Rev. sci. tech. Off. int. Epiz., 1991, 10 (1), 197-204

Satellite imaging as a technique for obtaining disease-related data M. HUGH-JONES *

Summary: Because of its initial expense, remote sensing imagery has been frequently ignored in studying the epidemiology and ecology of human and animal diseases. However, these digital images have many advantages when the theoretical restrictions and constraints on the data are understood. Remote sensing imagery has the potential significantly to improve the effectiveness and delivery of disease control programmes. As soon as it can be integrated into the operational aspects of programmes, remote sensing imagery will pass from being a data-driven research tool to being need-driven.

KEYWORDS: Disease - Habitat - Remote sensing - Satellites - Vectors.

Satellite imagery is best considered as a spatially orientated digital dataset obtained from orbiting satellites that can be used to display and compare vegetation classes and biomass, ground temperature and moisture. Because each scene or image is acquired at a known time and date, different datasets can be compared to demonstrate changes over time. A s the image can be registered to k n o w n m a p coordinates, it is possible to compare the image values of a number of k n o w n sites within the same scene, and sometimes between scenes. It can also be used with Geographic Information System (GIS) datasets that are similarly co-registered. A t its most m u n d a n e , it can be used to provide m a p s . This simple facility is frequently overlooked by those who have ready and free access to published m a p s . Unfortunately, a significant n u m b e r of countries, under the umbrella of " n a t i o n a l security", do not allow the publication of national maps and, in some instances, even the ownership of such m a p s . In addition, many ex-colonial countries have not updated their maps since independence. The digital data are derived from the reflected or emitted energy from the ground, water, or vegetation in various wavelengths (see Tables I and II). It is not the substance, but the platonic shadow of that object, that is recorded. While this might sound complex, it is really very similar to studying viruses by measuring the antibodies formed in response to an infection; instead of antibodies, however, we use the echo of objects on the surface of the earth. These echoes are affected by the internal structure of those objects and by the wavelength of the bands used. " S e e i n g " the same object over a wide range of wavelengths allows one t o observe and measure m o r e t h a n if one were just looking down from space or from a very high-flying airplane.

* Department of Epidemiology & Community Health, School of Veterinary Medicine, Louisiana State University, Baton Rouge, L A 70803, United States of America.

198 TABLE I

Thematic-mapper

(TM) spectral

bands Characteristics

Band

Wavelength (nm)

1

450

_

520

2

520

-

600

3

630

-

690

4

760

-

900

5

1,550

-

1,750

6

10,400

-

12,500

7

2,080

-

2,350

Blue-green [no multispectral scanner (MSS) equivalent]; maximum penetration of water; useful for bathymétrie mapping of shallow water, distinguishing soil from vegetation, and deciduous from coniferous plants Green (MSS-1); matches green reflectance peak of vegetation; useful for assessing plant vigour Red (MSS-2); matches chlorophyll absorption band; important for discriminating vegetation types Reflected infrared (MSS-3 & 4); useful for determining biomass and mapping shorelines, separating vegetation from dark soils Reflected infrared (IR); indicates moisture content of soil and vegetation; penetrates thin clouds and provides contrast between vegetation types Thermal IR; night-time images for thermal mapping and estimating soil moisture Reflected infrared (IR); coincides with absorption band caused by hydroxyl ions in minerals; useful for mapping hydro-thermally altered rocks and mineral deposits T A B L E II

Advanced

Very High Resolution Radiometer (A VHRR) and Operational Linear Scanner (OLS)

System

Channel

AVHRR

1

580

_

680

2

725

-

1,100

3

3,550

—

3,930

4

10,300

-

11,300

5

11,500

-

12,500

1 2

400 10,000

-

1,100 12,000

OLS

Wavelength (nm)

Comments Red-albedo For snow & ice, daytime clouds, and vegetation Reflected Infrared (IR) For shorelines & vegetation Thermal IR For fires & volcanoes, contaminated by solar radiation so nocturnal images are preferred Thermal IR For day & night land and water temperatures, volcanic plumes, sea surface temperatures and clouds Thermal IR Similar to Channel 4 Visible light Thermal infrared

199 For example, each of the seven Thematic M a p p e r bands used by L A N D S A T has specific uses, which are given in Table I. A p h o t o g r a p h of the ground using the Band 4 near-infrared reflectance will clearly show differences between water and soil/vegetation, as well as differences between various types of vegetation. The French SPOT satellite has three b a n d s , corresponding to Bands 2, 3, and 4 of L A N D S A T , but offering better resolution and more frequent viewing of the same g r o u n d . T h e Russian imagery covers a similar band range as S P O T with an even better resolution, though the availability has yet to m a t c h that of the French. As the light reflected from the earth enters the satellite, it is broken into its various wavelengths and passed over an array of electronic sensors. It is the structure of those sensors and mirrors, together with the satellite altitude, that determines the resolution of the individual d a t u m . With L A N D S A T - T M , resolution is 30m for bands 1-5 and 7,120m for the thermal band 6; with S P O T it is 10m or 20m, depending on the system being used (see Table III for details). The minimum unit of co-registered data in the image is k n o w n as a picture element, or pixel. They are set in regular rows (x) and columns (y) with a digital number (z) of the intensity of the electromagnetic energy measured for the area of ground represented by that pixel. T h u s , in a 185 k m by 170 km L A N D S A T - T M scene, there are 5,667 scan lines each with 6,167 pixels for each band, or 34.9 x 10 pixels per b a n d . With all seven bands there is a total of 244.3 x 10 per scene. The thermal data of lower resolution are replicated to provide a comparably structured dataset. 6

6

Because it is in digital form, the image is relatively easy to manipulate in the computer in order to highlight the contrast between, for example, different structures or vegetation, and to enhance linear structures. This capability can be used to produce map-like images. A most useful image can be obtained from the near-IR b a n d , as this frequency b a n d responds most strongly to vegetation and clearly differentiates land from water. A colour composite m a y be constructed from any three bands by allocating a colour to each band; by convention the near-IR b a n d is shown in red as in infrared photographic film. T h u s , actively growing crops will appear in the composite image in bright red. Because of their high albedo, roads are readily recognised and not only provide a geographic reference but also demonstrate access routes. Because water absorbs energy, water and humid soil will appear darker and thereby provide an additional ready reference. Experience has shown that new clearings in tropical forests can be recognised in SPOT or L A N D S A T imagery a n d are high risk sites for Leishmania infections. As these clearings are converted and expanded for coca planting, the risk is decreased. The various clearances in the A m a z o n forest have a characteristic RS "finger p r i n t " that can be reliably interpreted to indicate the reason for the forest clearing (mining, colonisation, cattle grazing, etc.) and each with its different but significant risk of malaria (2). The development of major irrigation and hydroelectric schemes o n the Mahaweli River near Kandy has resulted in a massive increase in malaria incidence in Sri L a n k a (17). Such sites can be readily monitored by L A N D S A T and S P O T imagery. Side-Looking Airborne Radar (SLAR) can provide equally accurate habitat identification, but at much higher resolution (and cost) for places obscured by clouds. Satellite-derived maps have been used in West Africa to find large numbers of villages missing from official maps so that they can be included in disease control programmes. As dams are built, the local flooding can be readily monitored as villages pass from the risk of onchocerciasis t o that of schistosomiasis. The size and management of these dams have an immediate impact on the risks that they present to surrounding

200 TABLE

Resolution Platforms

Resolution at nadir

Aeroplanes Low altitude (900 m-3,000 m)

< 2m

High altitude (13,650 m) (18-21,000 m) Satellites LANDSAT1-3 (918 km) LANDSAT-5 (705 km)

1

6 m

of

III

imagery Comments

Resolution ( < 1 m to 9 m) depends on camera, lens and film used (colour, B & W, CIR), or video characteristics. Resolution of Side-Looking Airborne Radar (SLAR) depends on wavelength, on 0.8 to 100 cms

9m 79 m 82 m or 30 m

TIROS-N/NOAA (850 km)

1.1 km

DMSP (850 km)

Fine 550 m smooth 2,700 m

SPOT (832 km)

10 m 20 m

KFA 1000

3-5 m

1972-1984; MSS only; 18 days to return over site; crosses 40 N at 9.30 h; 185 km swath 1984; muttispectral scanner (MSS) and TM; 16 days to return over site; crosses 40 N at 10.30 h; 185 km swath AVHRR; two polar satellites cross equator at 7.30 h & 19.30 h, and 1.00 h & 13.30 h daily; 2,700 km swath ÓLS; two polar satellites cross equator at 7.30 h & 19.30 h, and 10.30 h & 22.30 h; 3,200 km swath. Fine resolutions available for day-time visible and night-time IR bands; smooth resolutions for day­ time IR and night-time visible bands Panchromatic mode (510-730 nm); multispectral mode (500-590, 610-680, 790-890 nm); 26 days to return over site but any site can be reviewed seven times during this period; crosses 40 N at 10.00 h; 80 km swath Six bands (460-500, 520-560, 580-620, 640-680, 700-740 & 780-860 nm) available in SPOT or LANDSAT format; 80 km swath. Erratic availability

populations (13). The cumulative length of the peripheries of small dams and agricultural impoundments, which can be readily seen and measured by remote sensing, increase the risk of disease by four- to ten-fold in comparison to equivalent bodies of water held by large dams (8). The agricultural successes of these small dam schemes result in an explosive increase in their number. Remote sensing projects thus become important both for inventory and health p r o g r a m m e purposes. The great advantage of these data is that more information can be extracted from them t h a n from photographs of the same sites or regions. T h e red and near-infrared bands can be used to determine the plant biomass, leaf water and chlorophyll contents of the vegetation. A variety of combinations of these values can be used to do this, but the most commonly calculated is Tucker's normalised difference vegetation

201 index (NDVI) or, in L A N D S A T - T M band values, (Band4 - Band3)/(Band4 + Band3); refer to Tucker (15) for a comparison of these indices. It is extremely valuable for demonstrating and monitoring the greening-up and senescence of habitats. T h e seasonal N D V I responses of a specific vegetation are characteristic of that area and can be used for land-cover classification (16). If one knows the sites being monitored, the N D V I has some very useful predictive capacities for certain h u m a n and livestock disease and vector incidences; for example, for tsetse flies and h u m a n trypanosomiasis (14) and Rift Valley fever (11). It has also been used in mapping Theileria parva in Africa (10) and remotely predicting potential Amblyomma variegatum densities in Guadeloupe (7). However, the ease of N D V I calculation has resulted in extensive abuses, especially when used with Global Area Cover (GAC) 4-km resolution imagery. Daily G A C is derived by partly resampling Local A r e a Cover (LAC); this is done by averaging the first four out of five pixels in every third L A C scan line (each fifth pixel is ignored). This single subset of average values is used to represent a 3 X 5 picture element block of approximately 16 k m . A weekly, ten-day, 14-day, or 21-day composite is constructed by picking the picture elements with the highest N D V I values during that time period. This data sorting manœuver automatically minimises the number of pixels of clouds; 21-day composites will have less t h a n 2% cloud. For some purposes, the errors involved are negligible compared to the information derived while monitoring droughts or the gross seasonal changes in, say, the sub-Sahelian zone. G A C imagery for efficient locust control is based on ten-day G A C composites. A number of studies have used composite G A C because of the minimal cloud cover (and because it is readily available and relatively cheap), while individual L A C transmissions are unevenly available. The errors introduced by using G A C values for epidemiological studies for specific sites are obvious: the location of the four L A C pixel group is uncertain and satellites, drifting vertically and horizontally, make registration a case of "best-fit" so that the true location of any G A C pixel is only approximate. 2

All N D V I values will be affected by the angle of the sun, off-nadir viewing (the resolution and reflectance are affected as one moves away from a position directly under the track of the satellite), atmospheric moisture, the different calibration of each satellite (4), canopy morphology (e.g., within a species, whether the leaves are held vertically or horizontally) (1) and soil (e.g., brighter soils result in higher indices for a given quantity of incomplete vegetation cover). In areas where there are considerable variations in soil brightness from moisture differences, roughness variations, or organic content differences (quite apart from differing soils), these soil-related factors will influence the vegetation index. Readers are advised to read Huete (5), and related papers, for soil effects and how to calculate soil-corrected vegetation indices. At the present time, vegetation indices are a powerful tool for measuring vegetation response to moisture and temperature. The true utility of vegetation indices in this area of research is still under development and scrutiny. The ecology and epidemiology of a number of parasites and disease vectors have been studied by the use of remote sensing. These include riceland, puddle, forest and wetland breeding mosquitoes, tsetse flies, trematodes and ticks; a useful introductory review is Hugh-Jones (6). " G r e e n i n g - u p " dates can predict mosquito hatches of Anopheles freeborni. Further studies have shown how these numbers are affected by proximity t o grazing cattle and roosting areas (18). The importation of ecological

202 parameters, such as are described by Gabinaud (3), and of GIS will increase the accuracy of mosquito habitat maps. Not only are different Amblyomma variegatum habitats remotely identifiable, each with its own apparent tick-carrying capacity, but the latter tick densities also seem to be statistically related to b a n d reflectance variance and habitat heterogeneity and are thus remotely predictable (7). Meitzer (12) has shown that grass " p a t c h e s " of tick habitat, derived from aerial photographs, have a fractal dimension which is statistically related to the stability of their size and to the probability of disease risk. These studies demonstrate how remote sensing can increase the knowledge of vector habitats at the regional and national levels. In pragmatic terms, remote sensing should p r o m o t e the o p t i m u m location and delivery of control activities in relation to livestock and vector distributions and abundance. Density dependent and independent factors are sometimes better modelled separately. For example, in the centre of vector distributions there are multiple interacting factors; e.g., with tsetse flies and increased rainfall there are complex interactions from the responding wildlife, livestock and h u m a n populations quite apart from the effects of rainfall on the fly population itself. But at the margins of a vector population, where epidemics can be more common, the controlling factors are singular, such as temperature on tsetse flies in southwestern Africa; temperatures are either too cold, causing death, or warm enough to allow survival. Furthermore, Jovanovic (9) has proposed a range of health-related conditions based on anthropogenic, natural and environmental factors that may eventually be capable of remote monitoring. His suggestion that remote sensing should be used to map industrial pollutants demonstrates vision. Satellites can m a p crops and grazing both in time and space. Thus, by discovering the source points of pollutants, we should be able to predict which specific farm lands will be at risk as well as the probable level of risk. As satellite monitoring of the atmospheric column improves, it will be possible to plot winds and atmospheric moisture near the ground more accurately. At present, by using the TIROS Operational Vertical Sounder (TOVS), wind direction and speed is plottable at 3,000m; these data have the advantage of being obtained by the same satellite collecting A V H R R imagery. One must remember that insect disease vectors are not botanists. Disease vectors seek the optimum microclimate they can find and respond to the changes in that microclimate. It is hoped that such a microclimate, extrapolated into a habitat, will be recognised remotely. Similarly, it is the response of habitats to minerals, flooding or pollution that may indicate their potential risk to livestock and m a n . Remote sensing can both provide contemporary maps of countryside and identify the variable risk levels of different areas. By using imagery collected on different dates, it can demonstrate how those risk levels have changed, or will change, over time. Thus, the present operational value of remote sensing is the provision of satellite cartography and risk m a p s . The very concept of producing " r i s k " maps emphasises how uneven the present knowledge concerning the relationships of the environment to vectors and disease is, and how much more so is the remote monitoring of the environment. The advances will come by testing these " r i s k " maps against reality. It must also be appreciated that remote sensing data, useful though they may be in themselves, are only one or more dataseis inside a larger geographic information system. The greatest advances will come through combining RS and GIS databases in conjunction with careful field studies and ground truth. The most difficult data

203 to obtain are reliable field d a t a , especially of diseases. T h e mere production of pretty coloured pictures would be a very expensive waste of time. *

* * APPLICATION DE L'IMAGERIE PAR SATELLITE AU RECUEIL DES INFORMATIONS SUR LES MALADIES. - M. Hugh-Jones. Résumé: En raison de son coût initial, l'imagerie par télédétection est souvent ignorée dans les études épidémiologiques et écologiques des maladies humaines et animales. Cependant, sous réserve de comprendre les limites et les contraintes théoriques qui s'appliquent aux données, ces images numériques offrent de nombreux avantages. L'imagerie par télédétection permet d'améliorer significativement l'efficacité et la mise en application des programmes de prophylaxie. Dès qu'elle pourra être intégrée au volet opérationnel des programmes, l'imagerie par télédétection deviendra un outil de recherche commandé par les besoins et non plus par les données. MOTS-CLÉS : Habitat - Maladie - Satellites - Télédétection - Vecteurs.

* * * APLICACIÓN DE LA IMAGINERÍA POR SATÉLITE A LA OBTENCIÓN DE DATOS SOBRE ENFERMEDADES. - M. Hugh-Jones. Resumen: A causa de su costo inicial, la imaginería por teledetección suele ser ignorada en los estudios epidemiológicos y ecológicos de las enfermedades humanas y animales. Sin embargo, estas imágenes numéricas ofrecen numerosas ventajas a condición de comprender los límites y las restricciones teóricas que se aplican a los datos. La imaginería por teledetección permite mejorar significativamente la eficacia y la aplicación de los programas de control. En cuanto se pueda integrar en las fases operacionales de los programas, la imaginería por teledetección se convertirá en una herramienta de investigación controlada por las necesidades y no por los datos. PALABRAS CLAVE: Enfermedad - Habitat - Satélites - Teledetección Vectores.

* * * REFERENCES 1. BARTLETT D . S . , HARDISKY M.A.,

J O H N S O N R . W . , GROSS M.F.,

KLEMAS V . & H A R T M A N

J.M. (1988). - Continental scale variability in vegetation reflectance and its relationship to canopy morphology. International Journal of Remote Sensing, 9, 1223-1241. 2. CRUZ M A R Q U E S A. (1987). Parasitol. Today, 3, 166-170.

Human migration and the spread of malaria in Brazil.

3. G A B I N A U D A. (1987). - Ecological mapping to support mosquito control on the French Mediterranean coast. Parasitol. Today, 3, 317-320

204 K.P. & EIDENSHINK J . C . ( 1 9 8 8 ) . - Differences in visible and near-IR responses, and derived vegetation indices, for the NOAA-9 and NOAA-10 AVHRRs: a case study.

4 . GALLO

Photogrammetric 5. HUETE

A.R.

Engineering

Environment,

& Remote

Sensing,

54, 4 8 5 - 4 9 0 .

- A soil-adjusted vegetation index (SAVI). Remote

(1988).

Sensing of

25, 2 9 5 - 3 0 9 .

M.E. ( 1 9 8 9 ) . - Applications of remote sensing to the identification of the habitats of parasites and disease vectors. Parasitol. Today, 5, 2 4 4 - 2 5 1 .

6. H U G H - J O N E S

7 . H U G H - J O N E S M.E., BARRE N., N E L S O N G., W E H N E S K., W A R N E R J . , G A R V I N J . & GARRÍS G. ( 1 9 9 0 ) . - LANDSAT-TM identification of Amblyomma variegatum (Acari:

Ixodidae) habitats in Guadeloupe. Remote publication). J.M. & IMEVBORE A.M.A.

8. JEWSBURY

Today,

(1988).

Sensing

of Environment

(submitted for

- Small dam health studies.

Parasitol

4, 5 7 - 5 9 .

9. JOVANOVIC

Advances

P.

(1987).

- Remote sensing of environmental factors affecting health.

in Space Research,

7 ( 3 ) , 11-18.

1 0 . LESSARD P . , L ' E P L A T T E N I E R R., N O R V A L R.A.I., KUNDERT K., D O L A N T.T., CROZE H., WALKER J . B . , IRVIN A.D. & PERRY B.D. ( 1 9 9 0 ) . - Geographic information systems for

studying the epidemiology of cattle diseases caused by Theileriaparva.

Vet. Rec., 126,

2 5 5 - 2 6 2 , corrigenda 126, 5 8 2 .

C.L., G L Y N D A V I E S F. & TUCKER C . J . ( 1 9 8 7 ) . - Detection of Rift Valley Fever activity in Kenya by satellite remote sensing imagery. Science, 235,

1 1 . LINTHICUM K . J . , BAILEY 1656-1659.

M. ( 1 9 9 0 ) . - Livestock biotechnology: the economic and ecological impact of alternatives for controlling ticks and tick-borne diseases in Africa. Chapter 1 0 : a new methodology for assessing ecological impact. PhD thesis. Cornell University, 325-354. RIPERT C.L. & RACCURT C P . ( 1 9 8 7 ) . - The impact of small dams on parasitic diseases

12. MELTZER

13.

in Cameroon. Parasitol.

Today,

3, 2 8 7 - 2 8 9 .

D. ( 1 9 9 0 ) . - Tsetse flies and trypanosomiasis in Africa. In Proc. of International Conference on the Applications of Remote Sensing to Epidemiology & Parasitology. Prev. vet. Med. (in preparation). TUCKER C.J. (1979). - Red and photographic infrared linear combinations for monitoring

14. ROGERS

15.

vegetation. Remote 16. TUCKER

Sensing of Environment,

8, 127-150.

C . J . , T O W N S H E N D J.R.G. & G O F F T.E. ( 1 9 8 5 ) .

-

African land-cover

classification using satellite data. Science, 227, 369-375. 1 7 . WIJESUNDERA

Today,

M.

DE

S.

(1988).

- Malaria outbreaks in new foci in Sri Lanka. Parasitol.

4, 147-150.

18. W O O D R., W A S H I N O R., P A R I S J . , BECK L., HIBBARD K. & REJMANKOVA E. ( 1 9 9 1 ) .

-

Estimating ricefield mosquito production using remote sensing and geographic information systems. In Proc. of International Conference on the Applications of Remote Sensing to Epidemiology & Parasitology. Prev. vet. Med. (in preparation).