Content uploaded by Jerry Asaana
Author content
All content in this area was uploaded by Jerry Asaana on Jun 09, 2020
Content may be subject to copyright.
SPATIAL ANALYSIS AND MAPPING
OF CHOLERA CAUSING FACTORS
IN KUMASI, GHANA.
JERRY ASAANA ANAMZUI-YA
March, 2012
SUPERVISORS:
First Supervisor: Prof. Dr. Ir. A. Stein
Second Supervisor: Dr. N. A. S. Hamm
SPATIAL ANALYSIS AND MAPPING
OF CHOLERA CAUSING FACTORS
IN KUMASI, GHANA.
JERRY ASAANA ANAMZUI-YA
Enschede, The Netherlands, March, 2012
Thesis submitted to the Faculty of Geo-information Science and Earth
Observation of the University of Twente in partial fulfilment of the requirements
for the degree of Master of Science in Geo-information Science and Earth
Observation.
Specialization: Geoinformatics
SUPERVISORS:
First Supervisor: Prof. Dr. Ir. A. Stein
Second Supervisor: Dr. N. A. S. Hamm
THESIS ASSESSMENT BOARD:
Dr. Ir. R. A. de By (chair)
Dr. A. Dilo (External Examiner)
Disclaimer
This document describes work undertaken as part of a programme of study at the Faculty of Geo-information Science and Earth
Observation of the University of Twente. All views and opinions expressed therein remain the sole responsibility of the author, and
do not necessarily represent those of the Faculty.
ABSTRACT
Recent Global health reports show a continual vulnerability of large populations to infectious
diseases such as cholera in relation to our environments. Of particular concern is the spatial dis-
tribution of cholera incidences and its associated environmental risks factors. This can be used by
health officials and policy makers to make appropriate planning and resource allocation. Despite
the availability of remotely sensed data in various formats for mapping, few studies have utilized
the technology for mapping the environmental niche of V. cholerae. A key environmental factor
which predisposes persons to cholera infection is sanitation. Two identified important measures
of sanitation in an urban city, Kumasi are proximity to refuse dumps and water reservoirs within
a community. To this end a RapidEye image is exploited to map the potential cholera reservoirs
and compare with digitized reservoirs from a topographic map in spatial analysis. A spatial con-
ditional autoregressive (CAR) modeling was carried out to determine the spatial dependency of
cholera prevalence on (1) proximity to refuse dumps and digitized reservoirs and (2) proximity
to refuse dumps and classified reservoirs from RapidEye image. The results showed that there is
an inverse spatial relationship between cholera prevalence and proximity to both refuse dumps(p-
Value <0.0001) and classified reservoirs(p-Value <0.001). The model with digitized reservoirs
was found to be insignificant at α=0.05 (i.e, p-Value =0.07). The results of the spatial models
suggests a better fit with the classified reservoirs. Moran’s I analysis showed a significant spatial
association of cholera risk with neighboring communities. Probability and risks maps were also
generated to characterize the spatial patterns of cholera prevalence in the Kumasi Metropolis.
Keywords
disease mapping, spatial statistics, remote sensing, cholera, conditional autoregressive modeling, envi-
ronmental factors, GIS
i
ACKNOWLEDGEMENTS
I thank The Good Lord for His goodness, abundant grace and wonderful works in my life.
I acknowledge the Netherlands organization for international cooperation in higher education
(nuffic) for providing the funds for my study. Thanks to my employers Bolgatanga Polytechnic
for granting me study leave.
I am grateful to my supervisors Professor Alfred Stein and Dr. Nicholas Hamn for their con-
structive thoughts, critisms and remarks during my research work. My sincere thanks to Dr.
Frank B. Osei for his advice and allowing me use his data.
Special thanks to my family for their support, prayers and encouragements; Obviously, i would
not have come this far without them. Thank you so much Dad(who unfortunately passed away
during my study). You have been a pillar and an inspiration in every aspect of my life. May the
good Lord lay you to rest till we meet again. Thank you mum, uncle Eddy, Alhaji Osman, Rhoda,
Ruth, Judith, Frank, Jonas and Joshua; I love you all.
Finally, i salute all my friends here at ITC, especially my classmates GFM 2 2010-12, and all
the Ghanaian students who in one way or the other have touched my life, God bless you all.
ii
TABLE OF CONTENTS
Abstract i
Acknowledgements ii
1 Introduction 1
1.1 Motivation and problem statement . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 ResearchIdentification ............................... 2
1.2.1 Generalobjective.............................. 2
1.2.2 Specificobjectives.............................. 2
1.2.3 Researchquestions ............................. 2
1.3 Outlineofthesis .................................. 2
2 Literature review 5
2.1 Cholera ....................................... 5
2.1.1 Cholera Transmission . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.2 Factors influencing spread . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.3 The burden of cholera in Ghana . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Spatialepidemiology ................................ 7
2.2.1 Framework for spatial analysis . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 Statistical methods for spatial epidemiology . . . . . . . . . . . . . . . . 10
2.2.3 Spatial epidemiology of cholera . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Areal data and spatial autocorrelation . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Spatial weights and neighborhoods . . . . . . . . . . . . . . . . . . . . 12
2.3.2 Spatial autocorrelation test . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.3 Global Moran’s I.............................. 13
2.3.4 Local indicators of spatial autocorrelation(LISA) . . . . . . . . . . . . . 13
2.4 Modelingarealdata................................. 14
2.4.1 Spatial regression models . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.2 Conditional autoregressive (CAR) models . . . . . . . . . . . . . . . . . 14
2.4.3 Simultaneous autoregressive (SAR) models . . . . . . . . . . . . . . . . 15
3 Study area, datasets and data preparation 17
3.1 Thestudyarea ................................... 17
3.2 Datasets....................................... 18
3.2.1 Choleradata ................................ 18
3.2.2 Refusedumpsdata ............................. 18
3.2.3 Riversdata ................................. 20
3.2.4 RapidEye image data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Datapreparation .................................. 21
3.4 Software....................................... 21
iii
4 Research methodology 23
4.1 Introduction..................................... 23
4.2 Imageanalysis.................................... 24
4.2.1 Image pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2.2 Image classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Data Integration and Visualization . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3.1 Spatialfactormaps ............................. 24
4.3.2 Mapping and geovisualization . . . . . . . . . . . . . . . . . . . . . . . 27
4.4 Spatialanalysis ................................... 28
4.4.1 Autocorrelation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.4.2 Spatial autoregressive modeling . . . . . . . . . . . . . . . . . . . . . . 29
5 Results and analysis 31
5.1 Classification .................................... 31
5.2 Mapping and geovisualization . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.3 Spatialanalysis ................................... 35
5.3.1 Autocorrelation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3.2 Spatial autoregressive modeling . . . . . . . . . . . . . . . . . . . . . . 37
6 Discussion 41
6.1 Imageclassification ................................. 41
6.2 Autocorrelation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6.3 Spatial autoregressive modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.4 Remote sensing, GIS and spatial statistics in health studies . . . . . . . . . . . . 42
6.5 Limitations of the study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7 Conclusions and recommendations 45
7.1 Conclusions..................................... 45
7.2 Recommendations ................................. 46
A R codes used 53
iv
LIST OF FIGURES
2.1 Cholerariskfactors................................. 8
2.2 Conceptual framework of spatial epidemiological data analysis . . . . . . . . . . 9
2.3 John Snow’s 1854 cholera-outbreak map of London . . . . . . . . . . . . . . . . 12
3.1 Regional map of Ghana(left) and Kumasi(right) . . . . . . . . . . . . . . . . . . 17
3.2 A communal waste container, at KNUST Kentinkrono, Kumasi. . . . . . . . . 18
3.3 Solid waste dumped in a river . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4 Solid waste dumped in a gutter . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 True colour subset of RapidEye image . . . . . . . . . . . . . . . . . . . . . . . 21
4.1 Flow chart of Methodology in the study . . . . . . . . . . . . . . . . . . . . . . 23
4.2 Distance surface of nearest refuses dumps . . . . . . . . . . . . . . . . . . . . . 26
4.3 Kerneldensitysurface ............................... 26
4.4 Proximity to digitized reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.5 Proximity to classified reservoirs . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.6 Rook neighborhood structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1 Landcovermap ................................... 31
5.2 Choropleth maps showing showing cholera prevalence for each community . . . 32
5.3 Proportional symbol maps showing showing cholera prevalence for each commu-
nity ......................................... 33
5.4 Probability map of cholera cases; Choynowski’s approach . . . . . . . . . . . . 34
5.5 Poisson Probability and Relative risk maps of cholera incidence . . . . . . . . . 34
5.6 Raw(crude) rates and Expected count maps of cholera incidence . . . . . . . . . 35
5.7 Moran scatter and spatial correlogram plots . . . . . . . . . . . . . . . . . . . . 36
5.8 LISA significance and cluster maps . . . . . . . . . . . . . . . . . . . . . . . . . 36
v
LIST OF TABLES
3.1 RapidEye product specifications . . . . . . . . . . . . . . . . . . . . . . . . . . 20
5.1 Accuracyassessment ................................ 32
5.2 Summary statistics of probability mapping . . . . . . . . . . . . . . . . . . . . 33
5.3 Moran’s Index for spatial autocorrelation of cholera cases . . . . . . . . . . . . . 35
5.4 Spatial correlogram for CASES . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.5 Summary statistics of variables used in spatial modeling . . . . . . . . . . . . . 37
5.6 Results of OLS Regression, Model A . . . . . . . . . . . . . . . . . . . . . . . . 37
5.7 Results of OLS Regression, Model B . . . . . . . . . . . . . . . . . . . . . . . . 38
5.8 Results of CAR Regression, Model A . . . . . . . . . . . . . . . . . . . . . . . 38
5.9 Results of CAR Regression, Model B . . . . . . . . . . . . . . . . . . . . . . . 38
5.10 Results of Updated OLS Regression, Model A . . . . . . . . . . . . . . . . . . . 38
5.11 Results of Updated OLS Regression, Model B . . . . . . . . . . . . . . . . . . . 39
5.12 Results of Updated CAR Regression, Model A . . . . . . . . . . . . . . . . . . 39
5.13 Results of Updated CAR Regression, Model B . . . . . . . . . . . . . . . . . . . 39
vi
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 1
Introduction
1.1 MOTIVATION AND PROBLEM STATEMENT
Recent global health reports show a continual vulnerability of large populations to infectious dis-
eases in relation to our environments[54, 80]. Governments and international organizations have
extensively recognized the need to improve the well-being of its populations as well as to undertake
prompt preventive and control measures. Infectious diseases are complex to control and prevent,
leading to questions on how best to combat them through novel and creative solutions[35, 76].
Cholera is an epidemic and infectious disease which is of global and public health significance,
hence the need to recognize and address it accordingly[81].
Cholera is an acute intestinal infection caused by a bacterium (V. cholerae) leading to intestinal
infection and diarrhea. Infection is acquired chiefly by intake of contaminated water or food[16,
38]. Transmission is due to the fecal contamination of food and water as a result of poor sanita-
tion. The bacterium can live naturally in any environment[52]. Thus, it remains a global threat
especially in countries where access to clean safe drinking water and sufficient sanitation cannot
be assured. Nearly every developing country faces cholera outbreaks or the risk of a cholera
epidemic[47, 52].
The disease has been a public health burden in Ghana since 1970 when the first case was
reported[61]. Between 1999 and 2005, a total of 26,924 cases and 620 deaths were reported offi-
cially to the World Health Organization (WHO). In addition to human suffering and loss of lives,
a cholera outbreak causes panic, disrupts socio-economic activities and can hinder development
in the affected areas. It is more prevalent in the Kumasi metropolis than other districts within the
Ashanti region of Ghana. Various factors influence this, including overcrowding, urbanization,
proximity and density of refuse dumps, proximity to water sources and poverty[31, 59, 60]. Of
particular concern is the spatial distribution of cholera incidences and its associated risks factors.
This can be used by health officials and policymakers to make appropriate planning and resource
allocation. Also, it will be useful in limiting the severity and duration of an outbreak.
Spatial epidemiology is the study of the spatial distribution of disease incidence and its asso-
ciation to potential risk factors[34, 78]. Detail descriptions of spatial epidemiology and its ap-
plications can be found in[34, 62, 68, 78]. Most studies about cholera epidemiology use spatial
statistical methods and geographical information systems (GIS) to map the disease[39, 33, 61, 72].
Satellite images can greatly enhance mapping of the environmental factors associated with
cholera risk. Integrating satellite images, spatial statistics and GIS can provide public health offi-
cials with vital information needed to detect and manage cholera outbreaks. In order to correctly
plan, manage and monitor any public health system, it is important to have up to date, relevant
and complete information available to decision-makers.
Despite the availability of remotely sensed data in various formats for mapping, few studies
have utilized the technology for mapping the environmental niche of V. cholerae. Even though the
environmental variables have been identified[61], there is no map characterizing their distribution
in the Kumasi Metropolis. This thesis focuses on combining field observations on cholera and data
1
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
retrieved on environmental factors by remote sensing such as refuse dumps and water reservoirs.
Spatial statistical methods are used to map the disease.
1.2 RESEARCH IDENTIFICATION
RapidEye AG is a German geospatial information provider focused on assisting in management
decision-making through services based on their remotely sensed imagery. The company owns a
five satellite constellation producing 5 meter resolution imagery. These five identical Earth obser-
vation satellites allows for a continual capture of imagery. Any point on earth can be accessed,
which enables rapid response for crop, environmental and emergency monitoring[64].
Identified environmental variables such as water reservoirs and refuse dumps can be derived
from RapidEye images. Maps of these identified variables and cholera incidence data can be inte-
grated in a GIS and analyzed using spatial statistical tools.
1.2.1 General objective
The main objective is to identify and characterize the spatial distribution of environmental factors
that increase the risk of cholera infection in the Kumasi metropolitan area of Ghana using GIS,
remote sensing and spatial statistical methods.
1.2.2 Specific objectives
1. To map out potential cholera causing factors in the study area from a RapidEye image.
2. Visualize the relations between cholera incidence, water bodies and refuse dumps using a
GIS.
3. Determine the spatial relationship between cholera incidence and potential cholera reser-
voirs and refuse dumps using spatial statistics.
1.2.3 Research questions
1. Which environmental factors relevant to cholera modeling and mapping can be extracted
from a RapidEye image?
2. How can these environmental factors be extracted and what is their quality?
3. How can the derived remote sensing variables be combined with field data to produce maps
of environmental factors?
4. How can maps of cholera risks be visualized in a GIS?
5. Which models are most effective for modeling the effects of environmental risk factors on
cholera?
1.3 OUTLINE OF THESIS
The thesis comprises of seven(7) chapters.
Chapter 1 - Introduction
This chapter contains the motivation and problem statement, research objectives, research
questions and the methodology.
2
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 2 - Literature review
This chapter contains literature about cholera, statistical methods for spatial epidemiology,
areal data and applying spatial regression (Conditional autoregressive models) in modelling
areal data.
Chapter 3 - Study area, datasets and data preparation
This chapter consists of the study area, description of data sources and preparation.
Chapter 4 - Research methodology
This chapter describes the methods and and tools applied in the study to produce the out-
come.
Chapter 5 - Results and analysis
Chapter 6 - Discussion
Chapter 7 - Conclusions and Recommendations
3
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
4
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 2
Literature review
2.1 CHOLERA
Cholera is an acute diarrheal illness caused by a bacterium,Vibrio cholerae (V.cholerae) leading to in-
testinal infections[52, 13, 2]. There are numerous environmental strains of Vibrio cholerae, which
are found mainly in brackish waters and marine environments, but only two strains are responsi-
ble for cholera epidemics in humans, serogroups O1 or O139[47, 67]. Cholera mainly affects the
small intestines after ingestion of sufficient dose of the V.cholerae baterium through water and/or
food that is contaminated[47, 48, 67].
Cholera infection may often appear to be mild or even without symptoms, but can sometimes
be severe. Nearly one in 20 (5%) infected persons will have severe disease characterized by lots of
watery diarrhea, vomiting, and leg cramps[69]. For these people, rapid loss of body fluids leads to
dehydration and shock. Without immediate treatment, death can occur within a short time (few
hours). The symptoms typically start suddenly, between one to five days after consumption of
contaminated water or food. The watery diarrhea may have a fishy odor and an infected person
may produce 10 to 20 litres (0.01 to 0.02 m3) of diarrhea a day[69, 82]. Moreover, it causes intense
thirst, loss of skin turgor, wrinkled skin of hands and feet, sunken eyes, pinched facial expression,
thready or absent peripheral pulses, falling blood pressure, and inaudible hypoactive bowel sounds.
If the severe diarrhea and vomiting are not aggressively treated it can, within hours, result in life-
threatening dehydration and electrolyte imbalances[69, 48].
Though the bacterium can live naturally in any environment[52], studies have shown that
V. cholerae exist as natural inhabitants of aquatic ecosystems[42, 67, 21]. They usually occur as
part of flora of streams, riverine, brackish water, estuarine and coastal waters. They attach to
surfaces provided by plants, filamentous green algae, copepods (zooplankton), crustaceans, and
insects[21, 42]. Stagnant water and slow flowing water may also lead to an increasing exposure of
the organism. According to Islam et al.[42], the baterium can survive in almost all kinds of aquatic
environments including fresh water sources such as lakes, ponds, rivers and tanks. Cholera bateria
can also survive in non-aquatic environments such as refuse dump sites, fruits, fresh vegetables,
meat, cooked food[26], human and animal faecal waste, untreated or inadequately treated sewage.
2.1.1 Cholera Transmission
Sources of infection
Contaminated water with free-living V. cholerae cells are the main source of cholera, followed to a
lesser extent by contaminated food[22, 57, 13], particularly seafood likecrab, oysters, and shellfish.
Bad sanitation practices in highly populated areas harboring the bacteria are the source of intermit-
tent outbreaks due to contamination of drinking water and/or improper food preparation. The
source of the infection is usually other cholera sufferers when their untreated diarrheal discharge
is allowed to get into waterways or drinking water supplies. Drinking any of the infected water
and eating any foods washed in the water, as well as seafoods living in the affected waterbodies,
5
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
can cause a person to contract an infection. The infection is seldom spread directly from person
to person.
Transmission mechanism
There are two routes of cholera transmission namely primary transmission and secondary transmission[61,
20]. According to Osei[61]citing Hartley et al.[36], primary transmission happens through ex-
posure to an environmental reservoir of V.cholerae or contaminated water sources regardless of
previously infected persons or faecal contamination. Therefore, aquatic environments are essential
for the spread of cholera.
Secondary transmission route on the other hand occurs through exposure to fecally contami-
nated water sources, food or infected person. Osei[61]argued that this route reflect a complicated
tranmission pattern, since multiple factors may play a role in the spread of the disease. For in-
stance, faecal-oral transmission is increased by the degree of contamination of water supply as well
as frequency of contacts of these waters[19], which is in turn influenced by local environmen-
tal factors, socioeconomic, demographic as well as sanitation conditions. Both routes are clearly
important and useful to control epidemics.
2.1.2 Factors influencing spread
The cholera germ is passed in the stools of infected persons . It is widely spread by consuming
food or water which has been contaminated by the fecal waste/stools of an infected person. This
happens more often in developing countries. This is so because underdeveloped countries lack
adequate clean water supplies for drinking and proper sewage disposal systems as well as also prac-
tices poor sanitation and poor food hygiene [40, 41, 45, 47]. Once cholera is introduced to a
population in a specific location, numerous complex factors decisively influence its propagation
and may lead to prolonged transmission[70, 18, 45, 41]. Socioeconomic, environmental, demo-
graphic and climatic factors enhances the vulnerability of a population to infection and contribute
to the epidemic spread of cholera[40, 31, 14, 3, 56]. These factors include the following:
1. Poor sanitation
Cholera is hypothesized as a disease of deficient sanitation[2, 41, 45]. The lack of adequate
toiletry, cleaning, washing and drainage facilities results in sickness and increases the risk of
transmission.
2. High poverty and low income level
Borroto and Martinez-Piedra[14]and Talavera and Pérez[73]identified poverty as an impor-
tant predictor of cholera. Low income levels result in poor diet, malnutrition, poor housing
facilities and lack of access to education. Typically, the world’s poorest people obtain drink-
ing water from a river/streams or wells; in the absence of toilet facilities or public sewage
systems, people defecate near these rivers and streams allowing human waste to mix with
the same water used for drinking.
3. High migration
This plays a role by introducing cholera into new populations[3, 31]
4. Cooking practices
Light cooking of food which has been contaminated[45]
6
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
5. Overcrowding/high population
High population will lead to overcrowding putting strain on existing sanitation systems,
thereby putting the population at high risk[15, 71, 31, 51].
6. Lack of Clean drinking water
Unsafe water supply/contaminated water will increase the risk of cholera infection[72, 45]
7. Proximity and density of refuse dumps
According to Osei and Duker[59], there is a direct linear relationship between cholera preva-
lence and refuse dumps density and an inverse relationship with proximity to refuse dumps.
Two explanations given were (1) high rate of contact with filth breeding flies;they argued
that filth breeding flies serve as a carrier of the V.cholerae from refuse dump sites where all
kinds of human garbage and excreta is disposed to humans and (2) flood water contamina-
tion; in the event of heavy rains, runoff from open spaced refuse dumps serves as a pathway
for the distribution of the bacteria, washing infected excreta into wells, streams and surface
water bodies.
8. Proximity to surface water sources
Close proximity to contaminated drinking water bodies make inhabitants more prevalent
to cholera[3, 72, 60]
9. Climatic conditions
Studies have shown that there is direct correlation between cholera and sea surface temper-
ature, sea surface height, precipitation and chlorophyll concentrations[55, 21, 50].
10. Poor personal hygienic standards
Poor personal hygiene increases cholera propagation within a given environment.
Figure 2.1 is a table of a summary of cholera risk factors taken from Collins et al[20].
2.1.3 The burden of cholera in Ghana
The disease has been a public health burden in Ghana since 1970 when the first case was reported[61].
Between 1999 and 2005, a total of 26,924 cases and 620 deaths were reported officially to the World
Health Organization (WHO). In addition to human suffering and loss of lives, cholera outbreaks
causes panic, disrupt socio-economic activities and can impede development in the affected com-
munities.
2.2 SPATIAL EPIDEMIOLOGY
"Spatial Epidemiology is the description and analysis of the geographic, or spatial, variations in
disease with respect to demographic, environmental, behavioral, socioeconomic, genetic, and in-
fectious risk factors"[24]. The spread of infectious diseases is closely associated with the concepts
of spatial and spatio- temporal proximity, as individuals who are linked in a spatial and a temporal
sense are at a higher risk of getting infected[62]. Proximity to environmental risk factors is there-
fore important. Thus knowledge of the spatial and temporal variations of diseases and character-
izing its spatial structure is essential for the epidemiologist to understand better the population’s
interactions with its environment[61].
7
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Source: Collins et al[20]
Figure 2.1: Cholera risk factors
Spatial epidemiology dates back to the 1800s, when maps of disease rates in different countries
began to emerge to characterize the spread and possible causes of outbreaks of infectious diseases
such as yellow fever and cholera[79]. Spatial analysis in the nineteenth and twentieth century
was mostly employed by plotting the observed disease cases or rates. For example, Snow[72]
mapped cholera cases together with the locations of water sources in London, and showed that
contaminated water was the major cause of the disease. Recent advances in technology now allow
not only disease mapping but also the application of spatial statistical methods[23, 44], satellite
derived data[37]and Geographical information systems(GIS)[3, 60].
2.2.1 Framework for spatial analysis
Spatial epidemiology comprises of a wide range of methods. Determining which ones to use can
be challenging[62]. Fig. 2.2 is a diagrammatic representation of a spatial analysis framework taken
from Pfeiffer et. al[62]adapted from Bailey and Gatrell[9]. Pfeifferet. al[62]identified four groups
8
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
as illustrated in Fig. 2.2 that can be used to define a logical, sequential process for conducting spatial
analysis:
Source: Pfeiffer et. al[62], adapted from Bailey and Gatrell[9]
Figure 2.2: Conceptual framework of spatial epidemiological data analysis
1. Data
"The objectives of spatial epidemiological analysis are the description of spatial patterns,
identification of disease clusters, and explanation or prediction of disease risk"[62]. Cen-
tral to these objectives is the need for data. Geographic data systems include georeferenced
feature data and attributes, be they points or areas. These data are obtained by taking field
surveys, remotely sensed imagery or use of existing data generated either by government or-
ganizations or those closely linked to government such as cadastral, postal, meteorological
or national census statistics and health organizations.
2. GIS and DBMS
Management of the data is performed using GIS and database management systems (DBMS),
and is of relevance throughout the various phases of spatial data analysis. GIS provide a plat-
form for managing these data, computing spatial relationships such as proximity to source of
infection, connectivity and directional relationships between spatial units, and visualizing
both the raw data and results from spatial analysis within a cartographic context[62].
3. Visualization and exploration
Visualization and exploration cover techniques that focus solely on examining the spatial
dimension of the data. Visualization tools are used resulting in maps that describe spatial
patterns and which are useful for both stimulating more complex analyses and for communi-
cating the results of such analyses. Exploration of spatial data involves the use of statistical
9
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
methods to determine whether observed patterns are random in space. However there is
some overlap between visualization and exploration, since meaningful visual presentation
will require the use of quantitative analytical methods[53].
4. Modeling
Analytical procedures that simulates real-world conditions within a GIS using the spatial re-
lationships of geographic features. Modeling introduces the concept of cause-effect relation-
ships using both spatial and non-spatial data sources to explain or predict spatial patterns[62].
However, this is not a linear process, as presenting the results from exploration and modeling
requires a return to visualization.
2.2.2 Statistical methods for spatial epidemiology
There exist various epidemiological inquiry which include disease mapping, geographic correla-
tion studies and clustering/cluster detection.
Disease mapping
Disease mapping provide information on a measure of disease occurrence across a geographic
space. Disease maps are able to provide us a rapid visual summary of complex geographic infor-
mation. These maps may also identify subtle patterns in epidemic/health data that are sometimes
missed in tabular presentations[24]. The aims of disease mapping include:
• Simple description by showing or displaying a visual summary of geographical risk for ex-
ample, the map of Snow[72]in fig. 2.3.
• Hypothesis generation by giving clues to causes of diseases and or factors that influence
spread by informal examination of maps with exposure maps, components of spatial versus
non-spatial residual variability may also provide clues to source of variability. The formal
examination is carried out via spatial regression.
• Provide estimates of risk by area to inform public health resource allocation.
• Estimation of background variability in underlying risk in order to place epidemiological
studies in context.
Geographic correlation studies
The objective of geographic correlation studies is to examine geographic disparities across inhab-
itants in an exposure to environmental variables which may be measured in air, water, or soil,
socioeconomic and demographic measures such as race and income, or lifestyle factors such as
smoking and diet in relation to health outcomes measured on a geographic scale[24]. Correlation
studies also aims at:
• Examination of the association between disease outcome and explanatory variables, in a
spatial setting, using regression models.
• Conventional modeling approaches such as logistic regression for point data, and loglinear
models for count data.
• the examination of risk with respect to a specific point or line putative source of pollution.
10
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
• For count data, disease mapping models can be extended to incorporate a regression com-
ponent.
Correlation studies deals with the association between disease risk and exposures of interest. we
examine the association between risk and exposures at the area level via ecological regression using
Poisson regression as a framework for areal data and logistic regression for point data.
Clustering/Cluster detection
Clustering examines tendency for disease risk to exhibit "clumpiness", while the Cluster detec-
tion refers to on-line surveillance or retrospective analysis, to reveal "hot spots". The aim is to
investigate disease clusters and disease incidence near a point source[46].
2.2.3 Spatial epidemiology of cholera
John Snow was the first to map cholera[72]. In his study, Snow was able to assess the spatial pattern
of cholera cases in relation to potential risk factors, in this instance the locations of water pumps.
He furthermore made a solid use of statistics to demonstrate the connection between the quality
of the source of water and cholera incidence and used a dot map to illustrate how cases of cholera
clustered around the Broad Street water pump in London (See fig. 2.3).
After Snow’s work, some epidemiological studies on cholera have focused on pathogenesis and
biological characteristics of V.Cholerae[32, 67]. These studies have been useful, in understanding
the environments that are most suitable for the bacteria.
To be able to identify and map environmental factors that impact risk of cholera, spatial epi-
demiological tools have to be applied in cholera studies. Understanding the spatial relationship be-
tween cholera and environmental risk factors have been a challenge for long. Recent studies have
used GIS based and statistical methods in mapping the disease[3, 60, 25, 29]. Osei and Duker[59]
mapped locations of all an environmental risk factor (refuse dumps) using a Global Positioning
System (GPS) in Kumasi, Ghana. They created two spatial factor maps using GIS and spatial anal-
ysis, a spatial distance surface showing proximity of distances from community to refuse dumps
and a kernel density surface showing the number of refuse dumps per density area. Two spatial co-
variates were derived and used as explanatory variables in spatial regression model to relate cholera
incidence to refuse dumps in Kumasi. In a related study, Osei et. al[60], potential cholera reser-
voirs (rivers and streams) and elevation were digitized from a topographic map. Spatial distance
factor maps of nearest reservoirs to communities were created and used as covariates in spatial
regression modeling.
2.3 AREAL DATA AND SPATIAL AUTOCORRELATION
Area data are observations associated with a fixed number of areal units. The areas may form
a regular lattice, as with remotely sensed images, or be a set of irregular areas or zones, such as
countries, districts and census zones[27]. Data about individuals are often available only at an
aggregated areal level in order to protect personal information. For example, average income levels
for census tracts are readily available, but the income of an individual person in that census tract
is usually not available. Similarly, the total number of people with cholera in a health service area
might be known, but not each person’s individual location within that area.
Spatial autocorrelation statistics are used to measure and analyze the degree of spatial corre-
lation/dependency among observations in a geographic space[28]. The principle underlying the
analysis of spatial data is the proposition that values of a variable in near-by locations are more
similar or related than values in locations that are far apart. This inverse relation between value
11
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Source: Snow[72]
Figure 2.3: John Snow’s 1854 cholera-outbreak map of London (deaths shown as dots, water
pumps as crosses)
association and distance is summarised by Tobler’s first law that "everything is related to everything
else, but near things are more related than distant things"[74].
2.3.1 Spatial weights and neighborhoods
An important aspect of defining spatial association is the determination of the relevant neighbor-
hood of a given area, that is, those areal units surrounding a given data point (area) that would
be considered to influence the observation at that data point. This is a necessary step in using
areal data[27]. These neighboring areas are spatial units that interact in a meaningful way. This
interaction could relate, for example, to spatial spillovers and externalities[46].
Spatial autocorrelation measures require a weights matrix that defines a local neighborhood
around each geographic area/unit[5]. The value at each areal unit is compared with the weighted
average of the values of its neighbors. A weighting system is chosen and assigned to the neighbor-
hoods. Weights can be constructed based on either contiguity to the polygon boundary (shape)
files, or calculated from the distance between points (points in a point shape file or centroids of
polygons)[12, 78, 5]. The weights are usually row-standardized to ensure that the row members
for each observation sum to 1, with zero on the diagonal and some non-zero off-diagonal elements.
12
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
The formula for each weight is:
wij =Cij
PN
j=1 Cij
(2.1)
With
Cij = 1
when iis linked to j, otherwise
Cij = 0
The spatial weights reflects the strength of the geographic relationship between observations
in a neighborhood, e.g., the distances between neighbors, the lengths of shared border, or whether
they fall into a specified directional class such as "North".
2.3.2 Spatial autocorrelation test
Spatial autocorrelation statistics include Global Moran’s Iand Local Moran’s I(LISA). These mea-
sures compare the spatial weights to the covariance relationship at pairs of locations. A spatial au-
tocorrelation value observed to be positive than expected from random shows there is clustering of
similar values across geographic space, while significant negative spatial autocorrelation indicates
that neighboring values are more dissimilar than expected by random, suggesting there is a spatial
pattern similar to that of a chess board[5].
2.3.3 Global Moran’s I
Moran statistics are one class of measures of spatial autocorrelation. Global autocorrelation statis-
tics provide a single measure of spatial autocorrelation for an attribute in a region as a whole[5].
I=N
PiPjWij
×PiPjWij (yi−¯y)(yj−¯y)
Pi(yi−¯y)2(2.2)
where there are Nunits, the attribute value for each unit iis yi, and Wij is the weight (or con-
nectivity) for units iand j. The locational information for this formula is found in the weights.
Therefore, for non-neighboring tracts, the weight is zero, so these are not used in the calculation
of correlation. The expected value of Moran’s Iis
E(I) = −1
(n−1) (2.3)
A Moran Iof +1 indicates strong positive spatial autocorrelation (i.e., clustering of similar values),
0 indicates random spatial ordering, and -1 indicates strong negative spatial autocorrelation (i.e., a
checkerboard pattern)[4].
2.3.4 Local indicators of spatial autocorrelation(LISA)
Local spatial autocorrelation statistics provide a measure, for each unit in the region, of the unit’s
tendency to have an attribute value that is correlated with values in nearby areas.
Ii=ziX
j
wij zj(2.4)
Where ziandzjare standardized scores of attribute values for unit iand j, and jis among
the identified neighbors of iaccording to the weights matrix wij. The local spatial autocorrelation
analysis is based on LISA statistics and computes a measure of spatial association for each individual
location[4].
13
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
2.4 MODELING AREAL DATA
Analysing public health data involve disease counts, proportions, or rates. These counts or rates
are not continuous like the continuous outcomes familiar in linear regression. Whereas large
counts or rates may roughly follow the assumptions of linear models, spatial analyses often fo-
cus on counts from small areas with relatively few subjects at risk and few cases expected during
the study period. Such instances require models appropriate for count or rate outcome[27].
Modeling spatial interactions that arise in spatially referenced data is commonly done by in-
corporating the spatial dependence into the covariance structure either explicitly or implicitly
via an autoregressive model. In the case of lattice (areal) data, two common autoregressive models
used are the conditional autoregressive model (CAR) and the simultaneously autoregressive model
(SAR). Both of these models produce spatial dependence in the covariance structure as a function
of a neighbor matrix, W and often a fixed unknown spatial correlation parameter[78, 77].
2.4.1 Spatial regression models
Spatial regression models are statistical models that account for the presence of spatial effects, i.e.,
spatial autocorrelation or spatial dependence and/or spatial heterogeneity. This is not the case
with the standard linear regression model[7]. The standard linear regression using least squares
(OLS) is used to find a linear relationship between a dependent variable and a set of explanatory
variables. It is written in vector form as
y=Xβ +ε(2.5)
where ε∼N(0, σ2I)which can be written as:
y∼N(Xβ, σ2I)(2.6)
But with spatial data, the assumption that the error terms are independent and normally dis-
tributed may not hold. For spatial regression, the error terms are assumed to be correlated. More-
over it includes spatial dependency in regression analysis, in which case a general model adopted
is[30]:
y∼N(Xβ, A)(2.7)
where Ais a positive definite symmetric covariance matrix which allows non-zero covariances
amongst the error terms. Now, Ais chosen so that elements of ythat are closer to each other in
space also have higher covariance. Denoting X β as µ,(2.6) can be simplified as:
y∼N(µ, A)(2.8)
This represents a generic assumption for spatial regression models. Two ways in which such models
are specified are the conditional autoregressive(CAR) and the simultaneous autoregressive (SAR)
models. These Spatial autoregressive models were developed primarily for use with geographically
aggregated spatial data where measurements could be taken at any location in the study area in
contrast to the geostatistical models developed for spatially continuous data[78].
2.4.2 Conditional autoregressive (CAR) models
The primary purpose of CAR models is to provide a modeling mechanism to account for residual
spatial correlation i.e., spatial trends not explained by spatial patterns in covariate values [78]. For
14
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
CAR models, the y-variable is assumed to be dependent not only on explanatory x-variables but
also on other nearby y-variables. The model is specified as
yi|{yj:j i} ∼ N(µi+n
j=1cij (yi−µj), λ2)(2.9)
That is, the distribution of yiconditional on all the other y-values is normal. The distribution is
expressed in terms of yi−µi, the difference between the observed yiand the expected value of yi
obtained when considering the x-variables. Some restrictions are imposed on the spatial weight
values, the cij values. First, cij =cji and second,cii = 0. The former restriction means the weights
must be symmetrical; the latter restriction simply means that the conditional distribution of yi
cannot depend on yiitself - only on other y-values. Typically, the cij s are chosen to reflect the
spatial structure of the data. If the data are associated with a set of zones, then cij might be defined
as 1 if zones iand jare contiguous, and 0 otherwise. For point data, cij might be defined as a
continuous function of distance such as kd−α
ij where α=1 or 2 and dij is the distance between
points iand j(assuming that there are no coincident points). This latter scheme could also be
applied to zonal/area data using distances between zone centroids.
Equation (2.9) can be rewritten as:
y∼N(µ, (I−C)−1λ2)(2.10)
where λ2is the conditional variance, and the variance-covariance matrix C=cij , from which it
can be seen that, restricting cij =cji stops the matrix I−Cfrom becoming ill-defined.
2.4.3 Simultaneous autoregressive (SAR) models
A simultaneous autoregressive model can be defined as
y∼N(µi+n
j=1bij (yi−µj), τ 2)(2.11)
This differs from the CAR model because the distribution of yiis not conditional. In this case the
marginal distributions for all the yis are specified as a system of simultaneous equations. Again,
the restriction on the spatial weights that bii=0 is imposed but there is no longer a symmetry
constraint on the weight matrix. Equation (2.11) can also be re-written as;
y∼N(µ, τ 2(I−B)−1(I−BT)−1)(2.12)
similarly to equation (2.10), B=bij
Differences between SAR and CAR
The structure of Band Cis usually specified by the shape of the lattice. One common way to
construct Bor Cis with a single parameter that scales a user defined neighborhood matrix Wthat
indicates whether the regions are neighbors or not as described in Equation (2.1). Thus, for the
SAR model B=λsWand for the CAR model C=λcWwhere λsand λcare often referred to
as "spatial correlation or spatial dependence" parameters and are left to be estimated.
• Marginal variances differ, even if λc=λs, all variances are location independent and are the
same.
• The CAR model with constant variance requires that the matrix W is symmetric
• OLS is inconsistent for the SAR model
For this study, the CAR model was considered and used.
15
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
16
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 3
Study area, datasets and data preparation
3.1 THE STUDY AREA
The study area is the Kumasi metropolitan, which is the second largest urban center of Ghana.
Kumasi metropolis is one of 18 districts and capital of the Ashanti Region (See Figure 3.1). Ashanti
Region is centrally located in the middle belt of Ghana. The metropolis lies at the intersection of
latitude 6.040N and longitude 1.280W, covering an area of about 220km2[31, 83]. According to
Osei and Duker[31], Kumasi has a population of about 1.2million which accounts for just under
a third (i.e. 32.4%) of Ahanti’s region population. The metropolis is subdivided in communities.
Unfortunately, these communities have no established boundaries. For this study, 68 communities
were used.
There are two major seasons, the rainy and dry season. The rainfall pattern is bimodal with
long rainy season from April to July, sometimes with peaks in May/June and a short season
September and Mid-November[60]. As described by Osei[61], approximately 82% of the inhabi-
tants in Kumasi have access to portable, pipe-borne water, however surface water from rivers and
streams is still used largely for cooking, bathing and washing utensils due to rampant water short-
ages. According to Osei et al.[60], the coverage rate of safe house-to-house collection of solid waste
is very low as such a greater proportion of household, approximately 81.2%, dispose of solid waste
at open space refuse dumps[59]. Furthermore, most demarcated areas for public sanitation and
waste disposal facilities have been sold out due to high demand for land compelling inhabitants to
defecate at open space refuse dumps[61].
Figure 3.1: Regional map of Ghana(left) and Kumasi(right)
17
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
3.2 DATASETS
3.2.1 Cholera data
The cholera data was obtained from Osei[61]. The data was collected from the Kumasi Metropoli-
tan Disease Control Unit (DCU) in the year 2005 during which there were severe outbreaks of
cholera in Kumasi, Ghana. According to Osei and Duker[59], the outbreak lasted for 72 days,
which was within the rainy season. The data consist of number of reported cases per commu-
nity (spatial unit for reporting). Each community is represented as point shapefile feature with X,
Y coordinates in meters and has number of cholera cases reported in 2005, population estimates
for 2005 and Raw rates as attributes. The population estimates, obtained from the Ghana Statis-
tical service(GSS), were used in calculating the raw rates. Raw rates were calculated as number
of cholera cases in each community divided by the estimated population in 2005 and rescaled by
multiplying it by a factor of 10,000 to express the raw rates as per 10,000 people more intuitively.
This study utilized only cholera cases reported during the 2005 outbreak.
3.2.2 Refuse dumps data
The composition of waste in Kumasi is predominantly made of biodegradable(organic) materials
and a high percentage of inert materials as well as small amounts of paper,plastic and metals[8].
The inert material is mostly made of wood ash, sand and charcoal. Solid waste management is
contracted to a number of private companies by the Waste Management Department in Kumasi
(WMD). The collection system of the waste management in the metropolis is based on two sys-
tems which are house-to-house waste collection and communal solid waste collection[75, 85]. The
communal waste collection system consists of containers placed throughout the city (See Figure
3.2). The containers are being emptied by waste collection companies and transported to landfill
sites located in the outskirts of the metropolis in a regular basis.
Source: Wikner[85]
Figure 3.2: A communal waste container, at KNUST Kentinkrono, Kumasi.
18
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
With house-to-house waste collection, the waste is collected at the yard or door at the house-
holds. At least 5 out of 10 household dispose of their waste right besides their houses, instead of
finding the nearest waste dump[59]. Waste that is not being collected is being indiscriminately
dumped in rivers ( See Figure 3.3) and gutters/drains as shown in (Figure 3.4) or burned.
Source: Wikner[85]
Figure 3.3: Solid waste dumped in a river
Source: Wikner[85]
Figure 3.4: Solid waste dumped in a gutter
The refuse dumps data were obtained from Osei and Duker[59]. The data was collected in a
field survey in 2005 using a Global Positioning System(GPS). A total of 124 refuse dumps were
mapped. The refuse dumps data consist of only point shapefiles showing only the X and Y coor-
dinates in meters.
19
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
3.2.3 Rivers data
Included in the data obtained from Osei[61]is a layer of stream segments which were digitized
from a topographic map in 2005.
3.2.4 RapidEye image data
This research made use of a RapidEye sensor. The RapidEye sensor which captured the data used
for classification of the potential cholera reservoirs is briefly described in the section below.
RapidEye sensor and image capture
RapidEye AG is a German geospatial information provider focused on assisting in management
decision-making through services based on their own Earth observation imagery. RapidEye has
five satellite constellation producing 5 meter resolution imagery that was designed and imple-
mented by MacDonald Dettwiler (MDA) of Richmond, Canada. Each of the five satellites contain
identical sensors, equally calibrated and travel on the same orbital plane (at an altitude of 630km).
Together, the 5 satellites are capable of collecting over 4 million km2of 5 m resolution, 5-band
color imagery every day[66, 84]. A summary of the image specification is shown in Table 3.1.
Table 3.1 RapidEye product specifications
Source: RapidEye[65]
Digital Data Product Information
Specifications
Spectral Bands Capable of capturing any subset of
the following spectral bands:
Blue 440 - 510 nm
Green 520 - 590 nm
Red 630 - 685 nm
Red Edge 690 - 730 nm
NIR 760 - 850 nm
Ground sampling 6.5 m
distance (nadir)
Pixel size 5 m
(orthorectified)
Swath Width 77 km
On board data storage 1500 km of image data per orbit
Equator crossing time 11:00 am (approximately)
RapidEye image data of study area
Four image tiles covering the study area captured on 9th November 2009 were acquired. These
images were RapidEye ortho-Level 3A specification. Level 3A offers the highest level of processing
available. Hence, radiometric, sensor and geometric corrections have been applied to the data. It
is a 16 bit data and consist of an image file, a metadata file, a browse image file and Unusable Data
Mask(UDM) file. It has a UTM projection and WGS84 Horizontal datum system. The images
were first mosaicked into one large image using ERDAS Imagine 2011 and the study area of interest
subsetted from the larger image. A subset of the image is shown in Figure 3.5.
20
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 3.5: True colour subset of RapidEye image
3.3 DATA PREPARATION
The shapefiles for open space refuse dumps, community centroids and rivers/streams was adopted
from Osei[61]. All three feature layers cover the whole of the Kumasi Metropolitan. The shape-
files when loaded into ArcMap had no coordinate system nor a reference ellipsoid. However, the
data description from Osei[61]revealed that they were in the Ghana Transverse Mercator Sys-
tem(GTM). They were then reassigned to Accra Ghana Grid (Ghana coordinate system) and later
converted to Universal Transverse Mercator UTM_WGS 1984 and then overlaid with the Rapid-
Eye image during image processing.
3.4 SOFTWARE
The following softwares were used for the study:
• ERDAS imagine 2011 for image processing/classification
• ArcGIS (ArcMap 10) for database creation, geovisualization and geospatial analysis
• OpenGeoDa, version 1.0.1 [7]for exploratory data analysis and geovisualization
• R-software, version 2.13.2[63]for statistical analysis. The R packages used in this study
include; spdep[11], maptools[49]rgdal[43]and the RColorBrewer[58].
21
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
22
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 4
Research methodology
4.1 INTRODUCTION
The image of the study area was classified to obtain a land cover map. Classified rivers and water-
bodies were extracted and integrated with refuse dumps layer and cholera data and visulized using
GIS. Spatial analysis and regression modeling was then carried out on the data. The flow chart for
the research is shown in Figure 4.1.
Figure 4.1: Flow chart of Methodology in the study
23
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
4.2 IMAGE ANALYSIS
4.2.1 Image pre-processing
The RapidEye satellite image was imported into ERDAS imagine. The image has been orthorec-
tified with radiometric, geometric and terrain corrections in WGS1984(UTM Zone 300N) projec-
tion from the producer[66]. This implies, all pixels on the image has been assigned to real world
coordinates in the WGS1984(UTM Zone 300N) projection system. This projection system was
maintained, since Ghana uses the WGS1984(UTM Zone 300N) projection system.
4.2.2 Image classification
A pixel based approach (supervised classification) was used in classifying the image. The image
classification process apportions the pixels of an image to exact spectral behavior of the ground
data. Pixels are sorted into a finite number of individual classes, or categories of data, based on
their data file values. If a pixel satisfies a certain set of criteria, then the pixel is assigned to the
class that corresponds to that criteria. This process converts image data to thematic data. The
landuse/landcover of the study area was classified to identify water reservoirs using the RapidEye
image of 2009 using the maximum likelihood algorithm.
The results of the image classification were validated in order to assess their accuracy. For
this study, a random sampling scheme was used to select 85 points (pixels) from the classification
output and compared with the reference data. Comparison was done by creating an error matrix.
The image classification and accuracy assessment were done in ERDAS imagine 2011.
4.3 DATA INTEGRATION AND VISUALIZATION
The communities do not have established boundaries between them, though the cholera cases were
collected at the community level. Hence to analyze the data at this level, arbitrary boundaries were
created with the aid of the communities centroids and the RapidEye image. This was done visually
and arbitrary keeping in mind, natural boundaries such as rivers and that these communities form
clusters at certain intervals.
The classified water reservoirs were extracted and converted into shapefiles. Two sources of wa-
ter reservoirs are used in this study. They include water reservoirs obtained from the classification
of the RapidEye image and digitized rivers/streams data that were adopted from Osei et.al[60].
They are analyzed separately in this study since they are measured on different scales. Maps of
refuse dumps, community centroids and water reservoirs were overlaid on each other in ArcMap
10 and Spatial factor maps were created and visualized. A description of how the spatial factor
maps were generated is shown in the next section below.
4.3.1 Spatial factor maps
Based on the hypothesis that cholera is a disease of deficient sanitation and assuming all water
reservoirs (rivers/streams) are potential cholera reservoirs, the following predictions are made[59]
:
• Inhabitants living in close proximity to open-spaced refuse dumps should have higher preva-
lence than those farther
• Areas with high density of open-spaced refuse dumps should have higher cholera prevalence
than areas with lower density
24
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
• Inhabitants living in close proximity to to potential cholera reservoirs should have higher
prevalence than those farther
Therefore to determine the spatial relationships between cholera prevalence per community and
(a) distances to refuse dumps (b) density of refuse dumps and (c)distance to potential cholera reser-
voirs, four spatial factor maps were created using ArcMap. The spatial factor maps were overlaid
with point map (centroids) of communities to create four explanatory variables (independent vari-
ables);
1. Proximity to refuse dumps; spatial distance surface, showing distances of each point or pixel
to the nearest refuse dumps
2. Density of refuse dumps; kernel density surface, showing the number of refuse dumps per
unit area
3. Proximity to digitized reservoirs; spatial distance surface, showing distances of each point
or pixel to the nearest reservoir that describes all digitized streams and rivers from Osei
et.al[60].
4. Proximity to classified reservoirs; spatial distance surface, showing distances of each point or
pixel to the nearest reservoir that describes all water reservoirs extracted from RapidEye
image.
The difference between (3) Proximity to digitized reservoirs and (4) Proximity to classified reservoirs
is that (3) uses information of reservoirs that were digitized from a topographic map as adopted
from Osei et.al[60]whilst (4) uses information of reservoirs that was classified from the RapidEye
image. The reservoirs are from the same study area but measured separately. This way we can
compare their effects on cholera to see if there are any significant differences.
Finally, a database for cholera was generated for the study area consisting of polygon bound-
aries of communities and points of community centroids. The attributes of each community were;
number of cholera cases, cholera raw rate (per 10,000 people), population and the four explanatory
variables discussed above. A description of how each of the explanatory variables were derived is
discussed below.
Proximity to refuse dumps
Using Spatial Analyst extension and the Distance toolbox, Euclidean distance surface was gener-
ated in ArcMap, with refuse dumps layer selected as input feature source. This calculates for each
cell, the euclidean distance to the closet source. In this study, a search radius of 1km was used.
From the Analysis Tools extension, the Near tool from the Proximity toolbox was then used to
determine the distance from each community centroid to the nearest refuse dump. This proce-
dure adds a new field NEAR_DIST to the input community centroids attribute table and stores
the paths of the centroids that contain the nearest refuse dumps. A map showing the distances
from each pixel to the nearest potential cholera source is shown in Figure 4.2.
Density of refuse dumps
A kernel density surface, showing the number of refuse dumps per unit area (Figure 4.3) was cre-
ated. This was done using the Density toolbox. Kernel density calculates the magnitude per unit
area from point features using a kernel function to fit a smoothly tapered surface to each point.
The surface value is highest at the location of the point and decreases with increasing distance from
the point, reaching zero at a search radius distance from the point. A radius of 1km was used. The
25
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 4.2: Distance surface of nearest refuses dumps
Extract Values to Points tool under the Extraction toolbox in the spatial analyst extension was
used to extract the density values of community centroids and recorded into the attributes of the
community feature class.
Figure 4.3: Kernel density surface
Proximity to digitized reservoirs
The same approach as discussed in section 4.3.1, subsection Proximity to refuse dumps was ap-
plied in this case. Here the input layer used are the digitized streams adopted from Osei et.al[60].
Figure 4.4 is the output distance surface map showing the distance from each pixel to the nearest
potential cholera reservoir.
26
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 4.4: Proximity to digitized reservoirs
Proximity to classified reservoirs
Similarly to the method discussed in the section above, spatial distance surface (Figure 4.5) map,
showing distances of each point or pixel to the nearest reservoir that describes all water reservoirs
extracted from RapidEye image was created and overlaid with point map of communities to derive
the nearest distances to the reservoirs.
Figure 4.5: Proximity to classified reservoirs
4.3.2 Mapping and geovisualization
Before measuring and modelling spatial dependence, maps of cholera prevalence and risks were
generated. Typically, we have counts of the incidence of cholera by spatial unit, associated with
counts of populations at risk. The task is then to try to establish whether any spatial units seem
to be characterized by higher or lower counts of cases than might have been expected in general
terms[9].
27
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
The following maps were generated;
• Choropleth maps; these maps use different color and pattern combinations to depict differ-
ent values of cholera incidence associated with each community.
• Proportional symbol maps; maps showing cholera incidence was generated in ArcMap 10,
refuse dumps represented by dots and river segments layers were added to this map to visu-
alize and characterize their distribution spatially.
• Choynowski’s probability map[17]; based on the choynosky () function, providing the
probability values required for each community.
• Four probability maps; based on the probmap () function in R package as described by
Bailey and Gatrell[9]. The function returns a data frame of rates for counts in populations
at risk with crude rates, expected counts of cases, relative risks, and Poisson probabilities.
4.4 SPATIAL ANALYSIS
Spatial analysis was carried out in two principle steps. Firstly, autocorrelation analysis was carried
out to determine the extent of correlation among neighboring communities and secondly the four
spatial covariates described in section 4.3 used as inputs in spatial autoregressive modeling.
4.4.1 Autocorrelation analysis
The following steps were carried out to determine the extent of spatial autocorrelation among
neighboring communities:
1. choosing a neighborhood criterion;
2. creating a spatial weight matrix;
3. running statistical test, using weight matrix to examine spatial correlation.
Autocorrelation analysis was carried out in R. Before the analysis, the cholera polygon shape-
file was imported into R and a SpatialPolygonsDataFrame created. The R libraries used include
spdep[63], maptools[49], rgdal[43]and RColorBrewer[58].
Univariate (Local Indicators of Spatial Autocorrelation) LISA statistics developed by Luc Anselin[6]
were invoked by clicking on the Univariate LISA button on the Explore toolbar in Geoda. This
brings up a dialog that lets you specify which of the four output options you want to generate. The
most relevant of these options are The Significance Map and The Cluster Map, which are unique
to the LISA functionality. These two maps were generated using GeoDa.
Neighborhood criterion
The first step in the analysis of spatial autocorrelation is to define the neighborhood structure over
the entire area. The neighborhood structure describes the areas that are linked. A first order rook
contiguity criterion was used in this study. Rook contiguity uses only common boundaries to
define neighbors. The rook neighborhood structure used is illustrated in Figure 4.6.
28
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 4.6: Rook neighborhood structure
Creating a spatial weight matrix
A spatial connectivity matrix that contains information of the neighborhood structure of each
community was created and weights assigned to the areas that are linked. Row-standardized weights
were applied. Row standardization is used to create proportional weights in cases where features
have an unequal number of neighbors.
Running statistical test
A significance test against the null hypothesis of no spatial autocorrelation was used to test for
the significance of the statistic. Moran’s Index under randomization was calculated and a spatial
correlogram computed and plotted.
4.4.2 Spatial autoregressive modeling
To attempt to determine the spatial relationship between cholera incidence (the response vari-
able) and the four spatial covariates (explanatory variables) derived in section 4.3.1, i.e, proxim-
ity to refuse dumps(ND_RD), density of refuse dumps(Den_RD), proximity to digitized reser-
voirs(ND_DR) and proximity to classified reservoirs(ND_CR), two set of models were developed;
•Model A: relates cholera incidence with proximity to refuse dumps, dumps density and prox-
imity to digitized reservoirs.
•Model B: relates cholera incidence with proximity to refuse dumps, dumps density and prox-
imity to classified reservoirs.
The difference between these two models is that model A uses information of reservoirs that were
digitized from a topographic map as adopted from Osei et.al[60]whilst model B uses information
of reservoirs that was classified from the RapidEye image. The reservoirs are from the same study
area but measured separately. This way we can compare their effects on the model.
First, a standard linear regression, i.e. Ordinary least squares (OLS) model as described in
chapter 2, section 2.4.1 was fitted for models A and B to determine the linear relationship between
29
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
the response and the explanatory variables. This was done by running lm in R. However the as-
sumption that the error terms in the OLS have a zero mean and are identical and independently
distributed are usually violated due to the presence of spatial dependence (autocorrelation). There-
fore to include the spatial dependence, the conditional autoregressive model (CAR) as described
in chapter 2, section 2.4.1 was used to fit models A and B. This was also implemented in R us-
ing the spautolm function. The function take into account the neighbors and weights for the
autoregression model estimation by Maximum Likelihood arguments.
30
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 5
Results and analysis
5.1 CLASSIFICATION
After classification, the following landcover were mapped; waterbodies/streams (reservoirs), built
up areas, urban centers (heavily built up), forest, dense vegetation, grasslands and wetlands. Wet-
lands areas were characterized by flooded areas and or rivers with emergent vegetation. For this
reason, they were merged with watebodies. Forest was also merged with the dense vegetation
class resulting in five landcover classes. The landcover map resulting from the classification of the
RapidEye image is shown below in Figure 5.1. The rivers layer obtained from Osei et.al[60]was
Figure 5.1: Landcover map
used to validate the presence of the rivers on the image. The digitized rivers when overlaid on
the classified image lied on the classified streams thereby confirming their presence on the image.
Additionally, a random sampling scheme was used to select 85 points (pixels) from the classifica-
tion output and compared with the reference (true world class). The resulting classification was
assessed at 83.53% overall accuracy and an overall Kappa statistics of 0.80. A summary of the
accuracy assessment totals are shown in Table 5.1.
31
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
The refuse dumps could not be identified on the image. The 124 waste containers placed
throughout the city as described in chapter 3(See Figure3.2) are far less than 5m2making them
impossible to see in a 5m resolution image. Though the refuse dumps layer adopted from Osei
and Duker[59]were overlaid on the image, it could not really be confirmed on the image. This
study therefore adopted the refuse dumps data mapped by Osei and Duker[59].
Table 5.1 Accuracy assessment
Overall classification accuracy =83.53%, Kappa statistics =0.80
Class Reference Classified Number Producers Users
Name Totals Totals Correct Accuracy(%) Accuracy(%)
waterbodies/streams 10 8 8 80.00 100.00
urban 12 9 8 66.67 88.89
built area 23 22 20 86.96 90.91
dense vegetation 7 8 6 85.71 75.00
wetlands 7 10 7 100.00 70.00
forest 12 13 10 83.33 76.92
grasslands 14 15 12 85.71 80.00
Totals 85 85 71
5.2 MAPPING AND GEOVISUALIZATION
Figure 5.2 show choropleth maps of cholera count cases (5.2a) and raw rates per 10,000 people(5.2b)
for each community in the Kumasi Metropolis. Each community is colored according to the
category into which its corresponding attribute value falls. Here, communities with red color have
the highest counts/rates of cholera incidence; communities with the blue color have the lowest
counts/rates. The legend provides us indication of the overall magnitude of cholera incidence and
the magnitude of the relative differences in attribute values that correspond to the range of colors
used in the map. This map has been able to highlight two extremes; the high areas in red and the
low areas in blue as well as highlight the mean areas shown in orange. The community centroids
are shown in green dots.
(a) recorded incidence ( count cases) (b) incidence rate per 10,000 people
Figure 5.2: Choropleth maps showing showing cholera prevalence for each community
32
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 5.3 are thematic maps visually depicting how cholera incidence are shown by propor-
tions in each community. In Figure 5.3a, where only the count cases are represented, high cholera
cases appear only at few areas in the upper north and at the central parts of Kumasi with minor
reported cases at the outer communities. With the rates per 10,000 people, more communities are
now showing high risk (Figure 5.3b) than that shown in figure 5.3a. With a proportional symbol
map, the symbol size is proportional to the magnitude of the attribute values in each class. In
Figure 5.3, the size of each pink circle indicates the relative number of of count cases in (a) and
rates per 10,000 people in (b) per community.
Refuse dumps represented by the black dots and river segments in blue were added to this map
to visualize and characterize their distribution spatially.
(a) recorded incidence ( count cases) (b) incidence rate per 10,000 people
Figure 5.3: Proportional symbol maps showing showing cholera prevalence for each community
The probability map (Figure 5.4) of cholera cases based on Choynowski’s[17]approach folds
the two tails of the measured probabilities together, so that small values for a chosen α, occur for
spatial units (communities) with either unusually high or low rates. For this reason, the high and
low communities are plotted separately in Figure 5.4.
The probality map in figure 5.4 shows raw rates (assuming a constant rate across the study area),
relative risks, and Poisson probability map values calculated using the standard Poisson cumula-
tive distribution function. This does not fold the tails together, so that communities with lower
reported cases than expected, based on population size, have values in the lower tail, and those
with higher observed counts than expected have values in the upper tail, as Figure 5.4 shows.
Table 5.2. is a summary statistics of the results returned by running the the probmap ()
function in R. The function returns a data frame of rates for counts in populations at risk with
crude rates, expected counts of cases, relative risks, and Poisson probabilities.
Table 5.2 Summary statistics of probability mapping
Raw(crude) rate Expected count Relative risk Poisson probabilities
minimum 4.699 ×10−51.354 4.953 0.0000
mean 1.022 ×10−313.956 107.691 0.5155
stan. Dev. 6.895 ×10−413.103 72.684 0.3783
maximum 3.192 ×10−353.521 336.429 1.0000
33
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Figure 5.4: Probability map of cholera cases; Choynowski’s approach
(a) Poisson probability map (b) Relative risk map
Figure 5.5: Poisson Probability and Relative risk maps of cholera incidence
In Figure 5.5 (a), Poisson probability map values represents the probablility of getting a more
"extreme" count than actually observed whilst in Figure 5.5 (b), relative risks shows the ratio of
observed and expected counts of cases multiplied by 100.
34
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
(a) Raw rates map (b) Expected count map
Figure 5.6: Raw(crude) rates and Expected count maps of cholera incidence
Figure 5.6 (a) is simply the map of crude rates, i.e number of count cases in each community
divided by the total population in that community. Figure 5.6 (b) shows the expected counts of
cases assuming a global rate.
5.3 SPATIAL ANALYSIS
5.3.1 Autocorrelation analysis
The extent to which neighboring values of cholera cases are correlated was measured using Moran’s
Index. A significant assessment under randomization procedure was run in R to determine the sig-
nificance of the computed Moran’s Index. There is positive and spatial autocorrelation for cholera
incidence in the Kumasi metropolis (Moran’s I =0.138, p =0.045), as shown in Table 5.3. Table
5.4. shows the standard deviates of Moran’s I and a two-sided probability value for 6 lags.
Table 5.3 Moran’s Index for spatial autocorrelation of cholera cases
Moran’s I p-value expectation variance
0.138 0.045 -0.015 0.006
Table 5.4 Spatial correlogram for CASES
Spatial correlogram for CASES(method: Moran’s I)
estimate expectation variance standard deviate Pr(I) two sided
1 0.1382 -0.0149 0.0058 2.0083 0.04461 *
2 -0.0454 -0.0149 0.0027 -0.5784 0.56298
3 0.0200 -0.0149 0.0019 0.7918 0.42847
4 -0.0587 -0.0149 0.0018 -1.0310 0.30255
5 -0.0171 -0.0149 0.0019 -0.0494 0.96063
6 -0.1141 -0.0149 0.0031 -1.7757 0.07578 .
35
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Moran’s scatter plot and spatial correlogram are also shown in Figure 5.7. The slope of the
scatter plot corresponds to the value for Moran’s I. It is a measure of global spatial autocorrelation
or overall clustering in a dataset. The four quadrants of the scatter plot describe an observation’s
value in relation to its neighbors; starting with the x-axis, followed by y: High-high, low-low
(positive spatial autocorrelation) and high-low, low-high (negative spatial autocorrelation). These
quadrants correspond to the clusters and spatial outliers in the LISA maps in Figure 5.8.
Figure 5.7: Moran scatter and spatial correlogram plots
(a) LISA significance map (b) LISA cluster map
Figure 5.8: LISA significance and cluster maps
The significance map, illustrated in Figure 5.8(a) shows the locations with significant local
Moran’s I /LISA statistics in different shades of green (the corresponding p values are given in the
36
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
legend). Not much significance can be seen from the significance map as only 7 communities has a
LISA statistics of p =0.05. The LISA cluster map, shown in Figure 5.8(b) provides essentially the
same information as the significance map but with the significant locations color coded by type of
spatial autocorrelation. The four codes are shown in the legend. There is no high-low location in
the study area map. These four categories correspond to the four quadrants in the Moran’s scatter
plot shown in figure 5.7.
5.3.2 Spatial autoregressive modeling
A summary statistics of the four variables used in modelling are shown on Table 5.5. Cholera
incidence (count) for the period of study ranged between 1 and 76 recorded cases per community
(mean =13.96 and standard deviation =14.19).
Table 5.5 Summary statistics of variables used in spatial modeling
variable minimum mean maximum stan Dev
cases 1.00 13.96 76 14.19
proximity to refuse dumps(ND_RD)(m) 12.28 413.63 1033.6 223.35
density of dumps(Den_RD)(dumps per square km) 0.00 1.83 6.29 1.39
proximity to digitized reservoirs(ND_DR) (m) 56.72 293.72 700.51 137.92
proximity to classified reservoirs(ND_CR) (m) 18.75 309.15 1641.43 239.38
The results of the OLS regression models for Model A(where cholera reservoirs are the dig-
itized rivers adopted from Osei et.al[60]are shown in Table 5.6 and that of Model B(where cholera
reservoirs are extracted from RapidEye image) are shown in Table 5.7. Proximity to refuse dumps(ND_RD)
in both Models show significant correlation with cholera incidence. The preliminary analysis
shows that Density of dumps(Den_RD) is not a significant contributor in both Models(p-Value >
0.6). Also proximity to digitized reservoirs(ND_DR) in Model A is insignificant whilst the clas-
sified reservoirs(ND_CR) in Model B shows significant correlation with cholera. However using
ND_CR in model B has shown a significant improvement of Model A. (p-Value of Model B <
p-Value of Model A).
The results of the CAR models for Models A(Table 5.8) and B(Table 5.9) produce similar results
as the OLS models in Tables 5.6 and 5.7. Both OLS and CAR models show that Den_RD is not a
significant contributor of cholera incidence(p-Value=0.83 in Table 5.8 and p-Value=0.42 in Table
5.9). For this reason, Models A and B were updated to exclude the Den_RD variable.
Table 5.6 Results of OLS Regression, Model A
Parameter Estimate Std. Error t-Value p-Value
ˆ
β0(intercept) 28.3491 7.7320 3.666 0.00050 ***
ˆ
β1(ND_RD) -0.0223 0.0108 -2.065 0.04299 *
ˆ
β2(Den_RD) 0.6779 1.7122 0.396 0.69348
ˆ
β3(ND_DR) -0.0218 0.0114 -1.907 0.06098 .
Resid. Stan. error 12.69 on 64DF
Multiple R20.246
Adjusted R20.211 p-Value =3.973 ×10−4
After updating the OLS models to exclude the Den_RD variable; Both updated Models A (Ta-
ble 5.10)and B(Table 5.11) still show significant negative relationship for both ND_RD, ND_DR
37
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Table 5.7 Results of OLS Regression, Model B
Parameter Estimate Std. Error t-Value p-Value
ˆ
β0(intercept) 28.1687 7.2463 3.887 0.00024 ***
ˆ
β1(ND_RD) -0.0218 0.0104 -2.096 0.04007 *
ˆ
β2(Den_RD) 0.2474 1.6652 0.149 0.88232
ˆ
β3(ND_CR) -0.0183 0.0064 -2.832 0.00617 **
Resid. Stan. error 12.30 on 64DF
Multiple R20.292
Adjusted R20.259 p-Value =5.696 ×10−5
Table 5.8 Results of CAR Regression, Model A
Parameter Estimate Std. Error z-Value p-Value
ˆ
β024.4031 7.5280 3.2416 0.00118
ˆ
β1(ND_RD) -0.0182 0.0103 -1.7578 0.07878
ˆ
β2(Den_RD) 1.3867 1.6667 0.8320 0.40540
ˆ
β3(ND_DR) -0.0194 0.0111 -1.7408 0.08172
LR test value =0.63926
ˆ
λ=0.30089 ˆ
σ2=148.69 AIC =545.82
Table 5.9 Results of CAR Regression, Model B
Parameter Estimate Std. Error z-Value p-Value
ˆ
β025.7649 7.0933 3.6323 0.00028
ˆ
β1(ND_RD) -0.0191 0.0100 -1.9091 0.05625
ˆ
β2(Den_RD) 1.6261 1.6667 0.4214 0.67345
ˆ
β3(ND_CR) 0.0062 0.0111 -2.7550 0.00586
LR test value =0.35503
ˆ
λ=0.22493 ˆ
σ2=140.89 AIC =541.83
Table 5.10 Results of Updated OLS Regression, Model A
Parameter Estimate Std. Error t-Value p-Value
ˆ
β0(intercept) 30.9133 4.1957 7.368 <10−6***
ˆ
β1(ND_RD) -0.0255 0.0070 -3.631 0.00055 ***
ˆ
β2(ND_DR) 0.0113 0.0064 -1.916 0.05973 .
Resid. Stan. error 12.61 on 65DF
Multiple R20.244
Adjusted R20.221 p-Value =1.106 ×10−4
38
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Table 5.11 Results of Updated OLS Regression, Model B
Parameter Estimate Std. Error t-Value p-Value
ˆ
β0(intercept) 29.1234 3.3286 8.749 <10−6***
ˆ
β1(ND_RD) -0.0229 0.0069 -3.328 0.00144 **
ˆ
β2(ND_CR) -0.01842 0.0064 -2.877 0.00542 **
Resid. Stan. error 12.21 on 65DF
Multiple R20.292
Adjusted R20.270 p-Value =1.341 ×10−5
and ND_CR.
The results of the updated CAR Model A (Table 5.12) shows very significant correlation be-
tween cholera incidence and ND_RD (p-Value <0.001) and appear to have no significant effect
by the ND_DR variable (p-Value =0.073). However, the updated CAR Model B results (Table
5.13) shows very significant correlation between cholera incidence and both the ND_RD (p-Value
<0.001) and the ND_CR variables. Comparing λand the Akaike Information Criterion val-
ues(AIC), the updated CAR model B has better fit than the updated CAR model A. λis the spatial
correlation or spatial dependence and measures the strength of the relationship. As expected, an
inverse relationship between cholera prevalence and proximities to refuse dumps and classified
reservoirs was observed.
Table 5.12 Results of Updated CAR Regression, Model A
Parameter Estimate Std. Error z-Value p-Value
ˆ
β029.9900 4.2108 7.1220 <10−6
ˆ
β1(ND_RD) -0.0249 0.0068 -3.6316 0.00028
ˆ
β2(ND_DR) -0.0200 0.0111 -1.7891 0.07360
LR test value =0.34867
ˆ
λ=0.2144 ˆ
σ2=150.49 AIC =544.27
Table 5.13 Results of Updated CAR Regression, Model B
Parameter Estimate Std. Error z-Value p-Value
ˆ
β028.5056 3.3414 8.5310 <10−6
ˆ
β1(ND_RD) -0.0223 0.0067 -3.3115 0.00092
ˆ
β2(ND_CR) -0.0177 0.0062 -2.8319 0.00462
LR test value =0.28033
ˆ
λ=0.19442 ˆ
σ2=141.31 AIC =539.93
39
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
40
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 6
Discussion
6.1 IMAGE CLASSIFICATION
The overall accuracy orproportions correctly classified of the landuse/land cover was 83.53% with
a Kappa statistics of 0.80. The producer’s and user’s accuracies of the classified water reservoirs
class were assessed to be 80% and 100% respectively. The users accuracy is the probability that a
certain reference class has indeed actually been labelled as that class, whilst producers accuracy is
the probability that a sampled point on the map is indeed that particular class. The overall, users
and producers accuracies are important measures of mapping accuracy[1]. The digitized rivers
obtained from Osei et.al[60]when overlaid on the classified image lied on the classified reservoirs
thereby confirming their presence on the image. The classified reservoirs however contained ad-
ditional patches of water bodies(ponds/stagnant water) which are however missing from the digi-
tized rivers from Osei et.al[60]. Therefore, the classification of the water reservoirs is a validation
of the presence of the digitized rivers from the data of Osei et.al[60]on the image.
6.2 AUTOCORRELATION ANALYSIS
The extent to which neighboring values are correlated was measured using Global Moran’s In-
dex. There is positive and spatial autocorrelation for cholera incidence in the Kumasi metropolis
(Moran’s I =0.138, p =0.045), as shown in Table 5.3. This explains the spatial clustering of high
cholera cases recorded in the central and upper north parts of the Metropolis, and low cases around
the peripheries(see Figure 5.2(a)). However, further diagnostics reveal a fluctuation between nega-
tive and positive Moran’s I as shown in Table 5.4 and Figure 5.7. This could explain the co-location
of dissimilar values in some few areas, i.e. low values surrounded by high values or vice versa in
certain parts of the Metropolis, e.g. the south-west parts as shown in Figure 5.7. Generally the
high values occur at the central parts as expected and supports earlier findings by Osei et.al[31].
In their analysis, they observed clustering of high rates around Kumasi Metropolis, with Moran’s
I=0.271 and p <0.01. The difference between this study and that of Osei et.al[31]is that, they
analyzed the entire Ashanti region consisting of 18 districts as spatial units for cholera incidence
from a period of 1997 to 2001, whilst this study only focussed on the Kumasi Metropolis(one of
the 18 districts) with 68 communities as spatial units. Unlike the districts which have defined
boundaries, the communities did not. The boundaries of the communities were estimated from
the RapidEye image with the help of Google Earth image. These boundaries which describes the
neighborhood structure used in the analysis could also explain for the low p-Value obtained for
the Kumasi Metropolis.
The reasons for the clusters at the central parts being that the central parts are the most com-
mercialized areas and therefore, there is always a high daily influx of people from neighboring
communities. This high daily influx puts strains on existing sanitation systems, hence increasing
the risk of cholera transmission. Also due to overcrowding and high cost of housing, the migrants
end up settling in slummy areas where environmental sanitation is poor[31].
41
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
6.3 SPATIAL AUTOREGRESSIVE MODELING
In the results of the autoregressive models; updated CAR Model A (Table 5.12) shows very sig-
nificant correlation between cholera incidence and proximity to refuse dumps (p-Value <0.001)
and appear to have no significant effect by the proximity to digitized reservoirs (p-Value =0.073.
i.e. p >0.05). However, the updated CAR Model A(CAR A) results (Table 5.13) shows very sig-
nificant correlation between cholera incidence with both proximity to refuse dumps (p-Value <
0.001) and proximity to classified reservoirs(p-Value <0.01). This suggests a very high significant
inverse relationship between cholera incidence with the classified reservoirs as expected. The AIC
criteria are smaller for the updated CAR Model B(CAR B) indicating a better fit. The λvalues;(λ
=0.2144 for updated CAR Model A and λ=0.19442 for updated CAR Model B) measures the
strength of the autocorrelation(spatial dependence).
These findings confirms the research of Osei et.al[59, 60]which concluded that proximity to
refuse dumps and surface water influences the risk of cholera in the Kumasi Metropolis. The rea-
son being that, during outbreaks, runoff from open spaced dumps as a result of rains or flooding
may serve as major pathways for the distribution of the bacteria. Infected human excreta washed
away from these dump sites run into nearby wells, streams and surface water bodies thereby con-
taminating the water which can lead to cholera infection when used.
Comparing the results of CAR A to that of CAR B, we see how using information extracted
from a remote sensing image(RapidEye) affects the results and conclusions. The classified reser-
voirs however contained additional patches of water bodies(ponds/stagnant water) which are how-
ever missing from the digitized rivers from Osei et.al[60]. This reveals the importance of using
satellite imagery to detect environmental factors (cholera reservoirs). Satellite images are available
in various formats and provide high spatial and temporal coverage of the earth’s surface. The com-
bined use of remote sensing, GIS and spatial statistics provides a strong tool for assessing how land
cover variables relate to cholera incidence.
6.4 REMOTE SENSING, GIS AND SPATIAL STATISTICS IN HEALTH STUDIES
Satellite images can greatly enhance mapping of the environmental factors associated with disease
risk. With increasing higher spatial and spectral resolutions, more frequent coverage, lower price,
and increased availability of images offered by a wide range of new sensors[10], health researchers
should be able to extract many moreenvironmental variables associated with disease transmissions.
These will provide new opportunities to extend the uses of remote sensing technology beyond a
few vector and water borne diseases to studies of vegetation and climatic related diseases.
GIS plays a central role for providing a platform that allows us to store, integrate, manipulate,
visualize and perform exploratory analysis on data from various diverse sources[1]. It’s powerful
capabilities allows us to integrate data represented in different formats e.g., raster or image data,
vector or point, line and area data, at different spatial resolutions. GIS is an important tool used to
show the distribution of diseases within a community as well as highlight high spots where health
officials need to take prompt action. Basically a GIS should also be able to answer to "what?,
where? and when?" diseases occur.
In analyzing disease data (public health data), several disciplines come into play than just maps
and visual interpretations[78]. Medical science may only provide insights to some specific causes
of diseases (e.g., biological mechanisms of transmission and identification of infectious agents).
Moreover, not everybody exposed to a suspected cause contracts the disease. As such, it is impor-
tant to use statistical methods to analyze each person’s risk or probability of contracting a disease.
The main aim is to identify and quantify any exposures, behaviors, and characteristics that may
modify a person’s risk. Waller and Gotway[78]identifies four (4) usefulness why statistical and
42
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
spatial statistical methods should be employed to analyze public health data; (1) to evaluate dif-
ferences in rates observed from different geographic areas, (2) to separate pattern from noise, (3)
identify disease "clusters," and (4) assess the significance of potential exposures. With spatial sta-
tistical methods, we are also able to quantify uncertainty in our estimates, predictions, and maps
and provide the foundations for statistical inference with spatial public health data[78].
The general important message is that in addressing problems in health management for infec-
tious diseases, there exist a number of promising tools available for use. A combination of remote
sensing, geographical information systems (GIS) and spatial analysis provide important tools that
should be exploited in the fight against diseases.
6.5 LIMITATIONS OF THE STUDY
The limitations in this research include the following;
• The communities do not have defined boundaries. The boundaries of the communities
were estimated from the RapidEye image with the help of Google Earth image. This may
invariably affects the results of spatial autocorrelation since autocorrelation varies with the
neighborhood structures defined. As such the spatial autocorrelation results should be in-
terpreted with caution due to different shape and sizes of the communities.
• The cholera data used is only for a single year(2005) outbreak. Additional data from several
outbreaks could enhance the analysis.
• The cholera data are count data aggregated at community levels. This does not contain
spatial information about the exact locations of affected individuals. The assumption made
was that the population within a community has equal risk.
43
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
44
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Chapter 7
Conclusions and recommendations
7.1 CONCLUSIONS
This study utilized remote sensing image(RapidEye) to capture potential cholera reservoirs. Two
statistical models were developed to determine the spatial dependency of cholera incidence on; (1)
proximity to digitized reservoirs from a topographic map and refuse dumps and (2) proximity to
classified reservoirs from a RapidEye image and refuse dumps. The findings reveal a high significant
association between cholera cases and proximity to classified reservoirs from a RapidEye image
and refuse dumps than between the proximity to digitized reservoirs from a topographic map and
refuse dumps. Maps were produced to characterize the distribution of cholera prevalence and risk
in the Kumasi metropolis. The specific objectives and research questions in this study are addressed
below.
Objective 1 - To map out potential cholera causing factors in the study area from a RapidEye
image.
• Which environmental factors relevant to cholera modeling and mapping can be ex-
tracted from a RapidEye image?
–Water reservoirs, consisting of rivers, streams, ponds, lakes and wetlands.
• How can these environmental factors be extracted and what is their quality?
–A pixel based approach (supervised classification) using the maximum likelihood
algorithm. The classification was done using the ERDAS IMAGINE 2011 soft-
ware.
–The overall accuracy or proportions correctly classified of the landuse/land cover
was 83.53% with a Kappa statistics of 0.80. The producer’s and user’s accuracies of
the classified water reservoirs class were assessed to be 80% and 100% respectively.
Objective 2 - Visualize the relations between cholera incidence, water bodies and refuse dumps
using a GIS
• How can the derived remote sensing variables be combined with field data to produce
maps of environmental factors?
–By Integrating all datasets on a common platform in a GIS environment(ArcMap
10 software was used) with their attributes and applying mapping tools (See Fig-
ures 5.1 and 5.3).
• How can maps of cholera risks be visualized in a GIS?
–By computing the risks using spatial analyst tools and visualizing in GIS (See Fig-
ure 5.2). Probabilities and risks maps can also be also be computed and plotted in
R using the probmap () function (See Figures 5.4;5.5 and 5.6).
45
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Objective 3 - Determine the spatial relationship between cholera incidence and potential cholera
reservoirs and refuse dumps using spatial statistics
• Which models are most effective for modeling the effects of environmental risk factors
on cholera?
–Spatial autoregressive models such as the Conditional autoregressive model ap-
plied in this study.
–There is positive and spatial autocorrelation for cholera incidence in the Kumasi
metropolis (Moran’s I =0.138, p =0.045), as shown in Table 5.3.
–There are very significant and inverse relationship between cholera incidence and
proximities to both refuse dumps (p-Value <0.001) and reservoirs(p-Value <0.01).
7.2 RECOMMENDATIONS
Remote sensing data should be explored as a tool for monitoring the epidemiology and control of
infectious diseases. Research is needed to determine the appropriate level of geographic detail for
specific diseases.
In this study, refuse dumps could not be identified from the remote sensing image, because the
refuse dump bins were too small to be identify from a 5m resolution image. Further studies should
use a higher resolution image to map the refuse dumps, this will enhance the analysis. Other land
cover variables should also be explored to see whether if it has an effect on the disease.
The data used in this study are based on group of individuals at the community level. The
inclusion of detailed house hold data or a series of outbreaks over a certain time will be useful to
make inference on relatively smaller groups of individuals and/or spatio-temporal analysis.
46
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Bibliography
[1]GI Science and Earth Observation : a process - based approach : also as e-book. ITC Educational
Textbook Series. University of Twente Faculty of Geo-Information and Earth Observation
ITC, Enschede, 2010.
[2]Paul A. and Manning. Vibrio cholerae, infection and immunity. In Encyclopedia of Immunol-
ogy (Second Edition), pages 2476 – 2479. Elsevier, Oxford, second edition edition, 1998.
[3]M. Ali, M. Emch, J.P. Donnay, M. Yunus, and RB Sack. Identifying environmental risk
factors for endemic cholera: a raster GIS approach. Health & place, 8(3):201–210, 2002.
[4]L. Anselin. Local Indicators of Spatial Association (LISA). Geographical analysis, 27(2):93–
115, 1995.
[5]L. Anselin. An introduction to spatial autocorrelation analysis with GeoDa. Spatial Analysis
Laboratory, University of Illinois, Champagne-Urbana, Illinois, 2003.
[6]L. Anselin. GeoDa 0.9 User’s Guide. Spatial Analysis Laboratory, University of Illinois,
Urbana-Champaign, IL, 2003.
[7]L. Anselin, I. Syabri, and Y. Kho. Geoda: An introduction to spatial data analysis. Geograph-
ical Analysis, 38(1):5–22, 2006.
[8]M. Asase, E.K. Yanful, M. Mensah, J. Stanford, and S. Amponsah. Comparison of municipal
solid waste management systems in Canada and Ghana: A case study of the cities of London,
Ontario, and Kumasi, Ghana. Waste Management, 29(10):2779–2786, 2009.
[9]T.C. Bailey and A.C. Gatrell. Interactive spatial data analysis. Longman Scientific & Technical
Essex, 1995.
[10]L.R. Beck, B.M. Lobitz, and B.L. Wood. Remote sensing and human health: new sensors
and new opportunities. Emerging infectious diseases, 6(3):217, 2000.
[11]Roger Bivand, with contributions by Micah Altman, Luc Anselin, Renato Assunç˜ao, Olaf
Berke, Andrew Bernat, Guillaume Blanchet, Eric Blankmeyer, Marilia Carvalho, Bjarke
Christensen, Yongwan Chun, Carsten Dormann, Stéphane Dray, Rein Halbersma, Elias
Krainski, Pierre Legendre, Nicholas Lewin-Koh, Hongfei Li, Jielai Ma, Giovanni Millo,
Werner Mueller, Hisaji Ono, Pedro Peres-Neto, Gianfranco Piras, Markus Reder, Michael
Tiefelsdorf, , and Danlin Yu. spdep: Spatial dependence: weighting schemes, statistics and mod-
els, 2011. R package version 0.5-40.
[12]R.S. Bivand, E.J. Pebesma, V. Gómez-Rubio, and Inc NetLibrary. Applied spatial data analysis
with R, volume 747248717. Springer New York, 2008.
[13]Nathan S Blow, Robert N Salomon, Kerry Garrity, Isabelle Reveillaud, Alan Kopin, F. Rob
Jackson, and Paula I Watnick. Vibrio cholerae infection of drosophila melanogaster mimics
the human disease cholera. PLoS Pathog, 1:e8, 09 2005.
47
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
[14]R.J. Borroto and R. Martinez-Piedra. Geographical patterns of cholera in Mexico, 1991–
1996. International journal of epidemiology, 29(4):764, 2000.
[15]M. Bradley, R. Shakespeare, A. Ruwende, M. E. J. Woolhouse, E. Mason, and A. Munatsi.
Epidemiological features of epidemic cholera (El Tor) in Zimbabwe. Transactions of the Royal
Society of Tropical Medicine and Hygiene, 90(4):378–382, 1996.
[16]R. Campbell McIntyre, T. Tira, T. Flood, and P.A. Blake. Modes of transmission of cholera
in a newly infected population on an atoll: implications for control measures. The Lancet,
313(8111):311–314, 1979.
[17]M. Choynowski. Maps based on probabilities. Journal of the American Statistical Association,
pages 385–388, 1959.
[18]Deirdre L. Church. Major factors affecting the emergence and re-emergence of infectious
diseases. Clinics in Laboratory Medicine, 24(3):559–586, 2004.
[19]C. Codeço. Endemic and epidemic dynamics of cholera: the role of the aquatic reservoir.
BMC infectious diseases, 1(1):1, 2001.
[20]A.E. Collins, ME Lucas, MS Islam, and LE Williams. Socio-economic and environmental
origins of cholera epidemics in Mozambique: guidelines for tackling uncertainty in infectious
disease prevention and control. International journal of environmental studies, 63(5):537–549,
2006.
[21]Rita Colwell. Global climate and infectious disease: The cholera paradigm. Science,
274(5295):2025–2031, 1996.
[22]BC Deb, BK Sircar, PG Sengupta, SP De, SK Mondal, DN Gupta, NC Saha, S. Ghosh,
U. Mitra, and SC Pal. Studies on interventions to prevent eltor cholera transmission in urban
slums. Bulletin of the World Health Organization, 64(1):127, 1986.
[23]P. Elliott, JC Wakefield, NG Best, and DJ Briggs. Spatial epidemiology: methods and appli-
cations. Spatial epidemiology, 1(9):3–15, 2001.
[24]P. Elliott and D. Wartenberg. Spatial epidemiology: current approaches and future chal-
lenges. Environmental health perspectives, 112(9):998, 2004.
[25]M. Emch. Diarrheal disease risk in Matlab, Bangladesh. Social Science & Medicine, 49(4):519–
530, 1999.
[26]O. Felsenfeld. Notes on food, beverages and fomites contaminated with vibrio cholerae.
Bulletin of the World Health Organization, 33(5):725, 1965.
[27]Manfred M. Fischer, Jinfeng Wang, Manfred M. Fischer, and Jinfeng Wang. Modelling area
data. In Spatial Data Analysis, SpringerBriefs in Regional Science, pages 31–44. Springer
Berlin Heidelberg, 2011.
[28]M.M. Fischer and A. Getis. Handbook of applied spatial analysis: software tools, methods and
applications. Springer Verlag, 2010.
[29]G. Fleming, M. Merwe, and G. McFerren. Fuzzy expert systems and GIS for cholera health
risk prediction in Southern Africa. Environmental Modelling & Software, 22(4):442–448,
2007.
48
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
[30]A.S. Fotheringham, C. Brunsdon, and M. Charlton. Geographically weighted regression: the
analysis of spatially varying relationships. John Wiley & Sons Inc, 2002.
[31]O. Frank and D. Alfred. Spatial and demographic patterns of cholera in Ashanti region-
Ghana. International Journal of Health Geographics, 7, 2008.
[32]A.R. Ghosh, H. Koley, D. De, S. Garg, MK Bhattacharya, SK Bhattacharya, B. Manna, G.B.
Nair, T. Shimada, T. Takeda, et al. Incidence and toxigenicity of Vibrio cholerae in a freshwa-
ter lake during the epidemic of cholera caused by serogroup O139 Bengal in Calcutta, India.
FEMS microbiology ecology, 14(4):285–291, 1994.
[33]Peng Gong, Bing Xu, and Song Liang. Remote sensing and geographic information systems
in the spatial temporal dynamics modeling of infectious diseases. Science in China Series C:
Life Sciences, 49(6):573–582, 2006.
[34]A. J. Graham, P. M. Atkinson, and F. M. Danson. Spatial analysis for epidemiology. Acta
Tropica, 91(3):219–225, 2004.
[35]Margaret A. Hamburg. Considerations for infectious disease research and practice. Technol-
ogy in Society, 30(3-4):383–387, 2008.
[36]D.M. Hartley, J.G. Morris Jr, and D.L. Smith. Hyperinfectivity: A critical element in the
ability of v. cholerae to cause epidemics? PLoS Medicine, 3(1):e7, 2005.
[37]V. Herbreteau, G. Salem, M. Souris, J.P. Hugot, and J.P. Gonzalez. Thirty years of use and
improvement of remote sensing, applied to epidemiology: From early promises to lasting
frustration. Health & Place, 13(2):400–403, 2007.
[38]S.D. Holmberg, D.E. Kay, R.D. Parker, et al. Foodborne transmission of cholera in Microne-
sian households. The Lancet, 323(8372):325–328, 1984.
[39]M. Hugh-Jones. Applications of remote sensing to the identification of the habitats of para-
sites and disease vectors. Parasitology Today, 5(8):244–251, 1989.
[40]A. Huq, R.B. Sack, A. Nizam, I.M. Longini, G.B. Nair, A. Ali, J.G. Morris Jr, MN Khan,
A. Siddique, M. Yunus, et al. Critical factors influencing the occurrence of vibrio cholerae in
the environment of Bangladesh. Applied and Environmental Microbiology, 71(8):4645, 2005.
[41]E.O. Igbinosa and A.I. Okoh. Emerging vibrio species: an unending threat to public health
in developing countries. Research in microbiology, 159(7-8):495–506, 2008.
[42]M. S. Islam. The aquatic flora and fauna as reservoirs of vibrio cholerae: a review. Journal of
diarrhoeal diseases research, 12(2):87–96, 1994.
[43]Timothy H. Keitt, Roger Bivand, Edzer Pebesma, and Barry Rowlingson. rgdal: Bindings
for the Geospatial Data Abstraction Library, 2011. R package version 0.7-5.
[44]M. Kulldorff. A spatial scan statistic. Communications in statistics-theory and methods,
26(6):1481–1496, 1997.
[45]K.M. Kwofie. A spatio-temporal analysis of cholera diffusion in Western Africa. Economic
Geography, pages 127–135, 1976.
[46]A.B. Lawson and D.G.T. Denison. Spatial cluster modelling. CRC press, 2002.
49
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
[47]Kelley Lee. The global dimensions of cholera. Global Change & Human Health, 2(1):6–17,
2001.
[48]Myron M. Levine, Debasish Saha, A. S. G. Faruque, and Samba O. Sow. Cholera Infections,
pages 150–156. W.B. Saunders, Edinburgh, 2011.
[49]Nicholas J. Lewin-Koh, Eric Archer Adrian Baddeley Hans-Jórg Bibiko Jonathan Callahan
Stéphane Dray David Forrest Michael Friendly Patrick Giraudoux Duncan Golicher Virgilio
Gómez Rubio Patrick Hausmann Karl Ove Hufthammer Thomas Jagger Sebastian P. Luque
Don MacQueen Andrew Niccolai Tom Short Greg Snow Ben Stabler Roger Bivand, con-
tributions by Edzer J. Pebesma, and Rolf Turner. maptools: Tools for reading and handling
spatial objects, 2011. R package version 0.8-10.
[50]B. Lobitz, L. Beck, A. Huq, B. Wood, G. Fuchs, ASG Faruque, and R. Colwell. Climate
and infectious disease: use of remote sensing for detection of vibrio cholerae by indirect
measurement. Proceedings of the National Academy of Sciences, 97(4):1438, 2000.
[51]F.J. Luquero, C.N. Banga, D. Remartínez, P.P. Palma, E. Baron, and R.F. Grais. Cholera
Epidemic in Guinea-Bissau (2008): The Importance of "Place". PloS one, 6(5):e19005, 2011.
[52]S. Mandal, M.D. Mandal, and N.K. Pal. Cholera: A great global concern. Asian Pacific
Journal of Tropical Medicine, 4(7):573–580, 2011.
[53]A. Maroko, J.A. Maantay, and K. Grady. Using geovisualization and geospatial analysis to
explore respiratory disease and environmental health justice in New York city. Geospatial
Analysis of Environmental Health, pages 39–66, 2011.
[54]J. D. Mayer. Emerging Diseases: Overview, pages 321–332. Academic Press, Oxford, 2008.
[55]J. Mendelsohn and T. Dawson. Climate and cholera in KwaZulu-Natal, South Africa: the
role of environmental factors and implications for epidemic preparedness. International jour-
nal of hygiene and environmental health, 211(1-2):156–162, 2008.
[56]I. Mugoya, S. Kariuki, T. Galgalo, C. Njuguna, J. Omollo, J. Njoroge, R. Kalani, C. Nzioka,
C. Tetteh, S. Bedno, et al. Rapid spread of Vibrio cholerae O1 throughout Kenya, 2005. The
American journal of tropical medicine and hygiene, 78(3):527, 2008.
[57]E.J. Nelson, J.B. Harris, J.G. Morris, S.B. Calderwood, and A. Camilli. Cholera trans-
mission: the host, pathogen and bacteriophage dynamic. Nature Reviews Microbiology,
7(10):693–702, 2009.
[58]Erich Neuwirth. RColorBrewer: ColorBrewer palettes, 2011. R package version 1.0-5.
[59]F. Osei and A. Duker. Spatial dependency of v. cholera prevalence on open space refuse
dumps in Kumasi, Ghana: a spatial statistical modelling. International Journal of Health
Geographics, 7(1):62, 2008.
[60]F.B. Osei, A.A. Duker, E.W. Augustijn, and A. Stein. Spatial dependency of cholera preva-
lence on potential cholera reservoirs in an urban area, Kumasi, Ghana. International Journal
of Applied Earth Observation and Geoinformation, 12(5):331–339, 2010.
[61]Frank B. Osei. Spatial statistics of epidemic data : the case of cholera epidemiology in Ghana.
PhD thesis, 2010.
50
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
[62]Dirk Pfeiffer. Spatial analysis in epidemiology. Oxford University Press, GB, 2008.
[63]R Development Core Team. R: A Language and Environment for Statistical Computing. R
Foundation for Statistical Computing, Vienna, Austria, 2011. ISBN 3-900051-07-0.
[64]RapidEye. RapidEye: Dilivering the world. http://www.rapideye.net/about/index.
htm, 2011. [Online; accessed 29-November-2011].
[65]RapidEye. Satellite Imagery Product Specifications. http://www.rapideye.net/upload/
RE_Product_Specifications_ENG.pdf, 2011. [Online; accessed 3-January-2012].
[66]RapidEye. High resolution satellite imagery ortho product (level3A). http://www.
rapideye.net/products/index.htm, 2012. [Online; accessed 3-January-2012].
[67]Joachim Reidl and Karl E. Klose. Vibrio cholerae and cholera: out of the water and into the
host. FEMS Microbiology Reviews, 26(2):125–139, 2002.
[68]T. P. Robinson. Spatial statistics and geographical information systems in epidemiology and
public health, volume Volume 47, pages 81–128. Academic Press, 2000.
[69]David A. Sack, R. Bradley Sack, G. Balakrish Nair, and A. K. Siddique. Cholera. The Lancet,
363(9404):223–233, 2004.
[70]H. H. Shugart and Robert E. Shope. Factors Influencing Geographic Distribution and Incidence
of Tropical Infectious Diseases, pages 13–18. Churchill Livingstone, Philadelphia, 2006.
[71]AK Siddiquei, K. Zamani, AH Baqui, K. Akrami, P. Mutsuddy, A. Eusof, K. Haider, S. Islam,
and RB Sacku. Cholera epidemics in Bangladesh: 1985-1991. Diarrhoeal Diseases Research,
10(2):79, 1992.
[72]J. Snow. On the mode of communication of cholera. John Churchill, 1855.
[73]A. Talavera and E.M. Pérez. Is cholera disease associated with poverty? The Journal of
Infection in Developing Countries, 3(06):408–411, 2009.
[74]W.R. Tobler. A computer movie simulating urban growth in the Detroit region. Economic
geography, 46:234–240, 1970.
[75]Juliana Useya. Simulating diffusion of cholera in Ghana. Master’s thesis, Faculty of Geo-
information Science and Earth Observation of the University of Twente, 2011.
[76]Francis A. Waldvogel. Infectious diseases in the 21st century: old challenges and new oppor-
tunities. International Journal of Infectious Diseases, 8(1):5–12, 2004.
[77]M.M. Wall. A close look at the spatial structure implied by the CAR and SAR models.
Journal of Statistical Planning and Inference, 121(2):311–324, 2004.
[78]L.A. Waller and C.A. Gotway. Applied spatial statistics for public health data, volume 368.
Wiley-Interscience, 2004.
[79]SD Walter. Disease mapping: a historical perspective. Spatial epidemiology: methods and
applications, pages 223–239, 2000.
[80]WHO. Features: infectious diseases. http://www.who.int/features/infectious_
diseases/en/index.html, 2011. [Online; accessed 5-August-2011].
51
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
[81]WHO. Weekly epidemiological record. http://www.who.int/wer/2010/wer8531.pdf,
2011. [Online; accessed 5-August-2011].
[82]Wikipedia. Cholera. http://en.wikipedia.org/wiki/Cholera#cite_
note-Lancet2004-0, 2011. [Online; accessed 9-November-2011].
[83]Wikipedia. Kumasi. http://en.wikipedia.org/wiki/Kumasi, 2012. [Online; accessed
3-January-2012].
[84]Wikipedia. RapidEye. http://en.wikipedia.org/wiki/RapidEye, 2012. [Online; ac-
cessed 3-January-2012].
[85]Emma Wikner. Modelling waste to energy systems in Kumasi, Ghana. Master’s thesis,
Swedish University of Agricultural Sciences(SLU) in Uppsala, 2009.
52
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
Appendix A
R codes used
#######################################################
# I m p or t i ng c h o l e r a d a t a s h a p e f i l e t o R
#######################################################
l i b r a r y ( m a pt oo ls )
l i b r a r y ( s p )
l i b r a r y ( R Co lorB rew er )
l i b r a r y ( s p de p )
l i b r a r y ( r g d a l )
g e t i n f o . s ha p e ( " D: /Spatial_Analysis/c h o da t a . s hp " )
cho<−readShapePoly ("D:/Spatial_Analysis/c ho d at a . sh p " )
c l a s s ( c ho )
a t t a c h ( c ho )
#######################################################
# c r e a t e ro ok c o n t i g u i t y n e i gh b or s
#######################################################
cho_nbr<−po ly2 nb ( cho , qu een=FALSE ) # ro o k n e i gh b or ho o d
cho_nbr
#######################################################
# p l o t o f ro ok n ei gh bo rh oo d s t r u c t u r e
#######################################################
co o rds <−c o o r d i n a t e s ( c ho )
p l o t ( c ho , b o r d er =" b l a c k " , a x e s =TRUE )
p l o t ( c ho _n br , c o o rd s , a d d=T , c o l =" b l u e " )
t i t l e ( main=" F i r s t o r de r Rook c o n t i g u i t y n e i gh b ou r s " )
#######################################################
# n ei g hb or s l i s t t o we ig ht l i s t
#######################################################
cho_nbr_w<−n b 2 l i s t w ( c h o_ n br ) #Row s t a n d a r d i z e d w e i g h t s W
ch o_ nb r_w #Row s t a n d a r d i z e d w e i g ht s W
cho_nbr_wb<−n b 2 l i s t w ( c h o_ nb r , s t y l e ="B " ) # B i n a ry w e i g h t s B
cho_nbr_wb
#######################################################
# Moran ’ s I t e s t u nd er n or ma li ty
#######################################################
53
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
moran . t e s t ( cho$CASES , cho_n br_w , r an d om is a ti o n=FALSE ,
a l t e r n a t i v e ="t wo . s i d e d " ) # W
moran . t e s t ( cho$CASES , cho_n br_wb , ra nd om is at io n=FALSE ,
a l t e r n a t i v e ="t wo . s i d e d " ) # B
#######################################################
# Moran ’ s I t e s t u nd er r a nd o mi sa t io n
#######################################################
mora n . t e s t ( cho$CASES , l i s t w =c ho_ nbr_w , a l t e r n a t i v e =" two . s i d e d " ) # W
mora n . t e s t ( cho$CASES , l i s t w =c ho_ nbr_wb , a l t e r n a t i v e =" two . s i d e d " ) # B
#######################################################
# 99 9 Monte−C a rl o s i m u l a t i o n o f Mor an ’ s I
#######################################################
morpermCASES <−mo ran . mc ( cho$CASES , c ho_n br_w , 9 9 9 ) # W
morpermCASES <−mo ran . mc ( cho$CASES , c ho_n br_w b , 9 9 9 ) # B
morpermCASES
morpermCASES$res
#######################################################
#Moran s c a t t e r an d s p a t i a l co rr el og ra m p l ot s
#######################################################
pa r ( mfrow=c ( 1 , 2 ) )
mo ran . p l o t ( c ho$CASE S , c ho _nb r_ w , s pChk=NULL, l a b e l s =NULL,
x l a b ="CASES" , y l a b =" s p a t i a l l y la g ge d CASES" , q u ie t=NULL,
pch=19 , ma in ="Moran s c a t t e r p l o t ,
I=0 . 1 4 , p =0.04")
p l o t ( s pc o , m ai n=" S p a t i a l c o rr e lo gr am " )
#####################################################
# PROBABILITY MAPS
# c ho yn ow sk i ’ s a p p r oa c h
#######################################################
ch <−ch oy now sk i ( cho$CASES , cho$POP )
cho$ch_pmap_low <−i f e l s e ( ch$ t yp e , ch$pmap , NA)
cho$ch_pmap_high <−i f e l s e ( ! ch $t y pe , ch$pmap , NA)
pr bs <−c ( 0 , . 0 0 1 , . 0 1 , . 0 5 , . 1 , 1 )
cho$high =cu t ( cho$ch_ p map_hi g h , p rb s )
cho$low =cu t ( cho$ch_ pmap_low , pr bs )
s p p l o t ( c ho , c ( " l ow " , " h i gh " ) ,
c o l . r e g i o n s=g r e y . c o l o r s ( 5 ) )
# P oi ss on p r o b a b i l i t i e s
##################################################
pmap <−p robmap ( cho$CASES , cho$POP )
cho$pmap <−pmap$pmap
54
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
br ks <−c(0 ,0.001 ,0.01 ,0.025 ,0.05 ,0.95 ,0.975 ,0.99 ,0.999 ,1)
l i b r a r y ( R Co lorB rew er )
s p p l o t ( c ho , " pmap " , a t =br ks ,
c o l . r e g i o n s=g r e y . c o l o r s ( 5 ) )
s p p l o t ( c ho , " pmap " , a t =br ks ,
main=" Po is so n p r o b a b i l i t i e s " ,
c o l . r e g i o n s=r ev ( bre we r . p al ( 9 , " RdBu " ) ) )
# R e l a t i v e R is k
##########################################################
relRisk <−probma p ( cho$CASES , cho$POP )
cho$relRisk <−relRisk$relRisk
br ks <−c ( 5 , 4 2 , 7 9 , 1 1 5 , 1 5 2 , 1 8 9 , 2 2 6 , 2 6 3 , 3 0 0 , 3 3 6 )
l i b r a r y ( R Co lorB rew er )
# s p p l o t ( c ho , " pmap " , a t =br ks ,
c o l . r e g i o n s=g r e y . c o l o r s ( 5 ) )
s p p l o t ( c ho , " r e l R i s k " , a t =br k s ,
main=" R e l a t i v e R is k " ,
a x e s=TRUE, c o l . r e g i o n s =re v ( b re w er . p a l
( 9 , " RdBu " ) ) )
h i s t ( ch o$ re lR is k , m ain ="")
# Map o f Raw( cr ud e r a t e s )
##########################################################
raw <−probma p ( cho$CASES , cho$POP )
cho$raw <−raw$raw
br ks <−c ( 0 , 0 . 0 0 0 3 7 5 , 0 . 0 0 0 7 5 , 0 . 0 0 1 1 2 5 , 0 . 0 0 1 5 , 0 . 0 0 1 87 5 ,
0.00225 ,0.002625 ,0.003192)
l i b r a r y ( R Co lorB rew er )
# s p p l o t ( c ho , " pmap " , a t =br k s , c o l . r e g i o n s
=g re y . c o l o r s ( 5 ) )
s p p l o t ( c ho , " raw " , a t =br ks ,
main="Raw r a t e " , a x e s =TRUE, c o l . r e g i o n s
=re v ( b re w er . pa l ( 9 , "RdBu " ) ) )
h i s t ( c ho$raw , main ="")
# Map o f E x pec te d Count
###########################################################
expCount <−probma p ( cho$CASES , cho$POP )
cho$exp C ount <−expCount$expCount
br ks <−c ( 0 , 1 . 4 , 7 . 2 , 1 3 , 1 9 , 2 5 , 3 0 , 3 6 , 4 2 , 4 8 , 5 4 )
l i b r a r y ( R Co lorB rew er )
# s p p l o t ( c ho , " pmap " , a t =br ks ,
c o l . r e g i o n s=g r e y . c o l o r s ( 5 ) )
s p p l o t ( c ho , " e xp Co un t " , a t =br k s ,
main=" E x p ec t e d Co unt " ,
a x e s=TRUE, c o l . r e g i o n s =re v ( b re w er . p a l
( 1 1 , " RdBu " ) ) )
55
SPATIAL ANALYSIS AND MAPPING OF CHOLERA CAUSING FACTORS IN KUMASI, GHANA.
h i s t ( c ho $expCount , main ="")
##########################################################
# OLS Mod e ls
OLSmodela <−lm ( CASES ~ ND_RD +Den_RD +ND_DR, d a t a =cho )
OLSmodelb <−lm ( CASES ~ ND_RD +Den_RD +ND_CR, d a t a =cho )
summary ( OLSmodela )
summary ( OLSmodelb )
#############################################################
OLSmodelA<−l m ( CASES ~ ND_RD +ND_DR, d a t a =cho )
OLSmodelB <−lm ( CASES ~ ND_RD +ND_CR, d a t a =cho )
summary ( OLSmodelA )
summary ( OLSmodelB )
##############################################################
##CAR Mode ls
car1 <−s p au t o l m ( CASES ~ ND_RD +Den_RD +ND_DR, d a t a =cho ,
listw=ch o_ nbr _w , f a m i l y ="CAR" , method=" f u l l " , v e r b o s e=TRUE)
c a r 2 <−s p a ut o l m ( CASES ~ ND_RD +Den_RD +ND_CR, d a t a =cho ,
listw=ch o_ nbr _w , f a m i l y ="CAR" , method=" f u l l " , v e r b o s e=TRUE)
CARmodelA <−s p a u to l m ( CASES ~ ND_RD +ND_DR, d a t a =cho ,
listw=ch o_ nbr _w , f a m i l y ="CAR" , method=" f u l l " , v e r b o s e=TRUE)
CARmodelB <−s p au t o l m ( CASES ~ ND_RD +ND_CR, d a t a=cho ,
listw=ch o_ nbr _w , f a m i l y ="CAR" , method=" f u l l " , v e r b o s e=TRUE)
su mmar y ( c a r 1 )
su mmar y ( c a r 2 )
su mmar y ( CARmodelA )
summar y ( CARmodelB )
##############################################################
56
