Journal of the American Mosquito Control Association, 16(l):28-35, 2000Copyright @ 2000 by the American Mosquito Control Association, Inc.

APPLICATION OF GEOGRAPHIC INFORMATION TECHNOLOGY INDETERMINING RISK OF EASTERN EQUINE ENCEPHALOMYELITIS

VIRUS TRANSMISSION

ABELARDO C. MONCAYO,T JOHN D. EDMAN AND JOHN T, FINN,

Department of Entomology, University of Massachusetts, Amherst, MA 0IOO3

ABSTRACT Geographic information system (GIS) technology and remote sensing were used to identifylandscape features determining risk of eastern equine encephalomyelitis virus (EEE) transmission as defined bythe abundance of Culiseta melanura (the enzootic vector) and 6 putative epidemic--epizootic vectors in Massa-chusetts. Landsat Thematic Mapper data combined with aerial videography data were used to generate a mapof Iandscape elements at epidemic--€pizootic foci in southeastern Massachusetts. Geographic information systemtechnology was used to determine the proportion of landscape elements surrounding 15 human and horse casesites where abundance data were collected for Culiseta melanura, Aedes canadensis, Aedes vexans, Culex sal-inarius, Coquillettidia perturbans, Anopheles quadrimaculatus, and Anopheles punctipennis. The relationshipsbetween vector abundance and landscape proportions were analyzed using stepwise linear regression. Stepwiseregression indicated wetlands as the most important major class element, which accounted for up to 72.5Vo ofthe observed variation in the host-seeking populations of Ae. canadensis, Ae. vexans, and Cs. melanura. More-over, stepwise linear regression demonstrated deciduous wetlands to be the specific wetland category contributingto the major class models. This approach of utilizing GIS technology and remote sensing in combination withstreet mapping can be employed to identify deciduous wetlands in neighborhoods at risk for EEE transmissionand to plan more efficient schedules of pesticide applications targeting adults.

KEY WORDS Geographic information systems, eastem equine encephalomyelitis virus, remote sensing

INTRODUCTION

Recent studies have investigated the potential ofa variety of mosquito species to serve as the epi-demic vectors of eastern equine encephalomyelitisvirus (EEE) in Massachusetts (Vaidyanathan et al.1997; Moncayo and Edman, 1999). These potentialvectors include (in order of relative importance)Culex salinarius (Coq.), Aedes canadensis (Theo-bald), C oquille ttidia p e rturbans (Walker), Anop he -

les quadrimaculatus (Say), Aedes vexans (Meigen),and Anopheles punctipennis (Say). Controlling pop-ulations of these potential EEE vectors is of greatinterest to health officials, mosquito control work-ers, and the public at large. Identification of land-scape elements that predispose humans and horsesto risk of EEE transmission is important for under-standing and controlling eastern equine encepha-lomyelitis. Landscape epidemiology or the ecolog-ical approach to the study of infectious agents wasdeveloped more than 50 years ago (Pavlovsky1966, Fish 1996). In this study we adopted this ap-proach because of the relationship that exists be-tween the vectors that transmit EEE and the phys-ical and biological features of the landscape.Features in the landscape such as vegetation pro-vide food and other resources such as shelter andresting and developmental sites for mosquitoes.Geographic information system (GIS) technology

lPresent address: Center for Tropical Diseases, Depart-ment of Pathology, University of Texas Medical Branch,301 University Boulevard, Galveston, TX 77555-0609.

2 Department of Forestry and Wildlife, HoldsworthHall, University of Massachusetts, Amherst, MA 01003.

allows the examination of remotely sensed land-scape elements that relate to vector abundance, andtherefore transmission risk. This technology hasbeen used recently to answer epidemiologically rel-evant questions in a variety of arthropod-borne dis-ease systems (Linthicum et al. 1990, Randolph1993, Beck et al. 1994, Malone et al. 1994, Crosset al.1996, Rogers et al. 1996, Dister et al.1997,Gleiser and Gorla 1997, Thompson et al. 1997,Baylis et al. 1998, Estrada-Pena 1998, Hightoweret al. 1998, Kitron 1998, Maloney et al. 1998,Omumbo et al. 1998, Wilson 1998).

In this study we have used GIS technology as atool to study the landscape of epidemic foci of east-ern equine encephalomyelitis in southeastern Mas-sachusetts. The New England Gap Analysis Project(GAP) has developed base vegetation maps forNew England by using Landsat Thematic Mapper(TM) (Space Imaging Inc., Thorton, CO) data (mul-tispecral satellite data) combined with aerial vide-ography (Slaymaker et al. 1996). The GAP wasoriginally designed as a tool for determining theconservation status of components of biodiversity.The main purpose of the GAP has been to identifygaps (vegetation and animal species that are notrepresented in conservation areas) that may befllled through the creation of new reserves or im-provement of land management. The GAP data pro-vide many major forest, nonforest, wetland, urban,and water classifications that can be further sepa-rated into subtypes to obtain a more detailed per-ception of the landscape.

The New England landscape is 5O-957o forestedwith a wide variety of forest types in relatively

28

GIS ron EEE RIsK 29Marcn 2000

Table 1. Proportion of major landscape elements in the

studY area.

Landscape element Area (m'�) Proportion

fication data was verified by comparing LandsatTM classifications with ground cover examined byfield mapping. These data were in Arclnfo (Envi-ronmental Systems Research Institute Inc., Red-lands, CA) grid format (Raster data set; i.e., non-polygon). Landscape elements identified by thisdata set include the following 1l major type cate-gories: northern hardwoods, red maPle dominant'oak dominant. birch dominant, oak-maple-birchcodominant, conifer, maple-oak-{onifer codomi-nant, palustrine wetlands, nonforest cover, freshopen water, and urban elements. Each of these ma-jor types was further broken down into subtypes sothat a more detailed landscape could be visualized.A total of 64 such subtype elements were availablefrom this data set. Eight of the I I major type cat-egories were identified in our study area with onlynorthern hardwoods, birch dominant, and maple-oak--conifer codominant missing.

Mosquito abundance data: Population abundancedata were obtained through trapping performed frommid-July to mid-September, 1996. ABC traps(American Biophysics Corporation, East Greenwich,RI) traps were used to sample adult mosquito pop-ulations at 15 epidemic sites. Tfaps were equippedwith 2 attractants: a photosensitive flickering lightthat responds to changes in light intensity and startsoperating at dusk, and compressed COr, which wascontinuously emitted at 500 mVmin from a storagetank. This flow rate was intended to mirnic the av-erage CO2 discharge from a medium- to large-sizedmarnmal such as an adult human. Tlvo ABC trapswere placed at each case site during 2 consecutivenights each week. The average seasonal populationabundances obtained for each of the 6 putative EEEepidemic vectors (Ae. canadensis, Ae. vexans, Cq.perturbans, Cx. salinarius, An. quadrimaculntus, artdAn. punctipennis) ard the known enzootic vector(Cs. melanura) were used to construct our linear re-gression models.

Landscape-mosquito abundance analysis: Weprojected existing USGS case site coverages intothe GAP data coordinate system (UTM zone 19 toUTM zone l8). Each of the 15 case sites was buff-ered (delineated) around an area 1 km in radiusfrom the case site (Fig. 1). This l-km distance wasan arbitrary distance based on an average flightrange for mosquitoes (Hobbs et al. 1974). The buff-ered areas were clipped out and converted to poly-gons. The polygonal buffers were overlaid on thelandscape map to calculate the proportion of thearea that each landscape element occupied withinthe buffer. We calculated the area of each landscapecategory in acres for each buffered area.

Before variable selection tbrough stepwise linearregression, linear regression models of the nontrans-formed data were created. The I and log(Iz) were fit-ted in the proportions pl, . .. , p(c - 1), where pvalues, are untransforrned major class landscape ele-ment proportions, I is the average mosquito abun-dance per trap night during the sufilmer of 1996, and

Nonforest coverConiferWetlandsFresh open waterUrbanOak dominantOak-maple-birchRed maple dominantTotal

small stands interspersed throughout the landscape.In southeastern Massachusetts, wetlands make up2O7o of the land (Komar and Spielman 1994). For-ested wetlands, some of which are now conserva-tion areas, are important for a healthy water supplyand flood protection for local residents. Wetlandsalso provide an important habitat for wildlife, in-cluding many species of birds and mosquitoes im-plicated in the EEE cycle. Residential developmentwithin or adjacent to forested wetland areas hasbrought an increasing number of humans and hors-es in contact with mosquitoes capable of transmit-ting EEE (Komar and Spielman 1994).

The GAP data were used to describe the land-scape composition of 15 case sites resulting fromthe most recent major eastern equine encephalo-myelitis epidemics in Massachusetts in the 1980sand 1990s. Our objective was to identify landscapeelements that are highly correlated with populationabundance of the enzootic vector of EEE, Culisetamelanura (Coq.), as well as with the potential ep-idemic vectors in Massachusetts.

MATERIALS AND METHODS

Study site: This study focused on the southeast-ern corner of Massachusetts and specifically onBristol and Plymouth counties, where the majorityof human and horse cases have occurred histori-cally. In addition, these 2 counties have the greatestrecord of virus isolations from mosquito pools(Massachusetts Department of Public Health, StateLaboratory, Boston).

First, we developed a map based on locations ofhorse and human cases as well as virus isolates. Tocreate this coverage we used a U.S. Geological Sur-vey (USGS) 1:25,000 datalayer that was registeredto a universal transverse mercator (JTM) real-world coordinate system. This allows proper scaledorientation of case locations.

Landscape rnap generation.' A major componentof this study was the use of remotely sensed datathat were already available in the New EnglandGAP analysis homepage (Finn and'Griffin 1999).Particular attention was made to landscape ele-ments associated with mosquito abundance (e.g.,palustrine wetlands). The accuracy of GAP classi-

15,379,452r14,447,364

8,854,8363,262,3082,330,220r,398,t32

932,088932,088

46,604,400

0.330.310.190.070.050.03o.o20.021.00

GIS ron EEE Rrsx 3 lMencu 20O0

Table 2. proportion of major landscape categories within l-km radius from each trap site in each town.

Red maple Oak Oak-maPle- Conifer

Area (m2) dominant dominant birch dominant WetlandTown

Nonforestcover Water Urban Total

NortonMiddleboro-NWMiddleboro-NEMiddleboro-SLakevilleRochesterOnsetWarehamKingston-SKingston-NPembrokeAbingtonWhitmanBrocktonBridgewater

3,r25,7003,087,0003 , 1 3 1 , 1 0 03,1 23,0003,126,6003,121,2003,000,0003,000,0003,130,2003,127,5003,126,6003,126,6N3,128,4003,123,9003,126,600

0.000.01o.o20.000.020.010.000.000.010.000.000.06o.060.040.00

o.o20.030.040.010.030.040.010.00o.o2o.o2o.olo.09o .100.060.01

0.010.010.020.000.010.0r0.010.000.000.000.000.010.080.060.05

o.260.40o.450.51o.290.540.35o.490.40o.32o.320 . 1 60.08o.o20.01

o.220 . 1 6o.220.160.080 . 1 90.040.03o . 1 70.33o.07o. l6o . 1 10.060.80

o.250.38o.240.310 .19o.2ro.220 . t 90.35o.230.53o.480.530.68o . l 1

0 .160.000.000.000.380.00o.25o.240.000.010.030.000.000.000.00

0.080.01o.o20.010.000.0ro .17o.o20.040.090.04o.04o.040.100.01

1.001.00l . 0 l1.001.001 . 0 11.05o.970.99l.oor.001.001.00t.o20.99

of landscape variables were used as the independentvariables. Landscape proportions were thus added tothe equations for each of the 6 species until none ofthe remaining variables improved the model.

The lst regression models for the 6 species usedthe major landscape categories available in theGAP data set. Table I shows the distribution ofmajor landscape classes in the entire study area.Table 2 is a list of the proportions of the majorlandscape categories within a l-km radius fromeach trap site in each town surveyed.

Once a determination was made of which majortype categories were good predictors of speciesabundance, we perforrned another set of stepwiselinear regressions to tease out the major type sub-classes that most contribute to the maior class mod-els for each species.

RESULTS

Landscape description

The study area (combined buffered areas aroundeach trap site) was found to be composed of 8 dif-ferent major landscape classes as defined by theGAP data set (Table l). The predominant landscapefeature was nonforest cover (33Vo). This was fol-lowed by 3l%o conifer, l9Vo wetlands,7Vo freshopen wate! 5Vo urban,3Vo oak dominant, 2Vo oak-maple-birch mixture, and 2Vo red maple dominant.

The proportion that each landscape element con-

tributes to the total study area (Table 1) does notnecessarily follow the order of the contribution ofeach landscape element in a particular town (Table2). For example, in Bridgewater, the element thatmost contributes to the area I km around the trapsite is not nonforest cover, as would be expectedby looking at the combined landscape buffer zones,but rather wetlands (8OVo).

Stepwise linear regressions with maior classpredictors of mosquito abundance

We found that wetlands accounted for 72.5Vo ofthe observed variation in abundance of Ae. cana-densis populations (Table 3). This association waspositive (Fig. 2). The conifer-dominant element ac-counted for l2.5%o of the variation observed in thepopulation abundance of this same species. The co-nifer-dominant landscape proportion was positivelyassociated with abundance of Ae. canadensis b]utwas not highly correlated with it (r : 0.071; Table3 and Fig. 3). The regression analysis of varianceshowed the equation log(mosquito average abun-dance) : 0.188 + (2-598 X wetland proportion) *(6.716 x conifer-dominant proportion) to be a goodpredictor of abundance of Ae. canadensis duringthe summer season (F ratio : 10.094, P : 0.OOO).

Conifer dominant was the landscape element thatcontributed most to the Ae. vexans major classmodel, accounting for 52.5Vo of the observed var-

Table 3. Results of stepwise linear regression based on major type landscape categories.

Species Variable R R2 SEEI F ratio

Aedes canadensis

Aedes vexans

Culiseta melanura

WetlandWetland * coniferConiferConifer + wetlandWetland

0.851o.922o.725o.836o.694

o.7250.850o.5250.700o.482

o.2959o.22700.2801o.23190.1991

34.22110.09414.3796.966

I 2 .108

0.0000.0000.002o.0010.004

ISEE, standrd emor of the estimate.

J Z JoURNAL oF THE AMERTcIN Mosqurro Cowrnol AssocrarroN Vol. 16. No. I

y. 2-& + oas* - 0.76

0.1 0 .3 o.4 0.5

Fig.2. Linear regression of Aedes canadensis log av_erage abundance and wetland proportion.

iation in mosquito abundance (Table 3). The coniferelement was negatively correlated with mosquitoabundance (Fig. 5). When wetland was added tothis model, the R2 value increased from 0.525 to0.70O, which translates to an r value increase (abetter measure of the correlation between mosquitoabundance and landscape variables) from 0.725 to0.836. The standard error of the estimate, whichestimates the standard deviation around the regres-sion line, was decreased from 0.2801 to 0.2319with the addition of the wetland category. The wet-lands element was positively correlated with abun-dance (Fig. 4). The equation log(mosquito averageabundance) : O.625 + (-1.283 X conifer-domi-nant proportion) + (0.925 X wetland proportion.lwas a good predictor of abundance of Ae. vexansduring the summer season (F ratio : 6.966, p :0.001).

Wetlands was the major class element that ac-counted for most of the variation seen in the abun-dance of Cs. melanura populations (R, = 0.482).Again, wetlands were positively correlated with logaverage abundance of this species (r : 0.694; Table3 and Fig. 6). The equation log(mosquito averageabundance) : O.324 + (0.978 * wetland propor-tion) was an adequate predictor of abundance (Fratio : l2.lo8, P : 0.00a). Stepwise linear regres-sion with major types was not able to devise a mod-el that would predict the abundance of Cq. pertur-bans, Cx. salinarius, An. quadrimaculatus, or An.punctipennis,

y - 1.ga + 0.1S aRr .0 . f l5

. ' . j

0.3 0 .a 05 0 .6

w.d.nd ?rcrodon

Fig. 4. Linear regression of Aedes vexans log averaseabundance and wetland proportion.

Stepwise linear regressions with subclasspredictors of mosquito abundance

Stepwise linear regressions were then performedto tease out the wetland subclasses that most con-tributed to the major class models. In all cases, de-ciduous wetland was the specific wetland categorythat contributed most to the models that relate land-scape elements to abundance of the populations ofAe. canadensis, Ae. vexans, Cx. salinarias, and Cs.melanura (Table 4 and Figs. 7-10). Deciduous wet-lands best accounted for the observed variation inpopulations of Ae. canadensis (R2 : 0.834), fol-lowed by Cs. melanura (R, = O.527), Ae. vexans(R'� : O.442), and Cx. salinarius (R, : 0.317). Theregression equations for predicting mosquito abun-dance were as follows: for Ae. canadensis:log(mosquito average abundance) : -0.0137 -t(l.6ll X deciduous wetland proportion) + (2.827X scrub or shrub marsh proportion) * (1.944 xemergent persistent proportion) (F ratio : 5.O67, P: 0.000); for Cs. melanura: log(mosquito averageabundance) : 0.264 + (0.894 X deciduous wetlandproportion) (F ratio : 14.498, P = O.OO2); for Ae.yexansi log(mosquito average abundance) : O.292+ (1.091 X deciduous wetland proportion) *(-3.028 X mixed deciduous conifer wetland pro-portion) (F ratio : 8.113, P = O.0Ol); for Cx. sal-inarius: log(mosquito average abundance) : O.215+ (1.531 X deciduous wetland proportion) (F ratio- 6.O43, P : O.O29). Construction of stepwise lin-ear regression models was not possible for the wet-land subclasses for Cq. perturbans or the Anophelesspecies.

3

z z .

1 r zq ai ! r s

€ a 1

i o u

U 2.5

2 t z{ € r s

i I 'I 0.5

l

0.9 ;0.90.8o.10.60.2

0.8

I

! r .' . . -I ! e '

i € r . si !

i ! e 'l i o s

1 2

i 1 . 5

t 0 5

a

0.2 0.3 0.4AnbrhlDdPrctsdd

0.2 0 .3 04

6nhr bmimdPMon

' v = 0 . 7 & + 0 . 6 5 7*.0_129

Fig. 3. Linear regression of Aedes canadensis log av-erage abundance and conifer dominant proportion.

Fig. 5. Linear regression of Aedes vexans log averageabundance and conifer dominant proportion.

MancH 2000 GIS FoR EEE Rtsr 33

y-0 .976x+0.324

R , " 0 f i 2

-, I0.2 0.6 0.7 0.8 0.9

Fig. 6. Linear regression of Culiseta melanuralogav-erage abundance and wetland proportion.

DISCUSSION

The aim of this study was to use remotely sensedlandscape information to determine environmentalelements that predispose humans and horses to EEEtransmission. In Massachusetts, Atlantic white ce-dn (Chamaecyparis thyoides) and red rnaple (Acerrubrum) swamps provide optimal habitat for the de-velopment of larvae of Cs. melanura. Adult Cs. me-lanura are found in large numbers in light traps setnear these forested wetlands. This species is re-sponsible for the maintenance and amplification ofEEE in reservoir birds throughout the United States(Edman et al. 1993). Reservoir hosts, for example,American robins (Turdus migratorius) and red-winged blackbirds (Agelaius quiscula), are knownto form roosts in forested swamps (Komar andSpielman 1994). ln order for a bridge or epidemic-epizootic vector to obtain EEE, the vector must takea blood meal from the reservoir host. This infectiveepidemic--epizootic vector can then fransmit the vi-rus by taking a subsequent blood meal from a naivemammalian host. Although occupying differentniches, the 6 potential vectors of EEE in Massa-chusetts have been obtained from light traps at ep-idemic foci near forested wetlands. The focus ofthis study was to identify the type of habitat thatcontributes to the appearance and population sizeof Cx. salinarius, Ae. canadensis, Cq. perturbans,An. quadrimaculatus, Ae. vexans, and An. puncti-pennis. Because the population data used for thisanalysis were obtained from COr-baited ABC lighttraps, the data represent 6O-7OVa of the proportionof the population attracted to a human host during

l

y - 2 . 2 9 1 \ + O 2 T 3

o 0.1 0.2 0.3 0.4 0.5 0.0 0.7

&clduou.W.d.nd PrcFdon

Fig.7. Linear regression of Aedes canadensis log av-

erage abundance and deciduous wetland proportion.

a 2-h period beginning at sunset (Vaidyanathan andEdman 1997). The habitats identified in this studyrepresent habitats that best correlate with trappedadult populations.

Wetlands were found to account for 72.5Vo of theobserved variation in the abundance of Ae. cana-densis populations and 48.2Vo of the variation inCs. melanura populations. More specifically, thedeciduous (or leafy) wetlands accounted for thevariation seen in populations of these 2 species(83.4 and 52.7Vo, respectively). This is consistentwith the known biology of these 2 mosquito spe-cies. Wooded swamps and adjacent hardwood for-ests are the main types of habitats for both larvaland adult Cs. melanura (Joseph and Bickley 1969).Aedes canadensds develop in temporary woodlandpools containing decaying leaves. These sites havebeen documented as being closely associated withswamp habitats of Cs. melanura (Carpenter et al.1946, Main et al. 1968). In addition, female Ae.canadensis remain close to the forest margin evenwhen host seeking at dusk and in the early evening(Hayes 1962, Trueman and Mclver 1986).

Conifer dominant accounted for 52.5Vo of the ob-served variation of Ae. vexans populations and was72.5Vo conelated (F ratio : 14.379, P : 0.002)with these populations. However, this variation wasnegatively correlated with this landscape type. Ae-des vexans develops in open meadows or grassydepressions and river and stream floodplains (Car-penter et al. 1946, Horsfall et al. 1973, Komar andSpielman 1994). Conifer-dominant areas do notprovide the type of floodwater habitat that promotes

3

E 2.5

l s z

3 E r . s

; <E o u

F _ _

5

8 E; € 1 . 5B E

{ € 'f r o u

0.60.3 0.4 0.5W.trnd Prc@don

Table 4. Results of stepwise linear regression based on wetland subcategories.

Species Variable R R2 SEE F ratio

Aedes canadensis

Aedes vexans

Culex salinariusCuliseta melanura

Deciduous wetlandDeciduous wetland + scrub or shrub

marshDeciduous wetland f scrub or shrub

marsh + emergent persistentDeciduous wetlandDeciduous wetland f mixed Decid-

uous-ConiferDeciduous wetlandDeciduous wetland

65.499 0.0006.740 0.000

0.913 0 .8340.945 0.894

o.963

0.665o.817

0.563o.726

o.927

o.442o.667

o.317o.527

o.22950 . 1 9 1 I

0.1652

0.3036o.2441

0.50480.1903

5.067 0.000

r0.313 0.0078.113 0.OOl

6.043 0.029t4.498 0.OO2

34 JounNar- oF THE AMERTCAN Moseurro CoNrnor- AssocnnoN VoL. 16, No. I

!

! 0..

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

bcrduou.Wet.M prcFdon

y - 0 E g x + 0 . 2 i l

0 1 0.2 0,3 0.7 0.8 0.9

3g _ _E u..

J € 1 5

i 3 .i <9 ^ -

' t

y . 1 . 2 0 1 r + 0 . 0 8 2

* = 0 4 2

Fig. 8. Linear regression of Aedes vexans log averageabundance and deciduous wetland proportion.

the presence of Ae. vexans. This species was cor-related with wetlands and more specifically decid-uous wetlands (r : 0.665). Adults of this speciesrest in damp woods.

Derivaton of a major class model was not pos-sible for Cx. salinarius, Cq. perturbans, and the 2Anopheles species. However, we were able to detecta correlation between deciduous wetlands and var-iation in population sizes of Cx. salinarius.

We have identified deciduous wetlands as thespecific landscape category that contributes most tothe abundance of Cs. melanura, Ae. canadensis, Ae.yexans, and C.r. salinarius and, therefore. risk ofEEE transmission. Therefore, targeting of decidu-ous wetlands by mosquito control practitioners canreduce populations of enzootic and several potentialepidemic vectors of EEE. Landscape maps derivedfrom GAP data can be combined with street addressdata to better plan pesticide applications. Althougha stepwise linear regression did not correlate C4.perturbans abundances with any specific landscapefeature, their larval habitats are well known. Larvaepuncture roots of emergent vegetation (cattail, wa-ter lettuce, swamp loosestrife, arrow arum, andmaidencane) in permanent freshwater marshes(Crans and Schulze 1986, Morris et al. 1990).Emergent persistent vegetation is a distinguishablesubcategory of the wetlands GAP data classifica-tion and, therefore, can also be mapped based onthese TM data to help direct planning strategiesagainst this potentially important EEE vector.

ACI(NOWLEDGMENTS

We thank the Office of Geographic InformationAnalysis at the University of Massachusetts for the

I

Fig. 10. Linear regression of Culiseta melanura logaverage abundance and deciduous wetland proportion.

use of their personnel and facilities. Special thanksto Meghan S. Moncayo for assistance in obtainingmosquito abundance data used in this study. Thisstudy was funded by a grant from the Massachu-setts Department of Public Health and from Hatchsupport from the Massachusetts Agricultural Ex-periment Station.

REFERENCES CITED

Baylis M, Bouayoune H, Touti J, El Hasnaoui H. 1998.Use of climatic data and satellite imagery to model theabundance of Culicoides imicola Kieffer, the vector ofAfrican horse sickness virus in Morocco. Med Vet En-tomol 12:250-266.

Beck LR, Rodriguez MH, Dister SW, Rodriguez AD, Re-jmankova E, Ulloa A,Meza RA, Roberts DR, Paris JESpanner MA, Washino RK, Hacker C, Legters Ll. 1994.Remote sensing as a landscape epidemiological tool toidentify villages at high risk for malaria transmission.Am J Trop Med Hyg 5l:271-28O.

Carpenter SJ, Middlekauff WW, Chamberlain RW. 1946.The mosquitoes of the southern Uniled States east ofOklahoma and Texas. Monograph No. 3. Notre Dame,IN: The American Midland Naturalist.

Crans WJ, Schulze TL. 1986. Evidence incriminating Co-quillettidia perturbans (Diptera: Culicidae) as an epi-zootic vector of eastem equine encephalitis. I. Isolationof EEE virus from Cq. perturbans during an epizooticamong horses in New Jersey. BuIl Soc Vector Ecol Il:178-1 84.

Cross ER, Newcomb WW, Tucker CJ. 1996. Use ofweather data and remote sensing to predict the geo-graphic and seasonal distribution of Phlebotomus pa-patasi in Southwest Asia. Am J Trop Med Hyg 54:530-536.

Dister SW Fish D. Bros SM, Frank DH, Wood BL. 1997.Landscape characterization of peridomestic risk forLyme disease using satellite imagery. Am J Trop MedHyg 57:687-692.

Edman JD, Timperi R, Werner B. 1993. Epidemiology ofeastern equine encephalitis in Massachusetts. J FlaMosq Control Assoc 64(2):84-96.

Estrada-Pena A. 1998. Geostatistics and remote sensing aspredictive tools of tick distribution: a cokriging systemto estimate lxodes scapularis (Acari: Ixodidae) habitatsuitability in the United States and Canada from ad-vanced very high resolution radiometer satellite imag-ery. J Med Entomol 35:989-995.

Finn JT, Griffin CR. Southern New England GAP Anal-ysis homepage. http://bandersnatch.fnr.umass.edu/gap(accessed November 30, 1999).

i . "F E 'j ! . .! =

-

E e ,t

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

bdduowW.dand Prcpodon

Fig. 9. Linear regression of Culex salinarius Iog av-erage abundance and deciduous wetland proportion.

GIS FoR EEE Rtsx 35MARCH 2000

Fish D. 1996. Remote sensing and landscape epidemiol-

ogy. Adv Astronaut Sci 91:1057-1063.Gleiser RM, Gorla DE. 1997. Monitoring the abundance

of Aedes (Ochlerotatus) albifasciatus (Macquart 1838)(Diptera: Culicidae) to the south of Mar Chiquita Lake,

central Argentina, with the aid of remote sensing' Ann

Trop Med Parasitol 9l:917-926.Hayes RO. '1962. The diel activity cycles of Culiseta mel-

anura (Coquillett) and allied mosquitoes. Mosq News

22(4):352-357.Hightower AW, Ombok M, Otiendo R, Odhiambo R, Ag-

greg JO, Lal AA, Nahlen BL, Howley WA. 1998. Ageographic information system applied to a malariafield study in western Kenya. Am J Trop Med Hyg 58:.266-272.

Hobbs JH, Lowe RE, Schreck CF'. 1974. Studies on flightrange and survival of Anopheles albimanus Wiedemannin El Salvador. I. Dispersal and survival during the dryseason. Mosq News 34:389-393.

Horsfall WR, Fowler HW Jr, Moretti LI, Larsen JR. 1973.Bionomics and embryology of the inland fioodwatermosquito Aedes vexans Urbana, IL: University of Illi-nois Press.

Joseph SR, Bickley WE. 1969. Culiseta melanura (Co-quillett) on the eastern shore of Maryland (Diptera:Culicidae). Bulletin ,4'-161. College Park, MD: Univ.Maryland Agricultural Experiment Station.

Kitron U. 1998. Landscape ecology and epidemiology ofvector-borne diseases: tools for spatial analysis. J MedEntomol 35:435-445.

Komar N, Spielman A. 1994. Emergence of eastern en-cephalitis in Massachusetts. Ann N Y Acad Sci 740:157-168 .

Linthicum KJ, Bailey CL, Tucker CJ, Mitchell KD, LoganTM, Davies FG, Kamau CW Thande PC, Wagateh JN.1 990. Application of polar-orbiting, meteorological sat-ellite data to detect flooding of Rift Valley fever virusvector in mosquito habitats in Kenya. Med Vet Entomol4:433-438.

Main AJ, Hayes RO, Tonn RI. 1968. Seasonal abundanceof mosquitoes in southeastern Massachusetts. MosqNews 28:619-626.

Malone JB, Huh OK, Fehler DP, Wilson PA, WilenskeyDE, Holmes RA, Elmagdoub AI. 1994. Temperaturedata from satellite imagery and the distribution of schis-tosomiasis in Egypt. Am J Trop Med Hyg 5O:714-722.

Maloney JM, Kelly CS, Weinstein f; Maquire M, RitchieS. 1998. Domestic Aedes aegypti breeding site surveil-lance: limitations of remote sensing as a predictive sur-veillance tools. Am J Trop Med Hyg 59:261-264.

Moncayo AC, Edman JD. 1999. Toward the incrimination

of epidemic vectors of eastern equine encephalomyelitis

virus in Massachusetts: abundance of mosquito popu-

lations at epidemic foci. J Am Mosq Control Assoc 15:

479-492.Morris CD, Callahan JL, Lewis RH. 1990. Distribution

and abundance of larval Coquillettidia perturbans in a

Florida freshwater marsh. J Am Mosq Control Assoc 6:

452-460.Omumbo J, Ouma J, Rapuoda B, Craig MH, Le Sueur D,

Snow RW. 1998. Mapping malaria transmission inten-

sity using geographical information systems (GIS): an

example from Kenya- Ann Trop Med Parasitol 92:.7-22.

Pavlovsky EN. 1966. Natural nidality of transmissible dis-

eases Urbana, IL: Univ. Illinois Press.Randolph SE. 1993. Climate, satellite imagery and the

seasonal abundance of the tick Rhipicephalus appen-diculatus in southern Africa: a new perspective. Med

Vet Entomol 7:243-258.Rogers DL, Hay SI, Packer MJ. 1996. Predicting the dis-

tribution of tsetse flies in West Africa using Fourierprocessed meteorological satellite data. Ann Trop MedParasit 9O:225-241.

Slaymaker DM, Jones KML, Griffin CR, Finn JT, 1996.Mapping deciduous forests in southern New Englandusing aerial videography and hyper-clustered multi-tem-poral Landsat TM imagery. In: Scott JM, Tear TH, Da-vis FW, eds. Gap analysis: a landscape approach tobiodiversity pktnning Bethesda. MD: American Societyof Photogrammetry and Remote Sensing. p 87-101.

Thompson MC, Connor SJ, Milligan PJW Flossa S. 1997.Mapping malaria risk in Africa: what can satellite datacontribute? Parasitol Today 13:313-318.

Trueman DW, Mclver SB. 1986. Temporal pattems ofhost-seeking activity of mosquitoes in Algonquin Park,Ontario. Can J Zool 64:731-737.

Vaidyanathan R, Edman ID. 1997. Sampling with lighttraps and human bait in epidemic foci for eastern equineencephalomyelitis virus in southeastern Massachusetts.J Am Mosq Control Assoc 13:348-355.

Vaidyanathan R, Edman JD, Cooper LA, Scott TW. 1997.Vector competence of mosquitoes (Diptera: Culicidae)from Massachusetts for a sympatric isolate of easternequine encephalomyelitis virus. J Med Entomol 34:346-352.

Wilson ML. 1998. Distribution and abundance of Ixodesscapularis (Acari: Ixodidae) in North America: ecolog-ical processes and spatial analysis. J Med Entomol 35:446-457.