This item is only available as the following downloads:
1 THE ROLE OF THE GLOBAL AIR TRAVEL NETWORK IN VECTOR BORNE DISEASE CONNECTIVITY AND SPREAD By ZHUOJIE HUANG A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIRE MENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2013
2 2013 Zhuojie Huang
3 To my Mom and Dad
4 ACKNOWLEDGMENTS Firstly I would like to express my gratitude to my primary collaborator, my committee chair and my mentor, Dr. Andrew Tatem. I feel lucky to have landed in his group two years ago as he provided me invaluable opportunities to study the cutting edge geospatial methods in epidemiology and medical geography. I am looking forward to continuing our collaborations in the future. I would also feel indebted and grateful for the prolong support from my co chair and my friend, Dr. Timothy Fik. Special thanks are given to Dr. Liang Mao, Dr. Youliang Qiu and Dr. Xiaohui Xu, without them m y dissertation will not be successful. Also, I would like to acknowledge the rest of our group: Deepa Pindolia and Andres Garcia, as well as my collaborators from the computer science department: Aniruddha Das and Udayan Kumar. Their suggestions and advice shaped my thinking on geography and helped me design my research. Additionally, I would like to thank to my friends and colleagues: Renee Bullock, Erin Bunting, Jac lyn Hall, Ian Kracalik, Jing Sun Sam Schramski, Ying Wang Andrea Wolf and Yang Yang. They gave me inspiration and valuable advice on my analysis. Special thanks to Xiao Wu who gave me important support. I would like to acknowledge the support from the geography department. Many thanks are given to Dr. Jane Southworth and Dr. Peter Waylen. Six years ago their decisions have changed m y life trajectory. I would also like to thank Dr. Jason B lackburn, Dr. Michael Binford and Dr. Corene Matyas. In addition, I would like to thank Dr. Michael Daniels in the D ivision of Statistics and Scientific C omput ation at the University of Texas at Aust in for his valuable advices for my model building process Their support has been i mportant for my graduate research.
5 This research would also not be possible without support from the Malaria Atlas Project at Oxford University. In particular, I would like to thank Dr. Pete Gething and Dr. Simon Hay. They provided wonderful data which helps form the foundation of my graduate dissertation. Final ackno wledgement should be dedicated to both my parents, Mr. Zhicheng Huang and Mrs Meifang Liang. Though they never go to a college, they always give me the best support they can afford. Also I want to thank my grandmother, Ms. Yinghua Guan. I was so grieved when she passed away while I could not come back to China to see her.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 11 ABSTRACT ................................ ................................ ................................ ................... 12 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 2 WEB BASED GIS: THE VECTOR BORNE DISEASE AIRLINE IMPORTATION RISK (VBD AIR) TO OL ................................ ................................ ........................... 17 Chapter Summary ................................ ................................ ................................ ... 17 Background ................................ ................................ ................................ ............. 18 Data ................................ ................................ ................................ ........................ 21 Air Travel Data ................................ ................................ ................................ 21 Disease D istribution D atasets ................................ ................................ .......... 22 Vector Distribution Datasets ................................ ................................ ............. 24 Climate and O ther D atasets ................................ ................................ ............. 26 Method ................................ ................................ ................................ .................... 27 Combing Datasets and Creating Risk Metrics ................................ .................. 27 Risk Assessments ................................ ................................ ............................ 29 VBD AIR Tool Architecture ................................ ................................ ............... 32 Results ................................ ................................ ................................ .................... 35 Discussion ................................ ................................ ................................ .............. 37 Conclusions ................................ ................................ ................................ ............ 42 3 AN OPEN ACCE SS MODELED PASSENGER FLOW MATRIX FOR THE GLOBAL AIR NETWORK IN 2010 ................................ ................................ .......... 48 Chapter Summary ................................ ................................ ................................ ... 48 Introduction ................................ ................................ ................................ ............. 49 Materials and Methods ................................ ................................ ............................ 53 Airport L ocations and S cheduled R outes ................................ ......................... 53 G ross D omestic P rod uct and P opulation I nformation ................................ ....... 53 Actual T ravel P assenger F low ................................ ................................ .......... 54 Data Processing ................................ ................................ ............................... 55 Model ................................ ................................ ................................ ................ 57 Results ................................ ................................ ................................ .................... 59
7 Model Comparison ................................ ................................ ........................... 59 Pre diction and Interpretation of the O D Passenger F low M atrix ...................... 61 Discussion ................................ ................................ ................................ .............. 62 Conclusion ................................ ................................ ................................ .............. 65 4 GLOBAL MALARIA CONNECTIVITY THROUGH AIRTRAVEL ............................. 70 Chapter Summary ................................ ................................ ................................ ... 70 Background ................................ ................................ ................................ ............. 71 Methods ................................ ................................ ................................ .................. 74 Airport Locations, Flight Routes and Passenger Flow Matrix ........................... 74 Malaria D istribution ................................ ................................ ........................... 74 Weighted Network Analysis and Community Detection ................................ .... 75 Results ................................ ................................ ................................ .................... 78 Connectivity within E ndemic A reas and to N on E ndemic A reas ....................... 79 Connectivity to Southeast Asia ................................ ................................ ......... 82 Discussion ................................ ................................ ................................ .............. 82 5 CONCLUSION ................................ ................................ ................................ ........ 92 APPENDIX A SUPPLEMENTARY FIGURES FOR CHAPTER 2 ................................ .................. 94 B SUPPLEMENTARY INFORMATION FOR CHAPTER 3 ................................ ....... 108 C SUPPLEMENTARY INFORMATION FOR CHAPTER 4 ................................ ....... 116 LIST OF REFERENCES ................................ ................................ ............................. 123 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 135
8 LIST OF TABLES Table page 3 1 Descriptions of covariates used in the mod eling process ................................ ... 66 3 2 Comparison of the four models with respect to prediction accuracy (in percentages) ................................ ................................ ................................ ....... 67 3 3 Root Mean Squared Errors and Mean Absolute Errors for all models ................ 67 4 1 Top 10 airports based on estimated relative malaria importation rates .............. 87 4 2 Top 10 airports based on estimated relative malaria exportation rates .............. 88 4 3 Top 10 P.falciparum betweenness centrality airports with their degrees in a sub network that only contains dire ct links from airports in P.falciparum or P.vivax endemic areas. ................................ ................................ ...................... 89 4 4 Top 10 P.vivax betweenness centrality airports with their degrees in a sub network that only contains direct links from airports in P.falciparum or P.vivax endemic areas. ................................ ................................ ................................ ... 89 B 1 Fit characteristics and variable effects for model 4 ................................ ........... 111 B 2 Fixed eff ects for variables in model 4 ................................ ............................... 112 C 1 Top 10 P.f. Routes and P.v. Routes outside of the Great Mekong Sub Region. ................................ ................................ ................................ ......................... 116
9 LIST OF FIGURES Figure page 2 1 2011 Air network data used in VBD AIR.. ................................ ........................... 43 2 2 Example d isease and vector distribution maps from the VBD AIR tool .............. 44 2 3 The architecture of the VBD AIR tool ... ................................ ............................. 45 2 4 Flow of user input for the VBD AIR tool .. ................................ ............................ 46 2 5 User input example for the VBD AIR tool A sample user input for the VBD AIR tool for a user interested in imported dengue infection risks to Miami in January.. ................................ ................................ ................................ ............. 47 3 1 Diagnostic plots for all the models ................................ ................................ .... 68 3 2 Predicted air traffic flows ................................ ................................ ................... 69 4 1 Spatial distribution of P.falciparum/P.viva x network communities overlaid on P.falciparum/P.vivax prevalence maps ................................ ............................ 90 4 2 Estimated relative P.falciparum/P.vivax flows originating from the Great Mekong sub region overlaid on P.falcip arum/P.vivax prevalence maps.. ........... 91 A 1 P. falciparum prevalence ................................ ................................ ................... 95 A 2 P. vivax endemic areas ................................ ................................ ..................... 96 A 3 Dengue suitability map.. ................................ ................................ ..................... 97 A 4 Dengue suitability in areas of known outbreaks in 2010 ................................ ... 98 A 5 Dengue outbreak prone. Dengue outbreak areas (niche model) ....................... 99 A 6 Yellow fever suitability ................................ ................................ .................... 100 A 7 Yellow fever outbrea k prone ................................ ................................ ........... 101 A 8 Chikungunya outbreak prone ................................ ................................ ........... 102 A 9 Aedes aegypti presence ................................ ................................ .................. 103 A 10 Predicted Ae. albopictus distribution ................................ ................................ 104 A 11 Anopheles distribution ................................ ................................ ..................... 105 A 12 D ata f low in the VBD Air ................................ ................................ .................. 106
10 A 13 Global accessibility map. ................................ ................................ .................. 107 B 1 Plots for predicted value vs. the predicted value at a log scale for all models 110 B 2 Plots for residuals vs. the predicted values at a log scale for all models ......... 111 C 1 The travel time/distance mask to extra ct the prevalent rate of P.falciparum /P.v ivax ................................ ................................ ........................ 118 C 2 Air travel network communities weighted by directed estimate flow ................ 119 C 3 Communities for all possible connections originating from P.falciparum /P.v ivax endemic areas. ................................ ............................... 120 C 4 Spatial distributions of airports with P.falciparum /P.v ivax betweenness centrality scor es ................................ ................................ ................................ 121 C 5 Spatial distributions of airport nodes weighted by incoming P.falciparum /P.v ivax flows.. ................................ ................................ .............. 122
12 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy T HE ROLE OF THE GLOBAL AIR TRAVEL NETWORK IN VECTOR BORNE DISEASE CONNECTIVITY AND SPREAD By Zhuojie Huang May 2013 Chair: Andrew J Tatem Cochair: Timothy J Fik Major: Geography Over the past century, the size and complexity of the air travel network has increased dramatically. Nowadays, there are 29.6 million scheduled flights per year and around 2.7 billion passengers are transported annually. The rapid expansion of the network increasingly connects regions of endemic vector borne disease with the rest of the world, resulting in challenges to health systems worldwide in terms of vector borne pathogen importation and disease vector invasion events To face these challenge s multidisciplinary approaches that incorporate spatial and network information are advocated for better understanding the role of the air travel network, specifically its flows and architecture, on the spread of vector borne disease. The research provides a series of innovative geospatial analysis and complex network analysis aiming to 1 ) Create a Web GIS for visual analytics on the risk of vector borne disease importation via air travel 2 ) Provide a modeled traveler flow matrix that describes how the air travel network connects airports all around the world. 3 ) Examine how network metrics and communities rel ate to malaria exportation and importation
13 CHAPTER 1 INTRODUCTION We travel faster and farther than any time in history, and so do diseases. Everyday, the air travel network transports over 7 million people to more than 40 ,000 destinations all around the world  35,000 direct scheduled routes are established on the air travel network, with 865 new routes added in 2011 [1,2] Total passenger numbers will reach 1.6 trillion in 2016 accompanied by an annual growth rate of 5%  These represent just small facets of the conti nuous expansion of the air travel network in the last century Expansion of the airline network has profound impacts on the epidemiologic al landscape. Studies have s hown that the connectivity of the air travel network can facilitate the global spread of d irectly transmitted diseases such as influenza and severe acute respiratory syndrome (SARS ) [3 6] This connectivity also greatly affects the global propagation of vector borne disease s  E ndemic area s and non endemic area s of vector born e diseases are more connected than ever Air highways are provided for an infected person or vector carrying pathogens travelling between spatially distant locations with speeds of 600 miles per hour Today, vectors and vector borne diseases are moving bet ween different regions at exceptional rates, resulting adverse consequences, such as disease importation [7,8] and species invasion [5,9,10] Tr aditional drawbridge strategies that rely on spatial barriers to stop the transmission of diseases now play a less important role and the challenge now exists to improve global disease management and health surveillance  To face these challenges, multidisciplinary approaches, that incorporate spatial and network information with spatially referenced disease data and models can provide a useful
14 platform for more evidence driven surveillance design and mitigation pl anning [7,8,11,12] Substantial evidence now exists of the role the air travel network plays in carrying both vector borne diseases and their vectors between distant locati ons [7,9] This has resulted i n imported cases [13 15] local outbreaks [16 19] in non endemic areas disease resurgence and re emergence in endemic areas [20,21] and the spread of drug and insecticide resistance  Up to 8% of travelers to the developing world become ill enough to seek healthcare u pon returning home, with a significant proportion of these suffering from vector borne infections  Travel clinical studies have suggested that dengue infections are now considered to be the most common cause of fever in returning travellers [14,21,24] Around 10,000 cases of imported malaria in malaria free high income countries are reported each year, but the true number of cases is likely to be over 25,000  Due to infrequent encounters, physicians in non endemic countries often have difficulties in diagnosing and treating travelers that suffer from vector borne infections when they come back from endemic ar eas [23,26] Evidence ha s shown that t he flow of people via air travel from endemic areas may increase the risks of re emergence or resurgence [20,21,27 30] of disease in previously vector borne disease free or low transmission areas  With imported vector borne infections placing a financial and operational burden on health sy stems in both endemic and non endemic countries, as well as the risk of onward transmission and even establishment, geospatial methods for quantifying the spatiotemporal risks of importation of both vector borne diseases and the vectors that carry them can be valuable for strategic
15 planning for vector borne disease elimination and control, and the allocation of limited control and surveillance resources. The availability of global maps of vector borne disease presence and prevalance [7,19,32 34] and vector distributions [35 40] now provide evidence bases for understanding vector borne disease risks across the world. Incorporating these high resolution maps with air travel network d ata offers great potential for infection importation risk assessments and vector borne disease spread modeling  This dissertation presents an initial effort towards this integration for analyzing the risks of vector borne disease transportatio n Two innovative approaches are proposed T he first is a visual analytic approach that utilizes a web based GIS : the Vector Borne Disease Airline Importation Risk Tool (VBD AIR) This tool aims to help better define the roles o f airports and airlines in the transmission and spread of vector borne diseases In this dissertation a n interactive web portal is created allowing users to quantify seasonally changing risks of vector and vector borne disease importation and spread by ai r travel to a certain airport, with decision support for mitigation strategy planning. Also, a centralized geographical database is set up, storing world wide geographic information on disease distribution, vector distribution, climatic information, popula tion density, air travel connectivity and network capacity which lays out foundations for the research in the following chapters The second method used is complex network analysis, which further consist s of two parts: flow estimation and network investig ation. Firstly, a global modeled air travel database is constructed to quantify the volume of passengers on the air travel network between two airports within two stops of air line transportation In this database, a irport
16 (node) characteristics such as deg ree, centrality, city population, and local area GDP are utilized in a spatial interaction model framework to predict the air transportation flows between node pairs This database provides an eviden ce base for scientists and researchers who are aiming to understand the compl ex spatial interaction between ori gin and the destination airports. Secondly, using these predicted flows on the air network, a malaria weighted network is created to describe and quantify the patterns that exist in passenger flows weig hted by malaria prevalences. In addition the connectivity within and to the Southeast Asia region where artemisinin drug resistance emergence threats appear highest, was examined to identify and highlight risk routes for its spread. This dissertation cont ributes to an on going initiative, the human mobility mapping project ( www.thummp.org ), aimed at better modeling human and disease mobility, and will form part of the long spatial scale aspect of continued multi modal assessments of vector borne movements, and assessment of malaria elimination strategies [41,42]
17 CHAPTER 2 WEB BASED GIS: THE VECTOR BORNE DISEA SE AIRLINE IMPORTATION RISK (VBD AIR) TOOL 1 Chapter Summary The rapid expansion of the network increasingly connects regions of endemic vector borne disease with the rest of the world, resulting in challenges to health systems worldwide in terms of vector borne pathogen importation and disease vector invasion events. Here we describe the development of a user friendly Web based GIS tool: the Vector Borne Disease Airline Importation Risk Tool (VBD AIR), to help better define the roles of airports and airlin es in the transmission and spread of vector borne diseases. Spatial datasets on modeled global disease and vector distributions, as well as climatic and air network traffic data were assembled. These were combined to derive relative risk metrics via air tr avel for imported infections, imported vectors and onward transmission, and incorporated into a three tier server architecture in a Model View Controller framework with distributed GIS components. A user friendly web portal was built that enables dynamic q uerying of the spatial databases to provide relevant information. The VBD AIR tool constructed enables the user to explore the interrelationships among modeled global distributions of vector borne infectious diseases (malaria. dengue, yellow fever and chik ungunya) and international air service routes to quantify seasonally changing risks of vector and vector borne disease importation and spread by 1 A version of this chapter has been published as Huang Z, Das A, Qiu Y, Tatem AJ (2012) Web based GIS: the vector borne disease airline importation risk (VBD AIR) tool. Int J Health Geogr 11: 33. doi:10.1186/1476 072X 11 33.
18 air travel, forming an evidence base to help plan mitigation strategies. The VBD AIR tool is available for test ing at www.vbd air.com VBD AIR supports a data flow that generates analytical r esults from disparate but comple mentary datasets into an organized cartographical presentation on a web map for the assessment of vector borne disease movements on the air travel network. The framework built provides a flexible and robust informatics infrastructure by separating the modules of functionality through an ontological model for vector borne disease. The VBD Air tool is designed as an evidence base for visualizing the risks of vector borne disease by air travel for a wide range of users, including planners and decisions makers based in state and local government, and in particular, those at international and domestic airports task ed with planning for health risks and allocating limited resources. Background Air travel has changed the epidemiological landscape of the world, providing routes from one side of the Earth to the other that can be traversed by an infected person in signif icantly shorter times than the incubation period of the majority of infectious diseases. This epidemiological impact has led to a rethink in global disease management  with pandemic control being less reliant on conventional spatial barriers as the global air network continues to expand. Today, vector borne diseases and vectors are moving between different regions at unprecedented rates, resulting i n adverse ecological, economic and human health consequences [7,9,44] Reducing these problems requires examination of how humans facilitate the movement and establishment of diseases a nd their vectors in new areas. Moreover, the speed of air travel has meant that rapid reporting and surveillance now play an important role in preventing the spread of diseases. Finally, the cost of surveillance makes sampling
19 design and the development of cost effective monitoring and testing approaches vital in effective early warning systems  While work on these factors is becoming more sophisticated for directly transmitted infections [4,46,47] our understanding of the role of air travel in global vector borne disease epidemiology remains relatively incomplete  Significant evidence exists that documents examples of both vector borne di seases and the vectors that carry them being transported between distant locations via air travel [9,23] The global air network enables many of the w and diverse ecosystems t o become connected and aids the movement of organisms, including disease vectors, to new habitats where they can become damaging invasive species, economically and health wise [9,40,48] Mosquitoes can survive moderately high atmospheric pressures aboard aircraft  and can be transported alive between international destinations, even in wheel bays  However, air travel likel y plays a much more significant role in moving the vector borne disease (via infected passengers) than in moving the vector. It provides rapid and wide reaching connections between outbreaks or high levels of endemicity and susceptible vector populations e lsewhere in the world. Up to 8% of travellers to the developing world become ill enough to seek healthcare upon returning home, with a significant proportion of these suffering from vector borne infections  Around 10,000 cases of imported malaria to high income countries are reported each year, but the true figure may be over 25,000  With imported vector borne infections placing a financial and operational burden on health systems in non endem ic countries (e.g.  ), as well as the risk of onward transmission and even establishment, the development of tools for quantifying the spatiotemporal risks of importation of both vector borne diseases and the vectors that carry them can
20 be valuable for assessing and guiding the allocation of limited control and surveillance resources. Recent efforts in the field of global mapping of vector borne diseases (e.g. [32,34] ) and the vectors that carry them (e.g. [36,37,51] ), generally through using sample data of known disease or vector presence in combination with environmental covariates, now provide strong evidence bases for determining vector borne disease risks across the world. The linkage of these disease and vector distribution maps with air travel network data offers great potential for infection importation risk assessment and the modelling of vector borne disease spread. Such distribution maps, however, represent static pictures of relatively long term (>1 year) disease prevalence and vector presence. If importation and onward spread risks are to be accurately quantified, the substantial climate driven seasonal fluctuations in disease risk and vector densities need to be accounte d for  If a vector or an infected individual arrives in a new location via air travel, the risk of the vector establishing or the infection being passed on to local vector populations is often dependent upon the month of arrival. The arrival of an individual infected in a mosquito borne disease outbreak occurring in January in the southern hemisphere (e.g. Sao Paulo) into a city in the northern hemisphere (e.g. New Y ork) will present little risk of onward transmission, due to the cold January climate being not conducive for mosquito activity. However, a similar arrival from e.g. India (where climatic conditions may be suitable for year round transmission) in July pres ents a much greater risk  By utilizing gridded climate data to measure climatic similarity between origin and destination locations with known presence of a suitabl e vector, and adjusting for flight
21 passenger numbers as an additional measure of risk, these factors can be accounted for [9,10] Here we introduce the Vector Borne Disease Airline Importation Risk tool (VBD AIR; www.vbd air.com ), which brings together global vector borne disease and vector distribution maps, climate data and air network traffic information to inform on the spatiotemporal risks of disease and vector importation. The VBD AIR tool takes the form of an interactive online in terface and is targeted at users with interests in specific airports or regions, and the risks to those locations of vector borne disease importation and onward spread, or exotic vector importation and establishment. Data VBD AIR utilizes an entity relati onship model to integrate data sources from airport locations and air routes, disease and vector distributions, global climate data, and global land based travel time data. All of these datasets are described below. Air Travel D ata Information on a tota l of 3,632 airports across the world, together with their coordinate locations was obtained using Flightstats ( www.flightstats.com ) and is mapped in Figure 2 1a. Information on the airport name, IATA code, city an d country are all stored in the VBD AIR database that was constructed for the tool (see methods). Flight schedule and seat capacity data for 2010 and 2011 were purchased from OAG (Official Airline Guide, www.OAG.com). These included information on origin a nd destination airports, flight distances, and passenger capacity by month of each year. All routes used in the tool are mapped in Figure 2 1b.
22 Disease D istribution D atasets The current version of VBD AIR at the time of writing focuses on four vector born e diseases and their related vectors, chosen due to the availability of spatially referenced data for map production and rates of importation by air travel: Malaria ( Plasmodium falciparum and Plasmodium vivax ), dengue, yellow fever and chikungunya, all tra nsmitted by mosquitoes. The methods behind the construction of each of these datasets are described briefly here, with an example map that is shown in Figure 2 2a, and the remaining output maps and mapping process presented in Appendix A Plasmodium falci parum is a protozoan parasite, one of the species of Plasmodium that cause malaria in humans. P. falciparum is the most dangerous of these infections as P. falciparum (or malignant) malaria has the highest rates of complications and mortality, while P. viv ax is the most frequent and widely distributed cause of recurring (tertian) malaria. Non endemic countries often see many cases of imported malaria each year through travellers or returning migrants. The geographical distribution of predicted P. falciparum malaria endemicity in 2010 was obtained from the Malaria Atlas Project (www.map.ox.ac.uk) and the methods behind the construction of the map are presented in Gething et al  In brief, 22,212 community prevalence surveys were used in combination with model based geostatistical methods to map the prevalence of P. falciparum globally within limits of transmission estimated by annual parasite incidence and satellite covariate data  The mapping and modeling framework for P. vivax is presently not as advanced as for P. falciparum however, the limits of transmission and stable/unstable endemicity have been mapped  and this dataset was used within VBD AIR to define areas of P. vivax malaria endemicity.
23 Dengue fever is an infectious tropical disease caused by the dengue virus. The incidence of dengue fever has increased dramatically since the 1960s, with around 50 100 million people infe cted yearly, and imported cases through air travel to non endemic regions are on the rise  Global dengue distribution was mapped in three different ways: (i) a map of environmental 'suitability' for transmission (ii) the same map, but only with countries/regions of known recent transmission shown, and (iii) a map of environmental similarity to recent outbreaks. For construction of the first map, thousands of geographically located data points on dengue transmiss ion over the past decade were combined with climatic and environmental covariates within a boosted regression tree mapping tool  following the approaches of Sinka et al [36,37,51,56] to produce a global map of suitability for dengue transmission ( Appendix A ). To refine this map to be focused solely on regions of recent confirmed transmission, the second dengue map was constructed with dataset masked so that only regions cited by the CDC Yellow Book ( http://wwwnc.cdc.gov/travel/yellowbook/2012/chapter 3 infectious diseases related to travel/dengue fever and dengue hemorrhagic fever.htm ) as having transmission in 2010 were left ( Appendix A ). Finally, to produce an alternative, more contemporary dataset focused on regional suitability for significant outbreaks, geographical data on outbreaks occurring since 2008 were extracted from Healthmap ( www.healthmap.org ) and again combined with climatic and environmental covariates in a boosted regression tree tool  to map predicted dengue fever outbreak risk ( Appendix A ). Yellow fever is an acute viral h emorrhagic disease  The yellow fever virus is transmitted by Aedes a egypti (and other species) and is found in tropical and
24 subtropical areas in South America and Africa, but not in Asia. Since the 1980s, the number of cases of yellow fever has been increasing, making it a re emerging disease, and case numbers imported thr ough air travel to non endemic areas have been rising  Global yellow fever distribution was mapped in two different ways: (i) a map of environmental suitability for transmission and (ii) a map of environmental similarity to recent outbreaks. Hundreds of geographically located data points on yellow fever occurrence over the past twenty years were combined with climatic and environmental covariates within a discriminant analysis mapping framework to produce a global map of suitability for yellow fever transmission. The datasets and methods used are described in Rogers et al  The risk map of environmental similarity to recent outbreaks was produced using the same methods as f or the dengue outbreak similarity map described above, but using yellow fever outbreak data from Healthmap ( Appendix A ). Chikungunya virus is an insect borne virus, of the genus alphavirus that is transmitted to humans by virus carrying Aedes mosquitoes, p rincipally Ae.aegypti and Ae.albopictus Large sporadic outbreaks have occurred since around 2005, with spread associated with both the movement of infected air travelers and the spread of the range of Ae. a lbopictus  An outbreak risk map was produced using the same methods as outlined above for the dengue outbreak similarity map, but using chikungunya outbreak data since 2008 from Healthmap ( Figure 2 2a). Vector Distribution D atasets VBD AIR includes three global vector distribution maps: Aedes aegypti, Aedes albopictus and the dominant Anopheles vectors of malaria. An example map is shown in Figure 2 2b for Ae. albopictus and the remaining ma ps are shown in Appendix A with the mapping process described briefly here. More complete details on the mapping
25 process are provided in the VBD AIR user guide, available through the online tool (www.vbd air.com). The yellow fever mosquito, Aedes aegypti is a mosquito that can spread dengue fever, chikungunya and yellow fever viruses, as well as other diseases. The mosquito originated in Africa but is now found in tropical and subtropical regions throughout the world. Hundreds of geographically located da ta points on field caught Ae aegypti occurrence over the past 10 years [58,59] were combined with climatic and environmental covariates within a boosted regression tree mapping framework  following the approaches of Sinka et al [36,37,51] to produce a global map of predicted Ae aegypti presence ( Appendix A ). Ae. albopictus is of medical and public hea lth concern because it has been shown in the laboratory to be a highly efficient vector of 22 arboviruses, including dengue, yellow fever, and West Nile fever viruses  In the wild, however, its efficiency as a vector appears to be generally low, although it has been implicated in recent dengue fever and chikungunya outbreaks in the absence of the principal vector, Ae. aegypti. From its Old World East Asian distribution reported in 1930, Ae. albopictus expanded its range first to the Pacific Islands and then, within the last 20 years, to other countries in both the Old World and the New World, principally through ship borne transportation of eggs and larvae in tires [9,60] Here, hundreds of geographically located data points on field caught Ae. albopictus occurrence over the past 10 years [39,58,60] were again combined with climatic and environmental covariates within a boosted regression tree mapping framework [36,37,51,56] to produce a global map of predicted Ae. albopictus presence (Figure 2 2b).
26 Anopheles is a genus of mosquito. There are approximately 460 recognized species: while over 100 can transmit human malaria, only 30 40 commonly transmit parasites of the genus Plasmodium which cause malaria in humans in endemic areas. Anopheles gambiae is one of the best known, because of its predominant role in the transmission of the most dangerous malaria parasite species Plasmodium falciparum. Thousands of geographically referenced data points on the presence of the dominant Anopheles vectors of malaria have been gathered and used, in combination with env ironmental covariates, expert opinion maps and regression tree tools, to produce global maps of Anopheles distributions by the Malaria Atlas Project ( www.map.ox.ac.uk ) [36,37,51,56] Here, these maps were combined to produce a global map of dominant malaria vector presence ( Appendix A ). Climate and O ther D atasets The principal climatic constraints to vector survival (mosquitoes in this case), development and the vector borne disease life cycle within them are temperature, rainfall and humidity. Monthly gridded global data on each of these were obtained from the CRU CL 2.0 gridded climatology datasets ( http://www.cru.uea.ac.uk/cru/data/hrg/ )  The climate information from these gridded data was extracted for the locations of each of the airports. A dataset depicting travel time to the nearest major settlement (with population size > 50,000) was obtained (further details here: http://bioval.jrc.ec.europa.eu/products/gam/index.htm ) to provide contextual information on (i) airport disease accessibility at the origin and (ii) the potential ease of disease spread upon arrival. The risk of a disease being imported to a new location should not only be quantified by the level of predicte d risk at the location of the airport,
27 since travellers often travel long distances to get to an airport, often coming from more rural regions where disease risk may be higher. Therefore, the land based travel time dataset described above was used in combi nation with each disease distribution map to extract the maximum predicted disease risk within two hours and fifty kilometers of each airport (see Appendix A for more detail). There exists no globally comprehensive information on travel distances to airpor ts, thus we used this simple assumption here. The maximum level of disease risk within these travel times and distances were assigned to each Method VBD AIR is designed to be a flexible tool that combines multiple geospatial datasets to inform on the rela tive risks between differing airports, flight routes, times of year, diseases, and their vectors, in promoting the movement of passengers infected by vector borne diseases and the vectors that spread these diseases. In general, the tool relies on the assum ptions that the levels of imported vector and vector borne disease risk via air travel are related to (i) the presence of flight routes connecting to endemic regions (promoting the movement of people, pathogens and vectors), (ii) the level of traffic betwe en origin and destination (increasing the probability of infected passenger and vector carriage), and (iii) the monthly climatic similarity between origin and destination (since vector activity is required at both locations to firstly provide infected pass engers, and secondly prompt onward transmission or vector establishment at the selected destination). Combing D atasets and C reating R isk M etrics VBD AIR utilizes an entity relationship model to integrate the data sources from airport locations and air rout es, disease and vector distributions, global climate data,
28 and the global land based travel time data. VBD for knowledge dissemination between authoritative data producers and expert users, as it consumes data from a spatial data infrastructure and produces analytical output  To generate attributes for each airport from the spatial datasets, airports were setup as the report ing agents and implemented with an object oriented notation for extracting field data properties. A set of climate related indices that have been outlined elsewhere [48,63] was used here as metrics for imported vector and disease establishment risk. These rely on the assumption that the climatically sensitive disease vectors req uire similar climatic conditions in their new locations to that which they experienced and survived in at their previous home locations in order to successfully establish. Moreover, for imported cases of vector borne diseases to result in onward transmissi on in new locations, climatic conditions that promote vector activity, similar to the origin location where the disease was contracted, are required. Three simple indices were calculated: 1. Climatic Euclidean Distances (CEDs) are a measure of similarity in climatic regime between one location and another, and in this case were calculated through obtaining measures of rainfall ( r ), temperature ( t ) and humidity ( h ) for each airport location for each month from the gridded climatic datasets described previously The CED between airport i and j was then calculated by [48, 63] 2. Climatic Euclidean Distance scaled by passenger volumes (or traffic, t ) (CEDt), provides a more relevant relative measure of insect/disease introduction risk and consequent establishment/spread by route, through taking into account not only climatic suitability between regions, but the level of traffic on connecting flight routes. CEDt is calculated as in CED above, but the resulting values are normalized to the 0 1 range and multiplied by the traffic levels on the route in question, which have been normalized through dividing by the maximum traffic value in the database [4 8,63] 3. Climatic Euclidean Distance scaled by passenger volumes and 'risk' in terms of predicted disease endemicity or probability of vector presence (CEDtr) at the flight origin, provides a more relevant relative measure of insect/disease
29 introduction ris k to non endemic/vector free regions and consequent establishment/spread by route, through accounting for not only volume of traffic, but disease/vector prevalence at origin locations. CEDtr is calculated as in CED above, but the resulting value is normali zed and multiplied by the normalized traffic levels on the route in question and by predicted disease endemicity or vector presence probability at the origin location, again normalized to the 0 1 range [48,63] Risk A ssessments Three VBD AIR tool and the fo llowing paragraphs describe the rationale and content of each of these assessments, provide examples on getting user specified outputs and document s the caveats and limitations of each assessment. Imported vector borne disease case risk assessment : These are aimed at p roviding estimates for the relative risks between scheduled flights of incoming air passengers carrying cases of the user selected disease. Two simple measures are calculated for the selected airport, disease and month: (i) s cheduled passeng er capacity for 2011 on all routes coming from endemic or outbreak risk regions of the selected disease; (ii) these passenger capacity numbers normalized through dividing by the maximum traffic value in the database and rescaled by the disease risk value i n the region of origin The first metric provides a simple measure of the maximum number of passengers arriving each month from areas of the world where transmission of the selected disease is either known to be endemic, predicted to occur or has occurred in the past. Comparing these between routes and months provides a first pass measure of risk route and timing prioritization. The second one provides a refinement to the simple measure in (i) that incorporates information on disease endemicity (on a 0 1 sc ale) at the origin.
30 The risk values calculated are based solely on scheduled incoming flight routes in 2011, the traffic capacity on these routes and the estimated endemic disease risk at the origin airport region. These estimates do not take into account additional risk modifying factors such as actual passenger numbers, traveler activities and prophylaxis use, seasonal variations in disease transmission, chartered flights or multiple stopovers. Further details can be found in the VBD AIR user guide and th e user generated reports available from the online tool ( www.vbd air.com ) Onward transmission risk assessments are aim ed at providing basic estimates for the relative risks between scheduled routes of incoming fligh ts bringing infected passengers, and th o se passengers coming into contact with active, competent vectors to facilitate onward transmission. Two simple measures are calculated for the user selected airport, disease and month: (i) Flight capacities rescaled by climatic similarity between origin and destination regions for flights originating in disease endemic or outbreak prone regions (CEDt) This metric is based on the assumption that the risk of infected passenger arrival and onward disease spread is relat ed to the amount of traffic between locations (increasing the probability of disease carriage) and also the similarity of the climate at the destination to that of the origin, since vector activity is required at both locations to firstly provide infected passengers, and secondly prompt onward transmission at the selected destination. (ii) The previous metric rescaled by the disease endemicity/risk value r, at the origin region (CEDtr) [9,10] This provides additional refinement of the previous metric to account for spatial v ariations in disease risk across the world. Finally, the opportunity to overlay a map depicting travel time to the nearest major settlement is available, to provide contextual information on (i)
31 airport disease accessibility at the origin and (ii) the pote ntial ease of disease spread upon arrival. The risk values calculated are based on scheduled incoming flight routes in 2011, the traffic capacity on these routes, the climatic similarity to origin regions and the predicted presence of the disease at the or igin airport region and competent vector at the destination airport region. These estimates do not take into account additional risk modifying factors such as passenger numbers, vector preferences, passenger activities, seasonal variations in disease vecto r population sizes, or vector control measures in place. Imported vector risk assessments is aimed at providing basic estimates for the relative risks between scheduled routes of incoming flights bringing exotic disease vectors and their consequent establi shment. Two simple measures are calculated for the user selected airport, vector and month: (i) Flight capacities rescaled by climatic similarity between origin and destination regions for flights originating in vector presence regions. This metric is base d on the assumption that the risk of exotic vector arrival and establishment is related to the amount of traffic between locations (increasing the probability of carriage) and also the similarity of the climate at the destination to that in its home range, accounting for seasonal variations. (ii) The previous metric rescaled by the vector suitability value r at the origin region (CEDt r ) [7,9,10] This provides additional refinement of the previous metric to acco unt for spatial variations in vector suitability and abundance across the world. The risk values calculated are based on scheduled incoming flight routes in 2011, the traffic capacity on these routes, the climatic similarity to origin regions and the
33 which is a collection of data entit ies and their relationship s generated from the data server. This data assimilation process includes an object oriente d mapping procedure and add ons of analytical logics. The controller also has the responsibilities to call the map server for the overlays of base maps and raster spatial dataset layers. The data server is an SQL database employed in a snow flake structu re as the airport table is assigned as a fact table in the database. This table contains geographical information on the 3,6 8 2 airports. All disease, vector route capacity and climate spatial data are mapped into this relational database management syste m The data have been rectified for occasional misalignment error using ArcGIS v10 This schema simplif ies the data structure for queries on airports and facilitates the data exchange process and interoperation. The communication between the web server an d the data server is largely based on a repository pattern with a predefined data model. The repository pattern is a middleware container between the study objects and the data. It is capable of aggregating data collections using precompiled queries to the database  In general, the repository pattern provides a simple querying structure for intermediate data and facilitates the data exchanges The overlay of the world map from either Google Maps or OpenStreetMap, which illustrate the viewing extents for the significant air routes. Furthermore, it provides functions such as an auto zoom to the viewing extent when a map is generated or an airport is selected. The web mapping services for raster spatial datasets host the map files for disease and
34 vector distributions on a GeoServer. This service is able to stream the raster dataset to W eb M apping S ervices (WMS) and the user can then see the raster dataset overlay in their browser Once the user opens the VBD AIR website (w ww. vbd air .com), the Openlayer library with preset visualization is generated from the base map services and a raster dataset overlay is generated from the Geoserver. Feature layers for airports and a ir routes are populated from the JSON string streaming from the controller and th is JSON string is generated according to the user input on the html form of input choices, which contains airport information, disease/vector probabilities, flight passenger capacity information and climatic conditions. Finally, these feature layers are visualized and overlaid on the base map. Supported 'mouse events' include mouse driven map navigation and automated air route re rendering. Users can navigate the mouse pointer to an airport or an air route to obtain more detailed information and select different rendering scenarios on imported risks or onward transmission. When a metric is chosen, the routes are colored to distinguish between where the metric is larger than the average for all routes shown, and where it is smaller. The t op 10 routes ranked by the user's chosen metric are also displayed As the development of VBD AIR has and will continue to involve multidisciplinary effort, we have implement ed a test driven desi gn routine to utilize the MVC framework so as to minimize the cost for communication. T est Driven Development (TDD) permits flexibility for a more feature driven development. A feature c an be treated as a separable function in a project  T DD helps in separating the concerns of various
35 implementations of features for instance, the addressing of raster data ove rlay and PDF report generation should rely on different programming libraries, and th us, they have totally different input and output. TDD can state the objectives of these two features by dependency injection, which uses an interface as a contract to decl are functionalities T he contracting interfaces could be substituted with workable class es to implement the functionality. In the implementation, different programming resources can be imported to fulfill the requirements. TDD enables the design of reusabl e features that can be used across multiple development efforts. Results The test version of VBD AIR is available at www.vbd air.com. In the current version, VBD AIR enables users to explore the interrelationships between the global distribution s of the f our major vector borne infectious diseases, seasonal climatic changes and seasonally changing air traffic capacities, and the full set of user inputs available in outlined in F igure 2 4 The visualization of the air travel network within VBD AIR follows Schneiderman s pri nciple for data visualization : Overview first, zoom and filter, then details on demand  The w eb interface provides a map view of the flight connection s from endemic disease or vector presence area s to a user selected airport, with a predicted disease r isk or vector p resence map overlaid ( F igure 2 5) The user can view and navigate through the flight routes from each of the airport s with in the endemic or vector presence regions. As the user select s the ir choice of risk assessment type (imported disease r isks, onward transmission risks or imported vector risks see F igure 2 4) the view supports an automatic coloring scheme to render the origin airports and the route s based on the user input and the objective s for risk assessment. Moreover,
36 w hen the user clicks on an airport, detailed location information for that airport is provided The interface is also well equipped with a comprehensive user guide a short tutorial and integrated pop up help windows so that the user can get a better understanding of th e tool functionality, datasets and outputs For both the direct and indirect flight options ( F igure 2 4 ) user s are required to input the name or city of the ir airport of interest disease of interest, and the ir related choice of disease distrib ution or vector pre sence map. The direct flight option s provide imported disease risks and onward transmission risks functions if the user select s a disease map and imported vector risks if the user select s a vector map. The i mported disease risk op tion provide s the top 10 routes by route capacity f ro m disease endemic area s and the top 10 routes by risk scaled traffic from disease endemic areas. The onward transmission risks and imported vector risks provide top 10 routes by climatic similarity scal ed traffic and top 10 routes by climate similarity and disease /vector risk scaled traffic. The o ption to produce a more detailed PDF report output is provided to enable users to examine output statistics in greater detail and view longitudinal data An opt ion is also available to view indirect flights to the airport of interest, which highlights the possible routes from disease endemic areas via a single flight transfer, and a list of intermediate airports A sam ple user input is presented in F igure 2 5 I n this exa mple, the user is interested in the risks of dengue importation to Miami airport in January by direct flights. Once the user selects Miami airport through the auto complete text box, the map zooms to show the region (Figure 2 5a). The user then s elects dengue from the list of diseases available and chooses the spatial representation of dengue risk they are interested in
37 (see Data section above for information on each map), and can click on the question mark icon to get help in choosing, which in t urn links to the user guide to obtain further information. The user then selects January from the drop down month list and chooses direct flights, before choosing to display their selections. The next screen (Figure 2 5b) k map with all direct flights from endemic areas to Miami overlaid and colored by whether the traffic on the route is greater or less than the average traffic on all routes to Miami from endemic areas. The user can then select metrics from the imported dis ease risks box in the top left, choosing whether to view the top 10 routes by traffic from endemic areas or by traffic scaled by disease risks at the origin location. In each case, the top 10 risk routes are displayed in the box in the top right. For more detailed quantification of imported disease risks metrics, the user has the option to view a customized PDF format report. Finally, the user can view all those flights that are connected to Miami from endemic areas by a single flight change through returni (Figure 2 5c). Discussion With no apparent end in sight to the continued growth in global air travel, we must expect the continued appearance of disease spreading vector invasions an d vector borne disease movement. Approaches that can inform decision makers on the risk factors behind these importations and onward spread risks can be used to focus surveillance and control efforts more efficiently. This paper and the VBD AIR tool it des cribes show that multiple datasets on many aspects relating to the risk of movement of insect vectors and vector borne diseases through the global air network can be compiled to provide such information through a user friendly web tool. The principal
38 funct ion of the VBD AIR tool is to provide an evidence base for assessing the role of air travel in the spread of vector borne diseases and their vectors through available spatial data. The VBD AIR tool is designed with a wide range of potential users in mind. These include planners and decisions makers based in state and local government, in particular, those at international and domestic airports tasked with planning for and dealing with health risks and allocating limited resources. It is clear from explorati on of outputs from VBD AIR that each region, airport and flight route has a differing risk profile in terms of disease and vector importation, determined largely by the structure of the air network and its congruence with infectious disease distributions a nd outbreaks and vector distributions and seasonality, yet this is rarely quantified and used when control methods and surveillance are considered. The VBD AIR tool show s that a multi disciplinary approach, which draws on a variety of spatial data on facto rs known to infl uence the spread of vectors and the diseases they carry, offers potential for assessing the risk of disease im portation. A range of uncertainties and limitations do still exist in the datasets and outputs presented however, and users are ma de aware of these within the full user guide and throughout the information boxes within the tool. Firstly, VBD AIR considers only direct flights and their capacities within metric calculations rather than actual passenger numbers or stopovers, and users are made aware of the uncertainties that this entails [6 8] Within the disease and vector distribution modeling processes, uncertainties are inherent throughout  partic ularly in those regions with little field data to inform predictions. Moreover, we have treated the vector distributions as single homogenous type s of mosquit o, yet competition, competence, adaptation and pre ferences can vary widely
39 across their global dis tribution s [36,37,51,60] Accurate data on outbre ak locations and sizes, as with many diseas es, are also difficult to obtain to be sure of comprehensive assessments of risk, however, improvements in global surveillance and the rapid availability of data are improving (e.g.  ). Also, the distance between two airports and the population size of these airports have not been explicitly incorporated in the risk assessments here P roximity to endemic area plays an important role in vector borne diseas e importation  while city population size can be utilized to estimate the rate of disease movement between pairs of airports  Further the use of climatic similarity measures may not be appropriate to certain contex ts, such as those where arid climatic conditions prompt increased water storage, leading to rises in vector borne disease risks, rather than the decreases that may be indicated by CEDs. Finally, how to interpret and act upon the kind of relative risks ide ntified in VBD AIR is a challenge yet to be overcome but various approaches to mitigating risks are presented within the tool text boxes and user guid e  Future updates are planned to VBD AIR that will expand its ca pabilities as the MVC framework is designed to be flexible for robust expansion. These will include: (i) Regular updates of the disease and vector distribution maps, as new survey data and outbreak reports become available ; (ii) u pdates to the flight dat a as new information on flight capacities and routes becomes available ; (iii) a dditional scenario related functionality will be built into the tool based on a set of control and mitigation options. This will provide users with guidance on approaches to lim iting imported cases and vectors, and mitigating the effects of vector establishment or onward disease transmission ; (iv) t he interactive incorporation of the accessibility datasets. At present
40 this represents a simple visualizatio n of access to provide co ntext, and u pcoming extension s will focus on building in measures of access to better capture airport catchment areas and estimate likely regions impacted by imported cases or vectors ; (v) t he incorporation of extra vect or borne diseases and vectors The c hoice of additional diseases and vectors will depend upon availability of sufficient spatial data for mapping or validated global maps. Candidate diseases include leishm aniasis, Rift Valley fever and c hagas disease It is envisioned that future research be yond the simple updates and tool expansions described above will build upon VBD AIR to continue to improve quantification of these aspects, drawing on newly developed spatial datasets and mathematical models of transmission, to provide an evidence base to enable airports, airlines, and public health officials to assess the appropriateness and efficacy of current control, surveillance and treatment practices, and tailor strategies to these differing risk profiles for each disease, route and airport. Three sp ecific areas of research should be examined: (i) Constructing geospatial information databases on global endemic disease distributions, and building a framework for the rapid inclusion of outbreak reporting data from surveillance databases such as HealthMa p ( www.healthmap.org ) [70,73] Increasingly, spatial information on the prevalence of directly transmitted and insect borne diseases are being made available, and approaches for using these data to build distribution maps and dynamic transmission models are following. The potential of combining such data with air traff ic data for forecasting disease movements has been
41 shown for a handful of diseases in specific locations, but this potential has yet to be realized at global scales. (ii) Increasing the sophistication of flight passenger movement data and models Existing models of disease movement over air networks are generally driven by flight capacity and direct flight data  missing valuable information on stopovers, actual passenger numbers and lengths of stay. Sample datasets on ticket sales and flight occ upancy (e.g. US Transtats T100 and DB1B data http://www.transtats.bts.gov ) should be utilized to derive models that can better replicate realistic passenger flows, for integration with the disease risk data  Also, the incorporate of proximity measure (such as geograph ical distance or flight time between pairs of airports ) and population information which airports serve is likely to facilitate the estimations of the actual travel flows between two airports [71,72 ] (iii) The development of stochastic analogues of existing deterministic approaches to modeling of disease movement through air networks that are capable of handling input parameter distributions rather than simple mean values, and provide measures of u ncertainty with output forecasts. The process of disease importation is a stochastic process, and, depending upon the disease, each relevant variable (e.g. seasonal variations in transmission, passenger numbers, and infection risk) can exhibit substantial variations from the mean and include uncertainty in the way they are measured. By simulating risks of importation from literature derived probability distributions for each variable, improved and more informative model outputs could be produced that would enable the user to better understand and manage the uncertainties inherent in forecasts (e.g.  ).
42 Conclusions Incr eases in global travel are happening simultaneously with many other processes that favor the emergence of disease [75,76] Air travel is a potent force in disease emergence and spread, and the speed and complexity of modern aviation of disease control and quarantine increasingly irrelevant  With no apparent end in sight to the continued growth in global air travel, we must expect the continued appearance of disease vector invasions and vector borne disease movement. Approaches that can inform decision makers on the risk f actors behind vector borne disease importation and onward spread risk can be used to focus surveillance and control efforts more efficiently. The VBD AIR tool shows that multiple datasets on many aspects relating to the risk of movement of vector borne dis eases and their vectors through the global air network can be compiled to provide such information. VBD AIR is available at www. vbd air .com
43 Figure 2 1. 2011 Air network data used in VBD AIR (a) The location of the 3,632 airports across the world; (b) the flight routes for 2011 captured within the VBD AIR tool.
44 Figure 2 2. Example d isease and v ector distribution m ap s f rom the VBD AIR tool The predicted distribution of (a) chikungunya outbreak risk based on geolocated data on recorded outbreaks sin ce 2008 combined with satellite derived environmental covariates within a boosted regression tree species distribu tion prediction model. The colo r scale shows predicted unsuitable to suitable conditions for outbreaks as a continuous scale from yellow to da rk blue. (b) climatic and environmental suitability for Aedes albopictus presence based on field survey data combined with satellite derived environmental covariates within a boosted regression tree species distribution prediction model. The colour scale shows predicted unsuitable to suitable conditions as a continuous scale from yellow to dark blue.
45 Figure 2 3 The architecture of the VBD AIR tool VBD AIR adopts a three tier design: Web Server Tier, Data Server Tier and a Map Service Tier. The web in terface follows AJAX standard s The web server implements a Model View Controller (MVC) framework. The data server is an SQL database, employed in a snow flake structure and interpolated in the web server as data entities. The Web Mapping Services hosts th e coverage file for predicted disease and vector distributions on a GeoServer. T he b ase m ap s ervice retrieve s maps from Google and OpenStreetMap.
46 Figure 2 4. Flow of user input for the VBD AIR tool Risk assessment functions for the VBD AIR tool Imp orted disease risks onward transmission risks and imported vector risks are provided for the direct flight route scenario and the possible routes from endemic area via one transfer are provided for the indirect flight scenario
47 Figure 2 5. User i nput example for the VBD AIR tool A sample user input for the VBD AIR tool for a user interested in imported dengue infection risks to Miami in January a) The user selects their airport, disease and month of interest; b) The result shows the direct flight routes from den gue suitable area s to Miami and the user can select risk assessment s that include imported disease risk and onward transmission risks ; c ) The result shows the direct and one transfer flight routes from dengue suitable area s to Miami a nd t he user can navigate through all the results for more detailed information
48 CHAPTER 3 AN OPEN ACCESS MODELED PASSENGER FLOW MATRIX FOR THE GLOBAL AIR NETWORK IN 2010 2 Chapter Summary The expanding global air network provides rapid and wide reaching conne ctions accelerating both domestic and international travel. To understand human movement patterns on the network and their socioeconomic, environmental and epidemiological implications, information on passenger flow is required. However, comprehensive data on global passenger flow remain difficult and expensive to obtain, prompting researchers to rely on scheduled flight seat capacity data or simple models of flow. This study describes the construction of an open access modeled passenger flow matrix for all airports with a host city population of more than 100,000 and within two transfers of air travel from various publicly available air travel datasets. Data on network characteristics, city population, and local area GDP amongst others are utilized as covar iates in a gravity model framework to predict the air transportation flows between airports. Training datasets based on information from various transportation organizations in the United States, Canada and the European Union were assembled. A log linear m odel controlling the random effects on origin, destination and the airport hierarchy was then built to predict passenger flows on the network, and compared to the results produced using previously published models. Passenger flows between 1491 airports on 644,406 unique routes were estimated and analyses showed that the model presented here produced improved predictive power and accuracy compared to 2 A version of this chapter was submitted to Plos One as Huang Z, Wu X, Garcia AJ, Fik TJ, Tatem AJ (2013) An open access modeled passenger flo w matrix for the global air network in 2010. PloS One
49 previously published models, yielding the highest successful prediction rate in terms of prediction power and accuracy at the global scale. The airport node characteristics and estimated passenger flows are freely available as part of the Vector Borne Disease Airline Importation Risk (VBD Air) project at: www.vbd air.com /data Introduction Demand for travel has boosted the growth of the global air travel network at an unprecedented rate. In the past 20 30 years, the network has expanded dramatically with a steady growth rate of 4 5% per year  accompanied by a nearly 9% annual growth rate of passenger and freight traffic  In 2011, the worldwide internat ional and domestic passenger kilometers transported reached a record high of 5.2 trillion kilometers  The large volumes of air traffic, result in profound impacts on commodity trade  regional development  cultural communication  disease importation [7,8] and species invasion [ 5,9,10] As humans and commodities are transported at exceptional rates through aviation compared to other modes of transportation, how these patterns impact the socioeconomic, environmental and epidemiological landscape is of significant interest [5,7,9,81] Quantifying the volume of passengers on the air travel network is critical to understanding the complicated spatial interaction between origin and the destination cities [7,8] Previously, studies from a range of fields [3,5,9,10,82 84] ha ve made use of data from the International Air Transport Association (IATA) or the International Civil Aviation Organization (ICAO). These data are often restricted to scheduled flight plus seat capacity information on routes. However, not all commercial flights operate at full capacity, thus, such data often overestimate the passenger numbers on routes  Moreover, capacity data provide information on only point to point connection, thus,
50 travel patterns that require a stopover and transfer of planes are not captured [68 ] Although Origin Destination data derived from air ticket sales are available (e.g. http://www.iata.org/ps/intelligence_statistics/paxis/pages/index.aspx ), such data are expensive for research purposes, running to many tens of thousands of dollars, and can require significant legal and confidentiality agreements for data usage. Other databases of international flow by pair wise airports are held by private companies ( e.g. Marketing Information Data Transfer, http://ma.aspirion.aero/midt ) with costs, again, often high to obtain, with payment required repeatedly to maintain the latest data. Here we aim to outline a model frame work to produce open access estimates of global air traffic flow for research purposes that can be updated regularly. Spatial interaction models have been utilized to estimate the volume of passengers given an origin and destination city where data are la cking [3,72,78,82,83,85 90] The most common model used is the gravity model, which incorporates drivers such as the site characteristics of origin and destination areas, and destination to estimate the flow value. As Grosche et al  summarized, the d rivers in the spatial interaction model to estimate the air traffic include 1) socio economic characteristics of origin, destination, such as population, income, GDP, urban infrastructure, education level and 2) service related factors such as the qualit y (such as flight frequency, plane size and prices) and the market demand of airline service. The locational separation is usually calibrated by the distance or travel time separating origins and destinations. The gravity model provides a solid theoretica l and practical
51 background on understanding the movement of populations since it explicitly captures the absolute and relative spatial relationship of the origin and the destination  The utilization of network characteristics sheds l ight on the identification of air service factors in the gravity model for flow estimation, since 1) the layout of the global and the network topologies are indicated by the demands of air travel of the area where the airports serves. Firstly, large air travel companies in mature air travel markets adapt a hub and spoke model to achieve the balance of travel time for customer and the efficiencies in transportation infrastruct ures. In this model, a single airport is assigned to a single hub or multiple hubs to form a regional inter connected community [91 93] degree connections to a larger degree hub  The locations of airport hubs are selected as the optimum locations that satisfy the inter regional travel demands and minimize the total transportation cost [91,92] Moreover, the hub and spoke layout can Guimera et al  be reached from every other with only a small number of connections They also identified how central nodes with low degree connectivity play an important role for inter regional and intr a degrees of the air travel network follow a power law distribution is suggested by the nodal structure of flows clusters  as described by the hierarchical span for the major airports in the United States 
52 Secondly, the degree and centrality of air ports in the air travel network can act as indicators for air travel demand, since the local measurement of air passenger volume, population, and the level of economic activities at the periphery of the hub are highly correlated [97 99] Empirical research [100 102] have observed the increment of air passengers and increased passenger flow based on economic growth. Liu et al.  quantified the marginal effects of population i n a given metropolitan area on the air per 100,000 population growth. Wang et al  studied the air travel netw ork in China and found that cities in the more urbanized area of East China had a higher centrality score and a higher number of air passenger volumes compared to the more rural West China. These studies indicate the mutual correlation of network centrali ties and urban development level, which reflect the spatial agglomeration of economic activities and unequal air travel service demands. To study the movement of vector borne disease on the air travel network Johansson et al [68,74] modeled the actual passengers counts between 141 airports worldwide with epidemic significance Utilizing the air travel itineraries of the United States as a train ing set, they constructed a generalized linear model with a Poisson link to estimate worldwide passenger flows using nodes and routes characteristics as model covariates. Their results demonstrated good fits for the flow prediction on true origin destinati on travel. Our research follows the general modeling framework used in Johansson et al [68,74] but extends to a global model which includes: 1) all the nodes with a host city population of more than 100,000; 2) the routes between all airports that are within 0, 1 or 2 stops on the air travel network.
53 Materials and Methods Airport L ocations and S cheduled R outes Information on a total of 3,416 airp orts across the world, together with their coordinate locations was obtained using Flightstats ( www.flightstats.com ) for 2010. The connectivity and scheduled air travel network routes were defined by a 2010 schedu led flight capacity dataset purchased from OAG ( www.oag.com ). These included information on direct links (if a commercial flight is scheduled) of origin and destination airports, flight distances, and passenger capacity b y month for 2010. Directly connected airports pairs were utilized to construct a graph for the air travel network in 2010 with 3416 nodes and 37674 edges. The average degree of the network was 22.06, with the maximal degree recorded as 476 for Frankfurt A irport (IATA code: FRA). The topology of the graph exhibits both small world and scale free properties as already observed in similar global or regional air travel dataset analyses [95,105,106] The coefficients of the power law function fitting the scaled degree distribution was 1.010.1, which concords with a previous study  The average path length is 4.11, measured as the average number of steps travelling from one node to another node, while the diameter of this network was 14, which indicates the shortest path between the two most remote airports. Based on the network created by the flight statistics assembled, we calculated the degree, centrality and strength for each node and use these meas urements as covariates at the modeling stage. Gross Domestic Product and P opulation I nformation Generally, socio economic variables at a global scale are difficult to obtain. The G Econ data ( http://gecon.yale.edu/ ) provide indices representing both market exchange rates (MER) and purchasing power parity (PPP) at a 1 degree longitude by 1
54 degree latitude resolution at a global scale. Due to the large geographical coverage of the grid cells, we extracted the closest P PP value for an airport and calculated the PPP value per capita in 2005 by dividing the purchasing power parity by the population value in each grid cell. These data are utilized as local economic measurements for each airport. Given computing power limita tions on the modeling and matrix sizes, we selected the airports serving a city population number more than 100,000. To select these airports, a web crawler built on the WolframAlpha API ( http://products .wolframalpha.com/api/ ) was used to extract the city populations for each airport. Wolfram alpha is a knowledge engine which is capable of computing population information from various sources including: U.S census data, United Nations urban agglomeration and City Population ( http://www.citypopulation.de/ ) data. These data capture the most recent city population estimates from these data sources (for cities in United States, the US Census 2010 data are utilized ). In our database, there are 1491 airports satisfying these criteria. Actual T ravel P assenger F low Data on passenger origins and destinations on the air travel network were obtained from a variety of sources: The DB1B market data from the Airline Origin and Destination Survey (DB1B) provides a 10% sample of U.S. domestic passenger tickets from reporting carriers, including information such as the reporting carrier, origin and destination airports, prorated market fare, number of market coupons, market mi les flown, and carrier change indicators. To create a training dataset, these data were aggregated annually by the origin and the destination airport code with the sum of counts of itineraries. To protect the US air travel industry, the reported internatio nal Origin Destination data by the U.S carriers is strictly restricted to U.S citizens, and requires detailed statements on the use of the data. Hence, the research presented here did not take into account the international portion of the Origin Destinatio n data from DB1B.
55 The Canadian transport department provides statistics relating to the movement of aircraft, passengers and cargo by air for both Canadian and foreign air carriers operating in Canada ( http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=2703&la ng=en&db=imdb&adm=8&dis=2 ). This survey provides estimates of the number of passengers traveling on scheduled domestic commer cial flights by directional origin and destination city pairs. In this survey, significant numbers of Canada U.S trips were reported. The city pairs were matched to the airport pair that had the shortest routes defined by the OAG database with the passenge r number obtained from the above data source. For example, passenger numbers between Toronto and New York City were matched to the direct route of YYZ to JFK, since it is the shortest route between these two cities. Detailed route data for passenger numbe rs from EuroStat ( http://epp.eurostat.ec.europa.eu/portal/page/portal/transport/data/database ). This database presents passenger numbers between the main airports of reporting countries and their main partner airports in the European Union. All of these flow statistics were utilized to create a global prediction O D matrix, described below. Data Processing Cities are situated in a complex hierarchical network and the flows between cities are either constrained or facilitated by this hierarchical structure [78,107] We defined three levels of economic activity for each city per capita based on the 33% quartile of the distribution of PPP per capita. Thus, nine types of economic links are identified (low low, low medium, low high, etc.) to reflect the type of flow within/across the economic hierarchies. Similarly, we defined four levels of hierarchy based on the degree distribution of the airports, and sixteen types of flows are identified to reflect the type of flow within/across the air service hierarchies. A prediction dataset framework for routes was constructed based on the adjacency matrix defined by the OAG dataset. For each airport, destination airports via first order connection, second order connection and third order connections on the air travel network were identified. Along these routes, information on the minimum number
56 of stopovers and the maximum seat capacity were calculated. Moreover, following  we defined a categorical variable for distance classes to separate the markets by stage lengths, with 1 for short haul (20 00 kilometers or less), 2 for medium haul (between 2000 and 3500 kilometers) and 3 for longer hauls (3500 or more kilometers). We excluded routes less than 200 km since passengers are believed to have more efficient and effective land based methods to trav el such small distances. Note that only 3842 possible routes (<0.001%) are less than 200 km. Finally, an origin destination (OD) pair list with 1,295,752 rows was created. For analytical purposes, the global OD pair list was constructed following these as sumptions: The sum of passenger counts between an airport pair only represents the number of arrivals at a final destination. If the maximum capacity of a route is greater than the actual passenger number found in our datasets, the maximum capacity variab le will be replaced by the actual passenger number, which is the sum of passengers from all possible routes (including directed and undirected). stop at the connecting city. The data used for modeling is itinerary data which represents the minimum number of stops from one airport to another. Hence, passenger numbers in our database represent the flows for the first order, the second order and the third order of network connec tions. We assume that passengers choose the first shortest path found by a breadth first search algorithm, as the route is found by iterating all the neighboring nodes until a path from the origin and the destination is identified. If both the origin and t he destination cities have multiple airports, the passengers are assumed to take the shortest path from all possible routes between these airport pairs, which usually resulted in the path between the two largest airports in terms of capacity. This is suppo rted by Button et al  gers tend to choose a larger hub for their travel. Passengers do not choose routes with more than two stops. We used the number of stops as a categorical variable rather than a numeric variable since it is considered to be a measure of hierarchical accessi bility. In fact, for the air travel network in 2010, all of the possible calculated routes within two stops covered 83% of all the possible connections. Also, multiple stops (more than two
57 stops) are comparatively rare as a share of total passengers in our actual travel flow datasets. In DB1B domestic datasets, there are no itineraries for travels between cities with a population size more than 100,000 within two stops. Network characteristics were calculated using the igraph ( http://igraph.sourceforge.net/ ) library in R ( http://www.r project.org/ ). A summary of variables included in the model is presented in Table 3 1. Model The model takes the form of the general gra vity model: (2 1) where Pij is the annual aggregated passenger flow between Node i and j. Node i and Node j denote the collection of node characteristics, which are considered to be drivers of the size of the flow. Route ij denote t he collection of route characteristics, which are considered to be the proximity measurements. Interactions ij denotes the collection of two way interaction effects between categorical variables such as stops, country, degree link type, economic link type a nd haul type with other node and route characteristics. For the purpose of better estimation and thus prediction, we tested four model forms which include 1) a lognormal model for main effects only. This model adopts the general gravity model framework as the one described in Balcan et al  To utilize this model, a logarithm transformation is performed for each quantitative variable. The main effects include both node and route characteristics. 2) A generalized linear model for main effects and interactions wit h Poisson distribution and a log link. This model is the model utilized by Johansson et al [68,74] for predictions of the traffic flows between ep idemiologically significant cities. 3) A generalized linear model for main effects and
58 interactions with a negative binomial distribution and a log link. This model is similar to model 2 except that it utilizes a negative binomial distribution to account f or the possible over dispersion in the data. 4) A lognormal mixed model with main effects, interactions, and random effects on origin and destination city (note that a logarithm transformation is performed for each quantitative variable as well). This mode l assumes that the passenger flows are independent between different degree link type but correlated within the same degree link type, while model 1 3 make the assumption that all passenger flows are independent of each other, which is very strong and unre alistic in practice. Random effects are thus included to account for the dependence among passenger flows and the possible heterogeneity between levels of air travel services. More detailed model descriptions can be found in the Appendix B Apart from mod el fitting on the whole dataset, cross validation is performed to evaluate how accurately each model will predict in practice: firstly, the dataset is randomly partitioned into 10 subsets, each consisting of 10% of the observations. Then on each of the sub sets (the testing set), we validate the analysis performed on the remaining data (the training set). Lastly, the validation results are averaged over the rounds. Three criteria are chosen for model evaluation: 1) the coverage rate of the 95% prediction int ervals, which measures the percentage of the observations that fall into the corresponding 95% prediction intervals; 2) the coverage rate of the 30% observation intervals, which measures the percentage of the predictions that fall into the 30% intervals of the corresponding observations; 3) the successful prediction rate, which measures the percentage of predictions that fall into the same magnitude category as the corresponding observations. These magnitude categories are defined by dividing
59 the passenge r flow numbers into five groups: 10 2 and under, 10 2 10 3 10 3 10 4 10 4 10 5 and 10 5 +, each group represents one category. Results Model C omparison For each model, most coefficients are significant at the 0.05 significance level as the percentages of the si gnificant coefficients are about 90%, 100%, 96%, 95% respectively for model 1 to 4. For the purpose of prediction, we keep all the covariates in the model instead of removing the non significant ones. Not surprisingly, most of the interactions between node and route characteristics play an important role in model estimation as we treated the number of stops as a categorical variable. The interaction between haul types and inverse distance is also significant, which agrees with previous work  Both model 1 and model 2 provide narrow confidence intervals for predictions, while model 3 and model 4 provide wider intervals to accommodate variation in the data. All of these models h ave at least 68% successful prediction rates for predicting the magnitude of passenger flow. According to the results presented in Table 3 2 and Table 3 3 model 4 provides the most accurate prediction. For each of the models, we calculated the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). RMSE is a frequently used measure of the differences between estimate values and the values actually observed. A smaller RMSE suggests a better model fit. MAE is the average of the absolute value of the p rediction errors, which serves the same purpose as RMSE and is believed to be more robust in many situations. As shown in Table 3 3 model 4 yields the lowest RMSE and MAE for the majority of the data points except for extremely large observations. For the largest
60 observed passenger value category, model 2 gives the lowest RMSE and MAE, while model 4 gives the second lowest RMSE and MAE. Figure 3 1 presents the prediction and diagnostic plots for Model 4. Panel a) shows that most of the prediction values a re close to the y=x (prediction=observation) line. Panel b) shows that most of the residuals scatter along the y=0 (residual=0) line, yielding no obvious pattern. Both plots indicate that Model 4 is a plausible model for the passenger flows. However, the p rediction seems poor at the lower tail. This is expected, given likely randomness in the smaller amount of passenger exchanges between airports  Diagnostic plots for other models are presented in the Appendix B Alternative diagnostics for testing the model fit were performed for model 4 as well. Firstly a multilevel model described in Snijders et al  and implemen ted in the SAS code written by Recchia et al  to calculate r squared measures for the fourth model was utilized. The first level of the model was found to explain 84.0% of the variance in the data and the second level explained 98.7% of the variance, indicating a good model fit. Secondly, for the directly connected flights, we compare both the predicted value from model 4 and the capacity data from OAG to the observed passenger flows on a log scale using t he paired t test. The results show evidence of difference between the mean predicted passenger number and mean observed passenger number, and between the mean capacity number and mean observed passenger number, both at the 0.05 significance level. However, the geometric mean ratio of log (predicted value) to log (passenger number) was 1.01(panel c) in Figure 3 1), while the geometric mean ratio of log (capacity) to log (passenger number) was 1.08(panel d) in Figure 3 1). The predicted value shows more agree ment on the
61 observed value, while the capacity data represents a significant overestimation of flows between two directly connected airports. Hence, our predicted values provide a closer approximation of the traffic flows on the air travel network compared to the maximum seat capacity metric for the directly connected cities, as used in previous studies [3,5,9,10,82 84] In summary, our model (Model 4) outperformed the lognormal gravity model (Model 1) used in Balcan et al  and the Poisson model (Model 2) used in Johansson et al [68,74] Moreover, for direct flights, our estimates show more homogenous agreements with observed passenger numbers compared to simple seat capacity data. Prediction and Inter pretation of the O D P assenger F low M atrix Model 4 was applied on the estimation dataset to predict passenger flows. We have identified the over dispersed predictions that exceeded the maximum capacity on the routes (3% of the data) and replaced them with the product of the maximum capacity on the routes. According to the training dataset, the maximum numbers of itineraries for one stop a nd two stop connections were 140,086 and 8,060. Since these data are generated from the mature air travel market and constrained by the network structure, we considered them as the upper limits of the data distribution. Thus we adjusted the prediction of t he first order connection and the second order connection flights scaled by these two maximum numbers. Following this, we removed all the predictions with less than 1 person. Finally, 644,406 routes with origin, destination airport codes, number of stops a nd predicted passenger number were produced. As described before, the passenger counts were categorized into five categories as a test of successful prediction rate in magnitude: 1 10 2 ,10 2 10 3 ,10 3 10 4 ,10 4 10 5 and
62 10 5 and more. The first two categories pres ent small numbers of passenger exchanges, implying random flows between two airports, and the fourth and fifth categories indicate a higher probability representing steady flows between airports. Figure 3 2 a) shows all the flows with more than 10 5 predic ted passengers. Secondly, given an origin/destination, the dataset produced through the research outlined here can estimates the endpoints and starting points with passenger flows on the air travel network. Figures 3 2 b) d) illustrate the passenger flows and number originating from Atlanta, categorized by number of transfer. Figure 3 2 e) shows the distribution of airports with incoming passenger numbers over 5,000,000. This reflects the mature air markets of the United States and Europe, though noticeable concentrations of airports can be observed in the emerging markets such as India and China as well. Discussion With continuing growth of the global air travel network, we must expect continued socioeconomic, environmental, cultural and epidemiological imp acts. This research shows how network characteristics combined with multiple datasets on various perspectives relating to the movements of passengers of passenger flow on the global air network can be compiled to provide estimates that are more accurate th an previous modeling efforts. Such a dataset provides a valuable resource for scientists and decision makers to measure the global flow of air traffic and its potential influences. In the database outlined here, 644,406 unique routes spanning 1,491 airpor ts serving city populations of more than 100,000 are modeled based primarily on publicly available datasets. Our model has outperformed similar research at the global scale and can explain 98% of the variance in the data. Within the database, 23,785 routes follow a
63 direct connection, 291,745 routes are one stop connections and 328,876 routes are two stop connections. Using this route and airport information, anyone can construct flow matrices to describe the global air traffic flow and assess its multiple i mpacts. Due to data constraints, a range of uncertainties and limitations exist in the output modeled datasets. The first inconsistence comes with internal uncertainties within the DB1B dataset. To construct the DB1B dataset, Transtat only requires US carr iers to report O D pair data, hence the O D data is likely to be inaccurate in markets served with a significant number of foreign carriers (e.g., New York, Washington D.C., Chicago, and Los Angeles). If there is more than one airport in a city, each of t he airports is treated as a separate node. This may well result in overestimates of the flow to secondary airports in a city. The second set of inconsistency is the population data. Due to data availability, only city population data were utilized, when it is sometimes the case that people in neighboring metropolitan area can access the airport in question through other ground based transportation methods (for example, people in Gainesville FL are often likely to drive two hours to Jacksonville or Orlando t o take a plane, rather than utilize the Gainesville regional airport which is 10 miles away from the city center). As a result, our predictions may overstate the markets for small airports. The third source of uncertainty stems from the fact that the data we utilized for the training datasets were only from the United States, Canada and the European Union. Thus, international flights are less well represented in our dataset and most of the flight data describes the flows between airports in high income coun tries. Additionally, long haul international flights with more than three stops are absent.
64 The topology of air travel network is likely to vary at the regional level. Wang et al  found that in terms o f topological measurements, the Chinese air travel network is similar to the Indian one, but different than that of the US. As current air travel networks in low income countries usually feature point to point connections between city pairs  high income countries are increasingly prompted to utilize d a hub and spoke system due to their mature air travel markets. On the other hand, it is observed that some companies (such as Southwest Airlines and Jet Blue in United States) in high income countries also adopt spoke to spoke models to connect hot spots of air travel demand  This heterogeneity may affect the flow estimation country wise and overestimate the driving factor of hubs in both high and low income countries. The  Hence, it could be anticipated that the demand for air travel in each country varies and is correlated to GDP. Also, the demographic profile of passengers on the a ir travel network is likely different between countries. Under a regional context, this may affect the prediction of domestic passenger numbers, while international heterogeneities in traffic flows may be attributed to differing visa policies between count ries  Visa restrictions may reduce traffic flows substantially between countries  Moreover, cultural differences at a country level could represent indic ators of attraction and drivers of population movements [80,116] The potential limitations discussed above arise through the constraints of the data sources used. These may be alleviated through incorporation of more publicly accessible data in future work, including 1) More detailed economic indicator s (such as GDP, income etc.) at the city level: such measures could further describe drivers in the
65 gravity model. 2) Itineraries from low income regions of the world: such data would enlarge our training and testing databases to avoid sampling errors. 3) Hub characteristics (such as the number of enplanements, transfers and deplanements): these measures could help explain the function of the hubs in controlling network flows. Alternatively, transportation forecasting models [117,118] and radiation models  could be utilized to estimate the global O D matrix based on the traffic counts on nodes and edges. Conclusion The research presented here has documented the generation of a world wide Origin Destinatio n matrix of passenger flows in 2010 for airports with host city populations of more than 100,000. Results show that the modeled dataset improves substantially on the accuracy of datasets used in previous studies. The datasets are freely accessible for acad emic use and are published as part of the Vector Borne Disease Airline Importation Risk (VBD Air) project at www.vbd air.com/data/
66 Table 3 1 Descriptions of covariates used in the modeling process Variables Descriptions Node characteristics Pop i The population of the origin city Pop j The population of the destination city PPP2005 i The purchasing power index where the origin airport serves PPP2005 j The purchasing power index where the destination airpor t serves PDA2005 i The purchasing power per capita index where the origin airport serves PDA2005 j The purchasing power per capita index where the destination airport serves Strength i The sum of the edge weights of the adjacent edges for each vertex for the origin city Strength j The sum of the edge weights of the adjacent edges for each vertex for the destination city Degree_Out i The degree number of the origin city on the air travel network Degree_In j The degree number of the destination city on the air travel network Closeness_Centrality i The mean geodesic distance between a given node and all other nodes with paths from the given node to the other node. This variable is calculated according to the origin city. Closeness_Centrality j The closeness c entrality measure for the destination city. Betweeness_Centrality i The number of shortest paths going through a specific airport. In a weighted network, betweenness centrality is a useful local measure of the load placed on the given node in the network a s well as the node's importance to the network than just relying or connecting. Betweeness_Centrality j This is the calculation of betweeness centrality for the destination airport. Route characteristics Inverse Distance Inverse great circle distance between the origin and the destination airport Country Indicates whether the origin and the destination are in the same country. Alternative Number of alternative routes to the destination Stops Number of stops on the shortest route from the origin to the destination MaxC The maximum capacity along the shortest path Degree Link Type This variable identifies the types of flows between different hierarchies of airports defined by the air travel services level. Economic Link Type This variable identif ies the types of flows between different hierarchies of airports Haul Type This variable differentiates the effect of long haul flights. 1 for short haul (2000 kilometers or less), 2 for medium haul (between 2000 and 3500 kilometers) and 3 for longer haul s (3500 or more kilometers).
67 Table 3 2. Comparison of the four models with respect to prediction accuracy (in percentages) Coverage rate of the 95% prediction intervals Coverage rate of the 95% prediction intervals (cross validation) Coverage rate o f the 30% observation intervals Coverage rate of the 30% observation intervals (cross validation) Successful prediction rate Successful prediction rate (cross validation) Model 1 6.39 6.73 29.82 29.76 68.42 68.33 Model 2 4.80 4.79 31.48 30.63 69.16 6 8.80 Model 3 23.16 24.16 33.09 33.38 70.04 69.43 Model 4 52.11 49.86 47.79 31.17 79.72 70.41 Table 3 3. Root Mean Squared Errors and Mean Absolute Errors for all models Measurement Categories Number of Records Model 1 Model 2 Model 3 Model 4 RMSE Obse rved Passenger (OP)<10 2 2379 1680 2923 1947 726 OP in 10 2 10 3 6440 3536 5127 32405 1802 OP in 10 3 10 4 7314 7397 8771 10639 4346 OP in 10 4 10 5 4817 20780 23002 41585 21940 OP >10 5 1132 163352 85897 216610 127194 MAE Observed Passenger (OP)<10 2 23 79 286 538 402 120 OP in 10 2 10 3 6440 629 1073 1413 333 OP in 10 3 10 4 7314 2729 3218 3140 1621 OP in 10 4 10 5 4817 14697 14929 19689 13415 OP >10 5 1132 115710 61305 94447 89233
68 Figure 3 1 Diagnostic plots for all the models. a) Predicted vs. observed value of model 4. b) Residual vs. observed value of model 4. c) Distribution of ratio of predicted value vs. observed value in log scale with 95% confidence interval for geometric mean. d) Distribution of ratio of capacity vs. observed value in log scale with 95% confidence interval for geometric mean.
69 Figure 3 2 Predicted air traffic flow s a) Predicted flights with passenger flows of more than 100,000. b) All possible passenger flows throug h direct flights originating stop flights stop flights originating Atlanta. e) All airports with an incoming passenger numbers more than 5,000,000.
70 CHAPTER 4 GLOBAL MALARIA CONNECTIVITY THROUGH AIRTRAVEL 3 Chapter Summary Air travel has expanded at an unprecedented rate and continues to do so. Its effects have been seen on malaria in rates of imported cases, local outbreaks in n on endemic areas and the global spread of drug resistance. With elimination and global eradication back on the agenda, changing levels and compositions of imported malaria in malaria free countries, and the threat of artemisinin resistance spreading from S outheast Asia, there is a need to better understand how the modern flow of air passengers connects each P. falciparum and P.vivax endemic region to the rest of the world. Recently constructed global P. falciparum and P.vivax malaria risk maps along with d ata on flight schedules and modelled passenger flows across the air network, were combined to describe and quantify global malaria connectivity through air travel. Network analysis approaches were then utilized to describe and quantify the patterns that ex ist in passenger flows weighted by malaria prevalences. Finally, the connectivity within and to the Southeast Asia region where the threat of artemisinin resistance is highest, was examined to highlight risk routes for its spread. The analyses demonstra te the substantial connectivity that now exists between and from malaria endemic regions through air travel. While the air network provides connections to previously isolated malarious regions, it is clear that great variations exist, with significant regi onal communities of airports connected by higher rates of flow 3 A version of this chapter has been submitted to Malaria Journal as Huang Z. and Tatem AJ, Global Malaria Connectivity through airtravel.
71 standing out. The structures of these communities are often not geographically coherent, with historical, economic and cultural ties evident, and variations between P.falciparum and P.vivax cle ar. Moreover, results highlight how well connected the malaria endemic areas of Africa are now to Southeast Asia, illustrating the many possible routes that artemisinin resistant strains could take. The continuing growth in air travel is playing an impor tant role in the global epidemiology of malaria, with the endemic world becoming increasingly connected to both malaria free areas and other endemic regions. The research presented here provides an initial effort to quantify and analyse the connectivity th at exists across the malaria endemic world through air travel, and provide a basic assessment of the risks it results in for movement of infections. Background The worldwide air travel network has expanded at an exceptional rate over the past century. In ternational passenger numbers are projected to rise from 1.11 billion in 2011 to 1.45 billion by 2016, with an annual growth rate of 5.3%  Today there are 35,000 direct scheduled routes on the air travel network, with 865 new routes established in 2011  Malaria endemic areas are more connected to the rest of the world than at any time in history, with the disease able to travel at speeds of 600 miles per hour within infected passengers. The growth of the air travel network results in substantial concerns and challenges to the global health system with a need to place more emphasis on evidence driven surveillance and reporting that incorporates spatial and network information [7,8,11,12] Rising rates of travel between malaria free and endemic countries have led to general patterns of increased rates of imported m alaria over recent decades
72 [15,26,28,120] D ue to infrequent encounters [7,26] imported cases can challenge health systems in non endemic countries, with difficulties in diagnosis  misdiagnosis and del ays in treatment [121,122] as well as significant treatment expenses  where patients who do not have a foreign travel history become infected throu gh being bitten in the vicinity of international airports [10,124 126] Patterns in imported cases and airport malaria have been shown to be related to a combination of the number s of travellers and the malaria risk at the destination [7,10] and these relationships will continue to evolve as new routes become established. The flow of people via air travel between endemic areas may increase the risks of re emergence or resurgence  in previously malaria free or low transmission areas  The autochthonous malaria outbreaks in Virginia in 2002  Florida in 2003  and Greece in 2011  for example, demonstrate the continued risks of loc al outbreaks following reintroduction through air travel, though such occurrences are rare  Further, the examples of malaria resurgence in island nations such as Sri Lanka [12 7] Mauritius  and Madagascar  after control measures were relaxed reinforce the importance of vigilance and robust surveillance in terms of human movement in pre and post elimination periods  Identifying the risks of malaria movement through the air travel network can provide an evidence base through which public health practitione rs and strategic planners can be informed about potential malaria influxes and their origins [7,130] Meanwhile, growing concerns have been raised about the possible spread of artemisinin resistance from the Gre ater Mekong sub region in South East Asia to other
73 endemic areas. Recent research has highlighted increasing numbers of patients showing slow parasite clearance rates following treatment with artemisinin based drugs in the Cambodia Thailand border and Thai land Myanmar border regions [22,131,132] Tremendous health and socio economic costs occurr ed when chloroquine resistant parasites arrived in sub Saharan Africa from S outheast Asia and spread across the continent [133,134] Similarly, sulfadoxine and pyrimethamine resistance emerged in Asia and spread to Africa [135,136] The WHO reports that there countries that have adopted artesunate  and fear remains over the spread of artemisinin resistance from S outheast Asia to Africa, that could undermine current control and elimination efforts, with no alternative drugs coming in the foreseeable future. Rates of imported malaria, risks of resurgence and the spread of drug resistance are all today influenced by how the global air travel network c onnects up the malaria endemic regions of the world, and the numbers of passengers moving along it. Here we combine recently constructed global P. falciparum and P.vivax malaria prevalence maps with data on modelled passenger flows across the air network, to describe and quantify global malaria connectivity through air travel in 2010. We derive weighted network analysis statistics to examine (i) which regions show greatest connectivity to P. falciparum and P.vivax malaria endemic zones, (ii) where the larg est estimated whereby malaria infection flows within them are likely to be larger than between communities, and finally, (iv) we examine the connectivity within and to the S o utheast
74 Asia region where the threat of artemisinin resistance is highest, to explore risk routes for the spread of resistance. Methods Airport L ocation s, F light R outes and P assenger F low M atrix Information on the longitude, latitude, city name and airport code for a total of 1,449 airports which serve cities with more than 100,000 people and a modelled actual traffic flow connectivity list with 644,406 routes amongst these airports w ere obtained ( http://www.vbd a ir.com/data ) [8,138] A connectivity matrix was created from the connectivity list, quantifying the volumes and the directionalities of the passenger flows. Within this passenger flow matrix, 23,785 routes we re direct connections between two airports, 291,745 routes we re one stop connections and 328,876 routes we re two stop connections. The travel volumes on the routes were modelled based primarily on publicly available datasets under a generalized linear model framework. Full model details are provided in Huan g et al  but in brief, t o construct th e matrix, topological characteristics of the air travel network, city population, and local area GDP amongst others were utilized as covariates. Actual travel volumes for training and validation were e xtracted and assembled from various transportation organizations in the United States, Canada and the European Union. A log linear model controlling for random effects on origin, destination and the airport hierarchy was then built to predict passenger flo ws on the network. Th e model outperformed existing air travel passenger flow model s in terms of prediction accuracy  Malaria D istribution Global P. f alciparum and P. vivax prevalence maps were obtained from the Malaria Atlas Project (www.map .ox.ac.uk) and the methods behind their construction
75 are presented in Gething et al [32,33] In brief, 22,212 community prevalence surveys were used in combination with model based geostatistical methods to map the prevalence of P. falciparum globally in 2010 within limits of transmission defined by annual parasite incidence and satellite covariate data. Similarly, 9,970 geocoded P. vivax parasite rate ( P.vivax PR) surveys collected between 1985 and 2010 were utilized in a spatiotemporal Bayesian model based geostatistical approach to map endemicity , under the restrictions of a mask of the stable/unstable endemicity  and information on the prevalence of the Duffy blood group  We do not consider P. ovale, P.malariae or P.knowlesi here, since similar datasets on their distributions do not yet exist. Weighted N etwork A nalysis and C ommunity D etection Malaria prevalence can vary grea tly in the region around airports and the cities they serve, and travellers taking flights from a specific airport may reside many kilometres from the airports in higher transmission areas than found in the vicinity of the airport. Thus, simply assigning t he predicted prevalence from the malaria maps at the location of each airport could underestimate the risk and rate of infection exportation at the airport in question and under represent its contribution to global malaria connectivity. Therefore, followin g Huang et al  local accessibility to each airport was considered by assum ing that p assengers would travel less than 5 0 km with a travel time less than two hour s to access an airport to take a flight. Under this assumption, the P. falciparum and the P. vivax malaria prevalence s assigned to an airport were obtained as the maximum prevalence from the malaria maps within a mask of 5 0 kilometres and 2 hour travel time ( Figure C 1 ), in which the mask wa s generated using a global travel time map ( http://bioval.jrc.ec.europa.eu/products/gam/index.htm ). Likewise, an indicator that
76 defines whether an airport is located in the stable/unstable endemic zone wa s created according to the same mask, in which the indicator defines whether th e majority area of the mask is located in the stable/unstable zone. Figure C 1 in Appendix C shows this travel time/distance mask with the global travel time map. The above approach ensured that each airport had an assignment of a P. falciparum and P.vivax prevalence rate (or unstable/malaria free), which could then be network to analys e malaria connectivity. Thus, P. falciparum and P. vivax flow s we re calculated on each route as origin prevalence estimated passenger volume, to produce P. falciparum and P. vivax malaria network s A group of weighted centrality analyses and network com munity partition analyses we re performed on the malaria networks to quantify features of global malaria connectivity First, the in strength and out strength of each connection was calculated as ( 4 1) In which is the airport adjacency matrix and is the weighted malaria flow. This metric estimates the total weight of malaria flows that airports send and receive. Following this, weighted between n ess analys e s we re performed on t he malaria flow matrices Betwee n ness centrality measures the number of shortest paths going through a specific vertex  In a weighted network, betwee n ness centrality is a useful local measure of the load placed on the given node in the network as well as the node's importance to the network other than just conn ectivity  It i s often used in
77 transport network analysis to provide an approximation of the traffic handled by the vertices  Thus, here it provides an indication of the importance of each airport in the global flow of malaria infections via air travel i.e. a measure of how many infections likely pass through each airport each year, relative to other airports. The between n ess centrality is calculated as: ( 4 2) In which is the total number of shortest paths from node s to node t and is the number of those paths that pass through v. Note that on our weighted P.falciparum / P.vivax network s the distance between the two nodes s and t is defined by the sum of P.falcipa rum flows or P.vivax flows as the cost on this path  A normalized betweenness wa s used as ( 4 3) where n wa s the number of nodes (airports) in the air travel network  Communities in a network reflect the partition of nodes that are densely connected and separated from the other nodes in the network, thus these nodes  By mapping communities on the malaria networks defined here, we aimed to identify groups of airports that show strong li nks in terms of likely movements of infections. This potentially has utilit y in terms of providing evidence upon which regional surveillance strategies can be designed [42 ] Newman and Girvan  define a modularity score which measures the quality of network partitions as :
78 (4 4 ) In which, re presents the weight of the edge between i and j (here these are the P.falciparum and P.vivax flow matrices ), is the sum of the weights of the connections attached to airport i, is the community to which airport i is assigned ; is 1 if = otherwise A multilevel algorithm for community detection  was implemented This method utilizes an iterative approach that merges communities to maximize t he modularity score: Firstly modularity is optimized by allowing only local changes of communities, secondly the established communities are combined together to construct a new network. These two passes are repeated iteratively until no increase of modula rity is possible. The number of communities returned by this algorithm yields the maximum modularity score. We performed a simple Wilcoxon rank sum test  on the differences between "internal" and "external" degrees of a commun ity in order to test whether the establishment of communities was significan t We define d air connections within a community as "internal" and the connections connecting the airports of a community with the rest of the network as "external". The null hypot hesis of this test wa s that there wa s no difference between the number of internal and external routes incident to an airport of the community. Results The results of the global malaria connectivity analyses are presented in two sections: (i) analyses foc ussed on the connection of endemic malaria regions to each
79 other and to malaria free areas, that has particular relevance to imported malaria and malaria resurgence and re emergence; and (ii) analyses examining the connections between S outheast Asia and th e rest of the malaria endemic world, which are relevant to the spread of artemisinin resistance. C onnectivity w ithin E ndemic A reas and to N on E ndemic A reas Figure 4 1 show s the results of regional community structure analyses based on traffic flow data ove rlaid on the P. falciparum/P. vivax endemicity and stable/unstable transmission limit s maps. T he Wilcoxon test results show that the internal degrees for the airports within all communities are significantly different from the external degree s, with p valu es of < 0.01, thus the community partitions shown are significant. The maps highlight those countries that form communities linked by high levels of traffic scaled by P. falciparum/P. vivax prevalence at their origin endemic area. Figure C 2 describes simi lar analyses based on the travel network data from Huang et al . The communities detected reflect the architecture of the air network, and how this relates to malaria endemicity around the World. Geographical contiguity is clearly evident, as traffic l evels on shorter distance routes are generally higher than on longer distance routes, but interesting patterns relating to historical ties emerge. For instance, for P.falciparum London forms part of the Nigeria community, but Paris shows stronger ties to the remainder of sub Saharan Africa. These connections are often reflected in imported malaria statistics, with Nigeria being the main source of P.falciparum cases seen in the UK, but for France, the French speaking African countries are the main origin. S imilarly, UK airports also form part of the India/Bangladesh community, where historical ties exist, resulting in significant travel between the two regions, and consequent P.falciparum and P. vivax malaria importation to the UK. Ties also exist
80 between th e western US and E ast Asia, which form a single P.falciparum community ( F igure 4 2 a). Figure C 3 shows a community detection analysis for airports with direct connection s one transfer connection s and two transfer connections from endemic areas. To examine directional and net potential movements of people and parasites between airports in different countries we summed up the international route weightings to identify of malaria infections ( Table 4 1 and Table 4 2 ) He re, the weights of all possible incoming flows for airports in the non endemic area s and the weights of all possible outgoing flows from airports in endemic area s we re summed up to define strengths of importation and exportation (Note that we onl y considered the routes connecting two different countries regardless of the domestic routes). In this table, airports in the Far East and Middle Asia such as Singapore, Hong Kong, Dubai, and Sharjah display the highest importation values (Note that Singap ore ranked the first in both categories). Unsurprisingly, major air hubs in Europe (such as airports in London, Paris and Frankfurt) also showed high potential incoming P.falciparum flow s Miami is the only airport in the United States on the importation f low top 10 lists, with its strong connections to Central and South America In terms of exportation, the largest airports by traffic capacity and connections to the rest of the malaria endemic world were highlighted (Table 4 3 and Table 4 4 ) Mumbai was ra nked first as the largest exporter of P.falciparum and P.vivax flows, suggesting that it likely acts as an important portal for spreading malaria to the rest of the world. T he betweenness centrality metric was utilized to inspect the connectivity from ende mic areas. As the betweenness metric is defined as the number of shortest
81 paths connecting any two airports that involve a transfer at airport v high centrality airports in endemic area s provide hubs for people originating at le ss accessible airports in remote places to reach the rest of the world Table 4 3 and Table 4 4 show the top 10 highest between n ess centrality airports for transferring P.falciparum flow and P.vivax flow elsewhere For the P.falciparum flow, international airports in Africa play important roles as hubs for routing infections Some airports are observed to have small degrees (low numbers of connecting routes) and large centrality (importance as a hub) which can be considered as an abnormality  These airports connect less accessible and connected airports in endemic areas to other airports in the world. For P.vivax flow, Asian international hubs play more important roles. Of interest is Phoenix airport which ranked the 6 th in te rms of P.vivax centrality, suggest ing that it plays an important role as a gateway in linking P.vivax endemic areas to the United States Figure C 3 presents the spatial distribution of betweenness centrality scores for airports weighted by P.falciparum o r P.vivax flows As a comparison a similar figure based solely on the modelled passenger flow is presented in Figure C 4 To further investigate the effects of flows from endemic zones, Figure C 5 a) and b) shows the sums of international incoming risk flows for all the airports in those 36 countries that have national policies for malaria elimination, and are closest to eliminating the disease  Importation of infections threatens the success of elimination programs  and while air travel may not be the highest risk source for these introductions for most of these countries, it remains a potentially important source of incoming infections. From these two maps, it can be seen that China and countries in middle Asia are subjected to the greatest pressure of incoming flows relative to other
82 elimination countries, due to their larger incomi ng traffic volumes from endemic regions elsewhere around the world Further analyses on airport connectivity are provided in Appendix C Connectivity to Southeast Asia Figure 4 2 maps out the passenger flows scaled by origin prevalence for P.falciparum an d P.vivax from the greater Mekong sub region. Significant amount s of flow exchange within South e ast Asia can be seen in the close up subsets. For both P.falciparum and P.vivax it can be seen that the connectivity, through numbers of travellers, to Latin Am erican endemic regions is weak, but that much stronger connections to sub Saharan Africa and the Indian subcontinent exist. Increasing connections through trade and labour markets between Asia and Africa over the past decade is exemplified here in the stro ng connections between the S outh e ast Asian region and all of sub Table C 1 presents the top 10 risk routes spreading drug resistance of P.falciparum and P.vivax from the Great er Mekong Sub region to non Asian destinatio ns with estimated P.falciparum / P.vivax flow and the number of stops need ed to travel from the origin city to the destination city shown Discussion The continuing growth in air travel is playing an important role in the global epidemiology of malaria Flight routes now connect previously isolated malaria endemic regions to the rest of the world, and travelers on these routes can carry infections to the opposite side of the world in less than 24 hours. While many endemic areas still remain relatively iso lated, the malaria endemic world is becoming increasingly connected to both malaria free areas and other endemic regions. The impacts of this can be seen in
83 imported cases, vector invasions and the spread of drug resistant parasite strains. Here we present a spatial network analysis approach to demonstrate the connectivity that exists across the malaria endemic world through air travel, and provide quantitative indicators of the risks it results in for malaria movement. Results highlight the substantial co nnectivity that now exists between and from malaria endemic regions through air travel. While the air network provides connections to previously isolated malarious regions, it is clear that great variations exist, with significant regional communities of a irports connected by high rates of prevalence scaled flow standing out ( F igure 4 1). The structures of these networks are often not geographically coherent, with historical, economic and cultural ties evident. As new routes continue to be established, thes e communities will likely change, with new popular travel routes, such as those between China and Africa  likely altering global malaria flow routes, and new airports appearing in tables 1 and 2. These community maps (F igure 4 1) and lists of cities by likely import/export of infections ( T able 4 1 and Table 4 2 ) and hubs for infection flow (T able 4 3 and Table 4 4 ) provide a quantitative picture of how malaria infections are likely moving globally through air travel, and information from which global surveillance strategy design can draw upon. Tables 4 1 to 4 4 highlight that certain airports provide significant hubs and gateways for the movement of infections and their entry into countries, and that these are widely d istributed across the world. Their role in providing important nodes as both significant through flow of infections in the network, and entry and exit gateways for cases to/from regions means that they potentially represent valuable sentinel sites for focu sed surveillance. Finally, F igure 4 2 provides a stark reminder of how well connected the
84 malaria endemic areas of Africa are now to S outheast Asia, illustrating the many possible routes that artemisinin resistant strains could take. These routes can provi de a first step quantification to support the g lobal p lan against a rtemisinin resistance containment  and design of surveillance systems  and should be refined with information on the locations of resistance found. Such data could also inform decisions on where and how to limit the risk of spread, for example by pre travel or arrival screening and treatment A range of limit ations and uncertainties exist in the analyses presented here. In terms of the quantification of malaria transmission, the use of static maps of annual average prevalence [32,33] neglects the seasonality in transmission that is common to many areas, and also the substantial changes in transmission intensity seen in a variety of locations in recent years  Further, we have used parasite prevalence as our malaria metric, and while this may be an adequate measure of population p revalence at origin locations, it is not so appropriate for assessing the risk of infection acquisition for nave travellers, and entomological based indices are likely more appropriate here, as used in more local studies [130,144,150] Finally, our examination of relative artemisinin resistance spread risk focuses simply on all travel from four countries, and thus does not account for any heterogeneity in re sistance in the region. Uncertainties and limitations relating to the travel data used also exist. The modelled passenger flows represent just a 2010 snapshot, and thus routes and changes since then are not captured, while inherent uncertainty due to the m odelling process also exists  Moreover, the types of traveller and their activities during travel and their residential location are unknown, each of which contributes to differing
85 malaria infection risks. Finally, overland and shipping trav el flows are not considered here, which also contribute to local, regional and global malaria connectivity and flows. This work forms the basis for future analyses on imported malaria, elimination feasibility and the risks and potential routes of artemisin in resistance spread. Rates and routes of imported malaria have been shown to be significantly related to a combination of numbers of travellers to/from endemic destinations and the prevalence of malaria there  The potential thus exists to construct a model based on global malaria prevalence [32,33] transmission models for attack rate estimation  and traveller flow dat a  that can be used to forecast imported malaria rates, validated with imported malaria data. As nations make progress towards elimination  the importance of human movement and imported cases increases. The research presented here contributes to an on going initiative, the human mobility mapping project ( www.thummp.org ), aimed at better modelling human and disease mobility, and will form one aspect of continued multi modal assessments of malaria movements [31,130,143,144] and assessment of malaria elimination strategies [41,42] Finally, the potentially disastrous consequences of the rise and spread of artemisinin resistance requires that detailed and effective plannin g be implemented in preparation for containing and stemming any spread  We have presented a basic assess ment here of prevalence scaled travel from the four Southeast Asian countries where resistance has previously been observed, but significant refinements of these estimates and modeling methods should be undertaken. These may include improved tracking and m apping of observed resistance and human movement patterns in Southeast Asia, as is being undertaken by the TRAC project
86 ( http://www.wwarn.org/partnerships/projects/trac ), as well as scenario m odeling of the risks of resistance escape to Africa or Latin America. Further, the incorporation of accessibility [151,152] and travel data [144,150] with drug use data (e.g.  ), prevalence information [32,33] and models  all undertaken within a probabilistic modeling framework (e.g. [8,74] ), could aid in estimation of spread routes should resistance arise elsewhere.
87 Table 4 1 Top 10 airports based on estimated relative malaria impor tation rate s Malaria importation P.falciparum import P.vivax import City Country P.falciparum Flow In City Country P.vivax Flow In Singapore Singapore 366132 Singapore Singapore 175220 Bangkok Thailand 276563 Hong Kong Hong Kong 105261 Paris F rance 253291 Dubai United Arab Emirates 103790 Dubai United Arab Emirates 218599 Kuala Lumpur Malaysia 74976 London United Kingdom 212918 Seoul South Korea 58278 Hong Kong Hong Kong 183314 Sharjah United Arab Emirates 53548 Johannesburg South Africa 16 7734 London United Kingdom 48571 Casablanca Morocco 149298 Miami United States 47277 Kuala Lumpur Malaysia 141161 Taipei Taiwan 42840 Frankfurt Germany 132597 Mexico City Mexico 41710 Note: P.falciparum / P.vivax flow measures are calculated based on t he incoming and outgoing numbers of passengers travelling internationally, scaled by the malaria prevalence at the origin of the routes in the case of importation, and at t he airport listed in the case of exportation. The flows represent a relative measure of infection movement and are not designed to represent actual number of infections
8 8 Table 4 2. Top 10 airports based on estimated relative malaria exportation rates Malaria exportation P.falciparum export P.vivax export City Country P .falciparum Flow out City Country P.vivax Flow Out Mumbai India 974832 Mumbai India 214006 Ouagadougou Burkina Faso 513636 Bangkok Thailand 179633 Kinshasa Congo 467295 Manila Philippines 177760 Abuja Nigeria 461492 Delhi India 161249 Delhi India 44 7491 Ho Chi Minh City Vietnam 80725 Bamako Mali 439138 Bogota Colombia 72173 Douala Cameroon 382278 Ahmedabad India 70888 Manila Philippines 377384 Panama City Panama 70568 Lome Togo 267054 Guatemala City Guatemala 54788 Cotonou Benin 248226 Phnom Pen h Cambodia 53898 Note: P.falciparum / P.vivax flow measures are calculated based on the incoming and outgoing numbers of passengers travelling internationally, scaled by the malaria prevalence at the origin of the routes in the case of importation, and at the airport listed in the case of exportation. The flows represent a relative measure of infection movement and are not designed to represent actual number of infections
89 Table 4 3 Top 10 P.falciparum betweenness centrality airports with their degrees i n a sub network that only contains direct links from airports in P.falciparum or P.vivax endemic areas. Airport City Country Normalized Betweenness Centrality Degree NBO Nairobi Kenya 47.35 80 MBA Mombasa Kenya 32.44 27 JRO Kilimanjaro Tanzania 32.39 14 BOM Mumbai India 30.41 104 ADD Addis Ababa Ethiopia 28.21 64 DEL Delhi India 23.16 111 JIB Djibouti Djibouti 19.77 15 ADE Aden Yemen 18.63 15 MGQ Mogadishu Somalia 14.45 8 HRE Harare Zimbabwe 14.35 20 Table 4 4 Top 10 P.vivax betweenness centr ality airports with their degrees in a sub network that only contains direct links from airports in P.falciparum or P.vivax endemic areas. Airport City Country Normalized Betweenness Centrality Degree BKK Bangkok Thailand 96.43 146 ICN Seoul South Korea 78.12 150 DEL Delhi India 59.55 133 BOM Mumbai India 34.17 116 KMG Kunming China 30.79 90 PHX Phoenix United States 28.63 91 DPS Denpasar Bali Indonesia 27.94 34 SJO San Jose Costa Rica 27.72 37 DOH Doha Qatar 25.91 100 TAS Tashkent Uzbekistan 25.8 5 69
90 Figure 4 1 Spatial distribution of P.falciparum/P.vivax network communities overlaid on P.falciparum/P.vivax prevalence maps A) P.falciparum multilevel membership. B) P.vivax multilevel membership. These two maps show only airports that have d irect connections from endemic to non endemic areas. The inset maps present close up views of the United States and Western Europe. Airports with the same community membership (indicated by the same color ) display stronger links in terms of likely movement s of infections between them than to airports in other communities. Note that in the P.vivax map, communities with less than 10 airports are not shown.
91 Figure 4 2 Estimated relative P.falciparum/P.vivax flows originating from the Great Mekong sub reg ion overlaid on P.falciparum/P.vivax prevalence maps A ) P.falciparum flows originating from the Great Mekong sub region. B ) P.vivax flows originating from the Great Mekong sub region. The flows include estimated passenger numbers, including direct, one tr ansfer and two transfer flight routes. The inset maps show close up views for airports in the Great Mekong sub region.
92 CHAPTER 5 CONCLUSION The main purpose of this study is to examine how the interrelationships among t he global distribution of vector b orne infectious diseases, locations of known outbreaks, and international air service routes interact to result in seasonally changing risks of insect borne infectious disease transmission and spread by air travel. It presents initial efforts to integrate disparate datasets (disease/vector distribution, air travel network, and climate data etc.), to create a unified weighted network framework for vector borne diseases with spatial attributes. Up until now there has been no significant research on the comple x network analysis of air travel network changes, and how it connects endemic vector borne disease regions. This research lay s out analyses and data to facilitate a better understanding of the evolution of the risk of spread of vector borne disease s on the air travel network, coupled with the changes in the topographical structure of the network, climate, and the ecological niche of vector borne diseases. The research presented here has documented the generation of a world wide Origin Destination matrix of passenger flows in 2010 for airports with host city populations of more than 100,000. The datasets are freely accessible for academic use and are published as part of the Vector Borne Disease Airline Importation Risk (VBD Air) project at www.vbd air.com/data/ Researcher s can utilize this open database for further analyses on disease transmission on the air travel network. Moreover, t he analyses demonstrate the substantial connectivity that now exists between an d from malaria endemic regions through air travel. While the air network provides connections to previously isolated malarious regions, it is clear that great variations exist, with significant regional communities of airports connected by higher
93 rates of flow standing out. The structures of these communities are often not geographically coherent, with historical, economic and cultural ties evident, and variations between P.falciparum and P.vivax clear. Moreover, results highlight how well connected the mal aria endemic areas of Africa are now to Southeast Asia, illustrating the many possible routes that artemisinin resistant strains could take. The expansion of the air travel network is so far never anticipated to end, while our understanding of the role o f air travel in the transmission of vector borne diseases are still limited  We must prepare for this challenge by modeling and quantifying the dynamic s of human disease interaction s on the complex air travel networ k. This dissertation presents the first step towards better decision support for vector borne disease control, management, and ultimately eradication. To reduce the global burden of vector borne diseases and improve human health and wellbeing, we still hav e a long way to go.
94 APPENDIX A SUPPLEMENTARY FIGURES FOR CHAPTER 2 This document shows the supplementary figures used in chapter 2.
95 Figure A 1. P. falciparum prevalence : Plasmodium falciparum is a protozoan parasite, one of the species of Plasmodium th at cause malaria in humans. It is transmitted by the female Anopheles mosquito. P.falciparum is the most dangerous of these infections as P.falciparum (or malignant) malaria has the highest rates of complications and mortality. As of 2006 it accounted for 91% of all 250 million human malarial infections (98% in Africa) and 90% of the deaths. It is more prevalent in sub Saharan Africa than in other regions of the world; in most African countries, more than 75% of cases are due to P.falciparum whereas in mos t other countries with malaria transmission, other Plasmodia species predominate. Non endemic countries often see many cases of imported P.falciparum malaria each year through travelers or returning migrants. The Malaria Atlas Project (www.map.ox.ac.uk) re cently produced a global map of P. falciparum malaria endemicity in 2010, based on over 24,000 community prevalence surveys. The methods used behind construction of this dataset are described in more detail here: http://www.plosmedicine.org/article/info:doi/10.1371/journal.pmed.1000048
96 Figure A 2 P. vivax endemic areas: P. vivax is the most frequent and widely distributed cause of recurring (tertian) malaria, P. vivax is one of the four species of malarial parasite that commonly infect humans. It is less virulent than P. falciparum, which is the deadliest of the four, and is seldom fatal. P. vivax is carried by the female Anopheles mosquito. The Malaria Atlas Proj ect (www.map.ox.ac.uk) recently produced a global map of the limits of P. vivax malaria transmission in 2009, based on thousands of reports of cases. The methods used behind construction of this dataset are described in more detail here: www.plosntds.org/article/info:doi/10.1371/journal.pntd.0000774
97 Figure A 3 Dengue suitability map Dengue suitable areas (niche model): a map showing the predicted suitability for de ngue transmission based on thousands of reports of dengue cases. Dengue fever, also known as breakbone fever, is an infectious tropical disease caused by the dengue virus. Dengue is transmitted by several species of mosquito within the genus Aedes princ ipally Ae. aegypti but evidence also exists for Ae. Albopictus transmission. The virus has four different types infection with one type usually gives lifelong immunity to that type, but only short term immunity to the others. Subsequent infection with a d ifferent type increases the risk of severe complications. The incidence of dengue fever has increased dramatically since the 1960s, with around 50 100 million people infected yearly, and imported cases through air travel to non endemic regions are on the r ise.
98 Figure A 4 Dengue suitability in areas of known outbreaks in 2010 Dengue endemic areas (Yellow book): a map showing the predicted suitability for dengue transmission based on thousands of reports of dengue cases. This map is a refinement of the d engue suitability map in focusing solely on regions of recent confirmed transmission, as it was masked so that only regions cited by the CDC as having transmission in 2010 were left.
99 Figure A 5 Dengue outbreak prone Dengue outbreak areas (niche mo del): a map showing the predicted suitability for significant dengue outbreaks based on data on outbreaks since 2008 reported on Healthmap.
100 Figure A 6 Yellow fever suitability Yellow fever suitable areas (niche model): a map showing the predicted s uitability for yellow fever transmission based on hundreds of reports of yellow fever cases. Yellow fever is an acute viral hemorrhagic disease. The yellow fever virus is transmitted by the bite of female mosquitoes (the yellow fever mosquito, Aedes aegypt i and other species) and is found in tropical and subtropical areas in South America and Africa, but not in Asia. The WHO estimates that yellow fever causes 200,000 illnesses and 30,000 deaths every year in unvaccinated populations; around 90% of the infe ctions occur in Africa. A safe and effective vaccine against yellow fever has existed since the middle of the 20th century and some countries require vaccinations for travelers. Since no therapy is known, vaccination programs are, along with measures to re duce the population of the transmitting mosquito, of great importance in affected areas. Since the 1980s, the number of cases of yellow fever has been increasing, making it a reemerging disease, and case numbers imported through air travel to non endemic areas have been rising.
101 Figure A 7 Yellow fever outbreak prone Yellow fever outbreak areas (niche model): Yellow fever outbreak prone: a map showing the predicted suitability for significant yellow fever outbreaks based on data on outbreaks since 2008 reported on HealthMap.
102 Figure A 8 Chikungunya outbreak prone Chikungunya outbreak areas (niche model): a map showing the predicted suitability for significant chikungunya outbreaks based on data on outbreaks since 2008 reported on Health map. Chikungunya virus (CHIKV) is an insect borne virus, of the genus Alphavirus, that is transmitted to humans by virus carrying Aedes mosquitoes, principally Ae.aegypti and Ae.albopictus There have been recent breakouts of CHIKV associated with severe illness. CHIKV causes an illness with sympto ms similar to dengue fever. Large sporadic outbreaks have occurred since around 2005, with spread associated with both the movement of infected air travelers and the spread of the range of Ae. albopictus
103 Figure A 9 Aedes aegypti presence: a map show ing the predicted presence of Aedes aegypti based on hundreds of confirmed presence location data points. The yellow fever mosquito, Aedes aegypti is a mosquito that can spread the dengue fever, Chikungunya and yellow fever viruses, and other diseases. The mosquito can be recognized by white markings on legs and a marking in the form of a lyre on the thorax. The mosquito originated in Africa but is now found in tropical and subtropical regions throughout the world. Hundreds of geographically located data po ints on field caught Ae aegypti occurrence over the past 10 years were combined with climatic and environmental covariates within a regression tree mapping framework to produce a global map of predicted Ae aegypti presence.
104 Figure A 10 Predicted Ae albopictus distribution: a map showing the predicted presence of Aedes albopictus based on hundreds of confirmed presence location data points. Ae. Albopictus is of medical and public health concern because it has been shown in the laboratory to be a hig hly efficient vector of 22 arboviruses, including dengue, yellow, and West Nile fever viruses. In the wild, however, its efficiency as a vector appears to be generally low, although it has been implicated in recent dengue fever and chikungunya outbreaks in the absence of the principal vector, Ae. aegypti From its Old World east Asian distribution reported in 1930, Ae. albopictus expanded its range first to the Pacific Islands and then, within the last 20 years, to other countries in both the Old World and the New World, principally through ship borne transportation of eggs and larvae in tires. Hundreds of geographically located data points on field caught Ae albopictus occurrence over the past 10 years were combined with climatic and environmental covariates within a regression tree mapping fram ework to produce a global map of predicted Ae albopictus presence.
105 Figure A 11 Anopheles distribution: a map showing the predicted presence of one or more of the dominant Anopheles vectors of human malaria based on thousands of confirmed presence lo cation data and expert opinion maps. Anopheles is a genus of mosquito. There are approximately 460 recognized species: while over 100 can transmit human malaria, only 30 40 commonly transmit parasites of the genus Plasmodium, which cause malaria in humans in endemic areas. Anopheles gambiae is one of the best known, because of its predominant role in the transmission of the most dangerous malaria parasite species Plasmodium falciparum Thousands of geographically references data points on the presence of the dominant Anopheles vectors of malaria have been gathered and used, in combination with environmental covariates, expert opinion maps and regression tree tools, to produce global maps of Anopheles distributions by the Malaria Atlas Project (www.map.ox.a c.uk). Here, these maps were combined to produce a global map of dominant malaria vector presence.
106 Figure A 12 D ata f low in the VBD Air A data flow is presented here from raw data to decision maker. This data flow contains five stages: data collection allows business organization gathers information from various sources. Data stor age constructs an ontology model for a knowledge graph connecting different concepts in the study domain; the location information is assigned to the vertexes as attributes. Data interoperation establishes data channels for visualization. Data Visualizatio n provides interfaces to present spatial information on a web map or on a formatted report. The final object of this data flow is for decision making
107 Figure A 13 Global accessibility map a ) Global Accessibility Map to the nearest major settlement ( population size >50,000). b ) Accessibility masks to an airport generated from the Global Accessibility Map.
108 APPENDIX B SUPPLEMENTARY INFORMATION FOR CHAPTER 3 This appendix describes and compares the model s we utilize d in Chapter 3. To choose the best mo del for the predictions of air travel number, we adopted and compared these four models : Lognormal model : (1) Poisson model with variable interactions : (2) Negative Binomial Model with variable interactions : (3) Log linear model with variabl e effects and random effects on origin and destination airports (4) In these models, and describe the node characteristics for the origin airport and the destination airport. describes the route characteristics shows the interactions of routes and nodes characteristics. The first model utilizes a mixed model with all the variables transformed in a logarithmic scale. The second and the third model utilize the generalized linear model framework with link function of Poisson and
109 Negative Binomial Distribution. In model 4, denotes the random effects on the city and is identified as subjects in the mixed model. Figure B 1 and figure B 2 shows the diagnostic plots for all four models. Table 1 presents the detailed fit statistics.
110 A) B) C) D) Figure B 1 Plots for predicted value vs. the predicted value at a log scale for all four models A) Log normal model with main effects. B) Poisson prediction with variable int eraction. C) Negative Binomial model with variable interaction. D) Log normal model with variable information and random effect
111 A) B) C) D) Figure B 2 Plots for residuals vs. the predicted values at a log scale for all models A) Model 1. B) Mo del 2. C) Model 3. D) Model 4. Table B 1. Fit characteristics and variable effects for model 4 Fit Statistics 2 Res Log Likelihood 54094.8 AIC (smaller is better) 54100.8 AICC (smaller is better) 54100.8 BIC (smaller is better) 54103.1
112 Table B 2 Fixed effects for variables in model 4 Type 3 Tests of Fixed Effects Effect Num DF Den DF F Value Pr > F lop*stops 2 2E4 3.21 0.0403 ldp*stops 2 2E4 7.29 0.0007 lSX*stops 2 2E4 46.61 <.0001 lSY*stops 2 2E4 45.23 <.0001 lDOX*stops 2 2E4 2.24 0.1062 lDIY*stops 2 2E4 1.60 0.2024 lccX*stops 1 2E4 17.94 <.0001 lccY*stops 1 2E4 1.04 0.3073 lbcX*stops 2 2E4 2.43 0.0879 lbcY*stops 2 2E4 3.39 0.0339 li_d*stops 2 2E4 18.31 <.0001 lA*stops 1 2E4 13.68 0.0002 lPPPX*stops 2 2E4 4.90 0.0075 lPPPY*stops 2 2E4 5.81 0.0030 lPDCX*stops 2 2E4 1.15 0.3172 lPDCY*stops 2 2E4 2.07 0.1264 lop*dist 2 2E4 8.01 0.0003 ldp*dist 2 2E4 10.62 <.0001 lSX*dist 2 2E4 7.19 0.0008 lSY*dist 2 2E4 5.26 0.0052 lDOX*dist 2 2E4 2.51 0.0810
113 Table B 2. Continued Type 3 Tests of Fixed Effects Effect Num DF Den DF F Value Pr > F lDIY*dist 2 2E4 3.54 0.0291 lccX*dist 2 2E4 5.40 0.0045 lccY*dist 2 2E4 9.08 0.0001 lbcX*dist 2 2E4 0.78 0.4589 lbcY*dist 2 2E4 1.01 0.3652 li_d*dist 2 2E4 131.81 <.0001 lA*dist 2 2E4 11.75 <.00 01 lPPPX*dist 2 2E4 2.87 0.0567 lPPPY*dist 2 2E4 3.07 0.0464 lPDCX*dist 2 2E4 0.38 0.6867 lPDCY*dist 2 2E4 1.16 0.3123 lop*hub 15 2E4 2.55 0.0008 ldp*hub 15 2E4 2.65 0.0005 lSX*hub 15 2E4 8.77 <.0001 lSY*hub 15 2E4 10.01 <.0001 lDOX*hub 15 2E4 4.0 1 <.0001 lDIY*hub 15 2E4 4.40 <.0001 lccX*hub 13 2E4 2.09 0.0121 lccY*hub 14 2E4 1.68 0.0517 lbcX*hub 15 2E4 2.62 0.0006 lbcY*hub 15 2E4 2.90 0.0001 li_d*hub 15 2E4 34.09 <.0001
114 Table B 2. Continued Type 3 Tests of Fixed Effects Effect Num DF Den DF F Value Pr > F lA*hub 15 2E4 9.11 <.0001 lPPPX*hub 15 2E4 2.03 0.0102 lPPPY*hub 15 2E4 2.39 0.0018 lPDCX*hub 15 2E4 2.75 0.0003 lPDCY*hub 15 2E4 2.93 0.0001 lop*gdpl 8 2E4 2.53 0.0095 ldp*gdpl 8 2E4 2.50 0.0104 lSX*gdpl 8 2E4 6.33 <.0001 lSY*gd pl 8 2E4 5.60 <.0001 lDOX*gdpl 8 2E4 6.59 <.0001 lDIY*gdpl 8 2E4 6.99 <.0001 lccX*gdpl 8 2E4 2.01 0.0414 lccY*gdpl 8 2E4 1.84 0.0645 lbcX*gdpl 8 2E4 2.04 0.0382 lbcY*gdpl 8 2E4 2.10 0.0321 li_d*gdpl 8 2E4 5.07 <.0001 lA*gdpl 8 2E4 8.38 <.0001 lPPP X*gdpl 8 2E4 2.39 0.0144 lPPPY*gdpl 8 2E4 1.99 0.0433 lPDCX*gdpl 8 2E4 1.33 0.2220 lPDCY*gdpl 8 2E4 1.01 0.4262 stops 2 2E4 36.85 <.0001
115 Table B 2. Continued Type 3 Tests of Fixed Effects Effect Num DF Den DF F Value Pr > F dist 2 2E4 16.00 <.0001 hub 14 772 4.22 <.0001 Country 1 2E4 113.04 <.0001
116 APPENDIX C SUPPLEMENTARY INFORMATION FOR CHAPTER 4 This appendix presents the supplementary tables and figures for chapter 3. Table C 1 Top 10 P.f. Routes and P.v. Routes outside of the Great Mekon g Sub Region Values with (*) are returned as the adjusted largest flows between two connection flights. Outgoing P f Flow Origin City Destination City Source Country Destination Country Flow Value Number of Connections Transmission Type at Destination Yangon Harare Myanmar Zimbabwe 997.2 1 Unstable Yangon Brazzaville Myanmar Congo 287.8 2 Unstable Yangon Kinshasa Myanmar Congo 287.8 2 Unstable Yangon Mombasa Myanmar Kenya 287.8 2 Unstable Yangon Lome Myanmar Togo 287.8 2 Unstable Yangon Maputo Myanmar Mozambique 287.8 2 Unstable Yangon Ouagadougou Myanmar Burkina Faso 287.8 2 Unstable Yangon Djibouti Myanmar Djibouti 287.8 2 Stable Yangon Bamako Myanmar Mali 282.3 2 Unstable Ho Chi Minh City Harare Vietnam Zimbabwe 243.2 1 Unstable Outg oing P .v. Flow Origin City Destination City Source Country Destination Country Flow Value Number of Connections Transmission Type at Destination Bangkok Addis Ababa Thailand Ethiopia 1940.0 0 Unstable Bangkok Nairobi Thailand Kenya 1631.0 0 Unstable Ho Chi Minh City Harare Vietnam Zimbabwe 981.6 1 Unstable Yangon Harare Myanmar Zimbabwe 512.5 1 Unstable Bangkok Lusaka Thailand Zambia 434.1 1 Unstable
117 Phnom Penh Nairobi Cambodia Kenya 410.8 1 Unstable Phnom Penh Addis Ababa Cambodia Ethiopia 410.8 1 Unstable Phnom Penh Antananarivo Cambodia Madagascar 363.1 1 Unstable Siem Reap Addis Ababa Cambodia Ethiopia 275.5 1 Unstable Siem Reap Nairobi Cambodia Kenya 275.5 1 Unstable
118 Figure C 1. The travel time/distance mask to extract the prevalent ra te of P.falciparum /P.v ivax Inset map: t ravel time to the nearest major settlement (population size >50,000). Main map: e ach dot shows an airport location with a 50km buffer around it the raster is the global P.f. relevant map clip by the global access t ime map with value less than 2 hour These 2 hour and 50km thresholds were used to assign disease risks to airports (see main text).
119 Figure C 2. Air travel network communities weighted by directed estimate flow
120 Figure C 3. Communities for all poss ible connections originating from P.falciparum /P.v ivax endemic areas. A) P.falciparum multilevel membe rship. B) P.v ivax multilevel membership These two maps show direct connected, one transfer red and two transferred airports from endemic area.
121 Figur e C 4 Spatial distributions of airports with P.falciparum /P.v ivax betweenness centrality scores. A) Top airports with high betwee n ness score from P.falciparum endemic area, weighted by the P.falciparum Flow. B) Top airports with high betwee n ness score fr om P.v ivax endemic area, weighted by the P.v ivax Flow.
122 Figure C 5 Spatial distributions of airport nodes weighted by incoming P.falciparum /P.v ivax flows. A) Top airports with high incoming P.falciparum risk flows from P.falciparum endemic area, weight ed by the P.falciparum Flow. B) Top airports with high incoming P.v ivax risk flows from P.v ivax endemic area, weighted by the P.v ivax Flow.
123 LIST OF REFERENCES 1. IATA (2012) IATA 2012 Annual Review. B eijing: IATA. 2. IATA (2012) IATA Press Release: 50. Available: http://www.iata.org/pressroom/pr/Pages/2012 12 06 01.aspx. 3. Rvachev L, Longini IM (1985) A mathematical model for the global spread of influenza. Math Biosci 75: 3 22. doi:10.1016/0025 556 4(85)90064 1. 4. Colizza V, Barrat A, Barthelemy M (2007) Modeling the worldwide spread of pandemic influenza: baseline case and containment interventions. PLoS Med. 5. Tatem AJ, Rogers DJ, Hay SI, Graham a J, Goetz SJ (2006) Global transport networks an d infectious disease spread. Adv Parasit 62: 37 77. doi:10.1016/S0065 308X(05)62009 X. 6. Ruan S, Wang W, Levin S (2006) The effect of global travel on the spread of 7. Tatem AJ, Huang Z, Das A, Qi Q, Roth J, et al. (2012) Air travel and vector borne disease movement. Parasitology: 1 15. doi:10.1017/S0031182012000352. 8. Huang Z, Das A, Qiu Y, Tatem AJ (2012) Web based GIS: the vector borne disease airline importation risk (VBD AIR) tool. Int J Health Geogr 11: 33. d oi:10.1186/1476 072X 11 33. 9. Tatem AJ, Hay SI, Rogers DJ (2006) Global traffic and disease vector dispersal. Proc Natl Acad Sci 103: 6242 6247. doi:10.1073/pnas.0508391103. 10. Tatem AJ, Rogers DJ, Hay SI (2006) Estimating the malaria risk of African m osquito movement by air travel. Malaria J 5: 57. doi:10.1186/1475 2875 5 57. 11. World Health Organization (2007) The world health report 2007 A safer future: global public health security in the 21st century. Heymann D, editor. 12. Khan K, McNabb SJ, Memish ZA, Eckhardt R, Hu W, et al. (2012) Infectious disease surveillance and modelling across geographic frontiers and scientific specialties. Lancet Infect Dis 12: 222 230. doi:10.1016/S1473 3099(11)70313 9. 13. Svihrova V, Szilagyiova M, Novakova E, S vihra J, Hudeckova H (2012) Costs analysis of the treatment of imported malaria. Malaria journal 11: 1. doi:10.1186/1475 2875 11 1. 14. Jelinek T (2008) Trends in the epidemiology of dengue fever and their relevance for importation to Europe. Euro Surveil l: 1999 2001.
124 15. Odolini S, Parola P, Gkrania Klotsas E, Caumes E, Schlagenhauf P, et al. (2012) Travel related imported infections in Europe, EuroTravNet 2009. Clin Microbiol Infect 18: 468 474. doi:10.1111/j.1469 0691.2011.03596.x. 16. Centers for Dis ease Control and Prevention (2003) Local Transmission of Plasmodium vivax Malaria --Palm Beach County, Florida, 2003. MMWR Morb Mortal Wkly Rep 52: 908 911. 17. Centers for Disease Control and Prevention (2002) Local Transmission of Plasmodium vivax Mal aria --Virginia, 2002. MMWR Morb Mortal Wkly Rep 51: 921 923. 18. Danis K, Baka A, Lenglet A, Van Bortel W, Terzaki I, et al. (2011) Autochthonous Plasmodium vivax malaria in Greece, 2011. Euro Surveill 16: 1 5. 19. Powers A, Logue C (2007) Changing pa tterns of chikungunya virus: re emergence of a zoonotic arbovirus. Journal of General Virology. 20. Cohen JM, Smith DL, Cotter C, Ward A, Yamey G, et al. (2012) Malaria resurgence: a systematic review and assessment of its causes. Malaria J 11: 122. doi:1 0.1186/1475 2875 11 122. 21. Mackenzie JS, Gubler DJ, Petersen LR (2004) Emerging flaviviruses: the spread and resurgence of Japanese encephalitis, West Nile and dengue viruses. Nature medicine 10: S98 109. doi:10.1038/nm1144. 22. Phyo AP, Nkhoma S, Step niewska K, Ashley E a, Nair S, et al. (2012) Emergence of artemisinin resistant malaria on the western border of Thailand: a longitudinal study. Lancet 379: 1960 1966. doi:10.1016/S0140 6736(12)60484 X. 23. Freedman DO, Weld LH, Kozarsky PE, Fisk T, Robin s R, et al. (2006) Spectrum of disease and relation to place of exposure among ill returned travelers. New England Journal of Medicine 354: 119 130. 24. Stephan C, Allwinn R, Brodt HR, Knupp B, Preiser W, et al. (2002) Travel Acquired Dengue Infection: Cl inical Spectrum and Diagnostic Aspects. Infection 30: 225 228. doi:10.1007/s15010 002 3052 7. 25. Franco Paredes C, Santos Preciado JI (2006) Problem pathogens: prevention of malaria in travellers. The Lancet infectious diseases 6: 139 149. 26. Askling H H, Bruneel F, Buchard G, Castelli F, Chiodini PL, et al. (2012) Management of imported malaria in Europe. Malaria J 11: 328. doi:10.1186/1475 2875 11 328.
125 27. Charrel RN, De Lamballerie X, Raoult D (2007) Chikungunya outbreaks the globalization of vectorb orne diseases. New England Journal of Medicine 356: 769 771. 28. Checkley AM, Smith A, Smith V, Blaze M, Bradley D, et al. (2012) Risk factors for mortality from imported falciparum malaria in the United Kingdom over 20 years: an observational study. Bmj 344: e2116 e2116. doi:10.1136/bmj.e2116. 29. Kilpatrick AM (2011) Globalization, land use, and the invasion of West Nile virus. Science (New York, NY) 334: 323 327. doi:10.1126/science.1201010. 30. Shang C S, Fang C T, Liu C M, Wen T H, Tsai K H, et al. (2010) The role of imported cases and favorable meteorological conditions in the onset of dengue epidemics. PLoS neglected tropical diseases 4: e775. doi:10.1371/journal.pntd.0000775. 31. Pindolia DK, Garcia AJ, Wesolowski A, Smith DL, Buckee CO, et al. ( 2012) Human movement data for malaria control and elimination strategic planning. Malaria J 11: 205. doi:10.1186/1475 2875 11 205. 32. Gething P, Patil A, Smith D (2011) A new world malaria map: Plasmodium falciparum endemicity in 2010. Malaria J. 33. Ge thing PW, Elyazar IRF, Moyes CL, Smith DL, Battle KE, et al. (2012) A Long Neglected World Malaria Map: Plasmodium vivax Endemicity in 2010. PLoS Negl 6: e1814. doi:10.1371/journal.pntd.0001814. 34. Rogers DJ, Wilson a J, Hay SI, Graham a J (2006) The glo bal distribution of yellow fever and dengue. Adv Parasit 62: 181 220. doi:10.1016/S0065 308X(05)62006 4. 35. Sinka M, Bangs M, Manguin S (2010) The dominant Anopheles vectors of human malaria in Africa, Europe and the Middle East: occurrence data, distrib ution maps Vectors. 36. Sinka ME, Bangs MJ, Manguin S, Chareonviriyaphap T, Patil AP, et al. (2011) The dominant Anopheles vectors of human malaria in the Asia Pacific region: occurrence data, distribution maps and bionomic prcis. Parasites & Vectors 4: 89. 37. Sinka ME, Rubio Palis Y, Manguin S, Patil AP, Temperley WH, et al. (2010) The dominant Anopheles vectors of human malaria in the Americas: occurrence data, distribution maps and bionomic prcis. Parasites vectors 3: 72. 38. Fuller D, Troyo A, Calderon Arguedas O, Beier J (2010) Dengue vector (Aedes aegypti) larval habitats in an urban environment of Costa Rica analysed with ASTER and QuickBird imagery. International Journal of Remote Sensing 31: 3 11.
126 39. Benedict MQ, Levi ne RS, Hawley WA, Lounibos LP (2007) Spread of the tiger: global risk of invasion by the mosquito Aedes albopictus. Vector borne and zoonotic diseases (Larchmont, NY) 7: 76 85. doi:10.1089/vbz.2006.0562. 40. Lounibos LP (2002) Invasions by insect vectors of human disease. Annual review of entomology 47: 233 266. 41. Hay SI, Guerra CA, Tatem AJ, Noor AM, Snow RW (2004) Reviews The global distribution and population at 327 336. 42. Chiyaka C, Tatem A, Cohen J, Gething P, Johnston G, et al. (2013) The Stability of Malaria Elimination. Science in press. 43. IATA (2011) Annual Report 2011 International Air Transport Association. 44. Randolph SE, Rogers DJ (2010) The arrival, establishment and spread of exotic diseases: patterns and predictions. Nature Reviews Microbiology 8: 361 371. doi:10.1038/nrmicro2336. 45. Haggett P (2000) The geographical structure of epidem ics. Oxford University Press, USA. 46. Colizza V, Barrat A, Barthe M, Vespignani A (2005) The role of the airline transportation network in the prediction and predictability of global epidemics. Proc Natl Acad Sci. 47. Coelho FC, Cruz OG, Codeo CT (2008 ) Epigrass: a tool to study disease spread in complex networks. Source code for biology and medicine 3: 3. doi:10.1186/1751 0473 3 3. 48. Tatem AJ, Hay SI (2007) Climatic similarity and biological exchange in the worldwide airline transportation network. Proceedings Biological sciences / The Royal Society 274: 1489 1496. doi:10.1098/rspb.2007.0148. 49. Laird M (1984) Commerce and the spread of pests and disease vectors. 15th Pacific Science Congress, Dunedin, NZ (USA), 50. Russell RC (1987) Survival of insects in the wheel bays of a Boeing 747B aircraft on flights between tropical and temperate airports. Bull World Health Organ 65: 659 662. 51. Sinka ME, Bangs MJ, Manguin S, Coetzee M, Mbogo CM, et al. (2010) The dominant Anopheles vectors of human m alaria in Africa, Europe and the Middle East: occurrence data, distribution maps and bionomic precis. Parasites vectors 3: 117. doi:10.1186/1756 3305 3 117.
127 52. Charrel RN, De Lamballerie X, Raoult D (2008) Seasonality of mosquitoes and chikungunya in Ita ly. The Lancet infectious diseases 8: 5 6. doi:10.1016/S1473 3099(07)70296 7. 53. Guerra C a, Gikandi PW, Tatem AJ, Noor AM, Smith DL, et al. (2008) The limits and intensity of Plasmodium falciparum transmission: implications for malaria control and elimi nation worldwide. PLoS medicine 5: e38. doi:10.1371/journal.pmed.0050038. 54. Guerra CA, Howes RE, Patil AP, Gething PW, Van Boeckel TP, et al. (2010) The international limits and population at risk of Plasmodium vivax transmission in 2009. PLoS neglected tropical diseases 4: e774. 55. Whitehorn J, Farrar J (2010) Dengue. British medical bulletin. 56. Elith J, Leathwick JR, Hastie T (2008) A working guide to boosted regression trees. The Journal of animal ecology 77: 802 813. doi:10.1111/j.1365 2656.2008 .01390.x. 57. Monath T (2001) Yellow fever: an update. The Lancet infectious diseases. 58. Moffett A, Strutz S, Guda N (2009) A global public database of disease vector and 59. Foley DH, Wilkerson RC, Birney I, Harrison S, Christensen J, et al. (2010) MosquitoMap and the Mal area calculator: new web tools to relate mosquito species distribution with vector borne disease. Int J Health Geogr. 60. Gratz NG (2004) Critical review of the vector status of Ae des albopictus. Medical and veterinary entomology 18: 215 227. doi:10.1111/j.0269 283X.2004.00513.x. 61. Res C, New M, Lister D, Hulme M, Makin I (2002) A high resolution data set of surface climate over global land areas. 21: 1 25. 62. Budhathoki NR, Br uce B (Chip) C, Nedovic Budic Z (2008) Reconceptualizing the role of the user of spatial data infrastructure. GeoJournal 72: 149 160. doi:10.1007/s10708 008 9189 x. 63. Tatem AJ (2009) The worldwide airline network and the dispersal of exotic species: 200 7 2010. Ecography 32: 94 102. doi:10.1111/j.1600 0587.2008.05588.x. 64. Sanderson S (2010) Pro ASP.NET MVC 2 Framework. Apress. 65. Vlissides J, Helm R, Johnson R, Gamma E (1995) Design patterns: Elements of reusable object oriented software. Reading: Ad dison Wesley.
128 66. Cockburn A (2007) Agile software development: the cooperative game. 67. Shneiderman B (1996) The eyes have it: A task by data type taxonomy for information visualizations IEEE. pp. 336 343. 68. Johansson MA, Arana Vizcarrondo N, Bigger staff BJ, Staples JE, Gallagher N, et al. (2011) On the Treatment of Airline Travelers in Mathematical Models. PLOS ONE 6: e22151. doi:10.1371/journal.pone.0022151. 69. Patil AP, Gething PW, Piel FB, Hay SI (2011) Bayesian geostatistics in health cartogra phy: the perspective of malaria. Trends Parasitol. 70. Brownstein JS, Freifeld CC, Reis BY, Mandl KD (2008) Surveillance Sans Frontieres: Internet based emerging infectious disease intelligence and the HealthMap project. PLoS medicine 5: e151. doi:10.1371 /journal.pmed.0050151. 71. Gardner L, Fajardo D, Waller S (2012) A Predictive Spatial Model to Quantify the Risk of Air Travel Associated Dengue Importation into the United States and Europe. Journal of Tropical 72. Parker J, Epstein JM (2011) A Dist ributed Platform for Global Scale Agent Based Models of Disease Transmission. Acm T Model Comput S 22: 1 25. doi:10.1145/2043635.2043637. 73. Freifeld CC, Mandl KD, Reis BY, Brownstein JS (2008) HealthMap: global infectious disease monitoring through auto mated classification and visualization of Internet media reports. Journal of the American Medical Informatics Association 15: 150. 74. Johansson MA, Arana Vizcarrondo N, Biggerstaff BJ, Gallagher N, Marano N, et al. (2012) Assessing the risk of internatio nal spread of yellow fever virus: a mathematical analysis of an urban outbreak in Asuncion, 2008. Am J Trop Med Hyg 86: 349 358. doi:10.4269/ajtmh.2012.11 0432. 75. Wilson M (1995) Travel and the emergence of infectious diseases. Emerging infectious disea ses. 76. Wilson M (2003) The traveller and emerging infections: sentinel, courier, transmitter. Journal of applied microbiology. 77. Upham P, Thomas C, Gillingwater D, Raper D (2003) Environmental capacity and airport operations: current issues and futur e prospects. J Air Transp Manag 9: 145 151. doi:10.1016/S0969 6997(02)00078 9. 78. Smith D, Timberlake M (2001) World City Networks and Hierarchies, 1977 1997 An Empirical Analysis of Global Air Travel Links. Am Behav Sci 44: 1656 1678.
129 79. Marazzo M, Sc herre R, Fernandes E (2010) Air transport demand and economic growth in Brazil: A time series analysis. Transport Res E Log 46: 261 269. doi:10.1016/j.tre.2009.08.008. 80. Adey P, Budd L, Hubbard P (2007) Flying lessons: exploring the social and cultural geographies of global air travel. Prog Hum Geog 31: 773 791. doi:10.1177/0309132507083508. 81. Stoddard ST, Morrison AC, Vazquez Prokopec GM, Soldan VP, Kochel TJ, et al. (2009) The role of human movement in the transmission of vector borne pathogens. PLo S Negl 3: e481. doi:10.1371/journal.pntd.0000481. 82. Grais RF, Ellis JH, Glass GE (2003) Assessing the impact of airline travel on the geographic spread of pandemic influenza. European journal of epidemiology 18: 1065 1072. 83. Grais RF, Ellis JH (2004) Modeling the Spread of Annual Influenza Epidemics in the U.S.:The Potential Role of Air Travel. Health Care Manag Sci 7: 127 134. 84. Grubesic TH, Matisziw TC, Zook M a. (2008) Global airline networks and nodal regions. GeoJournal 71: 53 66. doi:10.1007/ s10708 008 9117 0. 85. Long W (1970) Air travel, spatial structure, and gravity models. Ann Reg Sci: 97 107. 86. Haynes K, Fotheringham A (1984) Gravity and spatial interaction models. 87. Brownstein J (2006) Empirical evidence for the effect of airline travel on inter regional influenza spread in the United States. PLoS medicine 3: e401. 88. Balcan D, Gonalves B, Hu H (2010) Modeling the spatial spread of infectious diseases: The GLobal Epidemic and Mobility computational model. J Comput Sci 1: 132 14 5. 89. Kenah E, Chao DL, Matrajt L, Halloran ME, Longini IM (2011) The global transmission and control of influenza. PloS one 6: e19515. doi:10.1371/journal.pone.0019515. 90. Grosche T, Rothlauf F, Heinzl A (2007) Gravity models for airline passenger vol ume estimation. J Air Transp Manag 13: 175 183. doi:10.1016/j.jairtraman.2007.02.001. 91. 106.
130 92. and Spoke Networks in Air Transportation: An Analytical Review. J Regional Sci 39: 275 295. doi:10.1111/1467 9787.00134. 93. Anal 18: 343 356. doi:10.1111/j.1538 4632.1986.tb00106.x. 94. Kuby MJ, Gray RG (1993) T he hub network design problem with stopovers and feeders: The case of Federal Express. Transport Res A Pol 27: 1 12. 95. Guimer R, Mossa S, Turtschi a, Amaral L a N (2005) The worldwide air transportation network: Anomalous centrality, community structur global roles. Proc Natl Acad Sci 102: 7794 7799. doi:10.1073/pnas.0407994102. 96. Nystuen J, Dacey M (1961) A Graph Theory Interpretation of Nodal Regions. Pap Reg Sci 34: 853 864. 97. Bhadra D (2003) Demand for air travel in the United St ates: bottom up econometric estimation and implications for forecasts by origin and destination pairs. J of Air Transp 8. 98. Taaffe E (1956) Air transportation and United States urban distribution. Geogr Rev 46: 219 238. 99. Brueckner J (2003) Airline T raffic and Urban Economic Development. Urban Stud 40: 1455 1469. doi:10.1080/0042098032000094388. 100. Jin F, Wang F, Liu Y (2004) Geographic Patterns of Air Passenger Transport in China 1980 Network Development: 37 41. 101. Goetz A (1992) Air passenger transportation and growth in the US urban system, 1950 1987. Growth Change 23: 217 238. 102. Irwin M, Kasarda J (1991) Air passenger linkages and employment growth in US metropolitan areas. Am Sociol Rev 56: 524 537. 103. Liu Z J, Debbage K, Blackburn B (2006) Locational determinants of ma jor US air passenger markets by metropolitan area. J Air Transp Manag 12: 331 341. doi:10.1016/j.jairtraman.2006.08.001. 104. Wang J, Mo H, Wang F, Jin F (2011) Exploring the network structure and nodal ex network approach. J Transp Geogr 19: 712 721. doi:10.1016/j.jtrangeo.2010.08.012. 105. Newman M (2004) Analysis of weighted networks. Phys Rev E 70: 056131.
131 106. Barrat A, Barthlemy M, Pastor Satorras R, Vespignani A (2004) The architecture of comple x weighted networks. Proc Natl Acad Sci 101: 3747 3752. doi:10.1073/pnas.0400087101. 107. Taylor PJ, Catalano G, Walker DRF (2002) Measurement of the World City Network. Urban Stud 39: 2367 2376. doi:10.1080/00420980220080011. 108. Button K, Lall S, Stou gh R, Trice M (1999) High technology employment and hub airports. J Air Transp Manag 5: 53 59. doi:10.1016/S0969 6997(98)00038 6. 109. Balcan D, Colizza V, Gonc B, Hu H (2009) Multiscale mobility networks and the spatial spreading of infectious diseases. Proc Natl Acad Sci 106: 21484 21489. 110. Snijders TAB, Bosker RJ (1994) Modeled Variance in Two Level Models. Sociological Methods & Research 22: 342 363. doi:10.1177/0049124194022003004. 111. Recchia A (2010) R squared measures for two level hierarchic al linear models using SAS. J Stat Softw 32: Code Snippet 2. 112. Trends. Eurasian Geography and Economics 48: 469 480. doi:10.2747/1538 7188.8.131.529. 113. Brons M, Pels E, Nijk amp P, Rietveld P (2002) Price elasticities of demand for passenger air travel: a meta analysis. J Air Transp Manag 8: 165 175. doi:10.1016/S0969 6997(01)00050 3. 114. Geogr 21: 76 90. 115. Neumayer E (2010) Visa restrictions and bilateral travel. The Professional Geographer: 37 41. 116. Crotts JC (2004) The Effect of Cultural Distance on Overse as Travel Behaviors. J Of Travel Res 43: 83 88. doi:10.1177/0047287504265516. 117. Bell M (1983) The estimation of an origin destination matrix from traffic counts. Transport Sci 17: 198 217. 118. Yang H, Zhou J (1998) Optimal traffic counting locations for origin destination matrix estimation. Transport Res B Meth 32: 109 126. 119. Simini F, Gonzlez MCM, Maritan A, Barabsi A L (2012) A universal model for mobility and migration patterns. Nature 484: 96 100. doi:10.1038/nature10856.
132 120. Abanyie FA, A rguin PM, Gutman J (2011) State of malaria diagnostic testing at clinical laboratories in the United States, 2010: a nationwide survey. Malaria J 10: 340. doi:10.1186/1475 2875 10 340. 121. Kain K, Harrington M, Tennyson S, Keystone J (1998) Imported mala ria: prospective analysis of problems in diagnosis and management. Clin Infect Dis 27: 142 149. 122. Nilles EJ, Arguin PM (2012) Imported malaria: an update. Am J Emerg Med 30: 972 980. doi:10.1016/j.ajem.2011.06.016. 123. Widmer LL, Blank PR, Van Herck K, Hatz C, Schlagenhauf P (2010) Cost effectiveness analysis of malaria chemoprophylaxis for travellers to West Africa. BMC Infectious Diseases 10: 279. doi:10.1186/1471 2334 10 279. 124. Isacson M (1989) Airport malaria: a review. Bull World Health Orga n 67: 737. 125. Thang H, Elsas R, Veenstra J (2002) Airport malaria: report of a case and a brief review of the literature. Neth J Med 60: 441 443. 126. Loupa C, Tzanetou K (2012) Autochthonous plasmodium vivax malaria in a Greek schoolgirl of the Attica region. Malaria J 11. 127. Wijesundera MDS (1988) Malaria outbreaks in new foci in Sri Lanka. Parasitology today (Personal ed) 4: 147 150. 128. Tatarsky A, Aboobakar S, Cohen JM, Gopee N, Bheecarry A, et al. (2011) Preventing the reintroduction of malar ia in Mauritius: a programmatic and financial assessment. PloS one 6: e23832. doi:10.1371/journal.pone.0023832. 129. Lepers J, Deloron P, Fontenille D, Coulanges P (1988) Reappearnce of Falciparum Malaria in Central Highland Plateaux of Madagascar. Lancet 331: 586. doi:10.1016/S0140 6736(88)91375 X. 130. Le Menach A, Tatem AJ, Cohen JM, Hay SI, Randell H, et al. (2011) Travel risk, malaria importation and malaria transmission in Zanzibar. Scientific reports 1: 93. doi:10.1038/srep00093. 131. Fairhurst RM Nayyar GML, Breman JG, Hallett R, Vennerstrom JL, et al. (2012) Artemisinin resistant malaria: research challenges, opportunities, and public health implications. Am J Trop Med Hyg 87: 231 241. doi:10.4269/ajtmh.2012.12 0025. 132. Anderson TJC, Nair S, Nkhoma S, Williams JT, Imwong M, et al. (2010) High heritability of malaria parasite clearance rate indicates a genetic basis for
133 artemisinin resistance in western Cambodia. J Infect Dis 201: 1326 1330. doi:10.1086/651562. 133. Dondorp AM, Yeung S, White L, Nguon C, Day NPJ, et al. (2010) Artemisinin resistance: current status and scenarios for containment. Nature Rev Microbiol 8: 272 280. doi:10.1038/nrmicro2331. 134. Trape JF (2001) The public health impact of chloroquine resistance in Africa. Am J Trop Med Hyg 64: 12 17. 135. Mita T, Venkatesan M, Ohashi J, Culleton R, Takahashi N, et al. (2011) Limited geographical origin and global spread of sulfadoxine resistant dhps alleles in Plasmodium falciparum populations. J Infect Dis 204: 1980 1988. doi:10.1 093/infdis/jir664. 136. Naidoo I, Roper C (2010) Following the path of most resistance: dhps K540E dispersal in African Plasmodium falciparum. Trends Parasitol 26: 447 456. doi:10.1016/j.pt.2010.05.001. 137. World Health Organization (2010) Global report on antimalarial drug efficacy and drug resistance: 2000 2010. Geneva: WHO. 138. Huang Z, Wu X, Garcia AJ, Fik TJ, Tatem AJ (2013) An open access modeled passenger flow matrix for the global air network in 2010. PloS One. 139. Howes RE, Patil AP, Piel FB Nyangiri O a, Kabaria CW, et al. (2011) The global distribution of the Duffy blood group. Nat Commun 2: 266. doi:10.1038/ncomms1265. 140. Freeman L (1977) A set of measures of centrality based on betweenness. Sociometry. 141. Brandes U (2001) A Faster Algorithm For Betweenness Centrality. J Math Sociol: 37 41. 142. Fortunato S (2010) Community detection in graphs. Phys Rep. 143. Tatem AJ, Smith DL (2010) International population movements and regional Plasmodium falciparum malaria elimination strategi es. Proc Natl Acad Sci 107: 12222 12227. doi:10.1073/pnas.1002971107. 144. Wesolowski A, Eagle N, Tatem AJ, Smith DL, Noor AM, et al. (2012) Quantifying the impact of human mobility on malaria. Science 338: 267 270. doi:10.1126/science.1223467.
134 145. Blon del V, Guillaume J (2008) Fast unfolding of communities in large networks. J Stat Mech Theor Exp: 1 12. 146. Bauer D (1972) Constructing confidence sets using rank statistics. J Amer Statist Assoc 67: 687 690. 147. UCSF Global Health Group (2011) Atlas o f Malaria Eliminating Countries. San Francisco. 148. Talisuna A, Karema C, Ogutu B (2012) Mitigating the threat of artemisinin resistance in Africa: improvement of drug resistance surveillance and response systems. Lancet Infect Dis 12: 888 896. doi:10.10 16/S1473 3099(12)70241 4.Mitigating. 149. burden of malaria in sub Saharan Africa. Lancet Infect Dis 10: 545 555. doi:10.1016/S1473 3099(10)70096 7. 150. Tatem AJ, Qiu Y, Smith DL, Sab ot O, Ali AS, et al. (2009) The use of mobile phone data for the estimation of the travel patterns and imported Plasmodium falciparum rates among Zanzibar residents. Malaria J 8: 287. doi:10.1186/1475 2875 8 287. 151. Tatem AJ, Hemelaar J, Gray R, Salemi M (2012) Spatial accessibility and the spread of HIV 1 subtypes and recombinants in sub Saharan Africa. AIDS. 152. Linard C, Gilbert M, Snow RW, Noor AM, Tatem AJ (2012) Population distribution, settlement patterns and accessibility across Africa in 2010. PloS One 7: e31743. doi:10.1371/journal.pone.0031743. 153. Cohen J, Woolsey A, Sabot O (2012) Optimizing Investments in Malaria Treatment and Diagnosis. Science 338: 612 614. 154. Flegg JA, Guerin PJ, White NJ, Stepniewska K (2011) Standardizing the mea surement of parasite clearance in falciparum malaria: the parasite clearance estimator. Malaria J 10: 339. doi:10.1186/1475 2875 10 339.
135 BIOGRAPHICAL SKETCH Mr. Zhuojie Huang is from Guangzhou, China. He did his undergrad study in his hometown and he finished his Master of Science in the department of geography, University of Florida. He can speak fluent Cantonese, Mandarin and English.