detection, identification, of sensing scanners

Bulletin ofthe World Health Organization, 63 (2): 361-374 (1985)

Detection, identification, and classification ofmosquito larval habitats using remote sensing scannersin earth-orbiting satellites*

RICHARD 0. HAYES,' EUGENE L. MAXWELL,2 CARL J. MITCHELL,3& THOMAS L. WOODZICK4

A method of identifying mosquito larval habitats associated with fresh-water plantcommunities, wetlands, and other aquatic locations at Lewis and Clark Lake in the states ofNebraska and South Dakota, USA, using remote sensing imagery obtained by multispectralscanners aboard earth-orbiting satellites (Landsat 1 and 2) is described. The advantages andlimitations of this method are discussed.

INTRODUCTION

Mosquitos and other insects connected withaquatic habitats and water impoundments have longbeen recognized as pests and a public health problem.These problems can be prevented or controlled byphysical (including water management), biological,and chemical means. Although chemical pesticideshave frequently been used to control mosquitos andother insect pests, the techniques preferred by mostmosquito control experts involve water management,source reduction, and other preventive measures.Mosquito control methods, except for chemical treat-ment against adult mosquitos, usually require aknowledge of the location of the aquatic habitats inwhich the mosquito larvae are developing("breeding"). Therefore, it is important thatmosquito control projects should begin with surveysto locate, characterize, and measure the extent ofthese larval habitats.

This article describes the results of using remotesensing imagery obtained by multispectral scanners

* This study was supported by the Water Resources Branch,Vector-Borne Diseases Division, Centers for Disease Control, PublicHealth Service, US Department of Health and Human Services, P.O.Box 2087, Fort Collins, CO 80522, USA. Requests for reprintsshould be sent to this address.

' Medical Entomologist Consultant, 11 18 Centennial Road, FortCollins, CO, USA.

2 Senior Scientist, Solar Energy Research Institute, Golden, CO,USA.

3 Chief, Vector Virology Laboratory, Division of Vector-BorneViral Diseases, Centers for Disease Control, Fort Collins, CO,USA.

4 Graduate Assistant, Department of Earth Resources, ColoradoState University, Fort Collins, CO, USA.

aboard earth-orbiting satellites (Landsat 1 and 2) tosurvey the aquatic habitats and vegetation associatedwith mosquito breeding in the vicinity of a largeimpoundment. Like all organisms, each species ofmosquito is adapted to a specific ecological niche.Larvae of some species can tolerate only a verynarrow ecological range, whereas larvae of otherspecies are more broadly adapted and may be widelydistributed in a great variety of habitats.

Bidlingmayer & Klock in 1955 reported thathydraulic tidal action and salt-marsh flora in thesouth-eastern United States were of primary impor-tance in influencing the multiplication of Aedestaeniorhynchus and Ae.sollicitans (3). Rioux andcoworkers in 1968 developed a phyto-ecological mapsystem in which they categorized four broad units and30 plant communities in salt-marsh mosquito areas insouthern France (16). In 1969, Cousserans et al. (6)showed a correlation between the various types ofplant communities in southern France and themosquito egg and larval habitats and indicated thevalue of that approach for mosquito control; Pautouin 1973 extended the findings and determined theaquatic habitat associations for a large number ofmosquito species in the Mediterranean coastal,Rhone-Alpes, and Atlantic coastal regions of France(15). Similarly, Jolivet et al. on Wido Island of theRepublic of Korea,a Provost in the eastern USA,b

a JOLIVET, P. ET AL. Application of phyto-ecological carto-graphy to detect the mosquito breeding places on an island in theYellow Sea, Korea. Unpublished document WHO/VBC/74.485(1974).

b PROVOST, M. W. Needle rush as an indicator of breeding ofsalt-marsh mosquitoes. In: 46th Annual Proceedings of the FloridaAnti-Mosquito Control Association, Vero Beach, Florida, April1975, pp. 23-28.

4533 361-

R. 0. HAYES ET AL.

and Maire (12) in Canada reported on plantassociations and mapping of different vegetationsassociated with mosquito larval habitats.

Barnes & Cibula described the use of aerial photog-raphy and satellite multispectral sensor imagery forclassifying vegetation and terrains that had implicitpublic health and insect control significance (1). Theyreported on the use of multispectral scanners (fromaircraft) to define the portions of a salt marsh nearHouston, Texas, that were suitable Ae. sollicitansbreeding habitats, and also on the use of twice dailyremote sensing (from satellites) to show the tem-perature, altitude, and vegetative cover for all ofMexico which was to support the screw-worm fly(Cochliomyia hominivorax) eradication pro-gramme.The Landsat 1 satellite, originally known as the

Earth Resources Technology Satellite (ERTS), waslaunched by the United States National Aeronauticsand Space Administration on 23 July 1972. It wasjoined by Landsat 2 on 22 January 1975. From thesatellite altitude of 917 km (569 miles), the multi-spectral scanner scanned a 184-km (114 miles)diameter area of the earth's surface. Remote sensingmultispectral scanners in Landsat satellites measurethe electromagnetic radiation emanating fromobjects. Each item in nature has a distribution ofreflected, emitted, and absorbed radiation, which canbe used to differentiate between items and obtaininformation about their shapes, sizes, and even aboutsome of their physical and chemical characteristics.Sunlight reflected from the earth is channelledthrough a telescope to detectors in the satellite that aresensitive to four different bands of the light spectrum.The detectors convert the light into electrical voltageswhich, in turn, are translated by a digitizer intonumber values from 0 to 63.The digitized data are beamed to receiving stations

on earth, recorded on magnetic tape, and stored in thelibrary of the US Department of the Interior's EarthResources Observation Systems (EROS) Data Centerat Sioux Falls, South Dakota. The data can be trans-formed by conventional computers to provide anarray of symbols representing reflectance values forthe different plant associations and other naturalfeatures and cultivated areas of the earth's surface.The smallest area that can be resolved is 0.45 hectares,or 1.11 acres. The reflectance values on the fourspectral bands can be combined to give a "spectralsignature" for homogeneous areas. Later, thecomposition of these areas can be verified by on-sitevisits to obtain "ground truth" identification andclassification of the areas. The reflectance valuesfrom these areas, designated as training sites, can

provide data which the computer can extrapolate anduse to classify other similar areas with the samespectral signature. After a sufficient number of

training sites representative of the major types ofplants or other features under investigation have beenselected, a classification map can be prepared for theentire area. This can be done by the computer on astandard topographic map scale.Remote sensing imagery from the Landsat 1 and 2

satellites, which each circle the earth every 18 days,has been considered for a wide variety of appli-cations, including uses for agriculture, hydrology,oceanography, geology, and geography (2). Sensordesign for monitoring vegetation canopies by Landsatdata was reported by Tucker & Maxwell (21).

Fig. 1. Map showing the locations of the three studyareas for detecting and measuring mosquito larvalhabitats in the Lewis and Clark Lake region of theMissouri River.

The present study was conducted around Lewis andClark Lake, an impoundment of the Missouri River,between South Dakota and Nebraska in the UnitedStates (Fig. 1). The area has had mosquito problemsfor many years, and many studies of the mosquitofauna and larval habitats have been made. The twomost abundant and most important mosquito speciesin the Lewis and Clark Lake region are Aedes vexansand Culex tarsalis. The former is known as a "flood-water mosquito", because it deposits its eggs on moistsoil above the waterline of groundpools, streams,impoundments, irrigated fields, floodplain bottomlands, and other intermittently flooded places.Ae. vexans can fly for several miles (10) and is avoracious daytime and evening feeder, the blood-meals being taken mostly from man and his domesticmammals (20). C. tarsalis differs from Ae. vexans bylaying its eggs directly on the water surface of morepermanent aquatic habitats, such as roadside ditches,waste lagoons, the margins of ponds, excess irrigationwater, marshes, and wetlands. C. tarsalis usuallydisperses only a few miles from its larval habitat (7),

SOUTH DAKOTAYako

Springfielda, PoowL aE 8

KIa.aa,, ~NEBRASKARIaa, )

Nioboro NiobrarcRiwN

LEGEND

(G NAaiK Shady eAra

() SaifAid Shd Aa

cG) San .lai Shady Araa

lok

362

DETECTION OF MOSQUITO LARVAL HABITATS BY REMOTE SENSING SCANNERS

and it is a persistent feeder on birds and mammals atdusk and after dark (20).

Certain limitations in methodology wererecognized at the outset, but we decided that thesewere outweighed by the need for preliminary studiesto adapt the Landsat technology for remote sensing ofmosquito larval habitats in the vicinity of a largefreshwater impoundment. The most obvious limi-tation was that the area of resolution was only 0.45 ha(1.11 acre) of Landsat imagery, precluding specificdetection of narrow streams, canals, ditches, etc., orponds, marshes, lagoons, and other bodies of watersmaller than 0.45 ha. We also recognized that areasflooded only for short periods of time may notdevelop specific "indicator vegetation". We alsoknew that mosquito larva breeding, itself, would notbe detected by the satellite imagery and that themosquito larvae may not be uniformly distributedthroughout the aquatic and vegetation habitats withwhich they might be associated.

METHODS

Ground cover classificationSeven study sites in Nebraska and seven in South

Dakota had been selected to delineate the mosquito-production areas around Lewis and Clark Lakeduring a previous study in 1975. They were selected onthe basis of their principal vegetation and theirpotential value as habitats for mosquito larvae. Adescription of the breeding sites, their location andsize, the species of mosquitos found in 1975, and therecommendations made for controlling the mos-quitos in the Lewis and Clark Lake region has beenpublished by Hayes et al. (9). These workers detectedand described two principal mosquito-producingregions in Nebraska, the Niobrara and the BazileCreek areas, which are located east of the town ofNiobrara and are adjacent to each other. TheNiobrara area had about 8 ha (20 acres) of mosquitolarval habitat, and the Bazile Creek area includedabout 35 ha (87 acres) of larval habitat. The RunningWater and adjacent Springfield Bottoms regions werethe principal mosquito larval habitats in SouthDakota, containing an estimated 24 ha (60 acres) oflarval habitat in the former and about 854 ha (2109acres) in the latter. Because mosquito breeding washigh in these areas in 1975, some larval collectingstations were established and mapped in each area,and these areas were selected for vegetation mappingin 1976 (using the 1975 satellite reflectance data). TheBon Homme Colony area, about 24 km (15 miles)east of Springfield, South Dakota, in which only 0.4ha (1.0 acre) of mosquito larval habitat was found in

1975, was also included in the vegetation mapping as anegative control.

Trips to Lewis and Clark Lake to identify andrecord the specific locations of vegetation groundcover were made on approximately the same dates in1976 as those on which the satellites had obtained theimagery in 1975. Leaves, stems, and roots of plantspecimens were collected, identified, labelled, placedin a plant press, and returned to the laboratory foridentity confirmation and storage. The 1975 studyarea and mosquito collecting stations also wereinspected. Samples of mosquito larvae were obtainedand identified as to species, and their nunmbers wererecorded.

In this study, the satellite reflectance data for theLewis and Clark Lake area were obtained from multi-spectral scanners aboard the Landsat I and 2 satel-lites, and the multivariate system analysis of themultispectral imagery methods of Maxwell (13) wasused for processing the data. Tapes containing thereflectance data were purchased from the USGeological Survey, EROS Data Center, at SiouxFalls, South Dakota. Three sets of data for bands 4, 5,6, and 7 were chosen for analysis; they included thewavelengths 0.5-0.6, 0.6-0.7, 0.7-0.8, and0.8-1.1 Am, respectively. Bands 4 and 5 are in thevisible range, and bands 6 and 7 are in the infra-redrange. The satellite overpass dates selected were 25June, 13 July, and 9 August 1975 because they had thebest quality reflectance data and the least cloud coverfor the summer months.The wild types of vegetation found in 1976 were

assumed to be the same as those present in 1975.Cultivated types of vegetation present in 1975 weredetermined by interviewing the landowners andfarmers. The "ground truth" data were developed inthese interviews by use of computer-generated grey-maps (13). On the greymap, each particular type ofvegetation generates a low reflectance in one bandand a high reflectance in another band; these reflec-tance variations, considered collectively across fourbands, indicated on the greymap a unique spectralsignature for each type of vegetation. These mapswere used to show the farmers the areas for which the1975 crop identifications were needed.The units of reflectance making up a greymap are

called picture elements, or pixels (0.45 ha). Thegreymaps were scaled so that they could overlay a 7.5minute topographic map (e.g., from 105037 '30"W to105045'W) with a scale of 1:24 000, allowing naturaland cultivated landmarks to be transferred to thegreymaps and the reflectance values for a given bandto be compared with the actual locations in the studyarea.

Extending ground truth data to correlations withreflectance data to allow interpretations and identifi-cations of ground cover on the basis of the spectral

363

R. 0. HAYES ET AL.

Table 1. Six habitat classes and their ground cover vegetation delineated by Landsat reflectance data classificationsymbols'

Water Wetland Periodically flooded Transitional Domestic Upland

River (W) Cattail (.) Bulrush (X) Bog marsh- Alfalfa (A) Bluff grass (+)crest (T)

Water, Cattail, Cattail (X) Corn (C)standing (U) bleached (.)

Corn (X) Foxtail Oats (0) Shrubs orbarley (T) bushes (B)

Duckweed (.) Polygonum (X) Johnson Sorgum ($ or=) Trees (*)grass (T)

Tree, dead I.) Kochia (T) Wheat (W)Polygonum (T) Wheat-

grass (W)Ragweed (T)

Reed canary (T)

a The computer symbols printed for classification mapping are shown in parentheses.

signatures alone is termed "classification". The clas-sification was done by selecting training fields for aparticular type of ground cover delineated on thegreymaps. With the ground truth information, thecomputer was provided the reflectance values of thepixels for each specific type of ground cover for thefour spectral bands. For example, if corn was grownin a specific area in 1975, the computer was given theinformation to classify the multispectral signature forthat corn. In that process, the identification for cornwas extended to every pixel in the study area that hada similar signature. Properly selected training fields(which may involve only a small part of the studyarea) ultimately accounted for a classification of theentire study area. Classification of the Landsat datawas based upon ground truth inspections of the grey-map reflectance data in training sites that totalled atleast 30 pixels. The training sites used for classifi-cation were located on both sides of Lewis and ClarkLake.The final product of the classification task was a

line printer map of symbols that classified the groundcover in each pixel. It resembled the greymap inappearance and scale, but depicted surface coverclasses such as water, wetland, and floodplains. Thehabitat class for water included the river and lakeareas, for which the classification symbol (W) wasused, and also the areas of nonturbid standing waterup to depths of 1.2-1.5 m that did not have aquaticvegetation, for which the symbol (/) was used; thewetland class symbol (.) represented various wetlandareas with water of the aforementioned depths withhydrophilic vegetation; the periodically flooded class

was the moist ground area with vegetation that wasindicated with the symbol (X). The periodicallyflooded habitat was flooded during 1975 but not in1976. The classification symbols for the vegetationground cover in the transitional, the domestic, andthe upland classes of habitats are identified in Table 1.The classification map comprehensively depictedsurface cover classes that were pertinent to the studyareas rather than reflectance values from oneparticular band.

In the terminology of remote sensing, the projectwas multidate because of the input on 3 dates, multi-spectral because each overpass generated reflectancedata from four bands of the electromagnetic spec-trum (two in the visible range and two near the infra-red range), and supervised because the categories ofinterest, e.g., cattails, river water, and bluff grass,were established before classification during fieldtrips to the area. Signatures were extracted forextension to the entire study area by analysis ofexisting aerial photography and maps and bydelineation of representative training fields for eachclass of reflectance patterns. The final process wasaccomplished digitally, i.e., on the CDC 6400-computer, through a set of routines known asLandsat Mapping System (LMS) at Colorado StateUniversity. These routines rely on a maximum likeli-hood discrimination once the training signatures havebeen established. The threshold levels for the classifi-cation maps were established to optimize thegoodness-of-fit parameter and to increase the level ofconfidence for comparisons between the single dateand multidate maps.

364


Table 2. Plants identified from the Lewis and Clark Lake study areas in Nebraska and South Dakota from June toAugust 1976

Botanical nanme Common name

Equisetum hyemale L.

Juniperus virginiana L.

Sagittaria spp.

Eleocharis macrostachya Britt.

Scirpus spp.Bromus inermis Leyss.Phragmites communis Trin.

Agropyron cristatus (L.) Gaertn.Agropyron intermedium (Host.) Beauv.Agropyron trichophorum (Link.) Richt.

Hordeumjabatum L.

Setaria lutescens (Weigel.) Hubb.Phalaris arunidnacea L.

Typha angustifolia L.

Typha latifolia L.

Lemna minor L.

Populus deltoides Marsh.Salix spp. (probably amygdaloides Anderss.and exigua Nutt.)

Quercus macrocarpa Michx.Celtis occidentalis L.

Cannabis sativa L.

Polygonum pennsylvanicum L.

Polygonum spp.

Kochia scoparia (L.) Schrad.Rorippa palustris (L.) Bess.Sisymbrium loeselii L.

Thlaspi arvense L.

Melilotus spp.Acer negundo L.

Ambrosia trifida L.

Ambrosia spp.Iva xanthifolia Nutt.Xanthiumn strumarium L.

Horsetail

Eastern red cedar

Arrowhead

SpikerushBulrush

Brome grass

Plume reed

Crested wheatgrassIntermediate wheatgrassPubescent wheatgrass

Foxtail barley

Foxtail

Reed canarygrass

Narrow-leaved cattail

Broad-leaved cattail

DuckweedCottonwood

Peached-leaved willowSandbar willowBur oak

HackberryMarijuana

Pennysivania smartweed

SmartweedFireweed

Bog marshcress

(No common name)Field pennycressSweet clover

Box elderGiant ragweedRagweedMarshelderCocklebur

RESULTS

Plant identificationA total of 33 types of plants were identified in our

study areas. The list of plants (Table 2) is incomplete,since we did not attempt to do an exhaustive floristicsurvey; however, most, if not all, of the dominantspecies associated with larval mosquito habitats areincluded.

Mosquito larva collectionsCollections of mosquito larvae from habitats

around Lewis and Clark Lake were made duringJune, July, and August 1976 (Table 3). Thirteencollections yielded 431 fourth-instar larvae repre-senting seven species. The vast majority of mosquitohabitats which yielded larvae in 1975 were dry duringvisits to the area in 1976, because drought conditionsprevailed throughout that summer.

365

366 R. 0. HAYES ET AL.

Table 3. Mosquitos identified from the Lewis and Clark Lake study in Nebraska and South Dakota from June toAugust 1976

Number of Aedes Aedes Culex Culex Culex Culiseta AnophelesDates in 1976 and place collections dorsalis vexans tarsalis restuans salinarius inornata walkeri Total

22-23 June

Niobrara, Nebraska 3 1 23 1 142 167Bon Homme Colony,South Dakota 1 2 6 16 24

13-14 JulyNiobrara, Nebraska 5 3 59 36 1 99Springfield Bottoms,South Dakota 2 1 3 17 21

1 0 August

Niobrara, Nebraska 2 53 11 17 39 120

Total 13 55 17 118 7 16 217 1 431

Classification mapping ofmosquito larval habitats numbered. It also indicates some of the topographicfeatures and the wetlands, grasslands, woods, and

Fig. 2, 3, and 4 are different types of maps of the cultivated fields. Fig. 3 is a portion of a US ArmyNiobrara study area. Fig. 2 is a map prepared during Corps of Engineers' topographic map of that areathe 1975 mosquito surveys, which depicts the indicating landmark features, such as the town ofmosquito collecting sites that are outlined and Niobrara, state highway number 12 leading into the

- ROADCITYWOODS

EELAKE ,,mCULTIVATED AREAJWETLAND VEGETATION - ** N

| GRASSES DRYLANDS 1 1

1 \H~~~~~~~~~~~~~1..-MISSOURI RIVER -

RN14-INS D D a 5) |RN49

ANt'

RRN

..KRN115~~~~~~~~N1

RN2S

SCALE IN METERS RN24 _I I jI I -

0 100 200 300 400 500

WHO S4136

Fig. 2. Habitat and land use map of the 1975 mosquito larval collection sites at Niobrara, Nebraska.


Fig. 3. Topographic map of Niobrara, Nebraska. (Source: U.S. Corps of Engineers)

town fom the east, and the roads in town, along theMissouri River, and surrounding the wildlife manage-ment site. Fig. 4 is a classification map of the samearea, which was computer-generated from the 12bands of reflectance data obtained by combining the 4bands of data obtained from each of the 3 satelliteoverpasses on 25 June, 13 July, and 9 August 1975.The computer symbols on the classification maps arethe same as those shown for the various ground covercharacteristics listed in Tables 1 and 4. Spaces on theclassification map without symbols result from nothaving coded information for the unidentified pixels,e.g., the reflectance values for the town of Niobrara,which is near the lower left-hand corner of Fig. 2-4.

The number of times each type of symbol is shown asa pixel on the classification map, its percentage com-position of the printout, and the area in hectares foreach symbol are shown in Table 4. The number ofhectares equals the number of pixels multiplied by0.45, the number of hectares per pixel. The number ofunclassified pixels (no symbol recorded) is also shownin Table 4 as "Other, unclassified".The Niobrara study area multidate classification

map (Fig. 4) was 24.0% wetlands, 9.5% flooded, and23.1% river (subtotal 56.67o), as shown in Table 4.The individual, 4-band classification maps also werecomputer generated, but are not shown, from the 23June, the 13 July, and the 9 August 1975 reflectance

367

R. 0. HAYES ET AL.

TTTYYXXX.XXXXXXXKX X..X ..XX##* *0##TT T.*TXT .KT XXXXXXXXXXXXXXKX.1.XX .t . * **@T .ITTTT.W**...T0XX.xxxxx..X.K. I@@.... . TTO*TY T*TX * XXXXXXXXKXXX:..T. ..TT-f;TTTTTXf ITKXXXXXXX.XKX ea. ....T.tw T.TTT1313333x3 X XXXX5.5115.X. eXXX.JXXX TC;TTT

"""| . x" 05eeKKXiX x xxx x ;,3, ,3,jqrii;ILIIIIUIx:ex:::IX::..X5 11331,I TJTTu,,,

ili,iil.I3133 3 13 333331313ii 3 iD!... 311 33333lioilII I

II 2..ro l i lIIZZIIIqE 111IIIIIIITXTgIXll"1§g'1

TCCTCITIt I ToTTTI TIT I TT , XTYTX.*.*T.11.1*.XTTT Y..TTT;TTIT TITTT CCTTTT I TTTTTTTTTTTXAKfXxTTXX . ITTTCCT.OTCCTCCa T T.TTf!TT .I TXTXXXXXXXXr.X.TTTTITCCTCC..is.T...IX TTIKKXK.X..T X..T.*** *IT

*T . [email protected], . TTTT TITT .TT,T . T TTIAXXX* 0 T ITTXTOVT*OTCCTTITTCI CXCTCXT.T...XT,X.,**..***...e.TTT

T.*TTTTTCTCTCXTTCCTX TXXTXCCTT TITITIK.xTTTTo#*IT.TTTCC CCCCCCCTfTCCTgCCCCCeeCTsT TTTX.. . TCCT IT

T t.IYTT* CTCTCCCCCC *OOCTCCCCCCCCCO.T IT*CIT ; c1(YTTT*#* * KT*TTTTCC CCCCCCCCCOCCccT cc CCTCCCTCCX fCCccc|- -l-I-4 It TTCCCCCCCCOOOT TCCCCCCCCC#CCCCTCCCCCCCCCCCCW.T+T1TTTT1TTTCC.00I.T CCCCCCC... CCCCCC CCCCCCCCCT#

00 ITITTTTTTTOOTIO..I....0O Toi.TI T.*.V.OTTTT+I,.sTTTTrTT# *WST*T&OW#+ VWTTO **T#+t V*#+O#+***W"

Fig. 4. Classification map of Niobrara, Nebraska,generated from multidate Landsat reflectance data inJune, July, and August 1975.

data for wetlands, flooded, and river areas. Thesecombined total percentages of the monthly maps were54.10No, 60.2%o, and 58.2%o, respectively. The

Missouri River area in the three monthly maps was489 ha (1208 acres) during June; 474 ha (1171 acres)during July, and 504 ha (1245 acres) during August.However, the amount of standing water and floodedlands changed from 720 ha (1778 acres) in June to 801ha (1978 acres) in July to 738 ha (1823 acres) inAugust, and the areas classified as wetlands increasedmonthly from 443 ha (1094 acres) to 497 ha (1228acres) to 603 ha (1489 acres), respectively. The trendof increasingly greater areas of wetlands over thesummer months was previously documented andreported for the 1975 field studies (9). The inundationof the region by water flowing through wetlands andcornfields in 1975 made many of the mosquito larvalcollecting sites unsuitable as mosquito larval habitatbecause the water was moving. It also was reportedthat the greatest production of Ae. vexans larvaeoccurred on both sides of the river during the periodof 23 June to 20 July 1975 when the river levels firstrose to flood the adjacent fields, whereas theC. tarsalis larval production indices increased later inthe flooded areas that developed into permanentwater habitats (9). The Niobrara and similar areas

had large numbers of C. tarsalis from 7 June throughAugust 1975.

Table 4. Reflectance data for the classification maps in Figures 4-6.

Niobrara a Springfield Bottoms' Bon HommebGround cover Symbol Pixels' %d Area (ha)e Pixels' %d Area (ha)e Pixels' %d Area (ha)e

Alfalfa A 20 0.4 9.0 14 0.1 6.3 53 2.9 23.8Bluff grass + 259 5.3 116.6 3017 24.5 1357.6 408 22.7 183.6Bushes B - - - - - - 9 0.5 4.0Corn C 239 4.8 107.6 341 2.8 153.4 - - -

Flooded X 471 9.5 212.0 769 6.2 346.0 0 0 0Oats 0 96 1.9 43.2 881 7.2 396.4 197 11.0 88.6River w 1140 23.1 513.0 1348 10.9 606.6 591 32.9 266.0

Sorghum $ or= - - - - - - 201 11.2 90.4Transitional T 994 20.1 447.3 1871 15.2 842.0 - - -

Trees mature * 144 2.9 64.8 43 0.4 19.4 1 0.1 0.4Trees, young Y 58 1.2 26.1 158 1.3 71.1 0 0 0Water, standing / 28 0.6 12.6 240 1.9 108.0 0 0 0Wetland 1187 24.0 534.2 1465 11.9 659.2 0 0 0

Wheat W 92 1.9 41.4 770 6.3 346.5 143 7.9 64.4

Other, unclassified 212 4.3 95.4 1397 11.3 628.6 195 10.8 87.8

Total 4940 100.0 2223.0 12314 100.0 5541.1 1798 100.0 809.0

a June, July, and August 1975; 12-band multidate reflectance data; 32.909 threshold value.b August 1976; 4-band reflectance data; 18.465 threshold value.' 1 pixel (picture element) = 0.45 ha 11.11 acres).d Percentage composition of the printout for each item of vegetation, water, etc., denoted by the symbol.' Area in hectares for each symbol.

368


31 CCCCCCCCCCCC.1 C*TCCOS*TIC TTITITT ICTTTTTTIIw*T**oWo**32 CCC CCECCCCCC..t *C 0 1.T ITTITTTOC C OTT TTWV oo0o*31 c CCCC C CC T?.. TC QO.1TTTOO C CC ...IOWVW #V*o*34 C TTTT W..CCCTCC CCC 00000. M.TV o0OOO0oo*.W*3 CITT TC ITT.T TsCCCCCECC 0 .W .0000 09000... seW.=TT T*** o es. OOOTT 0 O000. V**o 000q

.6 CTT .V sVOSo TO CCC

C C0 : .0T .CT *040 CCC .00000 ** *G... .WKVV.OOwWOWO .G# * 01

CCW C CCC .K...T.**V oW * * VKVO .0 ..*41 C~ V..., 6T.CCE CE .1.T1 .**W ;WVW WT* TOh* # *** *

.VC 0 00 4.

45 000 .0 * sV .V00 ..O.40 .OOcO VVW .V 0000 .*60 64 oo 0 OO VW*4 T 0 0 00 Wo+* .O Ks..TOOTT TI40 00 0 00000 C. ..V ....V..WW*#...........I. 000. TTIl49 0 W ...Kl 0 .w ..o* V ...s.T1.. 0.. 0 .** 00 CTIIr C CCCCCC

SS1 C CCC 00 00.0 C0CC CC CO OO.,.....V...TT1

T~COOQ 00 CC T TWW54 C CC 0005S 0TCC CTTII. T 0 0

0 .Ww TOOK.so51 .0 0* 000 C.WKAO.OW .*e o V.W.....W 01*O ITT.T T. **9TI...1.9 COTTT ....TTCCC V 0.T .0 .0 ..KVWWWWW VW fTOoT.I.. VV..

1.1 000000 0 OOCCCC .W..V W...V .V..TTTTITTIT ...*WO0 000 O11 CC V ooVw V ..TT OT* IOOOo

6143 00 0000000001TTCC A .o VV .A IW .Ve* VVKVWV V.I..T .111TT6M 000 00000. CCWWV K..VW .0o..TTTCCC 190** T W.T1T1TT

1 000000 c c .WVVA 100**** O ..VVV.W*W.W.0 ITTV T W.*V TTTITTC .0000000. OC.VWVWV * o o * T T T1C s.0K 000. WCC AKW .o* * o CT** 4 * TT VOQ.1X.

64 0 0.0 00.C CC ...A ...V 0T..ICo! C V.ooV .II..jT T;TT IT? TjTI.1.9 S.COS?O...0...TTO TO 10CC IIIII*#oT* TITITX

C _0 T.I0T0. TI.CCSCCC .o* O KQ*OW ..W TI.

E 1OW. ..01 ..TT14 ..0cc0CCCC C..00I3I.0 0oo**T fT TosetIi% * *.VVO.... COICTOCO .V Io* o* T 1 ff T *o f T e

It ..* V WA 0 C 0

....Wv A C .00 000 .T.. 0 1?79 000. C OTT 0 .000 It..

0O 00CC C OOOVW 0OO .o*** W .,** O WVC.V

sit C 00000. OW .#oo* *o o -V .*** WV..ITT If ...21.~ 0000 CC 4*....0 .TTS?T ITTTXI ..XX

SI 0o * *C o *.. T 0.9 O T.*11 f

Pt CTV 00 .T TTI II.. * . .-tAl

Ce *.. *I. 000 C C .TTW.. ? . TeT **..

11 0 W. ...?2.90 V CC ' . ? ?j ; 2

oweIC 010Tjj;111 1.1--8TT :...... 111W..T2T 2..C*

001.1111III..V.... 1111 ,1 *1CC

V... eM KK.o* VV`... .01TT 1T 111T.W.01.111TT 1T ..12..***e .TX9 C

. K ...K0f. I T ay99 ...eW . KAIC...C..I0. C . 0 C C C O . C I I 1 . 0 I I I 1 1 2: :;"I:2. 294 .V KM .A...O. 111CC.11OT* 0 T X ox2. . X .91 . . . . . C O I TI.I.01 1 1 T T0 1 "I1e22O.3399% 1* * WAT00 C . . . 0 0.T T 00 6**.IT 1WW 0 *0 T1 01 1X. X 3... . *

CAT .T OT :I * T T**T

f....C. TT... T0:11111 TT..1 T0T11 ...:vv :... SXP

...0e*o.CCK VTTTT I0TTo 1112TT 22C 1.*TV11121 1T2XX X I tO1 TI

.30Too* 9 1..o* .00 0.TTI 11T 11111* 14.. .121VI1T1XXX5ITO0o CCIKCCw 1111;T T 1. f xx xx xx

* * Ty . . .R X X X

Vs$To * 0TTO17T 1Te I TI~T X;1 ..TT;" * IXXXOT oCTKK O OT. I1 I To TY*1C TT I T T T XXIlif .v.v.VKV1O*o0W4V* T1 1111** TTIT-. .TXXV.h;wX X gwXXViV o01,*T * :e 0 e T T T X2 I T XolTtO

Ta It eaXV inXtance

ofV ul,adTu

The mean river water levels (elevations) at the

Niobrara gauging stations during 9-22 June, 7-20

July, and 4-17 August 1975, were 370.9 m (1216.6

feet), 371.4m (1218.2 feet), and 371.6m (1218.9

feet), respectively. At the Springfield station during

23 June-6 July, 7-20 July, and 4-17 August 1975, the

river levels were 368.2 m (1207.7 feet), 368.6 m

(1209.0 feet), and 368.9 m (1210.0 feet). Thus, the

image data for the areas of standing water, the

periodically flooded, and the wetland classes of

ground cover were directly associated with the

recorded rising elevations of the Missouri River

during June, July, and August 1975.

The study area in Springfield Bottoms, South

r I.. 1x

he, /. ..X x11.2... ..X/// I2 X $X../..2.

/ %X..1 O. 21 II/ /I/I//I

.low :a

H idhi.u,*1I2

Bottoms, South Dakota, generated from multidate

Dakota, was reported to comprise about 854 ha (2110

acres) and was the area where the most mosquitolarval habitats were detected during the 1975 study. A

total of 767.2 ha (1 895 acres) of wetland and'standing-water habitats were detected and classified

by using the reflectance data from the satellites (Table4 and Fig. 5). The size of the area determined from the

Landsat data for these two habitat classes that could

be considered as prime mosquito larval habitats is in

quite close agreement with the 197-5 study data,

especially considering that 1975 was a year with more

extensive flooding. The unclassified areas surrounded

by the classification map symbols for the river- or

lake-water symbols corresponded to sandbar areas in

369

R. 0. HAYES ET AL.

. , . . . . O fl =vU W IS% S% % S S...O ..^ WW.W.W4 +O4 -. +u AOVOOO*v *0A* ** W ++0 s%S%S S-*0**WWwvW.*+O *.OOVOOWv*###* #* * 0*o+#WW %+Sss iss So-Z6+#*iWWW *oo= +* ++oVowoooW*.....D.O.V.W#o*W.ASSSS$s SH-.0O.WAW*W00.,00000000000.**v**,OWW.*..W.*.s=$S%SSSSISSH-O3.* WOWOO*W*O*0000000000.***O*OW"WAAA+-S* W"OS- *"+0000W00W.OOOOOOOO.wVW*.*+OO.WWVW AAAA.BS!FlSAt$iS .* *AAW.WOS.*000000000AWWWW **W.W A. A,zsq %SS.OSBSSSS*5AAAW..SS O000000000A Vw

.#z*O. .A. ..2$%S,SSSY,$Sm. ..A. WOOOOOOOOOO. V.*.++ + + 0*+ **Sv * %s%9S§s;t%~* * A wJ~=

00000000004.W,W**V .$x,* ASSSSS 5SU...... .WV 00000000000*Wu

#V4 *, 'WA3,AAA rn*z S$SSS$ASSS.zV*.*z=A*WWWWOOOOOOOOOOWWA..A**...u SSISS SS.z.000.AAAWVOW0O00t)000WV.AI,, l, AA,. X VwWs*SSSS.S!3S,*..B mrSSSO wO *Owowv*.i"+*-I

U|||1A,IUl 1V+sso= - *OS. *+W*A S B2.* 000 V0*.V.VV

!!1!1111!1.1!f1w1ff.1 1WAAW1 0O-*WX1A. SS8 - . -0000+0 W*W#1WV#a#

Fig. 6. Classification map of the study area in BonHomme, South Dakota, generated from August 1975reflectance data.

the lake that resulted from siltation. Whether thesandbars were exposed or covered with water, theyproduced a reflectance value different from that ofthe river and lake water. Because the sandbar areaswere not classified during the ground truthing, theywere not printed on the classification maps. The samewas true for the shallow, flowing margins of wateralong the edges of the river and lake.The classification map of the study area in Bon

Homme, South Dakota, is shown in Fig. 6. The areawas considered to have only 0.4 ha (1.0 acre) ofmosquito larval habitat during the 1975 study. Thereflectance image resolution limits were 0.45 ha (1.11acres), and no pixel was recorded for either thewetland or the periodically flooded habitat class thatwould correspond to a potential mosquito larvalhabitat in this study area. However, types of groundcover such as bushes and sorgum were found thatwere not present in either the Niobrara or theSpringfield Bottoms area. The bluff grassessurrounding the Hutterite farm and its crops ofsorghum, oats, and wheat were the predominanttypes of ground cover in the Bon Homme area. Thecircular pattern of sorghum classification symbols($), which can be seen near the middle of the map,results from the crop being watered by a centre-pivotirrigation system. Another classification symbol (= ),also indicating sorghum crops, was sparsely distri-buted because of poorer irrigation.The elevations and the water table levels varied

across the river valley. Beginning at the river level wasthe water class; the elevation increased through thewetland, the periodically flooded, and the transi-tional classes into parts of the floodplain; the ele-vation continued to increase in most of the domesticclass croplands, and it rose abruptly into the uplandclass. Some of the bluff grass in the upland class was76.2 m (250 feet) above the Missouri River.

Classification accuracy

The observation that the use of multidate, orsequential monitoring of ground cover, classificationwas more sensitive to seasonal variations has beenreported (18). A special evaluation of seven classifi-cation categories (river and lake water, cattails,standing water, transitional, corn, mixed deciduous,and bluff grass) was made by using 15 pixels that werenot included in the initial ground-truth training fieldsfor each of the aforementioned categories. Theoverall accuracy of the single-date classification was72% for June, 60/o for July, and 83% for August.The multidate (three-month) classification accuracywas 95/o. This special evaluation also revealed thatlarge homogeneous categories were accuratelyclassified in both single-date and multidate maps; thatis, 55 and 60 of the 60 pixels for mixed deciduous andriver or lake water, respectively, were accuratelyclassified. Only 23 of 45 and 22 of 45, however, wereaccurately classified for the flooded and transitionalcategories, respectively, from the single-date data,whereas 13 of 15 and 14 of 15 were accurately clas-sified in the multidate map for these categories.

Project costs

This project was budgeted for US$40 000, andmost of the costs were in the categories of personnelsalaries and benefits and in travel. The expendituresfor the Landsat imagery items were only $738. Thoseitems were purchased from the US Earth ResourcesObservation Systems Data Center for the months ofJune, July, and August 1975. The costs for eachmonth's data were: one computer compatible tape at$200 (1980 price was $200); four black-and-whiteprints of the Landsat image at a scale of 1:500 000 at$8.00 each (1980 price was $12); four positive trans-parencies (5 x 5 cm) of the image at $3.00 each (1980price was $8), and one 18 cm positive transparency ofthe image at $5.00 (1980 price was $10). The computercost records were included in the project budget, butthese data are no longer available. Estimatedcomputer costs, however, for a 1980 project of similarsize and scope were $2250 for computer time, $2000for a computer programmer, $750 for keypunch time,and $400 for computer tapes, cards, binders, andrelated supplies.

DISCUSSION

Areas characterized by mosquito breeding or by nomosquito breeding were correlated with the types ofvegetation. The computer-programmed alphabeticcharacter printouts were used to identify and measurethe mosquito-breeding habitats. We found that the

370


wetland, periodically flooded, and transitionalhabitat classes were generally associated withmosquito breeding. Depending primarily on the riverand lake levels and their elevations, the three classeswould respectively correspond to permanentlyflooded, frequently flooded, and occasionallyflooded areas. By extrapolation, they could beexpected to have a slight, moderate, or highmosquito-breeding potential. Thus, for Culextarsalis, a permanent pool mosquito, the permanentlyflooded class of habitat would have the highestpotential and the occasionally flooded class wouldhave the lowest. The opposite would be true for theflood-water mosquito, Ae. vexans, for which theoccasionally flooded and frequently flooded habitatclasses would have much more mosquito-producingpotential than the permanently flooded habitat class.Collectively, these three classes would have muchmore mosquito-producing potential than any of theother classes characterized.The resolution, about 0.45 ha (1.11 acres), and the

rectilinear grid arrangement of the Landsatreflectance data are not ideally suited for preciselydelineating the complex growth patterns of many ofthe Lewis and Clark Lake plant communities andaquatic mosquito habitats, because they often aresmall, noncontiguous, irregularly shaped, hetero-geneous localities. Therefore, the classificationscheme of vegetation components comprised unitswith more than one species (Table 1), and it followedpreviously used concepts for classifying plant-mosquito associations (11). The technical refine-ments that will result in greatly improved satellitesensor resolution have already been developed (17). A30-metre resolution is now available from Landsat 4,which was launched in 1981. Further evaluation withcurrent technology, development of guidelines for itsuse, and determination of budgetary requirements arelogical prerequisites to implementing the use ofremote sensing as a tool in combating the vectors ofselected diseases. Improvements in resolution willgreatly increase the potential for using remote sensingdata for planning and conducting operationalmosquito control programmes. Including thecapabilities to monitor the water's depth would alsobe important in delineating the portions of aquatichabitats most likely to have mosquito larvae.Another technical refinement that will become

available for mosquito larval habitat detection byLandsat imagery is the capability of measuring thesoil's moisture content. Since most of the misclas-sified transitional pixels were indicated as flooded,the soil moisture data would minimize that misclas-sification. Including a parameter for groundelevation would be valuable for delineating areaswithin class categories that are the most susceptible toflooding. Contours at 0.3-m intervals could be

considered in flood-plain areas where the remotesensing data are to be repeatedly used in ongoingmosquito control projects, but this would not befeasible if satellite imagery were being used to obtainwide-area data for preliminary survey or reconnais-sance purposes. Where available, colour infra-redaerial photographs of portions of the ground-truthareas also facilitate category classification. Wheninternational political agreements permit, it seemslikely that future remote-sensing satellites can bedeveloped that will have a resolution on the order of0.02-0.09 ha (0.05-0.22 acres); this would greatlyimprove the ability to associate mosquito larvalhabitats with various plant and water classifications.However, long and narrow breeding sites, such asroadside ditches, still would not be detected, unlessthe water leaked or seeped onto larger areas.

In addition to the resolution limits, satelliteimagery has other limitations. Cloud cover whichobscures the satellite reflectance data obviouslywould be of more importance in some geographicalregions than in others. For example, in a region ofsouthern France that was considered for a studysimilar to the present study, it was found that cloudcover obscured the satellite pictures more than wasexpected for the region. This was because the satellitespass over that area at 09 h 30, and the cloud coveroften had not yet dissipated by that time of day, evenduring the summer months.The time lag between the date when the satellite

remote sensors record the reflectance data and thedate on which the magnetic tapes of data can beobtained for computer processing may be 3 months orlonger. This would not be a disadvantage if the datawere to be used for preliminary survey purposes,sufficiently projected in time, and the problem couldbe minimized by long-term studies to determineseasonal ground cover characteristics in areas wheremosquito control programmes operate continuously.However, the long time lag would become animportant problem in the case of ongoing operationalvector control programmes. Special arrangements forreducing the delay may be possible for suchprojects.The cost of the present project was modest, and

only about 15% of the costs were for satellite reflec-tance data and computer-related expenses. As in moststudies, the major portion of the budget wasconsumed by personnel and travel costs.

Additional applications for Landsat imagery ofmosquito habitats associated with impoundments,fresh- and salt-water marshes, and other types ofwetlands are needed in the different geographicalregions of the world to generate data on theappropriate habitat classes and their ground covercharacteristics. These data could be used in planningand conducting malaria, onchocerciasis, and possibly

371


other vector-borne disease control programmes. Suchinformation would allow remote sensing data to beused for delineating and measuring large potentialmosquito and other vector problem areas in thevarious regions of interest. The idea is not new. Theuse of remote sensing infra-red and reflectance datafor other veterinary and public health applicationshas also been postulated, for instance, by Cline (4) forthe control of schistosomiasis (Schistosoma mansoni)and its snail host (Biomphalaria galabrata), and forcontrol of several species of mosquitos that arevectors of malaria, filariasis, and arboviruses. c In areport of an evaluation of remote-sensing imageryproducts for range and water management problemsin West Africa, consideration was given to appli-cations for tsetse fly and blackfly control pro-grammes (5). The principal uses were for locatingareas with dense stands of trees that may harbour thetsetse flies and for locating bodies of water larger than200 m in diameter (3.1 hectares in area) that may beassociated with streams and rivers providing blackflybreeding sites. However, the resolution limits werenot sufficient to detail the specific breeding sites foreither type of fly, and it was stressed that adequate on-site investigations were essential for providing identi-fication, mapping, or monitoring of geohydrological,vegetative, and agricultural phenomena. Environ-mental data obtained from satellite imagery were usedfor the screw-worm control project in Mexico to

c BARNES, C. M. Extracts of remarks on potential contributionsof remote sensing to public health. In: Proceedings of the XIICongress of the International American Sanitary EngineeringAssociation and PAHO Symposium on Water Pollution, Caracas,Venezuela, 1970.

indicate areas in which the temperature was highenough to permit growth of the parasite (1).The use of the techniques described in this report

could be further refined by using classification mapsto identify and measure the potential mosquito larvalhabitat areas in other regions of the USA or theworld. Reservoir and other impoundments associatedwith urban and rural development create habitats thatfavour disease vectors such as Anopheles hyrcanus inAsia and Anopheles darlingi in South America (19).The development of water and related land resourceswithout plans to prevent and control the vector popu-lations results in increased risk of disease trans-mission (8), e.g., arboviral disease activity has beencorrelated with irrigation and rice culture in southernFrance, Japan, Korea, China (Province of Taiwan),Thailand, India, and the USA (14). We believe thatthe present study demonstrates that Landsat imagerycan be used to obtain maps of large-scale waterresource developments and that the maps can be usedto delineate, classify, and measure potential mosquitolarval habitats (and mosquito production). We alsobelieve that the study shows that the imagery could beused to evaluate the magnitude of a problem area inthe initial planning of disease control programmescovering a large area. Seasonal and annual imagery insuch programmes could provide larval habitat datathat would be of value in assessing progress of theprogrammes, especially in source reduction opera-tions. Finally, when the resolution of multispectralresource sampler imagery from satellites has beenimproved, this moderate-cost technology will findapplications in a large variety of operational vectorcontrol programmes.

RtSUME

DETECTION, IDENTIFICATION ET CLASSIFICATION DES GITES LARVAIRES DE MOUSTIQUES AU MOYEN DETELEDETECTEURS EMBARQUES SUR SATELLITE

Les images fournies par les teledetecteurs multibandesinstalles a bord de satellites a orbite circumterrestre(Landsat I et 2) ont permis d'etudier les biotopes aquatiqueset les communautes vegetales associes a la reproduction desmoustiques au voisinage d'une importante retenue d'eaudouce. Cette retenue, le lac Lewis and Clark, est situee sur leMissouri, entre le Dakota du Sud et le Nebraska (Etats-Unisd'Am6rique). Dans cette region, les moustiques,principalement Aedes vexans et Culex tarsalis posent desproblemes depuis de longues ann&es. Les methodes utiliseespour dresser la carte du biotope aquatique et de la vegetationsont decrites, la classification des types de sol et de couvertvegetal etant rendue possible par l'analyse, selon un systemea plusieurs variables, des images multispectrales etablies apartir des mesures de la r6flectance effectu&es depuis le

satellite. Les symboles employes sur la carte tracee parimprimante pour representer les divers types de sol et decouvert vegetal permettent de reperer les zones propices ounon a la reproduction des moustiques.

Les valeurs de la reflectance obtenues pour trois des zonesetudiees ont e utilisees pour distinguer les types suivants desol ou de couvert veg6tal: luzerne, gramin6es de pente,broussailles et buissons, mais, zone submergee, avoine,cours d'eau, sorgho, vegetation de transition, arbres(adultes), arbres (jeunes), eau (stagnante), marecages et ble.En plus de leur interet pour le trace des cartes, les valeurs dela reflectance permettent de connaitre la place occup&e, danschacune des zones etudi&es, par ces divers types de terrain etde vegetation (en pourcentage du total et en nombred'hectares). En se servant de donn&es relev&es a plusieurs

DETECTION OF MOSQUITO LARVAL HABITATS BY REMOTE SENSING SCANNERS 373

epoques (en juin, en juillet et en aout), on obtient uneclassification exacte a 95%7o. Les zones inondees de faqonpermanente, frequente ou meme occasionnelle, sont les pluspropices A la proliferation des moustiques. Dans le cas deC. tarsalis, moustique qui se reproduit dans les mares etetendues d'eau permanentes, les biotopes de la categorie deszones constamment submergees sont les plus propices a laconstitution de gites larvaires, tandis que les biotopes de lacategorie des zones occasionnellement submergees sont lesmoins propices. Dans le cas d'Ae. vexans, espece qui sereproduit dans les eaux d'inondation, la situation est exacte-ment opposee, les biotopes correspondant A des zonesoccasionnellement ou frequemment submergees etant plusfavorables A leur reproduction que ceux de la categorie deszones constamment submergees.

L'identification, le releve cartographique et la mesure dupotentiel larvaire des divers biotopes A partir des imagesobtenues par satellite se heurtent A plusieurs difficultes dontles principales sont la resolution (de l'ordre de 0,45 ha danscette etude), la couverture nuageuse et le dMlai qui s'ecouleentre la collecte des valeurs de la reflectance A partir dusatellite et le traitement de ces donnees apres report surbandes magnetiques. Le carroyage etabli A partir des valeursde la reflectance ne permet pas toujours de delimiter de

facon precise les petits biotopes heterogenes, non contigus etde forme irreguliere. La couverture nuageuse peut fausserles mesures de reflectance et le traitement des donnees prendjusqu'a trois mois ou plus. Cependant, il n'est pasimpossible sur le plan technique de nettement accroitre laresolution en atteignant 0,02-0,09 ha, et des dispositionsparticulieres peuvent permettre de reduire la duree du traite-ment.En conclusion, les auteurs estiment que l'imagerie a partir

de satellite est une bonne methode pour dresser la carte desgrands projets de mise en valeur des ressources hydriques envue de delimiter et de classer les biotopes et d'en mesurer lepotentiel larvaire (ainsi que l'importance effective despopulations de moustiques). De meme, la methode aurait saplace dans la planification des programmes de luttecouvrant une vaste superficie ou elle permettrait d'evaluerl'ampleur du probleme et de suivre les progres accomplisgrace aux operations de reduction du nombre de sources demoustiques. Quand les valeurs de la reflectance obtenuespar photographie aerienne multispectrale comporteront unemeilleure resolution, cette technique de cout modiquetrouvera de plus larges applications dans toute une serie deprogrammes operationnels de lutte antivectorielle.

REFERENCES

1. BARNES, C. M. & CIBULA, W. G. Some implicationsof remote sensing technology in insect control programsincluding mosquitoes. Mosquito news, 39: 271-282(1979).

2. BERNSTEIN, R. & STIERHOFF, G. C. Precision proces-sing of earth image data. American scientist, 64:500-508 (1976).

3. BIDLINGMAYER, W. L. & KLOCK, J. W. Notes on theinfluence of salt-marsh topography on tidal action.Mosquito news, 15: 231-235 (1955).

4. CLINE, B. L. New eyes for epidemiologists: aerialphotography and other remote sensing techniques.American journal of epidemiology, 92: 85-89 (1970).

5. COOLEY, M. E. & TURNER, R. M. Application ofERTS products in range and water managementproblems in Sahelian Zone, Mali, Upper Volta andNiger. Project Report (IR)WA-4, US GeologicalSurvey, Reston, Virginia, 1975.

6. COUSSERANS, J. ET AL. Les bases 6cologiques de lad6moustication: methodes de realisation et d'utilisationde la carte phyto-6cologique. Vieet milieu, C20(l): 1-20(1969).

7. Dow, R. P. ET AL. Dispersal of female Culex tarsalisinto a larvicided area. American journal of tropicalmedicine and hygiene, 14: 656-670 (1965).

8. HAYES, R. 0. Impact of water resources on vector-borne diseases. Journal of water resourcesplanning andmanagement, 102(WR2): 177-183 (1976).

9. HAYES, R. 0. ET AL. Lewis and Clark Lake mosquitocontrol recommendations. Journal of the Environ-mental Engineering Division, 104(EE4): 701-716(1978).

10. HORSFALL, W. R. ET AL. Bionomics and embryologyof the inland floodwater mosquito Aedes vexans.Urbana, IL, University of Illinois Press, 1973,pp. 64-78.

11. KUCHLER, A. W. Vegetation mapping. New York,Ronald Press, 1967.

12. MAIRE, A. Identification des biotopes a larves demoustiques des Tourbieres de la Bosse-Maurice (QuebecMeridional). Naturaliste Canadien, 104: 429-440(1977).

13. MAXWELL, E. L. Multivariate system analysis ofmultispectral imagery. Photogrammetric engineeringand remote sensing, 42: 1173-1186 (1976).

14. MITCHELL, C. J. Arthropod-borne encephalitis virusesand water resource developments. CahiersO.R. S. T. O.M., Entomologie medicale et Parasitologie.15: 241-250 (1977).

15. PAUTOU, G. ET AL. Cartographie 6cologique appliqu6ea la demoustication. Documents de cartographiee'cologique, 11: 1-15 (1973).

16. Rioux, J. A. ET AL. Phyto-ecological basis ofmosquito control: cartography of larval biotopes.Mosquito news, 28: 572-582 (1968).

17. SCHNETZLER, C. C. & THOMPSON, L. L. Multispec-tral resource sampler: an experimental satellite sensorfor the mid-1980s. Proceedings of the SPIE TechnicalSymposium, Huntsville, Alabama, 22-24 May 1979,Vol. 183: 174-176 (1979).

18. STEINER, D. Time dimension for crop surveys fromspace. Photogrammetric engineering and remotesensing, 36: 187-193 (1970).


19. SURTEES, G. Urbanization and the epidemiology ofmosquito-borne disease. Abstracts on hygiene, 46:121- 134 (1971).

20. TEMPELIS, C. H. Host-feeding patterns of mosquitoes,with a review of advances in analysis of blood meals by

serology. Journal of medical entomology, 11: 635-653(1975).

21. TUCKER, C. J. & MAXWELL, E. L. Sensor design formonitoring vegetation canopies. Photogrammetricengineering and remote sensing, 42: 1399-1410 (1976).