body top image

Satellite Imagery in the Study and Forecast of Malaria


Box 1: Satellite Sensors for Monitoring Diseases

Satellite sensor designs are rarely ideal for epidemiological studies because of trade-offs between spectral, spatial and temporal resolution, determined by constraints of the Earth's atmosphere, or the original requirements of commissioning agencies. Passive satellite sensor data (that is, reflections or emissions arising ultimately from the Sun) have been used most commonly for epidemiological studies, and are discussed here, but there is increasing interest in radar satellites with active sensors that can produce images even under cloudy conditions.

Spectral Resolution

Satellite sensors detect reflected sunlight or infrared radiation emitted by all bodies above absolute zero. Data are most readily available in three to seven wavebands or channels in the humanvisible and near-to-thermal infrared part of the electromagnetic spectrum (0.3-14-µm wavelengths).

Spatial Resolution

Earth-observing satellites produce data with spatial resolutions of 1-4 m (Ikonos-2), 10-20 m (Satellite pour l'Observation de la Terre; SPOT), 30-120 m (Landsats 1-5) or 15-60 m (Landsat 7). Images, made up of picture elements or 'pixels' of these sizes, have swath widths of ˜11 km (Ikonos), ˜60 km (SPOT) and 185 km (Landsat). The 'vegetation instrument' on SPOT-4 has a spatial resolution of 1 km and a 2,250-km swath width.

Meteorological satellites have lower spatial resolutions, with pixel sizes down to 1.1 km (National Oceanographic and Atmospheric Administration Advanced Very High Resolution Radiometer; NOAAAVHRR), and a correspondingly wider swath width of ˜2,400 km. Geostationary satellites maintain a constant position relative to the Earth, giving spatial resolutions of 1-8 km (Geostationary Operational Environmental Satellite for the Americas) or 2.5-5 km (Meteosat 4-6 for Europe/Africa) and images of the entire Earth halfdisk.

Temporal Resolution

Satellites with a higher spatial resolution have a repeat frequency of 11 (Ikonos), 16 (Landsat) or 26 (SPOT) days. Orbiting meteorological satellites produce two images per day of the entire Earth's surface, whereas geostationary satellites produce two images per hour to monitor weather systems.

New Satellites and Sensors

New systems promise greater spectral and spatial resolutions, and greater signal stability over time. These include the Moderate Resolution Imaging Spectrometer on the Terra spacecraft (, and the Spinning Enhanced Visible and Infrared Imager on the geostationary Meteosat Second Generation (

Images for Epidemiology

Imagery is adversely affected by atmospheric contamination such as clouds and other aerosols. The low repeat frequency of the higher spatial resolution satellites prevents the recording of important seasonal determinants of pathogen transmission rates. In contrast, frequent images from NOAA-AVHRR and Meteosat sensors can be combined to produce relatively cloud-free monthly images generated from the maximum values of each signal recorded during the period (assumed to reflect cloud-free conditions), called maximum value composites.[18] These are of much greater use in studying dynamical epidemiological processes.

Data from each satellite channel may be used directly to describe epidemiological events, or may be processed to produce indices related to ground-based variables such as soil surface temperatures. Commonly used products include the middle infrared band, derived from AVHRR channel 3, and land surface temperature (LST), derived from AVHRR channels 4 and 5, both of which are related to the Earth's surface temperature; the normalized difference vegetation index, derived from AVHRR channels 1 and 2 and related to plant photosynthetic activity; near-surface air temperature, derived from LST and vegetation index measurements; and cold cloud duration from Meteosat, which is correlated with rainfall in convective precipitation systems (all reviewed in refs 18, 45).

Data Processing and Application

Monthly composited imagery often shows strong serial correlations, and therefore data redundancy, which may be overcome in two ways. The data may be subjected either to principal components analysis (PCA) and the resultant significant principal components used in further analyses, or to temporal Fourier analysis that describes natural cycles in terms of annual, bi-annual, tri-annual, and so on, components with longer or shorter periods. Temporal Fourier processing removes data redundancy and produces a set of orthogonal (uncorrelated) outputs while retaining a description of seasonality (which is lost in PCA); this is of vital interest in epidemiology.[26][46][47] Windowed Fourier analysis overcomes the problem of serial changes in the mean and variance of the data. Trends are first removed by taking the difference of the time series from a moving average spanning a number of annual cycles, and this de-trended time series is then Fourier-analysed. To deal with changing variances, a fixed aperture window is moved progressively over the de-trended time series, and only the data within the window are analysed. By comparing such analyses across the entire time series, the periodicities contributing most to changes in the overall variance can be identified (Fig. 4 on page 3).

Composited, multi-temporal and Fourier-processed satellite sensor data may be used to describe epidemiological data statistically using linear and logistic regression techniques[48][49] or various discriminant analysis and maximum likelihood approaches.[26][50] Output maps record the similarity of each pixel to the satellite-determined environmental characteristics in a sample set of sites where the epidemiological situation is well documented (the training set). Such maps can record the predicted suitability of each pixel for the presence of a vector or disease (Fig. 2 on page 2), or quantitative data related to the burden of disease (Fig. 3 on page 2).

Predictive accuracy can be assessed using a contingency table that compares the training set data and the suitability category to which the pixels were assigned. From this is calculated the overall percentage of correct predictions, the percentage of false positives and false negatives (that is, false predictions of presence and absence, respectively), and the sensitivity and specificity (proportion of positives or negatives, respectively, correctly identified). The kappa index of agreement, k, measures predictive accuracy compared with a null model (that is, one with no predictive skill); values vary between 0 (fit no better than random) and 1.0 (perfect fit),[28] with a valuse of more than 0.75 regarded as excellent; confidence intervals can be attached to k values. Once robust and reliable correlations between the satellite and disease data are established, real-time monitoring of environmental conditions by satellites can provide valuable inputs into disease early-warning systems.[16]


S.E.R. is currently supported by a NERC Senior Research Fellowship. R.W.S. is supported as a Senior Research Fellow by the Wellcome Trust. S.I.H. is currently supported as an Advanced Training Fellow by the Wellcome Trust. We thank M. Coetzee for supplying geo-referenced observations on the African distribution of the A. gambiae complex and D. Shanks for providing malaria incidence and meteorological data from the Brooke Bond Kericho Tea Estate.


David J. Rogers
TALA Research Group
Department of Zoology
University of Oxford
South Parks Road
Oxford OX1 3PS, UK


  1. Snow, R. W., Trape, J.-F. & Marsh, K. Childhood malaria mortality in Africa: past, present and future.
  2. Trends Parasitol. 17, 593-597 (2001).
  3. Rogers, D. J. Satellite imagery, tsetse and trypanosomiasis in Africa. Prev. Vet. Med. 11, 201-220 (1991).
  4. Macdonald, G. The Epidemiology and Control of Malaria (Oxford Univ. Press, London, 1957).
  5. Onori, E. & Grab, B. Indicators for the forecasting of malaria epidemics. Bull. World Health Organ. 58, 91-98 (1980).
  6. Hay, S. I., Tucker, C. J., Rogers, D. J. & Packer, M. J. Remotely sensed surrogates of meteorological data for the study of the distribution and abundance of arthropod vectors of disease. Ann. Trop. Med. Parasitol. 90, 1-19 (1996).
  7. Anderson, R. M. & May, R. M. Infectious Diseases of Humans: Dynamics and Control (Oxford Univ. Press, Oxford, 1991).
  8. Yacob, K. B. M. & Swaroop, S. Investigation of long-term periodicity in the incidence of epidemic malaria in the Punjab. J. Malaria Instit. India 6, 39-51 (1945).
  9. Rogers, D. J. The dynamics of vector-transmitted diseases in human communities. Phil. Trans. R. Soc. Lond. B 321, 513-539 (1988).
  10. Christophers, S. R. Epidemic malaria of the Punjab, with a note on a method of predicting epidemic years. Paludism 2, 17-26 (1911).
  11. Gill, C. A. The prediction of malaria epidemics. Ind. J. Med. Res. 10, 1136-1143 (1923).
  12. Swaroop, S. Forecasting of epidemic malaria in the Punjab, India. Am. J. Trop. Med. 29, 1-17 (1949).
  13. Najera, J. A., Kouznetsov, R. L. & Delacollette, C. Malaria Epidemics: Detection and Control Forecasting and Prevention (World Health Organization, Geneva, 1998).
  14. National Research Council. Under the Weather: Climate, Ecosystems, and Infectious Disease (Committee on Climate, Ecosystems, Infectious Diseases, and Human Health, Board on Atmospheric Sciences and Climate, National Research Council, Washington DC, 2001).
  15. Snow, R. W., Craig, M., Deichmann, U. & Marsh, K. Estimating mortality, morbidity and disability due to malaria among Africa's non-pregnant population. Bull. World Health Organ. 77, 624-640 (1999).
  16. World Health Organization. The World Health Report 1999: Making a Difference (World Health Organization, Geneva, 1999).
  17. Myers, M. F., Rogers, D. J., Cox, J., Flauhalt, A. & Hay, S. I. Forecasting disease risk for increased epidemic preparedness in public health. Adv. Parasitol. 47, 309-330 (2000).
  18. World Health Organization. Malaria Early Warning Systems, a Framework for Field Research in Africa: Concepts, Indicators and Partners (World Health Organization, Geneva, 2001).
  19. Hay, S. I. An overview of remote sensing and geodesy for epidemiology and public health application. Adv. Parasitol. 47, 1-35 (2000).
  20. Hay, S. I., Omumbo, J., Craig, M. & Snow, R. W. Earth observation, geographic information systems and Plasmodium falciparum malaria in sub-Saharan Africa. Adv. Parasitol. 47, 173-215 (2000).
  21. Snow, R. W. & Marsh, K. Will reducing Plasmodium falciparum transmission alter malaria mortality among African children? Parasitol. Today 11, 188-190 (1995).
  22. Smith, T. A., Leuenberger, R. & Lengeler, C. Child mortality and malaria transmission intensity in Africa. Trends Parasitol. 17, 145-149 (2001).
  23. Mapping Malaria Risk in Africa/Atlas du Risque de Malaria en Afrique (MARA/ARMA) collaboration. Towards an Atlas of Malaria Risk in Africa. First Technical Report of the MARA/ARMA (Mapping Malaria Risk in Africa) Collaboration (MARA/ARMA, Durban, 1998).
  24. Craig, M. H., Snow, R. W. & le Sueur, D. A climate-based distribution model of malaria transmission in sub-Saharan Africa. Parasitol. Today 15, 105-111 (1999).
  25. Coetzee, M., Craig, M. H. & le Sueur, D. Distribution of African malaria mosquitoes belonging to the Anopheles gambiae complex. Parasitol. Today 16, 74-77 (2000).
  26. Lindsay, S. W., Parson, L. & Thomas, C. J. Mapping the ranges and relative abundance of the two principal African malaria vectors, Anopheles gambiae sensu stricto and An. arabiensis, using climate data. Proc. R. Soc. Lond. B 265, 847-854 (1998).
  27. Rogers, D. J. Satellites, space, time and the African trypanosomiases. Adv. Parasitol. 47, 129-171 (2000).
  28. Hay, S. I., Rogers, D. J., Toomer, J. F. & Snow, R. W. Annual Plasmodium falciparum entomological inoculation rates (EIR) across Africa: literature survey, internet access and review. Trans. R. Soc. Trop. Med. Hyg. 94, 113-127 (2000).
  29. Congalton, R. G. A review of assessing the accuracy of classifications of remotely sensed data. Rem. Sens. Environ. 37, 35-46 (1991).
  30. Hay, S. I., Snow, R. W. & Rogers, D. J. Predicting malaria seasons in Kenya using multitemporal meteorological satellite sensor data. Trans. R. Soc. Trop. Med. Hyg. 92, 12-20 (1998).
  31. Hay, S. I., Snow, R. W. & Rogers, D. J. From predicting mosquito habitat to malaria seasons using remotely sensed data: practice, problems and perspectives. Parasitol. Today 14, 306-313 (1998).
  32. Butler, R. J. Atlas of Kenya: A Comprehensive Series of New and Authenticated Maps Prepared from the National Survey and other Governmental Sources with Gazetteer and Notes on Pronunciation and Spelling (The Survey of Kenya, Nairobi, 1959).
  33. Hay, S. I. et al. Etiology of interepidemic periods of mosquito-borne disease. Proc. Natl Acad. Sci. USA 97, 9335-9339 (2000).
  34. Kovats, R. S., Bouma, M. J. & Haines, A. El Niño and Health (World Health Organization, Geneva, 1999).
  35. McGregor, G. R. & Nieuwolt, S. Tropical Climatology (Wiley, Chichester, 1998).
  36. Martens, W. J. M. et al. Climate change and future populations at risk of malaria. Global Environ. Change 9, S89-S107 (1999).
  37. Rogers, D. J. & Randolph, S. E. The global spread of malaria in a future, warmer world. Science 289, 1763-1766 (2000).
  38. Hay, S. I. et al. Climate change and the resurgence of malaria in the East African highlands. Nature (in the press).
  39. Shanks, G. D., Biomndo, K., Hay, S. I. & Snow, R. W. Changing patterns of clinical malaria since 1965 among a tea estate population located in the Kenyan highlands. Trans. R. Soc. Trop. Med. Hyg. 94, 253-255 (2000).
  40. Rogers, D. J. & Randolph, S. E. Mortality rates and population density of tsetse flies correlated with satellite imagery. Nature 351, 739-741 (1991).
  41. Rogers, D. J. A general model for tsetse populations. Insect Sci. Applic. 11, 331-346 (1990).
  42. Randolph, S. E. & Rogers, D. J. A generic population model for the Africa tick Rhipicephalus appendiculatus. Parasitology 115, 265-279 (1997).
  43. Green, P. E. Analyzing Multivariate Data (The Dryden Press, Hinsdale, IL, 1978).
  44. Wolter, K. The Southern Oscillation in surface circulation and climate over the tropical Atlantic, Eastern Pacific, and Indian Ocean as captured by cluster analysis. Clim. Appl. Meteorol. 26, 540-558 (1987).
  45. Wolter, K. & Timlin, M. S. Measuring the strength of ENSO-how does 1997/98 rank? Weather 53, 315-324 (1998).
  46. Goetz, S., Prince, S. & Small, J. Advances in satellite remote sensing of environmental variables for epidemiological applications. Adv. Parasitol. 47, 289-307 (2000).
  47. Rogers, D. J. & Williams, B. G. in Large-Scale Ecology and Conservation Biology (35th Symp. Br. Ecol. Soc./Soc. Conserv. Biol., Univ. Southampton, 1993) (eds Edwards, P. J., May, R. M. & Webb, N. R.) 249-273 (Blackwell Scientific Publications, Oxford, 1994).
  48. Rogers, D. J., Hay, S. I. & Packer, M. J. Predicting the distribution of tsetse flies in West Africa using temporal Fourier processed meteorological satellite data. Ann. Trop. Med. Parasitol. 90, 225-241 (1996).
  49. Augustin, N. H., Mugglestone, M. A. & Buckland, S. T. An autologistic model for the spatial distribution of wildlife. J. Appl. Ecol. 33, 339-347 (1996).
  50. Manel, S., Dias, J. M., Buckton, S. T. & Ormerod, S. J. Alternative methods for predicting species distributions: an illustration with Himalayan river birds. J. Appl. Ecol. 36, 734-747 (1999).
  51. Rogers, D. J. & Randolph, S. E. Distribution of tsetse and ticks in Africa: past, present and future. Parasitol. Today 9, 266-271 (1993).

page 1 | page 2 | page 3 | page 4 | top  

filler image
body bottom image