Ryan Staatz (05462043) and Tim Wolfe (05473345)

HUMBIO 153: Parasites and Pestilence

Interactive Disease Mapping – Malaria Atlas Project


Cartographic mapping has existed for centuries, but interactive mapping appeared only a few decades ago, when personal computers became powerful enough to run map imaging software. The combination of affordable computers and their widespread use, along with the new market of individual-consumer targeted software, created a niche for the development new applications like interactive mapping tools. Interactive mapping has enormous implications for many fields of study, but in particular, it has spurred the creation of an incredible array of new tools and technologies for the field of epidemiology. In conjunction with the digitalization of Geographic Information Systems and other stratigraphic data creation and analysis utilities, interactive mapping software has enabled professionals to produce and analyze disease patterns surprisingly fast, efficiently, and accurately. Recently, with increased dependency on internet access and the advent of cloud computing, many web-based applications for interactive disease mapping have developed that retrieve data from multiple web sources and offer a diverse amount dynamic display options. While these web applications are both powerful and useful, they lack tools and options for customizable data sets and statistical analyses. The ability of personal computers to process complex graphics combined with broadband connection speeds fostered the development of Google Earth, an interactive mapping program in the form of a virtual globe of the earth. While like the disease mapping web applications, it is lacking in statistical analyses, Google Earth has the ability to import custom data sets. Furthermore, a new file format was recently created, .KML, specifically for use in Google Earth and is being increasingly adapted as the standard for geographic overlays in interactive globe-based maps. Since a standard file format for stratigraphic data sets has been specified by this program, the provision of data sets in this format is being used by an increasing number of large research organizations, and several groups have made their data available to the public, one such group is the Malaria Atlas Project.

The Malaria Atlas Project has mapped global endemicity of Plasmodium falciparum malaria through collection of longitudinal study data from 1985 to 2008. They collected more than 8000 studies and have used these studies to predict the global parasitic rate of malaria. Their data has been publicized and available for download in .KML format for Google Earth so that it can be used in comparison and prevention efforts.




The Evolution of Disease Mapping

Historically, the idea of disease mapping has existed for over a hundred years, since John Snow’s geographic documentation of individual cases in the London cholera outbreak of 1854 (Stamp 1964). In fact, he used his mapped data (Figure 1) to determine the source of the outbreak: a well that was contaminated with cholera, and he subsequently removed the pump and stopped the outbreak (Stamp 1964). This rough archetype of geographic outbreak analysis is the cornerstone of modern epidemiology, and has led to the development of tools and techniques to assist in mapping, analyzing, and ultimately, halting the spread of the disease of interest. Additionally, John Snow’s cholera outbreak map was one of the first uses of a geographic information system (GIS), any generalized system of collecting, analyzing, and displaying geographic data, used to track the spread of a disease (Smith 2009). More recently, with the arrival of electronics, satellites, and personal computers came devices such as the handheld global positioning system (GPS), which, in addition to displaying the exact local time, can now consistently pinpoint the latitude and longitude of the holder’s current location down to a meter (United States 2010). Modern use of a GIS now almost always includes GPS coordinates in geographic data collection and construction, which has led to increased accuracy and precision of the model as a whole. Also, with the widespread use of powerful personal computers, a modern GIS is rarely strictly a physical map, and is usually constructed and displayed entirely in a digital form. However, just because it was created on a computer does not necessarily mean that the product is an interactive disease map.


Figure 1. John Snow’s Map of the 1854 Cholera Outbreak in London (Stamp 1964).


Interactive Maps

On the most basic level, interactive maps can be manipulated in real time, and more specifically, they confer user ability to change how the map, as well as what feature(s) on the map, is displayed. Google Maps is a good example of an interactive map, because the window size, map location, area size, zoom, type of map – traditional, satellite, or hybrid – can be changed very quickly. Waypoints can be designated to mark particular locations of interest and there is even a streetview option that displays street photographs from a point on a map (Google 2010). Most valuable, is the ability to display different layers of data, such as streets names, country borders, that can appear with the check of a filter (Google 2010). Thus, what distinguishes an interactive map from a traditional one is its immediate dynamicity of display. More pertinent to interactive disease mapping, however, is the relevance to epidemiologic analysis.

Spatial Analysis

For disease mapping in general, stratigraphic data has been analyzed using regression-based techniques and algorithms, but more recently, the focus has turned to spatial data analysis. Spatial data analysis, while similar to regression-based techniques in its pattern analysis, adds special weight to the spatial arrangement of data points (Gatrell 1999). There are three primary methods of spatial analysis: visualization, exploration, and modeling of data (Gatrell 1999). Visualization, how the data is displayed – usually in geographic map form, is crucial for elucidation of meaning behind the data and communication of the plot to others; exploration, how data maps relate to each other, is widely used in epidemiology, and in particular, this technique can determine spatial dependence and thus allows for density estimation of disease endemicity; modeling, the quantitative method of constructing relationships between data, is generally used for more formal ends, such as testing of experimental hypotheses (Gatrell 1999).

Early software that enabled users to perform these kinds of analyses includes Epi Info, a 20 year old program series originally designed for 1985 MS-DOS, was developed by the CDC for epidemiological analysis (CDC 2010).  While it is primarily designed for statistical analysis, a specific module of Epi Info, Epi Map, allows users basic options to overlay survey data and create GIS maps, laying the groundwork for later, more sophisticated disease mapping programs (CDC 2010).

More recently, the importance of interactive spatial analysis has been emphasized. Traditionally, plots were generated from lines of computer code as images that could not be easily changed, but more recently, the images themselves have become “interactive,” and can be manipulated in real time. Furthermore, in interactive maps, data points can be selected and immediately transformed into multiple plots or graphs that visually demonstrate the relationship of not only the relevance of one point to a plot, but the relation of that point to the same data point in another plot (Gatrell 1999). This ability to instantaneously change the way geographic disease data is presented has unbridled potential, because there are essentially an unlimited number of ways to display data, and because correlation between the same data point on each plot is preserved, disease patterns can not only be derived from correlative data between points, but also between plots. However, because this requires near-instantaneous change of appearance to user-initiated actions, dynamic graphics are necessary for this to be successful (Gatrell 1999). This combination of powerful analytical techniques and intuitive user interface immensely facilitates epidemiologic work. Some slightly older examples of software that do this are S-plus - statistical analysis software with dynamic graphics, SPLANCS - a point-based spatial epidemiologic pattern analysis, and REGARD – a plot generating application with mouse-over highlight of points linked between multiple windows of plots (Gatrell 1999).


Streaming Web-Based Interactive Maps

While many of these epidemiologic interactive mapping applications developed in the 1980s and 1990s that contain sophisticated software used for statistical analysis used by epidemiologists today, the widespread use and dependence on the internet has fundamentally changed the nature of interactive disease mapping. The advent of the internet and information age has created unparalleled speed and volume of data communication. With the ever increasing speed and capacity of both personal computers and internet connections, the amount of detail that can be communicated between persons in a very brief time also increases. Audio, images, video –nearly every possible digital data medium can now be quickly shared from person to person, and can be posted publicly on the World Wide Web for mass distribution and anonymous response. Moreover, access to both the internet and computers has increased worldwide, with programs like One Laptop per Child and the development of satellite based data connections (Negroponte 2010). This ability to quickly share and respond to limitless forms of epidemiologic data has shifted software design emphasis away from data collection, and more towards data analysis and distribution. GeoChat, a company that has created a program that links SMS messages with their geographic location, is commonly used by epidemiologists and humanitarian aid workers in undeveloped regions of the world (Instedd 2010). While this first appears to be a form of data collection, it is more importantly, a form of data organization and communication that can be used in real time analysis.

Cloud computing, the use of internet based software and storage, has gained increasing popularity and has affected the development of interactive disease mapping (Carey 2008). Nascent interactive disease maps are commonly internet-based, live streaming dynamic map web applets that collect disease data from multiple web sources. Since internet data collection is less a question of data production, and more a question of where the data can be consistently found and how reputable the source is, many interactive disease maps contain very similar data. For example, HealthMap, a global disease alert interactive map, takes streaming individual disease case data from multiple sources, such as news media, community reports, and scientific sources, and classifies them into different diseases, including parasitic ones such as salmonella, malaria, typhoid, Chagas, trypanomiasis, and trichinosis (Freifeld 2010).  As it is interactive, it has options to adjust the time range, source, disease, category, and type of map that allows near infinite display manipulation (Freifeld 2008). Manipulation by knowledgeable users can reveal disease patterns that otherwise may not have been recognized. Similarly, Biocaster’s Global Health Monitor also has the ability to adjust which internet sources, diseases, region, map type, etc are displayed, but in addition, provides trends based on monthly data for widespread disease as well as ontology and taxonomy searches (Biocaster 2010). While both of these tools are excellent examples of interactive disease maps, because while the data itself is dynamic in that it is constantly updating, the data sources are static, which means that other sources of data that are not already specified, are excluded from map display. While this is way to ensure the reliability and integrity of a source, it limits the interactive disease map to only pre-approved sources and creates a need for a more versatile interactive mapping program that can handle custom data sets, with capabilities similar to that of spatial analysis utilities mentioned previously.


Google Earth

Google Earth is a unique interactive mapping program that visualizes the earth as superimposed satellite images scaled appropriately to the spherical shape of the globe (Google 2010). While it can be considered a “virtual globe”, it still functions as a map, with roads, country borders, elevations and geographic labels. Other important features include three-dimensional urban buildings; weather patterns-specifically cloud formation, and humanitarian efforts, such as UNICEF’s water and sanitation projects (Google 2010). However, it is important to note that Google Earth, while rendered in a separate application and with the exception of user initiated data, is almost entirely streaming from the internet (Google 2010). With regard to interactive disease mapping, while epidemiologic data is not automatically retrieved from sources like HealthMap or Global Health Monitor, since the discontinuation of Google Earth Plus and addition of the associated features to the freely available downloadable version, these data sets can easily be imported (Google Earth Blog 2008).

Google Earth imports two types of customizable data files: .KML and .KMZ. KML stands for Keyhole Markup Language, which is based on the XML programming language and was originally created by Keyhole Inc, who developed the original Google Earth before it was bought by Google in 2004 (Google Earth Community 2004; Google 2010). KML encodes data that contains latitude and longitude, but may additionally have tilt, elevation, and other specifications depending on whether the geospatial object being described is a point, polygon, three-dimensional shape, etc (Google 2010). For distribution, KML files and their associated models, images, and overlays are zipped and compressed into a single archive in a .KMZ file that are commonly made available for download (Google 2010). As it pertains to interactive disease mapping, the most useful part of KML files is the ability to create geospatial overlays from stratigraphic data. In Google Earth, multiple overlays for the same geographic area and with opacity adjusted properly can provide a wealth of information about the epidemiology of a disease. This interactive spatial analysis, while it lacks modeling ability, has more than fulfilled the other two tenants of spatial analysis, visualization and exploration, because of both the three-dimensional nature of a visual earth and the near limitless array of display and filter options. Additionally, while Google Earth lacks rigorous statistical analyses and the ability to directly manipulate individual data points within the program, both of these tasks are easily accomplished within a spreadsheet program, such as Microsoft Excel that can then later be converted into the .KML or .KMZ file format for Google Earth. Thus, while .KML files in Google Earth cannot be manipulated on the level of individual data points within the program, it still is one of the most powerful interactive mapping tools in that it can import customized data points in a standard format and dynamically display multiple overlays of data.

In December of 2007, KML was announced by the Open Geospatial Consortium to be the international standard for GIS (Bacharach 2007). While it is not difficult to convert text files to excel spreadsheets to .KML files, since this standard is relatively recent, there is not a wide range of publicly available .KMZ formatted data for download. However, some agencies that collect and process large amounts of stratigraphic data, such as NASA or the NOAA, have started making their data sets available in .KMZ format so that it may be viewed on Google Earth (NASA 2010; NOAA 2010). In particular, these data sets are often weather, geologically, or ecologically related. While these are not directly related to disease mapping, these types of data are often used as correlates to disease and can be compared with actual disease overlays and valuable information concerning disease, such as prevalence and transmission, can be gleamed from this visual display. As for disease mapping, the availability of .KMZ formatted data is extremely limited, but is made available by larger research groups for diseases with widespread support, such as AIDS, tuberculosis, or malaria (WHO 2010; WHO 2010; Hay 2004). In particular, the Malaria Atlas Project stands out as prime example of a research group that not only provides publicly available .KMZ files, but also collects a vast amount data, uses verifiable sources, and has many publications concerning their work (MAP 2010).




Figure 2. Logo for the Malaria Atlas Project. (Hay 2006)


            Malaria serves as the focus of the mapping efforts by the Malaria Atlas Project (MAP). A basic understanding of the disease is necessary to understand the use of the map they created. Malaria is a parasitic disease that is present in many of the tropical areas of the world. According to the WHO there were an estimated 243 million cases of malaria and 863,000 deaths worldwide in 2008. Over 85% of the deaths were in children under 5 years of age.(WHO 2010) Several species of the protozoa Plasmodium serve as the disease agent for malaria and are spread through mosquitoes, the biological vector and definitive host. Plasmodium falciparum causes the most severe disease because of cytoadherence which allows it to enter the brain and promote neurological symptoms. In order to adequately fight malaria, it is important to understand its global endemicity. Despite affecting such a large population, the most recent global endemicity maps of malaria are from 1968. (Hay 2004).

The Malaria Atlas Project has sought to create an updated map of the relative risk of contracting malaria throughout the world. Initially, the MAP team focused on looking particularly at Plasmodium falciparum because of its high mortality rates and simple preventions measures. (Hay 2006) In order to map the disease risk, researchers used surveys of parasite rate measured by microscopy or rapid diagnostic test (RDT) between 1985 and 2008. These surveys were then filtered for strict data fidelity. The remaining surveys were then analyzed in several complex steps to yield a world malaria map that includes risk estimates, uncertainty, and is also based on population density. In the most recent map, 2007, the MAP team found 8,938 surveys and filtered them down to 7,953 for their analyses. (Hay 2009) These data were then overlaid on Google Earth to create an interactive map of malaria risk worldwide. Since the publication of the 2007 malaria endemicity map, the MAP team has begun to map Plasmodium vivax worldwide in a similar fashion. (MAP 2010)


Data Exclusion and Geospatial Predictions

In order to analyze the methods used by the MAP team (Figure 3), it is important to review some of the specific criteria that the group used for exclusion as well as the processes of data analysis leading to representation on the maps. (Hay 2009) One of the first criteria used by the team in filtering data was by the area of land that the study purported to cover. Studies that could not be geo-located were immediately discarded. Other studies which provided areas that were deemed “rough” were also excluded from the data.  In addition, studies that covered areas larger than a 5 km by 5 km area were excluded because the MAP team wanted their map to have this high spatial resolution. Immediate exclusion was also required for studies which did not provide the month they were conducted in.

Figure 3. A diagram showing the methods used by the team in data collection, exclusion, filtration, presentation, and validation. (Hay 2009)


After the data was initially filtered, it was important for the MAP team to standardize variables across sources so they could be compared as a single data set. Because the data was taken from the range 1985 until 2008, it was important to standardize the data set to a specific study age-range. The MAP data was all standardized to an age of 2 to 10 years through several complicated mathematical models that have been shown to be accurate for malaria. (Hay 2009) In order to ensure that that this transformation hasn’t produced unrealistic outliers, a geospatial filter was run on the data. The MAP team used this filter to eliminate outliers based on region and only marked outliers if their region contained relatively little standard deviation. This ensured that only implausible outliers were removed from the data set and completed the exclusion process.

            For the information to be valuable as a cohesive data set, it was important for the MAP team to create predictions based on the combined data. To do this, geospatial algorithms were used to predict values at unsampled locations based on linear combinations of the available sample data. These included parasitic rates of nearby locations, but also included information about whether a specific area is rural, peri-urban, or urban. Using these complex algorithms, a continuous overlay surface was created for the map. In order to test the process for accuracy, a random selection of 10% of the data set was removed and the algorithm was run again. (Hay 2009) The predicted values at the removed points were then compared to the recorded values and the difference was recorded as the variability in the predicting power of the model. For areas where more data was collected, there was greater accuracy in the endemicity of the parasite and the model was shown to be supported by the validation test.

            By examining the methods used by the MAP team, it is evident that the data they collected was filtered with accuracy and represents realistic findings. The data was controlled for time period as well as for the nature of the population as rural or urban in a way that allows for comparison between them. Overall, the data collection, exclusion, and filtration process does not raise serious questions about the data included.














Figure 4. Malaria endemicity map created by the Malaria Atlas Project. The maps show the locations and parasitic rates at each data source (A), classification of the parasitic rate predicted in each location as 0, <5%, 5-40%, and >40% (B), and the probability of the model to express the correct class of parasitic rate (<5, 5-40, >40) (C). (Hay 2009)



Analysis of the Endemicity Map

            The overlay of Google Earth that was created by the MAP team is the most complete model of malaria parasitic rate that has been created to this point. While it important to appreciate the large amount of data that was used in its creation, it is also important to examine concerns about the accuracy of predictions. Overall, the validation test predicted the correct class of endemicity (<5%, 5-40%, >40%) in 70.8% of the cases. (Hay 2009) In addition, only 1.1% of cases were predicted to be in a non-adjacent class. (Hay 2009) Though the model was able to accurately predict class of endemicity, it was not able to calculate the actual value of each location with much certainty. The uncertainty map (Figure 4) shows that some of the regions that were predicted to have the highest disease burden are also regions which have the most uncertainty. This is most evident in central Africa, particularly in the Democratic Republic of Congo and Angola, where the model predicted the region in the correct class of endemicity in as low as 33 percent of the time. (Hay 2009) The low accuracy reported in these regions is largely due to the lack of data collected at these locations. Generally there was more data collected on the coasts and as a result the certainty of predictions increased in these locations and reaches near 100 percent accuracy in some areas.


Implications of the Malaria Atlas Project

            Through the work of the MAP team it is possible to visualize the scale of the current malaria epidemic worldwide. The data compiled has several critical implications for the prevention and elimination of malaria. One of the most important repercussions of the endemicity map is that it shows how well the malaria epidemic has been characterized. In some regions, such as coastal regions and near major cities, the prevalence of Plasmodium falciparum infection has been shown clearly and reliably. Unfortunately, there is a severe lack of data shown by the MAP team in regions in central Africa and particularly in rural regions of the world. If the elimination of malaria is to be successful, it is important to fill in the gaps in the data that the MAP team has revealed. Without a clear picture of the epidemic, it is impossible to distribute resources to regions based on the level of the malaria epidemic present there. The framework that MAP has produced to create the malaria endemicity map is specifically designed so that it can be updated easily. The MAP team has requested submissions of additional data. (Hay 2009) They have also made their data and methods completely public so that their model can be replicated and built upon. While MAP has provided a significant improvement to the knowledge of the scale of the malaria epidemic, it is vitally important that the endemicity map created be updated frequently with a specific goal of filling in the gaps and creating an even more accurate picture of the epidemic.

            Using the map that does exist, however, there are several interesting and influential steps that can be taken towards malaria prevention and distribution of resources. Because the MAP team created their map for use in Google Earth, it is possible to very easily overlay different types of data that may have significance for combating the epidemic. The vector of malaria, the Anopheles mosquito, breeds primarily in areas of standing water where their larvae can develop. As a result, these areas seem to experience an increase in malaria disease burden. Increases in regional precipitation over the annual average can increase the population of the vector and as a result increase incidence of malaria in a given location. By overlaying the precipitation levels over the endemicity map on Google Earth it is possible to predict areas that may have an increased disease burden beyond their normal levels. Knowing this, it is possible to promote wider campaigns for vector control in these locations in order to prevent the parasitic rate of malaria from increasing in these areas. Temperature is also a determining factor in the ability of Plasmodium to develop in the mosquito. By overlaying temperature patterns, the increase or decrease of disease burden can be predicted. One major proxy for weather conditions is the normalized difference vegetation index (NDVI). Satellites detect the density and wavelength of green at different regions of the earth and this data is exported as NDVI by NASA. (NASA 2010) This can be overlaid on Google Earth as well and correlates with rainfall.

            In addition to these climate specific factors, prevention measures can be specific for certain species of the vector. The MAP team is currently working on a new project to map the endemicity of each species of mosquito that is capable of carrying malaria. (Hay 2010) Comparing the maps of mosquito and malaria endemicity can be influential for prevention measures. Many types of mosquitoes are resistant to certain pesticides or treated bed-nets. (S Smith 2010) Once this map is created it will be possible to target the areas which are most affected by malaria with the preventative tools that will be most effective against the specific type of mosquito that is present in those locations greatly increasing the efficacy of the intervention.

            Another article published by members of the MAP team seeks to map the percentages of children under 5 years of age sleeping under insecticide treated nets (ITNs). They gathered data from surveys done throughout Africa.  The percentage of children sleeping under ITNs was then predicted using statistical methods for each province. When overlaying this data on the malaria endemicity map, the MAP team predicted that in 2007 less than 40 percent of children under-5 living in endemic areas were sleeping under ITNs. (Noor 2009) By comparing the overlaid maps for ITN coverage and malaria burden, the humanitarian aid community can see the locations that they need to target with ITN delivery.


Potential Improvements

            The endemicity map of malaria created by the Malaria Atlas Project has uses beyond these few examples and has the potential to become an excellent tool for tracking the progress of elimination efforts and prevention coverage. The Malaria Atlas Project, however, could improve several aspects of their map. In its current state, the malaria endemicity map can only be updated by the creators and there are very strict limitations on what data can be included. The map currently shows parasitic rate of malaria, a statistic that must be measured through the use of a longitudinal study following a cohort and monitoring how many in that cohort develop disease. These studies are often expensive and time-consuming to conduct. As a result, there is a limited amount of data on parasitic rate of malaria. While the parasitic rate of malaria can adequately describe the number of cases expected in a given population, clinical data doesn’t necessarily relate to the population. It would be beneficial to find a way to express clinical cases of malaria that relates to population and is meaningful across the globe. If such a method were determined, then there would be much more data available on malaria because many health care providers keep records of these data. This would also create potential for a system in which the malaria endemicity map could be updated by persons other than the MAP team. While this would create issues requiring the monitoring of data validity, it has the potential to allow for a rapidly updated map of malaria endemicity. If this proposed system could be reliably maintained, it would be an even better tool for monitoring malaria disease prevalence worldwide.



The Malaria Atlas Project is a stellar example of parasite-specific interactive disease mapping, not only because of the high prevalence and worldwide burden of the disease itself, but also because this research group has compiled, and recompiled vast amounts of data to produce a most reliable, verifiable, and accurate representation of malaria endemicity. While some less accessible regions have a higher degree of uncertainty surrounding the associated data, the overall data scheme of MAP is incredibly detailed. With regard to spatial analysis, the data provided and its display within Google Earth provides extremely high resolution visualization and with this, comes high fidelity of exploration, and in particular predictive power. While the modeling aspect of spatial analysis is lacking from the .kmz data set itself, because published works on the statistical analyses and disease modeling based on this data have already been done within other spreadsheet and geospatial analysis software, this aspect of spatial analysis is not necessary for epidemiologic application. Also, the global scale and high spatial resolution of the data creates more of a gradient for analysis, rather than individual squares, further completes the visual modeling of malaria endemicity as a spectrum. Furthermore, the three-dimensional nature of Google Earth adds another dimension to terrain and geographical considerations involved in analysis of visual correlations between multiple overlays, contributing to more realistic visualization. Finally, the streaming nature of Google Earth and publicly available .KMZ data ensure that the data and geography are as current and as widely available as possible.

For epidemiology, the combination of MAP and Google Earth provides powerful insight into understanding new patterns surrounding the perpetuation and density of malaria that can lead to innovative prevention solutions. Since access to both the interactive mapping tool and the disease data are essentially uninhibited, Google Earth and .KMZ formatting have incredible potential for becoming the standard for interactive disease mapping. Since it is difficult to publicly add to the data MAP has provided, a consistent, easily accessible, accountable system for data addition and modification, especially for multiple types of data, needs to be implemented to overcome this limitation. If this limitation can be overcome for malaria, then this system has potential not only for a host of other parasitic diseases, but also for disease in general.


Works Cited

Bacharach, Sam. Open Geospatial Consortium Inc. The OGC Seeks Comment on OGC Candidate KML 2.2 Standard. 4 Dec 2007. 25 Feb 2010. <http://www.opengeospatial.org/pressroom/pressreleases/802>.

Biocaster. Global Health Monitor. 25 Feb 2010. <http://biocaster.nii.ac.jp/>.

Carey, W.P. Managing Technology. Cloud Computing: The Evolution of Software-as-a-Service. 24 June 2008. 25 Feb 2010. <http://knowledge.wpcarey.asu.edu/article.cfm?articleid=1614>.

Center for Disease Control. About Epi Info. 25 Feb 2010. <http://www.cdc.gov/epiinfo/about.htm>.

Freifeld, Clark C, Kenneth D Mandl, Ben Y Reis, and John S. Brownstein. “HealthMap: Global Infectious Disease Monitoring through Automated Classification and Visualization of Internet

Freifeld, Clark, and John Brownstein. HealthMap Global Disease Alert Map. 2007. 25 Feb 2010. <http://healthmap.org/en>.

Gatrell A.C. and T.C. Bailey. “Interactive spatial data analysis in medical geography.” Social Science and Medicine 42:6 (26 Feb 1999). 843-855. 25 Feb 2010.

Google. Google Earth. 26 Jan 2010. 25 Feb 2010.

Google. Google Maps. 25 Feb 2010. <http://maps.google.com/>.

Google. KMZ Files. 25 Feb 2010. <http://code.google.com/apis/kml/documentation/kmzarchives.html>.

Hay, Simon I. and Robert W. Snow. “The Malaria Atlas Project: Developing Global Maps of Malaria Risk.” PLOS Medicine 3.12 (2006). Web. 20 Feb. 2010. <http://www.map.ox.ac.uk/PDF/PLOSmed.pdf>.

Hay, Simon I. et al. “A World Malaria Map: Plasmodium falciparum Endemicity in 2007.” PLOS Medicine, 6.3 (2009). Web. 20 Feb. 2010. <http://www.map.ox.ac.uk/PDF/Hay_et_al_2009.pdf>.

Hay, Simon I. et al. “Developing Global Maps of the Dominant Anopheles Vectors of Human Malaria.” PLOS Medicine, 7.2 (2010). Web. 20 Feb. 2010. <http://www.map.ox.ac.uk/PDF/Hay_et_al_2010a.pdf>.

Hay, Simon I. et al. “The Global Distribution and Population at Risk of Malaria: Past, Present, and Future.” Lancet Infectious Diseases 4.6 (2004): 327–336. Web. 20 Feb. 2010. <http://users.ox.ac.uk/~hay/048.pdf>.

Instedd. Geochat. 25 Feb 2010. < http://code.google.com/p/geochat/>, <http://geochat.instedd.org/>.

Malaria Atlas Project, Oxford University. 20 Feb. 2010. Web. 20 Feb. 2010. <http://www.map.ox.ac.uk>.

Media Reports.” Journal of Medical Information Association 15 (2008) 150-157. 25 Feb 2010.

N.p. Google Earth Blog. “Google Earth Plus Discontinued.” 11 Dec 2008. 25 Feb 2010. <http://www.gearthblog.com/blog/archives/2008/12/google_earth_plus_discontinued.html>.

N.p. Google Earth Community. “Google Acquires Keyhole Corp.” 27 Oct 2004. 25 Feb 2010. <http://bbs.keyhole.com/ubb/ubbthreads.php?ubb=showflat&Number=12563&site_id=1#import>.

National Aeronautics and Space Administration. Intelligent Systems Division: Planetary Content. NASA. 20 Feb. 2010. Web. 20 Feb 2010. <http://ti.arc.nasa.gov/project/planetary/earth>.

National Aeronautics and Space Administration. Jet Propulsion Laboratory, California Institute of Technology. OnEarth. 25 Feb 2010. <http://onearth.jpl.nasa.gov/>.

National Oceanic and Atmospheric Administration. National Weather Service. National Weather Data in KML/KMZ formats. 25 Feb 2010. <http://www.srh.noaa.gov/gis/kml/>.

Negroponte, Nicholas. One Laptop per Child. 17 Feb 2010. 25 Feb 2010. <http://laptop.org/en/index.shtml>.

Noor, Abdisalan M. et al. “Insecticide-treated net coverage in Africa: mapping progress in 2000–07.” Lancet (2009); 373. 58–67. Web. 20 Feb. 2010. <http://www.map.ox.ac.uk/PDF/Noor_et_al_2009.pdf>.

Smith, Goodchild Longley. Geospatial Analysis - a comprehensive guide 3rd edition. (2006-2009). 25 Feb 2010. < http://www.spatialanalysisonline.com/output/>.

Smith, Scott. “Powerpoint Presentation on Malaria.” Parasites and Pestilence. 5 Feb. 2010.

Stamp, L. D. 1964, A Geography of Life and Death. 25 Feb 2010. <http://www.york.ac.uk/depts/maths/histstat/snow_map.htm>

United States. National Space-Based Positioning, Navigation, and Timing Coordination Office. Global Positioning System 25 Feb 2010. <http://www.gps.gov/ >.

World Health Organization. Global map of tuberculosis. 25 Feb 2010. <http://data.maplecroft.com/ge/tuberculosis.kml>.

World Health Organization. HIV/AIDS country information. 25 Feb 2010. <http://www.who.int/hiv/countries/en/>.

World Health Organization. Malaria: Data and Statistics. WHO. 20 Feb. 2010. Web. 20 Feb. 2010. <http://www.who.int/research/en/>.