks envisaged by a distributed

environment like the Geospatial Semantic Web.

Toovercome this problem, in this studywe adopt

a distributed local-responsibility service infra-

structure, which is an environment withmultiple

independent systems where each system has its

own local ontology. The advantage of this

approach is that the local ontology can be

designed to suit the geospatial data in each

system. However, this approach brings the

possibility of conflicts and mismatches among

different local ontologies. Thus, it is necessary

for this study to develop algorithms to integrate

the heterogeneous local ontology.

Although in the IT literature algorithms and

tools have been proposed to resolve the problem

of heterogeneous ontology integration (e.g.,

Castano et al. 2006), they are not developed for

dealing with spatial data. Hess et al. (2006)

recently proposed the G-Match algorithm for

geographic ontology integration. To match and

integrate two different geographic ontologies,

the G-algorithm measures the overall similarity

Sim(C1, C2) of their concepts by combining

similarity measures of concept names, attri-

butes, taxonomies, and conventional as well as

topological relationships in a weighted sum.

The G-algorithm rejects a matching of two

ontology classes if the overall similarity

measure is below a predetermined threshold.

One problem with this approach is that classes

or properties with very similar names could be

considered equivalent even though they are not

in reality or they are not compatible in structure

so that translation is impossible. Also, this

approach does not consider the range types of

relations in computing their similarity.

In this study, we adopt a partition

refinement algorithm (Zhang et al. 2010a) for

integrating heterogeneous ontology. The fol-

lowing is a basic version of the partition

refinement algorithm:

Input: A set of classes, datatype proper-

ties, and object properties.

Output: Partitions of classes PC, object

properties PR, and datatype properties PA,

where each partition in PC contains

equivalent classes and each partition in

PA(PR) contains equivalent datatype

(object) properties.

Step 1: initialisation:

a. Divide the set of datatype properties

into PA where each partition in PAcontains datatype properties with

similar names and the same range

type.

b. Put the set of object properties into PR

with one partition for spatial proper-

ties and one partition for the non-

spatial properties. Note that at initi-

alisation it is not necessary to separate

spatial properties from the rest, but

this can help increase the precision of

the algorithm.

c. Divide the set of classes into PC

such that C1, C2 are in a partition in

PC if and only if for each partition p

in PA, jp > attr(C1)j ¼ jp > attr

(C2)j, where jp > attr(Ci)j is the

size of the set p > attr(Ci).

Step 2: partition refinement:

a. Refine PR such that r1, r2 are in a

partition in PR if and only if for each

partition p in PC, jp > range(r1)j ¼ -

p > range(r2)j.b. Refine PC such that C1, C2 are in a

partition in PC if and only if for each

partition p in PR, jp > rel(C1)j ¼ -

p > rel(C2)j.Step 3: repeat Step 2 until PR and PC

stabilise.

Step 4: refine PC and PR even further based

on the name similarity of the classes and

object properties. If this step results in any

changes, then go back to Step 2.

Figure 3 illustrates the process of finding

equivalent ontology classes and properties

according to the algorithm shown above. First,

we pool all sets of classes, object properties,

and datatype properties from both local and

C. Zhang et al.8264

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

server ontologies. Then we initialise partitions

of classes, object properties, and data type

properties to form equivalence partitions.

After that we first refine the partitions of

classes and object properties based on their

structure. If too many equivalent classes and

object properties are found, we refine the

partitions based on names of classes and object

properties. The refinement process will be

continued until the partitions become stable.

Finally we obtain the final stable partitions,

which will not change and are considered to be

equivalent ontology classes and properties.

The main advantage of the partition

refinement algorithm is that it finds matching

ontology classes and properties based on their

structures. Unlike the G-algorithm, which

heavily relies on a string-based similarity

measure, the partition refinement algorithm

makes full use of the structures of the ontology

being mapped. Thus, it allows translation of

instances among different ontologies. Further,

the partition refinement algorithm can deal

with a recursive structure efficiently while the

G-algorithm cannot.

3. A case study

Connecticut has faced many significant storms

in the recent past. For example the August

Tropical Storm Irene hurricane and the

October Nor’easter hurricane in 2011 caused

widespread damage and left a record numbers

of residents (almost one million residents)

without electricity, heat or reliable supplies of

water for up to 9–12 days. Irene downed

approximately 1–2 percent of the State’s

trees. Damage from both storms was estimated

to be $750 million to $1 billion (McGee et al.

2012).

The significant impact of these storms has

served as a wake-up call to Connecticut. After

the storms, the State Governor created a Two

Storm Panel to review the preparedness,

response and recovery efforts during the

Irene and Nor’easter storms. Based on the

final report of the Two Storm Panel (http://

www.governor.ct.gov/malloy/lib/malloy/two_

storm_panel_final_report.pdf), two of the

important factors preventing a fast response

to the two disasters were (1) the inefficient

collaboration among municipalities, state

Figure 3. The process of finding equivalent ontology classes and properties through partitioning refinement.

Journal of Spatial Science 9265

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

resources, and electric utility service provi-

ders, and (2) the inadequate communication

between labour and management in all

utilities. One important reason for the

inadequate collaboration and ineffective com-

munication is the failure to share updated data/

information. The outage maps provided during

the two storms did not provide local details.

Questions as to which streets were blocked,

what poles and wires were down, where the

power was on and where it was off were

consistent complaints during the two storms.

Although the utility companies provided

information on line breaks based on their

existing grid system, they relied on consumers

to identify exactly where the resulting

problems were. However, there is no system

to allow consumers to share the information

with them. In addition, based on the final

report, some text messaging services offered

by utilities directed individuals to shelters that

had been closed or moved. The text infor-

mation was not updated in a timely fashion.

In this case study, we intend to develop a

prototype based on the aforementioned frame-

work to overcome some of the problems met

during the two storms in Connecticut.

We anticipate that the prototype will provide

useful tools to allow users to create and update

data/information over the Internet, thus they

can provide timely, accurate, and up-to-date

information for better disaster response. The

system should facilitate real-time sharing of

the existing legacy data and the new updated

information among different organisations,

such as town responders, utility companies,

medical departments, and police departments.

Thus they can work together and coordinate

seamlessly in the event of a disaster. The

contents of the implemented prototype should

also be machine-readable and computers

should be able to understand the content of

the databases; thus the required information

for disaster response can be readily retrieved

from multiple data sources.

The implemented prototype can be acces-

sible from the website: http://boyang.cs.uwm.

edu:8080/newHaven/. The prototype uses

several layers of map data for New Haven,

CT, and its surrounding areas. The map layers

include streets, roads, places, and city

boundaries. Both street and road features are

line geometries though street geometries are

mostly shorter than road geometries and street

features have more detailed attributes than

road features. The places are point geometries

for locations, such as schools, bridges, and

railroad stations.

For our experiments, we created only a few

ontology classes to demonstrate the feasibility

of the prototype. Figure 4 illustrates the

ontology classes created for the prototype.

The ontology class Line is the parent ontology

class of Streets and Roads while Point is the

parent ontology class of Places. The ontology

classes are used to overcome the semantic

heterogeneity problem in the original data sets.

For example, the ontology Streets is used to

overcome the semantic heterogeneity problem

in the original two Street data files – one file

called ‘path’ and another file called ‘lane’ for a

narrow road. In the future, additional classes

can be added to the hierarchy without any

change to the implementation.

The Streets ontology class is mapped to the

street feature type in the back-end web

services, while the Roads and Places ontology

classes are mapped to the road and place

feature types, respectively. The properties of

the ontology classes are mapped to the

attributes of the feature types. For example,

the following is the mapping between the

properties of the Road ontology class and the

nh:road feature type in OGC web services:

Figure 4. Ontology classes created for theprototype.

C. Zhang et al.10266

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

Attributes of the road feature type: Proper-

ties of the Roads ontology class

‘OBJECTID’ : ‘id’,‘FULLNAME’ : ‘name’,‘RTTYP’ : ‘type’,‘Shape_Leng’ : ‘length’,‘Blocked’ : ‘isBlocked’,

Similar mappings are defined for values of

some attributes to provide more readable query

results. For example, below is the mapping

between the values of the attribute ‘RTTYP’ in

the original data source and the values of the

property ‘type’ in the Roads ontology class:

Values of the attribute ‘RTTYP’: Values of

the property ‘type’ in the ontology class

‘C’ : ‘County’,‘I’ : ‘Interstate’,‘M’ : ‘Common name’,‘O’ : ‘Other’,‘S’ : ‘State recognised’,‘U’ : ‘US’

By using ontology classes, the

implemented prototype can allow different

stakeholder groups, such as local authorities,

relief specialists, nongovernmental agencies,

disaster responders, and the local residents, to

understand the vocabulary of the original data.

Thus it can deliver more meaningful spatial

information, which would be more useful in

the disaster response.

In addition to the datatype properties

mapped from web feature attributes, each

ontology class includes object properties such

as DWITHIN to allow users to specify spatial

queries in the form of ontology constraints. For

example, the following ontology query

retrieves streets near any high school:

select ?s ?p where?s rdf:type streets.?p rdf:type places.?p nh:category ?c.?c¼ ¼ High School.?s DWITHIN ?p

where the statement select ?s ?p specifies

the variables ?s and ?p as the queried ontology

instances. The constraints ?s rdf:type streets

and ?p rdf:type places are triples of the form

subject predicate object, which specify that

the variable ?s must be assigned instances of

the streets type and ?p must be assigned the

places type. The constraint ?p nh:category ?c

creates a new variable ?c, which is set to be

equal to High School in the constraint ?c ¼ ¼High School. The predicate nh:category is

based on the ontology property category of the

Streets class and nh: is a namespace prefix to

distinguish the property from other types of

predicates. Finally, the constraint ?s DWI-

THIN ?p says that the retrieved streets ?s must

be near the retrieved places ?p. The predicate

DWITHIN corresponds to an ontology prop-

erty that is not mapped to attributes since it

relates two spatial objects by the distance

between their geometries. In fact, there is no

explicit representation of the relationship since

it would be too inefficient to pre-compute and

store the relationships while only a small

fraction of them will be needed.

An additional benefit of the ontology-

based query is that users do not need to always

specify the types of the queried objects. For

example, the following query returns any

spatial objects by the name of Lawrence:

select ?s where?s nh:name ?name.?name , Lawrence*

where , is a predicate for the ‘LIKE’

relationship to match names such as Lawrence

St. The three ontology classes we defined all

have the property ‘name’. The above query is

in fact translated into three separate WFS

requests to retrieve features of the Web feature

types: streets, roads, and places. Each of the

WFS requests includes a filter with a different

name attribute: FENAME for streets, FULL-

NAME for roads, and NAME for places.

The prototype interface is a dynamic

HTML page embedded with about a thousand

lines of JavaScript code that use a geospatial

JavaScript library – OpenLayers to query/

update/render geospatial features through

WFS/WMS services. The Web service back-

Journal of Spatial Science 11267

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

end is an instance of a GeoServer program

running on a Linux workstation that provides

WFS/WMS services for several types of

geospatial features used in our experiments.

As shown in Figure 5, the client interface

includes an interactive map with several

control buttons and a form for ontology-

based queries.

The buttons ‘Sel streets’ and ‘Sel roads’

allow users to select streets or road features on

the map using the mouse. After having

selected some features, the users can click on

the buttons ‘Block streets’ or ‘Block roads’ to

save the selected streets or roads as being

blocked by the hurricane event. If the users

find out that they have made a mistake, they

can change the ‘blocked’ attributes to

‘unblocked’ status by selecting the ‘blocked’

streets or roads and clicking on the ‘Unblock

streets’ or ‘Unblock roads’ buttons.

Other than directly selecting streets or

roads, users can also click on the ‘Sel places’

button to enable a control to select point

features such as schools and bridges from the

map. After some places have been selected, the

interface client will display the attribute values

of the selected places below the map along

with two hyperlinks that allow users to select

the streets and roads near the selected places.

Figure 6 shows two selected places in blue

circles with their attributes being shown below

the map. The results indicate that one of the

places is a railroad station and the other place

is a high school. Users can click on the

dynamically generated hyperlinks ‘nearby

streets’ or ‘nearby roads’ to obtain line features

within a certain vicinity of the selected place.

For example, if a user clicks the hyperlink

‘nearby streets’ for the ‘New Haven Station’,

the resulting map will include the selected

streets, which are stored in a layer of vector

features that are displayed in a different colour

than the underlying street map layer (Figure 7).

Note that the displayed feature attributes

are transformed through ontology classes from

the original attributes stored in the data source.

Figure 5. The client interface of the implemented prototype.

C. Zhang et al.12268

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

Figure 6. Attributes and hyperlinks of the two selected places.

Figure 7. Results of clicking the hyperlink ‘nearby streets’ for the ‘New Haven Station’.

Journal of Spatial Science 13269

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

For example, the attribute for representing a

blocked street is called ‘Blocked’ while the

similar attribute for a blocked road feature is

called ‘Obstructed’. We create a mapping from

these attribute names to an ontology property

so that they are all accessed as the ‘isBlocked’

property.

The client interface also includes a textual

query form to allow users to select features

using the ontology-based query statements.

The textual query can be used to select features

using more fine-grained criteria that are more

difficult to express using the graphic interface.

For example, we may want to find out the

elementary schools and streets near the road

‘Connecticut Tpke’. We can write this query

using a SPARQL-like query statement to select

features of the ‘streets’, ‘roads’, and ‘places’

types so that the ‘category’ of the places is

‘Elementary school’, the road name is ‘Con-

necticut Tpke’ and the streets, the roads, and

the places are near each other (i.e., their

distances are within a certain threshold).

The implemented prototype not only is

able to dynamically update new sources of

information but also can automatically

integrate heterogeneous information at the

semantic level. It can, therefore, be used by

users including disaster responders and local

residents to report information about events

and damage in their neighbourhood and update

the related spatial information over the Web.

Users of the prototype need only to understand

the definitions for general concepts of streets,

roads, and places to be able to query and

update the spatial information, and they don’t

need to understand the underlying definitions

of the spatial data, which may be different

across regions or across data sources. As long

as the definitions of the original spatial data are

properly mapped to the ontology-based defi-

nitions, ontology-based queries can be under-

stood by the prototype.

The implemented prototype is capable of

turning the people with smart phones or

internet access into a big human sensor

network for fast updating of the existing

databases for disaster response. Using a mobile

client, a citizen can report disaster events such

as the failure of a bridge (for example,

Q Bridge) by simply touching a smartphone

screen. As shown in Figure 8, the system can

Figure 8. The affected roads across a failed bridge.

C. Zhang et al.14270

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

automatically highlight the affected streets and

roads that cross a failed bridge.

4. Discussion

The implemented case study in this paper

provides only a client interface for Internet

users. However, the VGI framework proposed

in this study can be implemented as a mobile

VGI system for smartphone users by modify-

ing the input control so it is suitable for touch

screens. The results show that the developed

prototype can allow users to add new spatial

information and update existing spatial infor-

mation over the Internet. Because the proto-

type was developed using ontology semantic

web technologies, it generates structured

information. The generated structured infor-

mation is machine-readable and can be

automatically processed by computers. The

information derived from the system can be

easily shared and used by different appli-

cations. Although the results of the prototype

show some promise towards an interoperable

online VGI system by using Geospatial

Semantic Web technologies for disaster

response, there are many issues such as

multilingual interoperability, fault/error hand-

ling and security issues still needing further

research to implement a fully workable system

for real-world applications. Here we address

four important issues in following paragraphs.

The first issue is performance. In fact, the

performance issue of OGC web services has

been recognised in the geospatial community

for a while and different approaches have been

suggested (e.g Zhao et al. 2008). Although the

prototype only involves data from one town in

Connecticut, New Haven, some layers

contain several thousand geospatial features.

For example, the street layer includes

almost four thousand geospatial features.

More geospatial features will be involved if

we implement the prototype for the whole

Connecticut state. Even for the New Haven

data sets, the system is slow to update the street

layer data. This is because the WFSs need to

transport the text-based GML data through the

network. When GML-coded geospatial data

are transported, all of the markup elements that

describe spatial and non-spatial features,

geometries, and spatial reference systems of

the data are also transported to the recipient.

This is important for data interoperability,

because the GML-coded data could be saved

and used by any other client-side applications

that can read GML data. However, this also

greatly slows down the performance of the

system. Compared with some binary GIS data

formats, the size of GML data files is large.

Large file sizes may hinder the use of GML

files as a means of data transport over the

Internet. Although solutions such as using

compression or sending the GML data to the

client in stages or progressively have been

proposed in the literature (Peng & Zhang

2004), issues still exist for improving the

performance.

In addition, currently the DL (Description

Logic)-based ontology knowledge base is

unlikely to efficiently handle large geographi-

cal knowledge bases. While DL is well suited

for representation of structured or semi-

structured attribute information, things

become complicated if a ‘non-abstract’ space

is considered. For example, for knowledge

bases it would be necessary to be able to

compute spatial relationships from the geome-

tries of objects. However, the DL-based

system is unlikely to efficiently compute

such spatial relationships. A huge ABox

containing all of the topological relationships

of the spatial data will be quite prohibitive for

querying or updating the spatial data without

using appropriate index structures or optimis-

ation techniques. There are many index

structures or optimisation techniques in the

computer literature. Which one is the best

index structure or optimisation technique for

handling spatial data? This is an important

issue that needs further study in the future.

The second issue is the data quality issue.

Compared with the traditional authoritative

geographic information, VGI lacks map

Journal of Spatial Science 15271

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

specifications, mechanisms or procedures to

assure quality. The main mechanism to assure

the quality of VGI is called Linus’s Law,

which states that ‘given enough eyes, all bugs

are shallow’ (Raymond 1999). If one volunteer

gives wrong information, others will be

expected to edit and correct the wrong

information. The success of this mechanism

depends on others who check the contribution.

Although the idea that ‘given enough eyeballs,

all bugs are shallow’ (Raymond 1999) was

proposed a decade ago, the reliability and

credibility of VGI data still face debates.

Because a VGI system does not use a strict

procedure to take the publication data quality

standards into account as done by traditional

authoritative sources, data quality is one of the

major issues faced by VGI. Several studies

have recognised the concerns in the literature

(e.g. Elwood et al. 2012; Ho & Rajabifard

2010; Flanagin & Metzger 2008) and several

approaches have been adopted to overcome the

limitations of the VGI data quality (e.g. Bishr

& Mantelas 2008; Zielstra et al. 2013;

Senaratne et al. 2013). In the implemented

prototype we developed an algorithm to first

evaluate the quality of the user’s input

information in the context of the existing

data before the system accepts the users’ input

information. In the future research, we may use

more elaborate methods for handling error

data, such as the quality filter method adopted

for Wikipedia. We may also use some

mechanisms to allow the related data publish-

ers under various departments to check the

volunteers’ input data and make sure that the

quality of the data conforms to the high

standards set forth by the original data

publishers. In addition, we also may develop

algorithms to automatically monitor and

correct errors. For example, address spelling

errors may be automatically recognised and

fixed by using approaches similar to those

employed by the geocoding functions in GIS

systems.

The third issue is how to create high-

quality ontologies. Currently, ontologies are

typically built by a small number of people, in

most cases researchers, using the existing

ontology tools and editors such as Protege.

These ontology tools and editors supporting

ontological modelling have improved over the

last few years and many functions are

available now, such as ontology consistency

checking, import of existing ontologies, and

visualisation of ontologies. Even so, ontology

building manually has proven to be a very

difficult and error-prone task and is becoming

the bottleneck of knowledge-acquiring pro-

cesses. For instance, it is unrealistic for non-

domain-experts to use these tools to build

high-quality ontologies. Although transform-

ation algorithms have been proposed by Zhang

et al. (2008) to automatically transform the

existing UML to OWL so as to avoid errors

and provide a cost-efficient method for the

development of high-quality ontologies, there

are many issues yet to be resolved due to the

differences between UML and OWL.

The fourth issue is the credibility of VGI.

VGI has brought many benefits to society.

However, it has also brought some challenges.

One of them is credibility. Traditional data

have been created by a small trusted group,

including mainly professionals and academics.

The creation of VGI is completely open.

Anyone can create VGI as long as he/she

can access the Internet or a mobile phone.

The contents published without control and

verification imply that users may post wrong

data or information. For example, a terrorist

may post false information about accessibility

of some important roads to create panic. How

to trust the authenticity and quality of the data

or information that has been published by

users? How to check the validity of contents

before confidently using the data/information?

How to provide a trusted VGI? Is the

traditional identity-based security possible to

provide a VGI user confidence about the

contents without personally knowing who

created the contents? How to warn users

against fraudulent, malicious, or incorrect

information? These are challenging questions

C. Zhang et al.16272

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

faced by a VGI system. Although techniques

have been proposed in the literature to create

trusted VGI (Havlik et al. 2013), they are still

in the initial research stage and further studies

are needed to answer the aforementioned

questions. In general, multiple sources of VGI

that ensure vast information availability also

make the verification of the credibility of

information very complex.

5. Conclusion

Considering the fact that Internet users are fast

to report disaster information, in this paper we

propose a framework of an interoperable

online VGI system based on Geospatial

Semantic Web technologies for timely updat-

ing, aggregating, disseminating, and sharing

information for disaster response. The import-

ant advantage of the proposed system is that it

can deliver more meaningful spatial infor-

mation to different stakeholder groups such as

local authorities, disaster responders, and local

residents by using ontology. People equipped

with smartphones or Internet access can use

the system to rapidly update the existing

heterogeneous databases in a shorter period of

time. The updated information can direct first

disaster responders to find the affected

locations (e.g., blocked roads, injured or

missing residents, damaged buildings,

or closed hospitals) and available relief or

services. In general, the proposed system

allows data to be updated not only by

professionals and official agencies, but also

by volunteers, who may not have much GIS

skill or knowledge. Although some geographic

information provided by such a system may

not have the same high quality as that from

traditional authoritative sources, it still will be

helpful to assist recovery workers because it

allows new information, which may reflect the

real disaster situation, to be incorporated and

distributed in nearly real time.

Because the system ensures the spatial

information is interoperable at the semantic

level, it can be reused across a wide variety of

stakeholders involved in relief efforts.

Although there are still many issues waiting

for further studies to fully implement such a

workable system, the implemented prototype

shows great promise for updating and integrat-

ing the existing legacy spatial databases over

the Internet.

References

Bahree, M. (2008) Citizen voices. Forbes MagazineWWW document. Available from: http://www.forbes.com/forbes/2008/1208/083.html(Accessed 1 October 2013).

Berikov, V., & Litvinenko, A. (2003) Methods forstatistical data analysis with decision trees,WWW document. Available from: http://math.nsc.ru/AP/datamine/eng/context.pdf (Accessed1 October 2013).

Bishr, M., & Mantelas, L. (2008) A trust andreputation model for filtering and classifyingknowledge about urban growth. GeoJournal,vol. 72, no. 3–4, pp. 229–237.

Bishr, Y. (1998) Overcoming the semantic and otherbarriers to GIS interoperability. InternationalJournal of Geographical Information Science,vol. 12, no. 4, pp. 299–314.

Castano, S., Ferrara, A., & Montanelli, S. (2006)Matching ontologies in open networkedsystems: techniques and applications, DataSemantics V, Lecture Notes in ComputerScience, Springer, Berlin, pp. 25–63.

Cova, T.J. (1999) GIS in emergency management.In: Longley, P.A., Goodchild, M.F., Maguire,D.J., & Rhind, D.W., eds. GeographicalInformation Systems: Principles, Techniques,Applications, and Management, John Wiley &Sons, New York, NY, pp. 845–858.

Craglia, M., Goodchild, M.F., Annoni, A., Camera,G., Gould, M., Kuhn, W., Mark, D., Masser, I.,Maguire, D., Liang, S., & Parsons, E. (2008)Next-generation digital earth: a position paperfrom the vespucci initiative for the advance-ment of geographic information science.International Journal of Spatial Data Infra-structures Research, vol. 3, pp. 146–167.

De Longueville, B., Smith, R.S., & Luraschi, G.(2009) “OMG, from here, I can see the flames!”:a use case of mining location based socialnetworks to acquire spatio-temporal data onforest fires, Proceedings of the 2009 Inter-national Workshop on Location Based SocialNetworks Seattle, WA, November 2009, ACM,7380.

Journal of Spatial Science 17273

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

De Longueville, B., Luraschi, G., Smits, P., Peedell,S., & De Groeve, T. (2010a) Citizens as sensorsfor natural hazards: a VGI integration work-flow. Geomatica, vol. 64, no. 1, pp. 41–59.

De Longueville, B., Annoni, A., Schade, S.,Ostlaender, N., &Whitmore, C. (2010b) Digitalearth’s nervous system for crisis events: real-time sensor web enablement of volunteeredgeographic information. International Journalof Digital Earth, vol. 3, no. 3, pp. 242–259.

De Rubeis, V., Sbarra, P., & Tosi, P. (2009) Webbased macroseismic survey: fast informationexchange and elaboration of seismic intensityeffects in Italy, Proceeding of the 6thInternational ISCRAM Conference Gothenburg,Sweden,May 2009.WWWdocument.Availablefrom: http://www.iscram.org/ISCRAM2009/papers/ (Accessed 1 October 2013).

Donkervoort, S., Dolan, S.M., Beckwith, M.,Northrup, T.P., & Sozer, A. (2008) Enhancingaccurate data collection in mass fatality kinshipidentifications: lessons learned from HurricaneKatrina. Forensic Science International: Gen-etics, vol. 2, no. 4, pp. 354–362.

Elwood, S., Goodchild, M., & Sui, D. (2012)Researching volunteered geographic infor-mation: spatial data, geographic research, andnew social practice. Annals of the Association ofAmerican Geographers, vol. 102, no. 3,pp. 571–590.

Flanagin, A.J., & Metzger, M.J. (2008) Thecredibility of volunteered geographic infor-mation. GeoJournal, vol. 72, no. 3–4,pp. 137–148.

Gao, H., Wang, X., Barbier, G., & Liu, H. (2011)Promoting coordination for disaster relief – fromcrowdsourcing to coordination, ProceedingSBP’11 Proceedings of the 4th InternationalConference on Social Computing, Behavioral-CulturalModeling and Prediction, Lecture Notesin Computer Science (LNCS) 6589 Springer,Berlin, pp. 197–204.

Goodchild, M.F. (2007a) Citizens as sensors: theworld of volunteered geography. GeoJournal,vol. 69, no. 4, pp. 211–221.

Goodchild, M.F. (2007b) Citizens as voluntarysensors: spatial data infrastructure in the worldof Web 2.0. International Journal of SpatialData Infrastructures Research, vol. 2,pp. 24–32.

Goodchild, M.F., & Glennon, J.A. (2010) Crowd-sourcing geographic information for disasterresponse: a research frontier. InternationalJournal of Digital Earth, vol. 3, no. 3,pp. 231–241.

Havlik, D., Egly, M., Huber, H., Kutschera, P.,Falgenhauer, M., & Cizek, M. (2013) Robustand trusted crowd-sourcing and crowd-taskingin the future internet. Environmental SoftwareSystems. Fostering Information Sharing IFIPAdvances in Information and CommunicationTechnology, vol. 413, pp. 164–176.

Hess, G.N., Iochpe, C., & Castano, S. (2006) Analgorithm and implementation for geoontologiesintegration, WWW document. Available from:http://www.geoinfo.info/geoinfo2006/papers/p46.pdf (Accessed 1 October 2013).

Ho, S., & Rajabifard, A. (2010) Learning from thecrowd: the role of volunteered geographicinformation in realising a spatially enabledsociety, WWW document. Available from:http://www.gsdi.org/gsdiconf/gsdi12/papers/45.pdf (Accessed 1 October 2013).

Hughes, A.L., & Palen, L. (2009) Twitter adoptionand use in mass convergence and emergencyevents, Proceedings of the 6th InternationalISCRAM Conference Gothenburg, Sweden,May 2009. WWW document. Available from:http://www.iscram.org/ISCRAM2009/papers/(Accessed 1 October 2013).

Jones, C.E., Mount, N.J., & Weber, P. (2012) Therise of the GIS volunteer. Transactions in GIS,vol. 16, no. 4, pp. 431–434.

Kawasaki, A., Berman, M.L., & Guan, W. (2013)The growing role of web-based geospatialtechnology in disaster response and support.Disasters, vol. 37, no. 2, pp. 201–221.

Laituri, M., & Kodrich, K. (2008) On line disasterresponse community: people as sensors of highmagnitude disasters using internet GIS. Sensors,vol. 8, no. 5, pp. 3037–3055.

McGee, J., Carozza, P., Edelstein, T., Hoffman, L.,Jackson, S., McGrath, R., & Osten, C. (2012)Report of the two storm panel, WWWdocument. Available from: http://www.governor.ct.gov/malloy/lib/malloy/two_storm_panel_final_report.pdf (Accessed 8 July 2014).

Morrow, N., Mock, N., Papendieck, A., & Kocmich,N. (2011) Independent evaluation of theUshahidi Haiti project, Tech. Rep. The UHPIndependent Evaluation Team. WWW docu-ment. Available from: http://ggs684.pbworks.com/w/file/fetch/60819963/1282.pdf (Accessed1 October 2013).

Palen, L., Vieweg, S., Liu, S.B., & Hughes, A.L.(2009) Crisis in a networked world: features ofcomputer-mediated communication in the April16, 2007, Virginia Tech event. Social ScienceComputer Review, vol. 27, no. 4, pp. 467–480.

Peng, Z.-R., & Zhang, C. (2004) The roles ofgeography markup language (GML), scalable

C. Zhang et al.18274

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016

vector graphics (SVG), and Web feature service(WFS) specifications in the development ofinternet geographic information systems (GIS).Journal of Geographical Systems, vol. 6, no. 2,pp. 95–116.

Poser, K., & Dransch, D. (2010) Volunteeredgeographic information for disaster manage-ment with application to rapid flood damageestimation. Geomatica, vol. 64, no. 1,pp. 89–98.

Rajabifard, A., Binns, A., & Williamson, I. (2006)Virtual Australia: developing an enabling plat-form to improve opportunities in the spatialinformation industry. Journal of SpatialScience, vol. 51, no. 1, pp. 63–78.

Raymond, E. (1999) The Cathedral and the Bazaar:Musings on Linux and Open Source by anAccidental Revolutionary, O’Reilly Media,Sebastapol, CA.

Senaratne, H., Broring, A., & Schreck, T. (2013)Using reverse viewshed analysis to assess thelocation correctness of visually generated VGI.Transactions in GIS, vol. 17, no. 3,pp. 369–386.

Stefanidis, A., Crooks, A., & Radzikowski, J.(2013) Harvesting ambient geospatial infor-mation from social media feeds. GeoJournal,vol. 78, no. 2, pp. 319–338.

U.S. House Select Bipartisan Committee toInvestigate the Preparation for and Responseto Hurricanes Katrina and Rita (2006) Washing-ton, DC: February 15, A Failure of Initiative:The Final Report of the Select BipartisanCommittee to Investigate the Preparation forand Response to Hurricanes Katrina and Rita.

White House (2006) The Federal Response toHurricane Katrina: Lessons Learned, ExecutiveOffice of the President, WWW document.Available from: http://www.whitehouse.gov/reports/Katrina-lessons-learned.pdf (Accessed1 October 2013) Washington, DC.

Zerger, A., & Smith, D.I. (2003) Impediments tousing GIS for real-time disaster decisionsupport. Computers, Environment and UrbanSystems, vol. 27, no. 2, pp. 123–141.

Zhang, C., & Li, W. (2005) The roles of web featureand web map services in real-time geospatialdata sharing for time-critical applications.

Cartography and Geographic InformationScience, vol. 32, no. 4, pp. 269–283.

Zhang, C., Li, W., Day, Z.-R., & Peng, M. (2003)GML-based interoperable geographical data-bases. Cartography, vol. 32, no. 2, pp. 1–16.

Zhang, C., Li, W., & Zhao, T. (2007) Geospatialdata sharing based on geospatial semantic webtechnologies. Journal of Spatial Science,vol. 52, no. 2, pp. 35–49.

Zhang, C., Peng, Z.-R., Zhao, T., & Li, W. (2008)Transformation of transportation data modelsfrom unified modeling language to webontology language. Transportation ResearchRecord: Journal of the Transportation ResearchBoard, vol. 2064, no. 1, pp. 81–89.

Zhang, C., Zhao, T., & Li, W. (2010a) Theframework of a geospatial semantic web-basedspatial decision support system for digital earth.International Journal of Digital Earth, vol. 3,no. 2, pp. 111–134.

Zhang, C., Zhao, T., & Li, W. (2010b) Automaticsearch of geospatial features for disaster andemergency management. International Journalof Applied Earth Observation and Geoinforma-tion, vol. 12, no. 6, pp. 409–418.

Zhang, C., Zhao, T., Li, W., & Osleeb, J. (2010c)Towards logic-based geospatial feature discov-ery and integration using web feature serviceand geospatial semantic web. InternationalJournal of Geographical Information Science,vol. 24, no. 6, pp. 903–923.

Zhao, T., Zhang, C., Wei, M., & Peng, Z.-R. (2008)Ontology-based geospatial data query andintegration, Lecture Notes in Computer ScienceLNCS: Geographic Information Science, 5266Springer, Berlin, pp. 370–392.

Zielstra, D., Hochmair, H.H., & Neis, P. (2013)Assessing the effect of data imports on thecompleteness of OpenStreetMap – a UnitedStates case study. Transactions in GIS, vol. 17,no. 3, pp. 315–334.

Zook, M., Graham, M., Shelton, T., & Gorman, S.(2010) Volunteered geographic information andcrowdsourcing disaster relief: a case study ofthe Haitian earthquake. World Medical &Health Policy, vol. 2, no. 2, pp. 7–33. doi:10.2202/1948-4682.1069.

Journal of Spatial Science 19275

Dow

nloa

ded

by [

Uni

vers

ity o

f C

onne

ctic

ut]

at 1

1:44

28

Janu

ary

2016