Table of contents

Monitoring and Prediction of Malaria Outbreaks

Jennifer Harlow, Petr Votava, Steve Running

Numerical Terradynamic Simulation Group

University of Montana

Missoula, MT 59812

jenlow321@hotmail.com, {votava, swr}@ntsg.umt.edu

Abstract

Malaria is caused by infection with Plasmodium species parasites, which are transmitted to humans through the bite of anopheline mosquitoes. Vector-borne diseases such as malaria are highly influenced by spatial and temporal changes in the environment, which are introduced mainly, but not exclusively by climate variability. Climate has been established as an important determinant in the distribution of vectors and pathogens. The purpose of my research was to study relationships between the social, climatic and economic variables, which currently best describe malaria transmission. There are few well-established methods for assessing the impacts on health due to climate change, since complex ecological systems involved are difficult to model and measure. Many factors other than climate, such as land-use, migration of people, and errors in water management influence the emergence of vector-borne diseases. Satellite observations allow a new method of health assessment due to the wide availability of climate data collected. Satellite data serves as an important source of continuous global information that can be used to monitor disease outbreaks. My objective is to use satellite data to look for relationships between climate fluctuations and malaria epidemics without overlooking important socio-economic factors that also may contribute to increased incidence of disease.

1. Background.

The complicated interaction between parasite, vector, environmental conditions and the human host contribute to the complexity of malaria. The evaluation of variables affecting malaria infection and transmission will best describe how changes in weather affect the distribution of the disease. Meteorological, biological and epidemic factors have been studied and each realm provides some evidence to the relationship between malaria outbreaks and climate change. Climatological evidence suggests that rainfall, maximum and minimum temperature and relative humidity are causal aspects in the prevalence of the mosquito vector. Biological evidence supports the observed effects of climate, for instance, increased rainfall and higher humidity lead to a higher incidence of mosquitoes, which in turn leads to an increased frequency of malaria. Vector abundance, the duration of the extrinsic incubation period and the survival rates of the vector are biological parameters that are directly influenced by meteorological variables. Evidence of epidemiological change in association with climate further links the prevalence of disease with fluctuations in the environment. Biological, epidemiological and climatic factors, together in combination, determine the stability of disease transmission and the effects of these variables result in seasonal patterns that can be monitored.

2. An evidence-based approach

2.1 Climate - A Closer Look

Statistical analysis of climate data demonstrates that disease incidence varies over time with the weather patterns. For instance, interannual climate variability is a result of climate anomalies either linked by direct evidence to El Nino Southern Oscillation (ENSO) or observed through vegetation trend variation over a span of years. These short-term interannual climate fluctuations are often unexplainable, but results are documented, and often variations in rainfall and land surface temperature are observed (Anyamba et al. 1996). El Nino is a strong determinant of interannual variability and has been associated with changes in the transmission of vector-borne diseases. El Nino events result in weather patterns directed in the Pacific, in which warming of regional land and sea surface is observed. Changes in storm tracks and changes in rainfall patterns such as floods and droughts are observed effects of an El Nino event (Kovats et al. 2001). No two El Nino events are alike, but most seem to follow a general pattern occurring every 2-7 years and lasting for approximately 12 to 18 months per cycle. Sea surface temperature (SST) increase and changes in the Southern Oscillation Index (SOI), which measures pressure differences straying from the norm across the equatorial Pacific, are indicators of an El Nino event. Sea surface temperature decrease, coastal upswelling increase and the strengthening of easterly winds mark the decay phase of an El Nino event (WHO 1999). The cooling phase of an El Nino is called La Nina.

2.2 Biological and climate evidence

Temperature effects malaria transmission for two reasons: either the minimum temperature is so low that it prevents parasite and vector development or else the temperature is too high resulting in increased mortality of the vector. A minimum temperature of 16 degrees C restricts parasite development and also prevents the development of the vector in its aquatic stages. At 17 degrees C parasites develop but not rapidly enough to cause an epidemic (Lindsay et al. 1998). Maximum temperatures in the neighborhood of 40 degrees C have been found to seriously reduce the daily survival probability of the mosquito (Martens et al. 1995). Temperature is an important factor when determining the distribution of mosquitoes since research has made it quite apparent that regions with temperatures dropping to 16 degrees C are rarely at risk for malaria epidemics and can be disregarded. Temperature can be measured at ground stations or by satellite instruments, which have the ability to measure the land surface temperature; this is an important advantage when meteorological stations are non-existent. Another climate variable that plays an important role in the mosquito life cycle is rainfall, because the majority of a mosquito’s life is spent in aquatic stages. Rainfall provides the breeding sites for mosquitoes and increases relative humidity necessary for mosquito survival (Lindsay et al. 1998). Without sufficient rainfall or water storage mosquitoes are unable to survive and as a result parasites cannot infect humans and thus malaria is not a dilemma.

2.3 Nonclimatic evidence.

Climate is an important factor in the transmission of malaria, however human activities and socio-economic factors also have a major impact on malaria transmission and should not be overlooked. The creation of new habitat for mosquitoes is observed when changes in land use, such as forest clearance and agriculture irrigation are part of the region. Forest clearance is a problem since it provides the vector with additional open sunlit pools for breeding. The modification of local microclimates further heightens mosquito population growth because shade is reduced, rainfall patterns are changed and air movement is increased (Reiter 2001; Kovats et al. 2001). Modifying microclimates can extend the life of mosquitoes by providing them with habitat that is accommodating to their growth and survival. Dams and irrigated crops are ideal places for mass production of mosquitoes, and result in more human-vector interactions, which cause an increase in malaria incidence in the human population. Ecological disturbances are unavoidable as humans are changing the environment in which they and other biological systems exist. The growth of population and urbanization is a non-climatic factor in the determination of health events such as epidemics of malaria. Issues with urbanization include open water storage and the inadequate disposal of water (Reiter 2001). When population is rapidly expanding, it opens doors to new habitats for the malaria vector, because not only are there more people in the city, but there are more places to breed in the city as well. Also of importance is the movement of people, infected people travelling to non-endemic areas can introduce disease where it would not normally occur. Conversely non-immune people are at a high risk if they move to a region where transmission is high. The periodic nature of malaria epidemics is closely related to the waxing and waning of herd immunity. The change in population immune status is a key determinate for the likelihood of an epidemic in the study of climate related health (WHO 1999). One last notable non-climatic factor associated with malaria transmission is the increased incidences of antimalarial drug and insecticide resistance. Resistance is widespread and is one of the most significant factors in the worldwide resurgence of malaria (Reiter 2001;Kovats et al. 2001). These disturbances have a profound effect on the incidence of human disease and should not be overlooked when trying to find health-climate relations.

2.4 Epidemiological evidence.

The study of drought events has led some researchers to further review the relationship between El Nino events and the occurrences of drought. Climatologists have correlated that disasters associated with drought are twice as frequent in the year following an El Nino as during other years. It has also been determined that the annual rate of persons affected by natural disasters is significantly associated with the El Nino cycle (WHO 1999). Drought has been identified as a contributing factor in increased malaria incidence, because it is believed that drought results in a reduction in herd immunity in the human population. This means that the size of the vulnerable population has increased over the course of a drought-ridden year. A second reason behind the increase of malaria after a drought may simply be a change in ecology of the natural predators of mosquitoes such as fish. A decrease in the fish population after a hot dry spell may in turn affect the vector’s dynamics allowing mosquito populations to recover more quickly following a dry year, which can result in higher transmission of malaria (WHO 1999).

3. Satellite data contribution.

Satellites are very valuable instruments in today’s world. They are used to observe climate changes as well as vegetation indexes and world population growth. Research conducted through the assistance of satellites has many applications. Satellites have expanded the capability of researchers for investigating risk factors in disease transmission. Once unconceivable, today with satellite technology we are able to obtain detailed environmental data in space and time to develop and validate models for specific disease systems in Africa. Satellites can overcome limitations presented by sampling inadequacies because instruments within the satellite incorporate variation in weather patterns. In the analysis of diverse epidemiological patterns it is important that the data used are from the same place and time. An analysis of archived satellite imagery can provide the historical environmental data reverent to previous outbreaks (Thomson et al. 1997). Satellites are continuously being improved for spectral, spatial and temporal capabilities that deliver more accurate results (Beck et al. 2000).

3.1 The use of NDVI data

The Normalized Difference Vegetation Index (NDVI) measures reflected solar radiation in the near infrared (NIR) and visible (VIS) wavelengths using the following formula

NDVI = (NIR-VIS)/(NIR+VIS)

NDVI is a measure derived by dividing the difference in infrared and red reflectance measurements by their sum. NDVI provides a measure of the amount and vigor of vegetation at the land surface. NDVI products allow for the study of short and long-term variability in climate by providing an effective measure of photosynthetically active biomass (ADDS). Thus it is also used as a measure of drought monitoring. The NDVI data set has advantages such as high temporal resolution, global coverage, and a uniform sampling scheme (Anyamba et al. 1996). NDVI is a nonlinear function that varies between –1 and +1. Values for vegetated land generally range from about 0.1 to 0.7 with values greater than 0.5 indicating dense vegetation (ADDS). NDVI patterns do not relate to any permanent spatial variation of vegetation types, but instead to the local occurrence of rainfall (Townshend et al. 1986). NDVI information relates to climate variables in a range of environmental conditions including evapotranspiration and rainfall. Over the period of the last twenty years one the main sources of NDVI data has been NOAA’s AVHRR satellite instrument. The data therefore provide a good base for evaluating the spatio-temporal response of vegetation to variation in climate over time (Anyamba et al.1996).

3.2 The use of EVI data

A new product called Enhanced Vegetation Index (EVI) is being derived from data obtained from MODIS instrument on board the Terra satellite. The new EVI product will improve upon currently available indices and will give a more precise measure of spatial and temporal vegetation change. The product uses surface reflectances, corrected for scattering, ozone absorption and aerosols. EVI is a vegetation index algorithm that has improved sensitivity into high biomass regions (NASA). MODIS EVI provides not only better vegetation measurements, but at 500 and 250 m resolution greatly increases the level of detail of spatial coverage.

4. Methods to measure

4.1 Statistical Models

The general goal of statistical models is to establish the relationship between malaria epidemics and mosquito abundance in relation to seasonal and local environmental conditions. Statistical techniques provide scientists the ability to observe patterns in empirical data in three different ways, namely regression analysis, multiple regression analysis and simulations. A form of statistical analysis called regression analysis describes the relationship between two quantitative variables as a mathematical function (Allen 1997). The environmental and climatic variables for analysis with malaria epidemics must be chosen carefully to reflect the proper factors that are significantly related to malaria transmission (Kleinschmidt et al.2000). A second method called multiple regression analysis is similar to regression analysis with one primary difference, it allows the user to estimate the form and accuracy of a relationship between a dependent variable and several independent variables at once (Allen 1997). Multiple regression analysis is a mathematical model for describing and analyzing particular types of patterns in empirical data. Models are representations designed to display the basic structure of a more complex set of phenomena, which can help researchers more easily determine the relationships between observed variables (Schafer 2000). Simulation is a third process of data analysis, in which a real phenomenon is imitated with a set of mathematical formulas. Simulations can show how a system will act differently under changed conditions or it can be used to test new theories. Useful simulations are developed when the most important factors of a weather condition or biological process are determined. Analysis by simulation is a valuable tool because it is easier to implement conceptually in complex problems (Schafer, 2000). Simulations are important in the course of study of relations between malaria and climate because they focus on the whole system, not only small sections.

4.2 Analysis - A Closer Look

The results of several studies have indicated that the basic epidemiology of malaria is determined by transmission intensity. The risk of malaria has been documented by the frequency with which a host can encounter infection; it seems that if this was the only factor affecting malaria, then an intervention aimed at reducing parasite exposure would work equally well everywhere. However, this is not the case (Omumbo et al. 1998). Evidence based on laboratory and field studies, which evaluate the sensitivity of vectors to environmental factors, shows that disease transmission alone is affected mainly by three properties; EIR, Parasite Ratio and vector survival and reproduction rates. Vector activity in the course of a year is quantified based on temporal pattern and intensity; this measure is called the Entomological Inoculation Rate (EIR) which determines the number of infectious bites an individual is exposed to in a given time period. The EIR is a form of empirical data however the sampling tools for this method are inconsistent and the number of mosquitoes collected may not all be equally infected. Therefore this data is rarely used (MARA 1998). A second measure includes the rates of development, survival and reproduction of the pathogen within vectors. This property can be measured by microscopy, which is slightly more reliable than the previous EIR. The Parasite Ratio, as this measure is called, helps relate the intensity of transmission due to climate and interannual variability (MARA 1998). The third way to measure field evidence is to determine the distribution and abundance by observing vector survival and reproduction rates (Kovats et al. 2001). Many factors define disease incidence such as malaria distribution, population mortality rates, and climatic factors. Experimental data collection is often difficult because of infrequency of surveys and also because of random distribution of data in libraries and universities. This is why satellite data, providing continuous coverage, is so helpful for researchers determining more exact relationships in disease transmission. The ultimate goal however, is to bring field evidence together with climate evidence to better illustrate how climate does affect the incidence of infectious diseases.

4.3 Biological Models vs. Climate Models. The War.

The correlation between climate variables and the distribution of vectors may be analyzed either using explicit statistical techniques or through the use of biological models based principally on the temperature dependence of mosquitoes (Rogers et al. 2000). Many researchers have created models that attribute malaria prevalence strictly due to climate. Global emergence of infectious disease however is multifactorial; these factors include land-use change, population growth, migration, control measures, and socio-economic development. Biologists studying the complexity of vector-borne diseases are cautious of existing predictions that forecast how disease distribution would change in the future, especially when predictions are based on an alteration of just one or two variables (Dye et al. 2000). Biological models are often based on transmission rates of the mosquito and often these transmission patterns cannot be satisfactorily modeled because of weakly defined parameters (Rogers et al. 2000). It is difficult to simplify biological models because of the numerous components with the vector and pathogen life cycle. Vector components consist of abundance, longevity, choice of host, and blood-feeding frequency not to mention vector susceptibility to the parasite (Dye et al. 2000). Many researchers describe biological approaches as false predictions because they cannot even accurately describe the global present day malaria situation. These biological models suggest that a large portion of Europe and America should be affected by malaria. Then why is it that malaria is no longer observed in the United States and Europe? The answer is that there are more complex variables such as health infrastructure and socio-economic conditions that cannot be described by a model that only calculates temperature, which is condusive to mosquito growth. Statistical models are gaining wider acceptance because their products better describe current distribution of malaria and uses variables that are accurately measured (Rogers et al. 2000). Recent climate model improvements have resulted in an enhanced ability to simulate many aspects of climate variability and malaria incidence. However climate models are still characterized by systematic errors and limitations in accurately simulating regional climate conditions. Multivariate statistical models give more accurate predictions because they are based on mean, maximum and minimum of three climatic variables: temperature, precipitation, and saturation vapor pressure (Rogers et al. 2000). This form of analysis provides a greater match between reality and prediction of modern day climate. It is therefore becoming more widely accepted that the application of a statistical approach is more precise than a biological approach when biological knowledge is incomplete (Rogers et al. 2000).

5. Summer Research Results and Images.

Accurately depicting events that contribute to malaria outbreaks is definitely not easy, the fact is it hasn’t been done. The majority of my summer research project was spent reading the trials and tribulations of several experiments involving research, which tries to relate climate variability to malaria distribution. Many researchers have failed miserably and are harshly criticized by their

colleagues for falsely portraying malaria distribution. I have no true professional colleagues to criticize me so this is what I learned and these are the methods I have chosen to take for observation of malaria in relation to climate.

I chose to use primarily climate data in the form of NDVI and rainfall anomalies. Since the disease reports and surveys from most countries are too incomplete to be used I steered away from using country level data. This is a major pitfall many researchers fell into, so I decided to go with climate data from the NOAA-AVHRR and NASA-MODIS satellite instruments, which are universally praised for providing data that is consistent. I did choose to use United Nations Common Database for statistics of crude death rates in African countries. The collection of data that is associated with malaria is often unreliable, so it is important to choose sources carefully.

The first stage of my analysis involved ordinary logistic regression analysis to determine the relationship between malaria prevalence and ecological predictors of malaria. The African country of Malawi was chosen for analysis, because of the severity of malaria outbreaks at that region. Graphs depicting the crude death rate for the last fifty years were created. Precipitation anomalies and amounts of monthly rainfall occurring in the last twenty years in Malawi were also depicted in separate graphs. The precipitation data was collected from the National Climatic Data Center (NCDC).

Figure 1 - Malawi Precipitation Anomaly 1980 – 2001 (source: NCDC)

Figure 2 - Malawi Precipitation Anomaly 1997 (source: NCDC)

Figure 3 - Precipitation, NSANJE, MALAWI from January 1980 till December 1990 (Source:NCDC)

Figure 4 - Malawi Crude Death Rate Trend 1953 - 2000 (source: UN Common Statistical Database)

Floods began in Malawi in February 1997 as a result of heavy rainfall. The precipitation anomaly in Figure 1 shows that an extreme amount of rainfall was received during this time period, which resulted in flooding. Floods have often been sources of malaria epidemics in the past since mosquitoes are able to reproduce rapidly in large quantities due to the increased amount of water. Figure 2 shows a further increase in a precipitation anomaly at the end of 1997. The El Nino event began in June 1997 so perhaps increased rainfall is a direct result since it has been documented that Eastern Africa receives above normal rainfall during El Nino events. Figure 3 illustrates the amount of precipitation a district in Malawi received by measuring the difference in millimeters of rainfall. The large amount of precipitation in Nsanje, Malawi during the year 1988, correlates with the increasing death rate at about the same time. The purpose of the last two graphs is only to give an example of how one can visibly see relationships between two quantitative variables. This kind of regression analysis leaves many variables out of the picture, so it shouldn’t be regarded as the only measure for determining malaria outbreaks.

An El Nino event causes a distortion in sea-surface temperatures in the Pacific Ocean. This oceanographic phenomenon is observed by warming, which creates nutrient

poor surface water across the tropical Pacific. The shift in the location of warm surface water creates a shift in the location of rain producing cloud formation. Monitoring capabilities verify precipitation and temperature anomaly patterns are relatively consistent from one El Nino episode to another (WMO 1997). The teleconnections between El Nino events, climate-related anomalies and their societal and environmental impacts are continuing to be investigated (Glantz 1999). Several studies stemming from teleconnection investigation have demonstrated that changes in the incidence of disease can occur in parallel with the extreme weather conditions associated with the El Nino Cycle. Climatic factors, such as changes in temperature and humidity, are known to be capable of assisting or interrupting the ability of insect vectors to transmit disease to humans. Many areas experience dramatic increases in the incidence of malaria during extreme weather events related to El Nino. The increased incidence of malaria results in larger more severe outbreaks, which is also a heightened threat because of fluctuating immunity due to inconsistent weather patterns (WHO 1998). Rains starting in October 1997 due to the El Nino warm event continued in Eastern and Central Africa until the end of December 1997. The observed ranges of precipitation were more than five to ten times above the normal amount of rainfall, which resulted in mass migration, damage to life and property and the emergence of infectious disease (WHO 1998).

Predicting El Nino events would be a very helpful planning tool for national health services. Through observed sea surface temperature and air pressure anomalies researchers can help save lives by informing countries of climate change. The monitoring of El Nino can provide the earliest warning of seasonally based, climate-related problems. The focus of my study was not the relation of El Nino to increased incidence of disease, but I want to point out the importance of satellite monitoring of El Nino events and the relationship between El Nino and malaria outbreaks. Figure 5 represents sea surface temperature anomalies associated with the initiation and decline of an El Nino event. The SOI graph in Figure 6 shows a form of monitoring an El Nino event based on Pacific basin pressure differences.

Figure 5 - Sea Surface Anomalies during 1997-1998 ENSO (source: NOAA)

Figure 6 - Southern Oscillation Index based on the normalized pressure difference

between Tahiti and Darwin (source: CM-Online)

Above normal NDVI anomalies persisted from October 1997 through the May 1998 over equatorial Eastern Africa. December to February is normally considered a dry season however as a result of an El Nino event it was very wet. Correlations between the temporal NDVI anomalies and ENSO indices shows that the rainy conditions over Eastern Africa were a direct result of sea surface temperature increase (Anyamba et al. 2001). Some labs are using NDVI data to describe biosphere responses to irregular climatic conditions. Through the examination of NDVI anomaly patterns researchers can begin to observe land surface reaction patterns as a result of an El Nino event. Thus, NDVI data can be used as an indicator of the biosphere response to climate variability over time (Anyamba et al. 2001).

Figure 7 - Composite Outgoing Longwave Radiation (OLR) anomalies (Wm^-2) for December-February 1997/98 (source: A. Anyamba at al.)

In Figure 7 positive anomalies indicate dry conditions while negative anomalies indicate areas of deep, cold clouds where precipitation is likely to occur. These pictures and past evidence shows that classic ENSO patterns over Africa tend to cause East Africa to receive above normal rainfall and cause South Africa to fall into a dry drought season (Anyamba et al. 2001).

My last investigation was an NDVI analysis of what is being called the largest malaria epidemic in ten years in Africa. The WHO fact sheet reports a whopping 1.4 million reported cases from November 2000 to January 2001 in Burundi, a small country in central Africa. This particular epidemic is thought to have been the consequence of the October rains, and its circumstances were similar to a recent major malaria epidemic that occurred in the western highlands of Kenya in the first part of 1999 (WER 2001). The bad set of weather conditions piled on top of a large number of internally displaced residents due to war, have created a truly catastrophic event. Burundi has been undergoing large-scale political and communal violence since the 1993 assassination of their president and the malaria epidemic only contributes to the instability of Burundi’s residents dealing with malnutrition, forced migration, and the scarcity of potable water (IDP). The Global IDP completed a profile summary of the situation in Burundi; it states that a combination of bad climate conditions, insecurity and massive forced displacement, malnutrition and disease are starting to cause more victims than war. They state that thirty percent of the population is affected by the malaria epidemic. Their report also states that the 2000 annual number of registered cases of malaria has increased five times from the 1990 figures. It has been reported that 3,018,995 people were affected by malaria in the same year.

Figure 8 - Reported cases of malaria in Burundi (1999- 2000) (source: IDP)

The epidemiology of malaria in Burundi is varied, depending a large extent upon altitude (Webster 2001). MARA an institute dedicated to mapping malaria incidence has created maps of each country showing the regions of climate suitability for malaria. The maps were developed using climatic variables only and may not provide a true representation of the malaria situation on the ground, since there are so many factors that can locally exert an influence (Webster 2001).

Figure 9 - Distribution of Stable Malaria Transmission in Burundi (source: MARA/ARMA)

The climate of Burundi is equatorial and the temperature varies with altitude, average rainfall is about 1500 mm. The wet seasons are typically from February to May and September to November, and the dry seasons are from June to August and December to January. Climatic change is one form of epidemic potential within Burundi the other form exists in the movement of non-immune people to regions of high probability for malaria epidemics. The levels of immunity depend on previous exposure, which for some internally displaced people means they have never in their whole life been exposed to malaria so when this happens later in life it results in higher mortality. Control measures have been put in place since the last epidemic; these include early diagnosis and prompt treatment, indoor insecticide spraying, increased response measures and epidemic preparedness. Any research that can contribute sound relationships with environmental variables would help decision makers and health care workers be more prepared for situations in the future (Webster 2001).

Figure 10 - NDVI Anomaly for Burundi (1981 - 2001)

The NDVI anomaly observed in 2000-2001 as shown in Figure 10 correlates well with the increased incidence of malaria during the same time period as depicted in Figure 8.

Figure 11 - Comparison of 8-km AVHRR NDVI and 500m MODIS EVI products

Improvements in satellite imagery can greatly increase the precision of monitoring and prediction of malaria outbreaks. Figure 11 show a difference between 8 km NDVI obtained from AVHRR and much more detailed 500 m EVI obtained from MODIS. Since NDVI/EVI are important variables in modeling malaria outbreaks, a better resolution and higher quality imagery provide tools for improved monitoring and prediction capabilities. NDVI and EVI by themselves are not accurate measurements, but in combination with other climate variables that can be measure by a wide range of satellite products, scientist can begin to model the environment and measure ecological changes.

6. Future Research Direction

The Howard Hughes Medical Institute Summer Research Fellowship allowed me to gain knowledge in the climate-modeling field due to the opportunity it gave me to work under Steve Running and his lab at the University of Montana. I would really like to further study climate relationship associated with malaria epidemics in Africa. Of particular interest to me is to potentially study climatic factors that may have had an impact on chloroquine drug resistance observed in the early 1980’s in much of Africa. Perhaps the late 1970’s and early 1980’s El Nino events resulted in an excess of chemoprophylaxis drugs being prescribed and as a result much of Africa began to observe the inefficiency of the drug to prevent and cure malaria. I think climate monitoring has much to offer to the human health field and it should continue to be researched by both Climatologists and Epidemiologists.

This work was funded by an IBS-CORE Undergraduate Research Fellowship to Jennifer Harlow through a grant from the Howard Hughes Medical Institute to The University of Montana.

References

Africa Data Dissemination Service (ADDS)

<http://edcsnw4.cr.usgs.gov/adds/adds.html

Allen, M.P. Understanding Regression Analysis. Plenium Press: New York. 1997.

Anyamba, A. and J.R. Eastman. 1996. Interannual variability of NDVI over Africa and its relation to EL Nino/ Southern Oscillation. Int. J. Remote Sensing.

17:13, 2533-2548.

Anyamba, A., C.J. Tucker, J.R. Eastman. 2001. NDVI anomaly patterns over Africa during the 1997/98 ENSO warm event. Int. J. Remote Sensing. 22:10, 1847-1859.

Beck, L.R., B.M. Lobitz and B.L. Wood. 2000. Remote Sensing and Human Health: New Sensors and New Opportunities. Emerging Infectious Disease.

6:3 pp. 217-227.

Climate Monitor Online. www.cru.uea.ac.uk/cru/climon/.

Dye,C. and P. Reiter. 2000. Climate Change and Malaria: Enhanced: Temperatures Without Fevers? Science. 289 (5485): 1697.

Glantz, M.H. Climate Environment and Disaster: The Case of El Nino.

www.fas.harvard.edu/~asiactr/fs_glantz2.htm. 17 May 1999.

Global IDP Project. IDPs in Burundi. www.idpproject.org.

Kleinschmidt, I., M. Bagayoko, G.P.Y. Clarke, M. Craig, D. LeSeur. 2000. A spatial statistical approach to malaria mapping. International Journal of Epidemiology. 29:355-361.

Kovats, R.S., D.H. Campbell-Lendrum, A.J. McMichael, A. Woodward and J.ST H. Cox. 2001. Early effects of climate change: do they include changes in vector-borne disease? Phil. Trans. R. Soc. Lond. B 356: 1057-1068.

Lindsay, S.W. and W.J.M. Martens. 1998. Malaria in the African highlands: past, present and future. Bulletin of the World Health Organization. 76 (1): 33-45.

MARA/ARMA. “Towards an Atlas of Malaria Risk in Africa”

http://www.mara.org.za/trview_e.htm. December 1998.

Martens, W.J.M., L.W. Niessen, J. Rotmans, T.H. Jetten and A.J. McMichael. 1995. Potential Impact of Global Climate Change on Malaria Risk. Environmental Health Perspectives. 103:5 pp. 458-464.

Medilinks - Malaria. www.medilinks.org.

NASA. Office of Satellite Data Processing and Distribution (OSDPD).

www.osdpd.noaa.gov/.

NASA/USGS EDC DAAC

http://edcdaac.usgs.gov/modis/mod13a1.html

National Climatic Data Center (NCDC). http://lwf.ncdc.noaa.gov/oa/ncdc.html.

Omumbo, J., J. Ouma, B. Rapuoda, M.H. Craig, D. Le Sueur and R.W. Snow. 1998. Mapping malaria transmission intensity using geographical information systems (GIS): an example from Kenya. Annals of Tropical Medicine & Parasitology. 92 (1): 7-21.

Reiter, P. 2001. Climate Change and Mosquito-Borne Disease. Environmental Health Perspectives. 109:1 pp.141-161.

Rogers, D.J. and S.E. Randolph. 2000. The Global Spread of Malaria in a Future, Warmer World. Science. (Reports) 289: 1763-1766

Schafer, J.L. Analysis of Incomplete Multivariate Data. Chapman & Hall/CRC. 2000.

Thomson, M.C., S.J. Conner, P. Milligan, and S.P. Flasse. 1997. Mapping Malaria Risk in Africa: What Can Satellite Data Contribute? Parasitology Today. 13:8 pp 313-318.

Townshend, J.R.G. and C.O. Justice. 1986. Analysis of the dynamics of African vegetation using the normalized difference vegetation index. Int. J. Remote Sensing. 7:11 pp. 1435-1445.

United Nations Statistical Division Common Database.

http://unstats.un.org/cgi-bin/uncdb.exe/cdb_login.

USGS EROS Data Center. Distributed Active Archive Center.

http://edcdaac.usgs.gov/modis/mod13a1.html.

Webster, Jayne. 2001. RBM Complex Emergency Malaria Data Base Burundi.

www.shtm.ac.uk/itd/dcvbu/malcon/Burundi.pdf. 3 March 2001.

Weekly Epidemiological Record. 2001. No. 1 5 January 2001.

World Health Organization. “El Nino Update”

www.reliefweb.int. 17 December 1997.

World Health Organization. “El Nino and its health impacts” Fact Sheet No 192.

www.reliefweb.int. 30 May 1998.

World Health Organization. “El Nino and Health – Protection of the Human Environment” Task Force on Climate and Health. 1999.

www.who.int/environmental_information/Climate/el_nino_and_health_contents.htm.

World Health Organization Factsheet.

www.who.int/inf-fs/en/fact090.html