Journal of the American Control Association, 19(4):382-391,2003 Copyright @ 2003 by the American Mosquito Control Association, Inc.

SPATIOTEMPORAL OVPOSITION AND HABITAT PREFERENCES OF TRISERIATUSAND ALBOPICTUS IN AN EMERGING FOCUS OF LA CROSSE VIRUS'

CHRISTOPHER M. BARKER,, CARLYLE C. BREWSTER3 rNo SALLY L. PAULSON

Department of Entomology, 216 Price HaII, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061 -03 I 9

ABSTRACT. The number of cases of encephalitis caused by La Crosse virus recently has increased in southwestern Virginia counties. This article presents results of a study conducted from May to September 2000 in Wise County, VA, that examined the area-wide oviposition activity and habitat preferences of Ochlerotatus triseriatus and Aedes albopictus, potential vectors of La Crosse virus in the region. Data from 490 ovitrap collections made throughout the county showed that mean oviposition activity throughout the study was higher for Oc. triseriatus (2O.4 eggs/trap-day) than for Ae. albopictus (3.7 eggs/trap-day). The 2 species also had distinct habitat preferences for oviposition, with Oc. triseriatus favoring forested habitats and Ae. albopictus favoring urban/residential habitats. A landcover map of 6 habitat types derived from Landsat satellite imagery of the county showed that 63Vo of the county was forested and lSVo was urban/residential. A Bayesian decision-rule model that incorporated the ovitrap data and landcover map was moderately successful at predicting the occur- rence of high oviposition activity and abundance of the 2 species. The predictions reflected seasonal and spatial fluctuations in oviposition activity, with accuracies between 55 and79Vo for Oc. triseriatus and7O and94Va for Ae. albopictus. Kappa (1fl, a measure of the predictive power of the model, varied from poor (K < 0.4) to good (O.4 < K < 0.75) for both species, and was highest during periods when actual egg abundance was high. This suggests that the predictions were most accurate during periods when the risk for La Crosse virus transmission is greatest. Limitations and suggestions for improving the model are discussed.

KEY WORDS Aedes albopictus, Ochlerotatus triseriatus, La Crosse virus, oviposition, remote sensing, Bayes- ian modeling

INTRODUCTION out southwestern Virginia and that their abundance, as measured by oviposition, varies with habitat The mosquito species Ochlerotatus triseriatus type. For example, at a hu- (Say) is a known vector of La Crosse virus (LAC, man case site, Ae. albopictus was found to oviposit Bunyaviridae: California serogroup) (Sudia et al. preferentially in the yard surrounding the home, 1971; Watts et al. 1972, I973a), which causes La while Oc. triseriatus showed no preference, ovi- Crosse encephalitis, the most prevalent pediatric positing equally in the yard and adjacent forests mosquito-borne disease in the USA during most (Barker et al. 2003). In a nearby area of West Vir- years. Aedes albopictus (Skuse) also is a potential ginia endemic for LAC, Nasci et al. (2000) found accessory vector in the LAC cycle (Grimstad et a1. that the Oc. triseriatus population densities were 1989, Gerhardt et al. 2001). Recently, there has related to habitat type, with higher densities occur- been an increase in the number of cases of La ring in mixed northern hardwood habitats, contain- Crosse encephalitis in southwestern Virginia. Dur- ing primarily large maples than in sites with hem- ing the 5-year period from 1994 to 1998, a total of locks mixed with hardwoods or in an abandoned 13 cases were recorded in 4 counties in the Appa- orchard site containing a large number of small ma- lachian Mountains of Virginia (Fig. 1); I case was ple trees. While local case studies on mosquito-hab- recorded in the entire state during the 19 years prior itat associations will continue to provide useful in- to this period. In 1997 alone, 5 confirmed cases of formation for evaluating disease risk, modern La Crosse encephalitis were reported in Wise pest-management County, VA, constituting a rate of approximately ecologically based tenets suggest l3 cases/100,000. that it is equally important to recognize that the pests Case site studies suggest that populations of Oc. local occurrences of usually are related to presence geographic (Na- triseriatus and Ae. albopictus may occur through- their over a larger area tional Research Council 1996). Therefore, for in- sects such as mosquitoes that utilize human beings I An abbreviated description of this research was pub- and other vertebrates as hosts, there is a need also lished previously in the Proceedings and Papers of the to develop an understanding of the spatial interac- 70th Annual Conference of the Mosquito and Vector Con- tions among the , its breeding habitats, and trol Association of California (Barker et al.2OO2). populations geographic 2 Present address: Arbovirus Field Station, 4705 Allen host over broader areas. Road, Bakerslield, CA 93312. Because populations of Oc. triseriatus and Ae. 3 To whom correspondence and reprint requests should albopictus are associated with characteristic habi- be addressed. tats, geospatial approaches such as remote sensing, Dscevesn 2003 SpATrorEMpoRALOvposrrroN oF LAC VEcroRs 383

average temperatures in January and July are O and 22oC, respectively (National Oceanic and Atmo- spheric Administration 2OO3). For this study, areas within the county were cat- ' ":l-.' egorized into 6 broad landcover types or habitats | (forest, urban/residential, shrub and brush range- t, ta- l;la at ;.dtr.]-a land, herbaceous rangeland, barren land, and water) .. , I based on knowledge of the landscape and the land use and landcover scheme suggested by Anderson et al. (1976). This stratification was required for designing the ovitrapping survey and for classify- ing satellite imagery to create a landcover map.

Egg survey Mosquito eggs were collected throughout the county by ovitrapping during a 16-wk period from Virginia (shaded)re- Fig. l. Counties in southwestern May 29 to September 18, 2000 (Fig. l). Ovitraps porting casesof La Crosseencephalitis between 1994 and were made using 450-ml black plastic cups with 1998. Wise County (enlarged)showing the distribution of ovitrap collection sites. several drain holes punched halfway down the sides. The inside of each cup was lined with a 5- cm-wide strip of seed germination paper as an ovi- position substrate (Steinly et al. 1991). The cups geographic information systems, and spatial sur- were filled paftially with water before placement at veys and modeling can be used to study regional each sampling site. Ovitraps were distributed and landscapes and to identify preferred breeding hab- collected weekly at representative sites in each of itats for these species. Other studies already have the 6 landcover types at a variety of elevations shown that geospatial technologies can be used to (range for the study period: 440-1,271m). Ovitrap assess the relationships between environmental and sites generally were located within walking dis- habitat variables and disease vector abundance tance of roads or other automobile-accessible areas (e.g., Rogers and Randolph 1991, Wood et al. 1992, and were distributed throughout the county during Beck et al. 1994, Cross et al. 1996, Roberts et al. each sampling period. The geographic location of 1996, Baylis et al. 1998, Moncayo et al. 2000). each ovitrap site was recorded with a hand-held The objective of this study was to characterize GPS III Plus receiver (Garmin International, Inc., ovipositional habitat preferences of Oc. triseriatus Olathe, KS) so that trap and egg data could be de- and Ae. albopictus in Wise County, VA, where La picted spatially. During the first 4 wk of the survey, Crosse encephalitis has emerged recently. Data 22 new ovitraps were set out each week (21 during from geographically referenced ovitrap collections wk 1), with a single ovitrap placed at each sampling in various habitat types then were combined with site. These ovitraps were sampled repeatedly at 4- remotely sensed satellite data to develop a Bayesian wk intervals through the end of the l6-wk period decision-rule model to classify Wise County into of the study, with the exception of 6 sites from the probabilities of high oviposition activity for Oc. first week that were continuously sampled weekly triseriatus and Ae. albopictus. throughout the study period. In addition, 6 new sites were set out and collected each week after the MATERIALS AND METHODS first 4 wk in habitats where lower numbers of mos- quitoes were expected (shrub and brush rangeland, Study area herbaceous rangeland, and barren land) to obtain The study was conducted in Wise County, VA an accurate representation of oviposition in these (36"44' and 37'14'N, and 82"14' and 82'54'W), in habitats. A total of 16O sites in the 5 landcover the southern Appalachian Mountains (Fig. 1). Wise types, excluding water, were visited during the County is -1,O72 km2 and the landscape is domi- study, resulting in 490 ovitrap collections. Eggs nated by mixed hardwood forests, including oak- collected weekly from ovitraps were identified to hickory, maple-beech-birch, and white pine-hem- species (Linley 1989a, 1989b) and counted. lock forest types (Johnson 1992). Small towns, barren land cleared by coal mining, and occasional Statistical analysis pastureland are interspersed among the forest stands. Elevation within the county ranges from A generalized linear model analysis of vari- -400 to 1,287 m, with most populated areas locat- ance (GLM ANOVA) was used to test for differ- ed at elevations between 400 and 800 m. Average encjr_ 3rneng means of square root-transformed annual precipitation recorded at the Wise 3 E [V(y + 0. 1)] egg counts in relation to landcover weather station in Wise, VA, is ll9 cm, and the and week of sampling. Fisher's least significant dif- Jouwar- or rrrr, Avr,nlclIl Moseuno CoNrnol AssocI.q.nolr Vol. 19, No- 4 ference (LSD) multiple comparison test (ct : 0.05) P(H I M*s) is the conditional probability of finding was used for mean separation. All statistics were a specific habitat given high oviposition activity, calculated in NCSS Statistical Software 2000 (Hin- P(H I Mb) is the conditional probability of finding tze 1998). a specific habitat given low oviposition activity, and P(Mn"n I f4 ls tne posterior probability of high ovi- position activity for a particular habitat. Actual high Landcover map or low oviposition activity was defined by whether A landcover map of Wise County was developed the count for a particular trap was above or below from Landsat 7 Enhanced Thematic Mapper the seasonal mean number of eggs collected for the (ETM+) imagery (path 18: row 34), which was ac- individual species. As such, the prior probabilities quired by the satellite on July 3, 2000. The data of high and low oviposition activity for each spe- were classified into 6 predefined landcover types cies were calculated by dividing the number of sites (Anderson et al. 1976) using the TNTmips map and with egg numbers above and below the seasonal image processing system (Microimages, Inc., Lin- mean, respectively, by the total number of collec- coln. NE) and the classification method described tions during the period of interest. Conditional by Brewster et al. (1999). Six of the seven Landsat probabilities were calculated by dividing the total ETM+ spectral bands were used in the classifica- number of sites above and below the threshold in tion: blue, green, red, near-infrared, and 2 mid-in- each habitat by the total number of sites that were frared bands. The thermal infrared band was not above and below the threshold for all habitats. Prior used because of its coarser spatial resolution (-60 and conditional probabilities were calculated for m) compared with that of the other bands (28.5 m). each 28.5-m pixel in the landcover map to create Prior to image classification, the geographic coor- maps of these variables that were entered into equa- dinates of representative sites in each of the land- tion I to produce a map of the posterior probabil- cover types were recorded with the GPS receiver. ities for high mosquito oviposition activity for the Using landcover descriptions at these sites as a ref- entire county. erence, known areas were defined for each land- The accuracy of the predictions was tested by cover type, which then were used as a training da- creating a buffer zone of 200 m around each ovitrap taset for the supervised maximum likelihood location and comparing the average of posterior classification of the satellite image. The accuracy probabilities for high oviposition activity within the of the landcover map was evaluated primarily using zone with the actual oviposition at the location. The the error matrix generated by the classifiet which radius of 200 m was chosen because it encom- compared pixels of known landcover in the training passed the flight range for most Oc. triseriatus data with the pixels in landcover classes predicted (Mather and DeFoliart 1984) and Ae. albopictus by the classification (Congalton 1991). The result- (Bonnet and Worcester 1946, Hawley 1988, Nie- ing landcover map had a spatial resolution equal to bylski and Craig 1994). The numbers of occurrenc- that of the original satellite imagery (28.5 m). es of actual high and low oviposition activity, de- fined by the seasonal mean threshold described earlier, and the numbers ofoccurrences ofpredicted Bayesian classification high and low oviposition activity, defined by av- erage posterior probabilities >5OVo or

Table 1. Error matrix and accuracy measures used for assessing predictive capabilities of the Bayesian model. Error matrix' Measure2 Calculation3 Accuracy @+atN - + (b + + Kappa (iC) @ + A (@ + c)(a + b) AG d))ttD Actual N- (((d* c)(a+ b) + (b + Ak + A)/M High Low .Hiehab Prevalence a/(a+b+c+d) Preolcteo L; c d False-positive rate b/(b + o False-negative rate cl(a + c)

rAdapted from Fielding and Bell (1997). , Accuracy is the proportion of sites that are classified cotrectly as having high or low oviposition activity. K is a more comprehensive measure of prediction success. 3 N is the total number of evaluated sites. bers of Ae. albopict .t eggs collected were low sometimes was higher in other habitat types (Fig. during June (<1.0 eggltrap-day), increased during 3B). Aedes albopictus had approximately equal ovi- July" and peaked in late August (8.9 eggs/trap-day; position preference for barren land, shrub and brush Frg.2). rangeland, and herbaceous rangeland, with the low- Oviposition preferences for both species differed est activity occurring in forested habitats (Table 2). significantly ,rmong the 5 landcover types for the season (Oc. tiseriatus.' F : 54.38, df : collection Landcover classifi cation 4, P < 0.001; Ae. albopictus: F : 27.61, df : 4, P < 0.001; Fig. 3). Oviposition activity for Oc. Figure 4 shows the landcover map that was de- triseriatus was significantly higher in forested than veloped from the satellite imagery of Wise County. in urban/residential habitats. Activity in shrub and Ot 22,303 pixels in the training dataset that was brush rangeland, herbaceous rangeland, and barren used to develop the landcover map of Wise County, land did not differ significantly, but the numbers of 98.l%o were classified correctly, and Kappa (K, an eggs were significantly lower than in forested and indicator of prediction success above that expected urban/residential sites (Table 2; Fig. 3A). In con- by chance) was 0.95. Commission errors, the un- trast, the mean numbers of Ae. albopictus eggs certainties in the association of pixels on the land- were significantly higher in urban areas than in all cover map with classes on the ground, ranged from orher habitats ('fable 2), although relative oviposi- O.O to 9.7Eo. From the map, it was estimated that tion activity of Ae. albopictus among landcover -63.27o of the landscape in the county is forested, classes varied over time (F : 1.75, df : 6O, P : l8.OVo is urban/residential, lO.6Vo is shrub and 0.001; Fig. 38) and oviposition in particular weeks brush rangeland, 4.9Vo is herbaceous rangeland,

40 *Oe tisqisfu$ 35 +Ae albopiar e30 #r.

o 7zo 0!! o o E15 iJ E,o

5

| 2 3 4 5 6 7 8 9 l0 11 12 13 14 15 16

Collection Week Fig.2. Mean eggs per trap-day from all ovitrap sites for Ochlerotatustriseriatus and Aedesalbopictus. Eggs were collected weekly between May 29 and September18, 2000. 386 JoumnL0FTHE AMERTcAN M0sQUnoColrunoL Assocnnoll

60

€b{o

bJJ40 () 3o :O bo 0 820

t0

0 20

€15 I F

o *10 b0 bo o ()

10 ll t2 13 t4 l5 l6 Jun Jul Aug sep

Collection Week

Fig. 3. Mean eggs per trap-day for Ochlerotatus triseriatus (A) and Aedes albopictu.r (B) by landcover type in Wise County. Eggs were collected weekly betweenMay 29 and September 18, 2000.

3.1Vo is barren land, and O.2Vois water. Because of Bayesian classification the 28.5-m resolution of the Landsat ETM* im- agery, areas classified in the water landcover type Maps of the posterior probabilities of high ovi- represented large bodies of water such as lakes, position activity for both Oc. triseriatus and Ae. which were not considered suitable habitats for the albopictus were created for each of the eight 2-wk treehole and container-breeding mosquito species periods during the l6-wk collection season (e.g., and were not sampled in this study. Fig. 5). The maps reflected the fluctuations in ovi-

Table 2. Mean 1+58; eggs per trap-day for Ochlerotatus triseriatus and Aedes albopictus collected weekly in ovitraps placed in 5 landcover types in Wise County, VA, between May 29 and September 18, 2000.'

Landcover Oc. triseriatus Ae. albopictus Forest 200 35.51+ 2.36a l.O3+ O.2lc Urban/residential t47 t6.72 + 1.80b 6.84 + 0-8la Herbaceous rangeland 60 4.66 + l.6oc 4.35 + 0.78b Shrub and brush rangeland 56 4.32 + O.96c 4.23 + l.O'tb + Barren land -l -) 0.81 + 0.54c 4.34 1.32b

,Means within a column followed by the same letter are not significantly different (P > 0.05) when analyzed by GLM (ANOVA) and Fisher's LSD multiple comparison tests.

388 JourNx- or rnp AurrucAN MoseulTo CoNrnol AssocIATroN Vol. 19,No.4

May 29-June12 June2GJuly l0 SeptemberGSeptember 18 AccuracY= 6,714; l(:0.773; Accuracy= 9.790;K = 0.533; Accuracy= 0.712;(* 0.337; Prevalence= 0.306: Prevalence: 0.387; Prevalence= 0.364: FPR* 0.079;FNR = 0.417 FPR=0.143; FNR= 0.542

May 29-June12 July24-August 7 August2l-September 4 Accuracy= 0.939;K= 0.000; AccuracY= 6.760;K= 0.398; Accuracy= 0.750:K= 0.490; Prevalence= 0.061; Prcvahnce= 0.450; Prevalence= 0.411; FPR= 0.303:FNR = 0.296

Fig. 5. Selected maps of Wise County showing posterior probabilities of high oviposition activity from the Bayesian model for Ochlerotatus triseriatus (A) and Aedes albopiozs (B). Grayscale levels from black to white represent the range of probabilities (O-8OSo) of high oviposition activity. K = Kappa, FPR : false-positive rate, and FNR : false- negative rate.

human habitation, usually a house with a manicured Aside from habitat-related variation in oviposi- lawn within a residential Eueaor rural housing clus- tion intensity, another factor that influences ovipo- ter or a business district within a small town. Large sition activity in an individual trap is competition cities are not found in Wise County, so areas clas- from other oviposition containers. In an effort to sified as urban/residential are similar to suburban minimize this source of variation. sites with anom- areas near major cities where trees and other veg- alously high numbers of competing oviposition etation are interspersed among the houses and other sites, such as tire dumps, were avoided because of buildings. Nevertheless, the preference of Ae. al- the problem of competing oviposition containers bopictus for urban areas and the relatively low and because such sites generally were not represen- abundance found in forests agree well with the pro- tative of the broad landcover classes used in the posed origin of this species in the USA from used study. It was impossible to equalize the numbers of tire shipments from northem Asia (Hawley et al. competing oviposition sites in all habitat types, 1987). Studies of macrohabitat preferences of Ae. such as treeholes in forested areas and peridomestic albopictus in Japan showed that this mosquito is a containers in urban areas, and some variation in cornmon or regularly occuffing species in urban, oviposition intensity within and among habitat suburban, and rural areas but is rare in forest (Haw- types undoubtedly resulted from these differences. ley 1988). However, studies in other parts of Asia However, because the highest numbers of eggs for found that Ae. albopictu,r was common in forests both species were collected in habitat types that (Hawley 1988), which led Mogi (1982) to suggest presumably also had the highest numbers of com- that forest-dwelling populations of Ae. albopictus peting ovitrap sites (treeholes for Oc. triseriatus in some areas of Asia may have resulted from rea- and peridomestic containers for Ae. albopictus) and daptation of anthropophilic strains of the species to because lower levels of oviposition activity were the forest habitat. Thus, it is possible that, because observed in habitat types with the fewest competing of the predominance of forest in Wise County, Ae. oviposition sites (barren land, herbaceous range- albopictus populations eventually may adapt to use land, and shrub and brush rangeland), equalizing forested habitats more frequently and compete for the number of competing oviposition sites among resources more directly with native Oc. triseriatus. habitat types is likely to have the effect of increas- Decpunen2003 SpenorevrponeI- OvposITIoN or LAC VEcroRs 389 ing the observed differences between preferred and low oviposition activity in the next iteration of the alternative habitats without appreciably changing model when new information is acquired. Despite the relative preference for individual habitat types. these advantages, there were several factors that For the Bayesian modeling portion of the study, could account for the moderate success of the mod- it was assumed that oviposition activity at an ovi- el in this study. The ovitrap data were not extensive trap site was correlated positively with the abun- enough to allow partitioning for prediction and val- dance of female Oc. triseriatus and Ae. albopictus idation, and a suitable retrospective mosquito abun- at that site. Although adult mosquitoes were not dance dataset was not available for the study area. collected in this study, females of Oc. triseriatus Validation of model predictions based on data used and Ae, albopictus are known to remain close to to create a model generally is considered question- their larval sites, (Horsfall 1955) and because their able or even invalid, but in this study, the input flight range is low (Bonnet and Worcester 1946, probabilities for the model were calculated inde- Mather and DeFoliart 1984, Hawley 1988, Niebyl- pendently of the spatial arrangement of the data and ski and Craig 1994), activities such as host-seeking the assessment was based on the locations of indi- and resting would be expected to take place within vidual sites and their surrounding landcover. There- a short distance of sites used for oviposition. Fur- fore, although the evaluations of predictions may ther studies incorporating adult collections in the not be as robust as they would have been if they various habitat types along with ovitrapping would had been tested with an independent dataset, the provide greater insight into differential habitat us- current assessment still provides a reasonable de- age patterns. piction of the utility of the Bayesian model. The The Bayesian model was moderately successful moderate levels of accuracy achieved also suggest in predicting the occurrence of high oviposition ac- a need to consider additional ecologically relevant tivity for Oc. triseriatus and Ae. albopictus factors. Elevation was considered. but no relation- throughout Wise County. Kappa (IQ, which takes ship to mosquito oviposition during the study pe- into account both correct and incorrect predictions, riod was found, agreeing with a study conducted provides a measure of prediction success of the by Ballard et al. (1987) in woodlands of eastern model above that expected by chance (Fielding and Kentucky. Rainfall also was considered, but no sig- Bell 1997). K was highest for both species during nificant correlation between this variable and ovi- periods of peak oviposition activity, suggesting that position activity was found at varying lag periods the predictive power of the model was highest dur- (0-6 wk) to account for the generation time from ing periods when LAC transmission risk also was egg hatching to oviposition by female mosquitoes. likely to be high. Overall accuracy levels, particu- The lack of correlation might have resulted from larly for Ae. albopictus, were very high (>90Vo) the coarse spatial resolution of the rainfall data that during the early season, but accuracy levels are de- were available from weather stations scattered pendent on prevalence (proportion of sites that has throughout Wise County. Because rainfall can vary actual high oviposition levels). Thus, during the dramatically, even over short distances, a better es- early season, when oviposition rates are low, very timate of rainfall might have been obtained by plac- high accuracy levels can be achieved even with a ing a rain gauge at each ovitrapping site. Rainfall poor predictor if nearly all sites have low actual was relatively frequent during the collection period, and predicted oviposition activity (Fig. 5B). Rates with measurable rainfall recorded during 90 of 183 for false negatives were lowest during periods of days from April through September 2003 in Wise, peak abundance for both species, while rates for VA (National Oceanic and Atmospheric Adminis- false positives were highest during these periods. tration 2003), so it might be more closely correlated False-positive and false-negative rates are measures with mosquito abundance during years when it is of the proportion of actual low or high abundance more sporadic and serves to limit mosquito pro- sites, respectively, that were not predicted as such. duction. For prediction of disease-vector abundance, false- In conclusion, the study showed that Oc. triser- negative rates must be minimized because false iatus and Ae. albopictus in southwestern Virginia negatives result in neglect of areas with high dis- have a clear oviposition preference for different ease risk by public health and vector-control per- habitats. Spatially referenced ground surveys and sonnel. Decisions based on false-positive predic- predictive modeling using data on oviposition ac- tions may result in wasted resources, but not in tivity as an indicator of mosquito abundance could increased risk of human disease. provide timely predictions of vector abundance An advantage of using Bayesian models is that over broad areas. Because vector abundance is re- prior probabilities can be estimated subjectively or lated to risk for virus transmission to humans, par- based on user-defined thresholds (e.g., the thresh- ticularly for a vertically transmitted virus such as olds for high and low oviposition activity used in LAC (Watts et al. 1973b, Miller et al. 1977), pre- this study). These models also can be updated eas- dictions of vector abundance could be useful for ily based on new information. The posterior prob- targeting public awareness campaigns and vector- ability of high oviposition activity, for example, can control measures in areas with the greatest risk for be used to derive the prior probabilities of high and virus transmission to humans. These approaches, 3q0 JounNar-op rsr AulrrcaN Moseurro CoNTRoLAssocleuoN Vor-.19, No.4 along with vertebrateserosurveys and virus testing weather data and remote sensing to predict the geo- of mosquitoes,would provide a better understand- graphic and seasonal distribution of Phlebotomus pa- ing of LAC distribution within the county that patasiin southwesrAsia. Am J TropMed Hyg 54:530- would improve the assessmentof diseaserisk as- 536. sociatedwith predicted mosquito distributions. Fielding AH, Bell JF. 1997. A review of methods for the assessment of prediction errors in conservation pres- ence/absence models. Environ Cons 24:38-49. ACKNOWLEDGMENTS Gerhardt RR, Gottfried KL, Apperson CS, Davis BS, Er- win PC, Smith AB, Panella NA, Powell EE. Nasci RS. We sincerely thank Charles Apperson (Depart- 2001. The first isolation of La Crosse virus from natu- ment of Entomology, North Carolina State Univer- rally infected Aedes albopictus. Emerging Infect Dis 7: sity) for helpful comments on the manuscript. We 807-81 l. also would like to thank the Wise County Health Grimstad PR, Kobayashi JEZtllang M, Craig GB Jr. 1989. Department for their invaluable assistance. Recently introduced Aedes albopiclus in the United States: potential vector of La Crosse virus (Bunyaviri- dae: California serogroup). J Am Mosq Control Assoc REFERENCES CITED 5:422-427. Hawley WA. 1988. The biology of Aedes albopictus. J Anderson JR, Hardy EE, Roach JT Witmer RE. 1976. A Am Mosq Control Assoc 4(Suppl l):l-40. land use and land cover classification for use with re- Hawley WA, Reiter B Copeland RS, Pumpuni CB, Craig mote sensor data. U.S. Geological Survey Professional GB Jr. 1987. Aedes albopictus in North America: prob- Paper 964. Washington, DC: U.S. Govemment Printing able introduction in used tires from Office. northern Asia. Sc! ence 236:1114-ll16. Aspinall R, Veitch N. 1993. Habitat mapping from satel- Hepinstall JA, Sader SA. 1997. Using Bayesian statistics, lite imagery and wildlife survey data using a Bayesian thematic mapper satellite imagery, and breeding bird modeling procedure in a GIS. Photogrammetric Engi- survey data to model bird species probability neering and Remote Sensing 59:537-543. of occur- rence in Maine. Photogrammetric Aspinall RL 1991. Use of an inductive modeling proce- Engineering and Re- mote Sensing 63:1231-1237. dure based on Bayes theorem for analysis of pattem in Hintze JL. 1998. NCSS statistical sofhuare 200O. Kays- spatial data. In: Farmer DG, Rycroft MJ, eds. Computer ville. UT: NCSS. modeling in the environmental sciences. Institute of Horsfall WR. 1955. Mosquitoes, their bionomics and re- Mathematics and its Applications Conference Series lation to disease New York: Hafner Publishing No. 28. Oxford, United Kingdom: Oxford University Com- pany. Press. p 325-339. Johnson TG. 1992. Forest statistics Ballard EM, Waller JH, Knapp FW 1987. Occurrence and for Virginia, 1992 Forest ovitrap site preference of tree hole mosquitoes: Aedes USDA Service, Southeastern Forest Experiment triseriatus and Aedes hendersoni in eastern Kentucky. Station Resource Bulletin SE-131. J Am Mosq Control Assoc 3:42-44. Landis JR, Koch GC. 7977 . The measurement of observer Barker CM, Brewster CC, Paulson SL. 2OO2. Geospatial agreement for categorical daLa.Biometrics 33:-159-174. and statistical modeling of mosquito distribution in an Linley JR. 1989a. Scanning electron microscopy of the (Protomacleaya) emerging focus of La Crosse virus. Proc Mosq Vector egg of Aedes tiseriatus (Diptera: Cu- Control Assoc Calif 7O:103-107. licidae). J Med Entomol 26:474-478. Barker CM, Paulson SL, Cantrell S, Davis BS. 2003. Hab- Linley JR. 1989b. Comparative fine structure of the eggs itat preferences and phenology of Ochlerotatus triser- of Aedes albopictus, Ae. aegypti, and Ae. bahamensts iatus and Aedes albopioas (Diptera: Culicidae) in (Diptera: Culicidae). J Med Entomol 26:5lO-521. southwestern Virginia. J Med Entomol (in press). Mather TN, DeFoliart GR. 1984. Dispersion of gravid Ae- Baylis M, Bouayoune H, Touti J, El Hasnaoui H. 1998. des triseriatus (Diptera: Culicidae) from woodlands into Use of climatic data and satellite imagery to model the open terrain. J Med Entomol 2l:384-391. abundance of Culicoides imicoLa, the vector of African Miller BR, DeFoliart GR. Yuill TN{I. 1977. Vertical trans- horse sickness virus, in Morocco. Med Vet Entomol 12.. mission of La Crosse virus (California encephalitis 255-266. group): transovarial and filial infection rates in Aedes Beck LR, Rodriguez MH, Dister SW, Rodriguez AD, Re- triseriatus (Diptera: Culicidae). J Med Entomol 14:437- jmankova E, Ulloa A, Meza RA, Roberts DR, Paris JE 440. Spanner MA, Washino RK, Hacker C, Legters LJ.1994. Mogi M. 1982. Variation in oviposition, hatch rate and Remote sensing as a landscape epidemiologic tool to setal morphology in laboratory strains of Aedes albop- identify villages at high risk for malaria transmission. ictus. Mosq News 42:196-201. Am J Trop Med Hyg 5l:271-28O. Moncayo AC, Edman JD, Finn JT. 2000. Application of Bonnet DD, Worcester DJ. 1946. The dispersal of Aedes geographic information technology in determining risk albopictus in the territory of Hawaii. Am J Trop Med of eastern equine encephalomyelitis virus transmission. 26:465-476. J Am Mosq Control Assoc 76:28-35. Brewster CC, Allen JC, Kopp DD. 1999. IPM from space: Nasci RS, Moore CG, Biggerstaff BJ, Panella NA, Liu using satellite imagery to construct regional crop maps HQ, Karabatsos N, Davis BS, Brannon ES. 2000. La for studying cropinsect interactions. Am Entomol 45: Crosse encephalitis virus habitat associations in Nich- 105-117. olas County, West Virginia. J Med Entomol 37:559- Congalton R. 1991. A review of assessing the accuracy 570. of classification of remotely sensed data. Remote Sens- National Oceanic and Atmospheric Administration. 2O03. ing Environ 37:35-46. National Climatic Data Center. http://www.ncdc.noaa. Cross ER, Newcomb WW Tucker CI. 1996. Use of gov. Last accessed: March 16,2003. Dscnrasnn2003 Splrrorsr*4poRAlOvposnoN oF LAC Vscrons 391

RA, Martin EB' Blaik- National Research Council. 1996. Ecologically based pest Tucker K, Rushton SB Sanderson bird distributions-a combined management: new solutions for a new century Wash- lock J. 1997. Modelling I'