environmental factors and risk areas of west nile virus in...

12
Environmental Factors and Risk Areas of West Nile Virus in Southern California, 20072009 Hua Liu & Qihao Weng Received: 2 February 2011 /Accepted: 5 December 2011 /Published online: 29 December 2011 # Springer Science+Business Media B.V. 2011 Abstract The West Nile virus (WNV) may post a significant health risk for mammals, including humans and insects. This study examines the spatialtemporal effects of environmental factors on WNV dissemination with a case study of ten counties in the southern California, where the epidemic was recently most prevalent within the USA. WNV surveillance data were obtained from the California Vectorborne Disease Surveillance System and Centers for Disease Control and Prevention. Remote sensing and Geographic Information Sys- tems (GIS) techniques were combined to derive environmen- tal variables. Principal component analysis was performed to select the most relevant environmental variables. Two ecolog- ical zones were identified based on the selected variables. Identification of risk areas for WNV was limited to a zone with 95% mosquitoes surveillance records. Three time windows, the epidemiological weeks 1826, 2735, and 3644 in each year of 20072009, were examined in details with risk area mapping. It is found that the southern part of San Joaquin Valley in Kern County and Los Angeles County (especially its southern part) were the most vulnerable locations for WNV outbreak. Main factors contributing to the WNV propagation included summer mean temperature, annual mean deviation from the mean temperature, land surface temperature, elevation, land- scape complexity, landscape diversity, and vegetation water content. The result of this study improves understanding of WNV ecology and provides tools for detecting, tracking, and predicting the epidemic. The holistic approach developed for this study, which integrated remotely sensed, GIS-based, and in situ-measured environmental data with landscape metrics, may be applied to studies of other vector-borne diseases. Keywords West Nile virus . Environmental factors . Risk areas mapping . Urban remote sensing . Spatial modeling . California 1 Introduction West Nile virus (WNV) is a mosquito-borne virus with the first human case discovered in Uganda in 1937. Its natural transmission cycle is birdmosquitobird, with humans or other mammals and reptiles becoming incidental hosts when bitten by infected mosquitoes. It first appeared in North America in the early summer of 1999 in New York City and then spread prolifically to the rest of the continental USA in a period of 5 years. The west coast and central states have become the most prevalent regions for WNV epidem- ics. Its dissemination varies seasonally in response to the characteristics of environmental and socioeconomic condi- tions and climate [2, 32, 41, 42]. Nonhuman WNV surveil- lance can be used as an early warning for WNV activity [10, 35]. The abrupt appearance of WNV in North America indi- cates arboviruses may continue to emerge in new regions [20]. Numerous studies have been made on the transmission ecology and epidemiology of WNV across the USA and in other countries, e.g., California, Indiana, and Texas [27, 32, 39, 44, 54] and Ferlo, Senegal [9], and Var and Camargue areas in France [38]. In the continental USA, Reisen et al. [40] testified the importance of mosquito and anthropogenic factors in WNV dissemination in California by using ento- mological methods. Bertolotti et al. [2] conducted a WNV H. Liu (*) Department of Political Science and Geography, Old Dominion University, Norfolk, VA 23529, USA e-mail: [email protected] Q. Weng Center for Urban and Environmental Change, Department of Earth & Environmental Systems, Indiana State University, Terre Haute, IN 47809, USA e-mail: [email protected] Environ Model Assess (2012) 17:441452 DOI 10.1007/s10666-011-9304-0

Upload: others

Post on 16-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

Environmental Factors and Risk Areas of West Nile Virusin Southern California, 2007–2009

Hua Liu & Qihao Weng

Received: 2 February 2011 /Accepted: 5 December 2011 /Published online: 29 December 2011# Springer Science+Business Media B.V. 2011

Abstract The West Nile virus (WNV) may post a significanthealth risk for mammals, including humans and insects. Thisstudy examines the spatial–temporal effects of environmentalfactors on WNV dissemination with a case study of tencounties in the southern California, where the epidemic wasrecently most prevalent within the USA. WNV surveillancedata were obtained from the California Vectorborne DiseaseSurveillance System and Centers for Disease Control andPrevention. Remote sensing and Geographic Information Sys-tems (GIS) techniques were combined to derive environmen-tal variables. Principal component analysis was performed toselect the most relevant environmental variables. Two ecolog-ical zones were identified based on the selected variables.Identification of risk areas for WNV was limited to a zonewith 95% mosquitoes surveillance records. Three timewindows, the epidemiological weeks 18–26, 27–35, and36–44 in each year of 2007–2009, were examined indetails with risk area mapping. It is found that thesouthern part of San Joaquin Valley in Kern Countyand Los Angeles County (especially its southern part)were the most vulnerable locations for WNV outbreak.Main factors contributing to the WNV propagation includedsummer mean temperature, annual mean deviation from themean temperature, land surface temperature, elevation, land-scape complexity, landscape diversity, and vegetation watercontent. The result of this study improves understanding of

WNVecology and provides tools for detecting, tracking, andpredicting the epidemic. The holistic approach developed forthis study, which integrated remotely sensed, GIS-based, andin situ-measured environmental data with landscape metrics,may be applied to studies of other vector-borne diseases.

Keywords West Nile virus . Environmental factors . Riskareas mapping . Urban remote sensing . Spatial modeling .

California

1 Introduction

West Nile virus (WNV) is a mosquito-borne virus with thefirst human case discovered in Uganda in 1937. Its naturaltransmission cycle is bird–mosquito–bird, with humans orother mammals and reptiles becoming incidental hosts whenbitten by infected mosquitoes. It first appeared in NorthAmerica in the early summer of 1999 in New York Cityand then spread prolifically to the rest of the continentalUSA in a period of 5 years. The west coast and central stateshave become the most prevalent regions for WNV epidem-ics. Its dissemination varies seasonally in response to thecharacteristics of environmental and socioeconomic condi-tions and climate [2, 32, 41, 42]. Nonhuman WNV surveil-lance can be used as an early warning for WNVactivity [10,35]. The abrupt appearance of WNV in North America indi-cates arboviruses may continue to emerge in new regions [20].

Numerous studies have been made on the transmissionecology and epidemiology of WNV across the USA and inother countries, e.g., California, Indiana, and Texas [27, 32,39, 44, 54] and Ferlo, Senegal [9], and Var and Camargueareas in France [38]. In the continental USA, Reisen et al.[40] testified the importance of mosquito and anthropogenicfactors in WNV dissemination in California by using ento-mological methods. Bertolotti et al. [2] conducted a WNV

H. Liu (*)Department of Political Science and Geography,Old Dominion University,Norfolk, VA 23529, USAe-mail: [email protected]

Q. WengCenter for Urban and Environmental Change, Department of Earth& Environmental Systems, Indiana State University,Terre Haute, IN 47809, USAe-mail: [email protected]

Environ Model Assess (2012) 17:441–452DOI 10.1007/s10666-011-9304-0

Page 2: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

study in suburban Chicago, IL with results showing thatbirds in urban-vegetated areas and vegetative landscapingpossessed a higher probability of virus propagation than thosefound in less vegetated areas. Time series data of WNVinfection are frequently used for studies in USA. Liu et al.[32] examined the WNV dissemination in Indianapolis,Indiana based on positive records of mosquito poolsduring the years of 2002–2007 by using remote sensing,geographic information system (GIS), ecological, andstatistical modeling. Winters et al. [53] employed regressionanalysis to develop spatial models for predicting high-riskareas of exposure to WNV in western and eastern Coloradobased on the human WNV cases from 2002 to 2006. Studies,at various scales, include the consideration of static anddynamic variables on environmental and socioeconomicconditions in relation to WNV epidemics in mammals,including humans and insects [3, 10, 29, 46]. Greennesslevel, distance to a WNV-positive dead bird species,geological factors, and socioeconomic conditions (e.g.,age of population and age of housing) are shown to bethe risk factors for human WNV infections in the greaterChicago area [46]. Road density, stream density, slope, andvegetation are closely related to avian WNVepidemics in thestate of Mississippi [10]. In addition, a simulation model hasbeen developed to describe the population dynamics ofmosquitoes and American crows, two major factors of WNVdissemination in North America [3].

Although previous studies have linked the WNV propa-gation with physical environment factors, these studies havenot employed a holistic approach by integrating remotesensing and GIS variables with high temporal resolution.The effect of landscape complexity, as indicated by land-scape metrics, has not been examined in detail. This studyinvestigated the effects of terrain and landscape patterncharacteristics derived from remote sensing images, histor-ical weather observation, and real-time remotely sensed landsurface temperature and vegetation water content on WNVinvasion in ten southern California counties. Risk areas werenext identified based on the aforementioned environmentaland weather factors for three time periods: late spring/earlysummer, summer, and late summer/early fall. The signifi-cance of the study is that it may produce a more accurate andreliable disease surveillance method by incorporating remotelysensed environmental conditions and landscape metrics aspotential environmental factors of WNVoutbreak.

2 Materials and Methods

2.1 Study Area

The following ten counties in California constitute the studyarea: San Luis Obispo, Kern, San Bernardino, Santa Barbara,

Ventura, Los Angeles, Orange, Riverside, San Diego, andImperial. All the islands (e.g., Santa Catalina Island, SantaClemente Island, and Santa Barbara Island) belonged to thesecounties were excluded from the study because few incidentswere reported from these islands. Figure 1 illustrates thegeographical location of the study area. The total land areaof these ten counties is about 56,500 square miles and isinhabited by over 22 million people based on US CensusEstimation 2006. The temperature difference between imme-diate coastal counties and inland counties in the studyarea is about 4°F (2°C) in winter and approximately 23°F (13°C) in summer (National Weather Service). The south-west monsoon increases the humidity in arid areas of southernCalifornia in the summer in general. There is a mixture ofheavily developed urban settings and undeveloped arid lands.

California has a long history of mosquito-borne diseaseand has become one of the most affected states for WNVepidemics in the USA, since the virus was initially detectedin a mosquito pool near Imperial County in summer of2003. The moderate Mediterranean climate provides afavorable habitat for mosquitoes and thus contributes to thedissemination ofWNV in the state [19, 43]. A great number ofWNV infections have been detected in birds, mosquitoes, andsome other mammals in the state according to the statisticspublished by the Centers for Disease Control and Prevention.The severe status of the WNV epidemic in California,especially in the study area, makes this study critical.California started to develop mosquito control programsin the early 1900s to defend mosquito-borne diseases, e.g.,arboviruses and malaria. Its strong surveillance program pro-vides enriched data for this study.

2.2 Data Collection and Processing

The geographical dissemination of WNV is related to theoccurrence of both avian reservoir host and mosquito vector,which is potentially associated with the environmental andsocial–economic conditions that can affect the abundancesof related birds and mosquito species [4, 18, 36, 55]. Wefirst collected necessary WNV surveillance data and nextcreated environmental variables using remote sensing andGIS technologies to assess the high-risk areas of WNVinfection combined with statistical analysis. In order tocontrol the error of localization of WNV incidence and tointegrate environmental variables with various spatial resolu-tions, we constructed a hexagonal network covering all tencounties (total 42,560 hexagons). The area of each hexagon(2 km side length) is close to the average size of census blockgroups in those counties, forming more neutral landscapeunits in comparison to the inconsistent shapes and unevenareas of block groups. Both WNV surveillance data andenvironmental variables discussed below were integrated intoindividual hexagonal geographic units for further analysis.

442 H. Liu, Q. Weng

Page 3: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

2.2.1 WNV Surveillance Data

WNV surveillance data for the whole state of Californiawere collected from the Arbovirus Bulletins published inthe California Vectorborne Disease Surveillance System(CALSURV). The records include the species of testedpositive mosquitoes, the number of mosquitoes in each pool,collected dates, surveillance site ID, and the method of virusdetection for years 2004–2009. We collected the geographicalcoordinates for each surveillance site in ten counties from theinteractive surveillance map published in CALSURV. Yearlyhuman WNV infection data were provided by Centersfor Disease Control and Prevention. As a result, everytested positive mosquito pool was linked to a specifichexagon using the geographical coordinates of surveil-lance sites. We focused our study of risk areas for years2007–2009 due to the lack of data availability, althoughthe WNV infection (in humans) has been reported fromCalifornia since 2002. More specifically, we grouped themosquito surveillance data into three periods, epidemio-logical weeks 18–26 (mainly in May and June), 27–35(mainly in July and August), 36–44 (mainly in September and

October) for each year of 2007–2009 in the identification ofWNV-risk areas.

2.2.2 Climate Factors

Climate is an important driver in the reproduction andsurvival of mosquitoes [16, 52]. Higher temperatures con-tribute to the faster distribution and wider spread of WNVthroughout North America [30, 43, 45, 50]. Because mos-quitoes thrive in hot, humid environments [12], this exacer-bation in temperature demonstrates the potential role ofurban environment in WNV dissemination. Precipitationshows close relationship to WNV incidence in the regionalthough its role seems to be inconsistent among studies [15,28, 50]. We obtained the historical temperature and precip-itation factors (monthly mean temperature, departure fromnormal temperature, total monthly precipitation, and depar-ture from normal monthly precipitation) from NationalOceanic and Atmospheric Administration (NOAA) NationalClimatic Data Center. The timeframe of weather records wasfrom years 2004–2009 since WNV infections became signif-icant starting in 2004. We calculated the mean values of

Fig. 1 Geographical location of the study area with two identifiedecological zones. Islands were not included in the analysis sincelimited WNV surveillance data were reported from those areas. Sur-veillance sites (islands excluded) are displayed at three levels, having

WNV in all years of 2007–2009 (red), having WNV in either years of2007–2009 (yellow), and having WNV infection in none years of2007–2009 (black)

Environmental Factors and Risk Areas of West Nile Virus 443

Page 4: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

temperature and precipitation in years 2004–2009 before com-puting their means for hexagons, the units of study. Theinverse distance weighting spatial interpolation method waschosen to create four weather maps (30 m spatial resolution),each corresponding to one of the temperature and precipitationvariables. In order to control the error caused by edge effectsin the spatial interpolation, we selected a total of 135 weatherstations within ten counties and their surrounding areas inthree states (California, Arizona, and Nevada). To identifythe possible influence of land surface temperature on WNVdissemination in three studied periods (weeks 18–26, 27–35,and 36–44) in years 2007–2009, we collected TerraModerate-Resolution Imaging Spectroradiometer (MODIS) Land SurfaceTemperature and Emissivity 8-day Level 3 Global images(1 km spatial resolution) acquired between April andNovember in years 2007–2009. The mean land surface tem-perature was calculated for each period in assessing theWNV-risk areas, and the average land surface temperature per hexa-gon (study units of risk analysis) was computed accordingly.

2.2.3 Terrain Conditions

Elevation and slope are found to be related to the dissem-inations of WNV in different regions [36, 53]. Winters’sstudy was done in Colorado where there are real differencesin climate due to great differences in elevation within thestudy area. We used US Geological Survey Digital Eleva-tion Model (DEM) data (30 m resolution) to create a rasterdataset of slope. The average elevation and slope were thenderived from the DEM and slope maps for each hexagonalunit.

2.2.4 Landscape Pattern

Landscape patterns, such as landscape complexity andaggregation, were reported to be strongly associatedwith the distributions of virus [30, 55]. The presenceof grass and other vegetation helps to retain moisture inthe neighborhood which provides a favorable habitat formosquito survival. Forest and other green spaces enhancelocal bird abundance and species diversity [6, 14], which inturn can contribute to the spread of WNV. Higher landscapediversity is associated with multiple land cover types such asurban, grass, and water. The ten studied counties, especiallythe areas along the coast, are highly urbanized with diversehabitats which can contribute to bird abundance [48]. In thisstudy, land cover data in remotely sensed datum format foryear 2001 were collected from the Multi-Resolution LandCharacteristics Consortium (MRLC). Those land coverimages were created by using the unsupervised image classi-fication method on the Landsat TM imagery [51]. The initialland cover images were encoded to only include seven differ-ent land cover types: open water, developed open space (e.g.,

parks) and low- to medium-intensity residential areas, high-intensity developed areas (e.g., commercial places), forest andscrub, grassland/pasture, wetlands, and barren lands.

We used Patch Analyst 4, an extension to the ArcGISapplication that facilitates the assessment of landscape patternto derive more than 15 relevant landscape metrics from theencoded land cover image. Due to possible correlation amongthese metrics, we performed bivariate Pearson correlationanalysis and principal component analysis and selected twolandscape metrics, Shannon’s diversity index (SDI) and meanshape index to represent the whole landscape pattern. The SDImeasures the landscape diversity introduced by Shannon [49]:

SDI ¼Xmi¼1

Pi lnPið Þ ð1Þ

where Pi is the proportion of patch type (class) i in thelandscape, and m is the number of patch types (classes) [34].

MSI ¼P

mi¼1

Pnj¼1 0:25Pij=

ffiffiffiffiffiAij

p� �

Nð2Þ

where Pij is the perimeter of patch ij, Aij is the area of patch ij,m is the number of patch types (classes), n is the number ofpatches per class, and N is the total number of patches in thelandscape [34]. Themean percentage of canopywas calculatedfor each hexagon based on the MRLC canopy product 2001[23].

2.2.5 Vegetation Water Content

The normalized difference water index is a metric measuringvegetation water content, derived from the near-infrared(NIR) and short wave infrared (SWIR) channels of theremote sensing satellite imagery [17]. The possible varia-tions caused by leaf internal structure and leaf dry mattercontent can be eliminated by combining the NIR with theSWIR reflectance, which better measures the vegetationwater content [7].

NDWI ¼ PNIR � PSWIR

PNIR þ PSWIRð3Þ

where PNIR is the reflectance in the NIR region and PSWIR isthe reflectance in the SWIR region.

Multiple studies have applied MODIS normalized differ-ence water index data to investigate vegetation water con-tent (e.g., [11, 21, 24]). In order to investigate the possibleinfluence of vegetation humidity to mosquito vector com-petence, we calculated normalized difference water indexbased on Terra MODIS Surface Reflectance 8-Day Level 3images (500 m spatial resolution) acquired in the sameperiod of time as that of MODIS land surface temperaturedatasets. In the calculation of normalized difference waterindex, band 2 was selected for use as the near-infrared

444 H. Liu, Q. Weng

Page 5: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

channel and band 6 was chosen as the short wave infraredchannel. As what we did for MODIS land surface tempera-ture images, the mean normalized difference water indexwas calculated for each studied period in years 2007–2009and the average normalized difference water index perhexagon as well.

As a result, we carefully chose a group of environmentalfactors as expositive variables for WNV dissemination:annual average temperature, summer (May–August) aver-age temperature, annual average departure from normaltemperature, summer (May–August) average departure fromnormal temperature, annual total precipitation, summer(May–August) total precipitation, annual average departurefrom normal precipitation, summer (May–August) averagedeparture from normal precipitation, land surface tempera-ture, elevation, slope, Shannon’s diversity index, meanshape index, and normalized difference water index.

2.3 Statistical Analysis

A principal component analysis (Quartimax rotation meth-od) was conducted to reduce the number of environmentalvariables addressed in the last section. Principal componentanalysis is a statistical analysis that selects a group ofindependent variables based on a set of original variables[25]. Since ten studied counties contains diverse landscapeand terrain conditions, it is necessary to generate homoge-neous ecological zones for those counties [47]. A K-meanscluster analysis was performed to group the hexagons withsimilar environmental conditions together by using the var-iables selected from principal components. Cluster analysisis a processing of grouping the objects that possess similarcharacteristics into the same cluster [1]. Each group indi-cates a distinguished ecological zone. The initial surveil-lance data were transformed into a series of Euclideandistance maps (30 m spatial resolution) corresponding tothree periods, epidemiological weeks 18–26, 27–35, and36–44 for individual years of 2007–2009. The correlationsbetween the Euclidean distance and environmental variablesselected by principal component analysis, land surface tem-perature, Shannon’s diversity index, mean shape index, andnormalized difference water index were examined using thebivariate Pearson correlation analysis for different ecologi-cal zones. Factors at 0.01 significance level were used tocalculate Mahalanobis distance in order to assess the riskarea of WNV.

Mahalanobis distance is a distance index based on corre-lation between variables by which different conditions canbe analyzed [33]. It has been used to examine the similarityof one environmental/habitat setting to another one (e.g., 26and 32). We used Mahalanobis distance to identify habitatsassociated with the dissemination of WNV in mosquitoesfor epidemiological weeks 18–26, 27–35, and 36–44 in

2007–2009. We assumed the hexagons containing testedpositive mosquito infection as one type of landscape. Thosewithout incident were expected to have different character-istics of landscape. Since some hexagons contained morethan one tested positive mosquito pools, we substituted aHorvitz–Thompson weighted [22] mean and covariancematrix in the calculation of Mahalanobis distance with hex-agons as study units. The formula can be defined as:

DðxÞ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiðx� μÞTS�1ðx� μÞ

qð4Þ

where D(x) is Mahalanobis distance, x is the vector ofenvironmental variables for each hexagon, μ is the vectorof Horvitz–Thompson weighted mean values of environ-mental factors for the hexagons with tested positive mos-quito infections, S−1 is the inverse Horvitz–Thompsonweighted covariance matrix of environmental factors forthe hexagons with tested positive mosquito infections, andT indicates a transposed matrix. Small Mahalanobis distancevalues suggest more favorable habitats for WNV-infectedmosquitoes, while greater values associate with less suitablehabitats. Assuming that Mahalanobis distances follow thechi-square (χ2) distribution, we calculated the p value foreach Mahalanobis distance and considered hexagons with pvalue between 0.9 and 1.0 as high-risk areas and ones with pvalue between 0.6 and 0.9 as moderate risk.

3 Results

3.1 Examination of WNV Dissemination

The investigation of WNV surveillance data helped tounderstand the general pattern of virus dissemination.Figure 2 shows the total tested positive mosquito andhuman surveillance records in ten counties in years2004–2009. The figure reveals a repeating pattern forevery year: the mosquito infection started to expand inMay (mainly corresponding to epidemiological weeks18–22) with a limited number reported in earliermonths. The amplification reached a peak in early ormiddle August (e.g., week 33 in 2005 and week 32 in2008) except that year 2004 had a crest in week 21(late May) and year 2006 peaked in week 37 (earlySeptember). The infection generally ended in late Novemberor December. Years 2004, 2005, and 2008 received the great-est total number of positive mosquito pools, while year 2006received the fewest mosquito infections. The conformableincrease of WNV infection in mosquitoes in late summeracross the years 2004–2009 may associate with post-nestingmigrations of summer and year-round resident birds [43]. Thisincrease may contribute to the outbreak of the virus in humansin the following weeks. The summer outbreak of WNV in

Environmental Factors and Risk Areas of West Nile Virus 445

Page 6: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

2004 may be linked closely to positive departure from normaltemperature [42]. Overwintering of WNV during the winters(e.g., 2004–2005) might contribute to the subsequent ampli-fication of virus in the following year 2006 [41]. The contin-uous amplifications of WNV in years 2007–2009 may beassociated with the combining effects of drought conditionsin the previous year and a departure from normal temperaturesin the early season.

Human andmosquito infection incidence followed a similarpattern throughout the study years; human infections peaked inAugust or September in years 2004–2009, e.g., week 31 foryear 2005 and week 37 in 2008 and mainly ended in

November. Human incidence followed a nearly identical trendas that of mosquito incidence: the peak years were 2004, 2005,and 2008, and year 2009 recorded the smallest number ofhuman cases. Figure 3 presents the yearly surveillance dataof mosquitoes and humans per county in years 2007–2009.Limited by data availability and completeness, years 2003–2006 were excluded in the figure. According to Fig. 3, twocounties, Kern and Los Angeles, received the maximumnumbers of WNV infections in both mosquitoes and human,followed by Orange, Riverside, and the rest counties with SanLuis Obispo, Santa Barbara, and Ventura possessing theminimum numbers (0–1) of infections.

Fig. 2 Total tested positive mosquito and human surveillance records in ten counties in years 2004–2009. X-axis associates with epidemiologicalweeks, Y-axis is the raw number of tested positive mosquito pools, and Z-axis is the raw number of human incidence

446 H. Liu, Q. Weng

Page 7: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

3.2 Environmental Factor Analysis

Ten components were generated as the result of principalcomponent (PC) Analysis, among which the first threerepresenting about 71% of the total variance (Table 1).The PC1 included the monthly temperature and annualprecipitation variables explaining 38% variance. It cantherefore be defined as a weather factor. The PC2 (20%variance) possessed the departures from normal temperaturesand precipitations, which seems to reflect an abnormal climatecondition. The PC3 summarized 13% variance with twovariables, elevation and slope, relating to the terraincharacteristics. Three representative variables with thehighest loading values, average summer (May–August)mean temperature for years 2004–2009, average annualmean departure from normal temperature for years 2004–

2009, and elevation were selected as input of K-means clusteranalysis.

Two clusters were generated by using K-means clusteranalysis with hexagons as units (Fig. 1), each indicating adistinguishable ecological zone. Ecological zone 1 (in darkgreen) contained 24,766 hexagons and 17,794 in zone 2 (ingray yellow). Table 2 records the environmental statistics fortwo zones. Zone 1 falls in coastal Central Coast and SouthCoast bioregions (total 34% area percentage), ColoradoDesert (28%), central and southeastern Mojave (26%), south-ern San Joaquin valley (11%), and southern Sierra (2%),whereas zone 2 comprised southernMojave (61%), inner areasof Central Coast and South Coast (total 26%), southern Sierra(7%), eastern Colorado Desert (6%), and southern border ofSan Joaquin Valley (3%). Zone 1 possessed less departurefrom normal temperature, lower summer temperature based

Fig. 3 Yearly surveillance dataof mosquitoes and human percounty in years 2007–2009.X-axis represents individualcounties and Y-axis shows theraw numbers of tested positivemosquito pools (left) andhuman incidents (right)

Table 1 Rotated component matrix as results of the principal component analysis (rotation method: Quartimax with Kaiser normalization)

Variables Loadings

PC1 PC2 PC3

Average summer (May–August) mean temperature for years 2004–2009 0.906 0.279 −0.042

Annual mean temperature for years 2004–2009 0.835 0.221 −0.267

Average annual mean monthly precipitation for years 2004–2009 −0.817 −0.101 0.272

Average annual mean departure from normal monthly precipitation for years 2004–2009 0.741 −0.308 0.018

Average summer (May–August) mean monthly precipitation for years 2004–2009 0.509 0.094 0.431

Average annual mean departure from normal temperature for years 2004–2009 0.239 0.892 −0.106

Average summer (May–August) mean departure from normal temperature for years 2004–2009 0.315 0.847 0.014

Average summer (May–August) mean departure from normal monthly precipitation for years 2004–2009 0.209 −0.627 −0.371

Elevation −0.119 0.141 0.880

Slope −0.273 −0.081 0.734

Mean values of all the listed variables in each hexagonal unit were used in the statistical analysis. Variables in PC1 were shown in bold, italics forPC2, and bold italics for PC3

Environmental Factors and Risk Areas of West Nile Virus 447

Page 8: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

on the climate records in 2004–2009, and lower elevationcompared to those of zone 2. Since 95% collected surveillancesites in ten counties fell inside zone 1, we conducted the riskanalysis mainly for zone 1. Some environmental factors dem-onstrated a closer relationship to WNV propagation thanothers, which include summer (May–August) mean tempera-ture, annual mean departure from normal temperature, landsurface temperature, elevation, landscape complexity, land-scape diversity, and vegetation water content.

3.3 Mapping of Risk Areas

According to the results of correlation analysis betweenselected environmental variables (three selected by principalcomponent analysis in Section 3.2, land surface tempera-ture, normalized difference water index, mean shape index,and Shannon’s diversity index) and Euclidean distance totested positive mosquito pools, all those environmental fac-tors were consistently and significantly related to WNVpropagation in mosquitoes in three studied periods in years2007–2009. Figure 4 demonstrates the WNV-risk areas inmosquitoes in those time windows based on Mahalanobisdistance p values. As observed in the figure, both high(Mahalanobis distance p values, 0.9–1.0) and moderate(Mahalanobis distance p values, 0.6–0.9) risk areas ofWNV infection in mosquitoes constantly appeared in/around southern San Joaquin Valley in Kern County andsouthern Los Angeles County in almost all the studiedperiods and years. The coastal areas of Orange and SanDiego also possessed moderate to high risk in some periodsof years, e.g., weeks 36–44 in all years, 2007–2009. Thespread pattern of identified WNV-risk areas seemed to beconsistent in 3 years: the risk areas appear in southern SanJoaquin Valley and southern Los Angeles County inweeks 18–26 (mainly May–June), and then expand to itssurrounding areas in weeks 27–35 (mainly July–August)and 36–44 (mainly September–October). The last map high-lighted the moderate- to high-risk areas of WNV. As acontrast to those vulnerable coastal lands, areas in theMajave Desert showed lower risk of WNV infection.

The risk analysis was conducted based on the environ-mental factors and surveillance data in years 2007–2009

without the records from 2003–2006. The risk map indicatesthe spatial distribution of possible invasion of WNV instudied counties. High-risk areas were identified in southernSan Joaquin Valley in Kern County and southern LosAngeles County, along with some coastal areas of Orangeand San Diego and their surrounding areas. The risk areasidentified in the City of Los Angeles area seem to becompatible with the results of another WNV study con-ducted by the authors with higher spatial resolution images[31]. The overall risk assessment is not contradictory to theexisting findings done by Reisen et al. [35], in which infec-tion rates in Culex pipiens quinquefasciatus, one of the mostfrequently infected mosquito species in the area werereported to be higher in both Kern and Los Angeles countiesthan in Coachella Valley, Riverside County. In addition totheir study, our assessment provides particular spatial loca-tions of possible vulnerability.

4 Discussion and Conclusions

We conducted a retrospective study of WNV spread for tencounties in the southern California based on surveillancedata from 2004–2009. Particularly, we assessed the WNV-risk areas based on the mosquito surveillance records inyears 2007–2009. Before conducting environmental analysisof WNV infections, we classified the whole study area intotwo zones since the area contains diverse climate conditionsand elevations. The construction of hexagon network simulat-ed more neutral landscape units than those of geographical orcensus units (e.g., census block groups). It is found thatsouthern San Joaquin Valley in Kern County and the southernLos Angeles County were the most vulnerable locations forWNV outbreak. Main factors contributing to the WNVpropagation included summer mean temperature, annualmean deviation from the mean temperature, land surfacetemperature, elevation, landscape complexity, landscapediversity, and vegetation water content. The result of thestudy improves understanding of WNVecology and providestools for detecting, tracking, and predicting the epidemic. Theholistic approach developed for this multidisciplinary study,which integrated remotely sensed, GIS-based, and in situ-

Table 2 Environmental statistics for two ecological zones (units, hexagons)

Ecological zones Average annual mean departure fromnormal temperature in 2004–09 (°F)

Average summer (May–August) meanmonthly temperature in 2004–2009 (°F)

Elevation (meters)

MIN MAX MN SD MIN MAX MN SD MIN MAX MN SD

Zone 1 −4.12 2.96 0.47 0.57 26.82 93.63 71.92 9.06 −74.3 717.6 337.62 219.20

Zone 2 −3.95 2.88 0.79 0.57 52.48 92.38 76.87 6.05 717.56 3269.98 1097.72 334.40

MIN minimum, MAX maximum, MN mean, SD standard deviation

448 H. Liu, Q. Weng

Page 9: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

measured environmental factors with landscape metrics, maybe applied to studies of other vector-borne diseases.

In order to better understand the variation of WNV out-breaks between 2004 and 2009, we examined the droughtconditions in those years based on Palmer drought severityindex (PDSI). PDSI is a drought index responding to abnor-mally dry or abnormally wet weather conditions, calculatedbased on temperature, precipitation, and local water contentof the soil [37]. Its value 0 indicates normal condition, whilenegative values associated with drought (−4 considered tobe extreme drought) and positive values implying excessrain. Figure 5 presents the PDSI and departure from normal2-month (February and March) average temperature (from1895 to 2009) in four climate divisions (CDs) that the tencounties fall in. These four CDs include Central CoastDrainage (CD4), San Joaquin Drainage (CD5), South CoastDrainage (CD6), and Southeast Desert Basins (CD7), amongwhich CDs 5–6 contained 90% tested positive mosquito

pools. The remaining records were in CD7 for years 2007–2009. No mosquito infections were documented for CD4 inthose 3 years. So, our focus of discussion was mainly on CDs5–7. The CD boundary information was obtained fromNational Climatic Data Center. Although year 2006 wasexcluded in our analysis of risk area, it is necessary to dem-onstrate the drought condition in that year because droughtcondition in the previous year may affect the outbreak ofmosquito population in the following year [8].

According to the PDSI values as shown in Fig. 5, CD5possessed slightly wet conditions (PDSI01.24) while CDs6–7 had incipient dry spell (PDSI0−0.82) to moderatedrought conditions (CD6, −0.82; CD7, −2.49) in 2006.The situation became worse in year 2007 with severedrought in CD5 (−3.57) and extreme drought in CDs 6–7(CD6, −5.04; CD7, −4.63). The drought conditions in CDs5–7 were consistent with slight relief in years 2008–2009.Based on the departure from normal 2-month average

Fig. 4 WNV-risk areas for three periods in years 2007–2009, indicatedby Mahalanobis distance p-value calculated based on mosquito surveil-lance records. In each map red hexagons indicate high risk (Mahalanobis

Distance p-value: 0.9–1.0) and yellow ones associate with moderateWNV risk (p-value: 0.6–0.9). Please refer to Fig. 1 for the rest symbolsin the maps

Environmental Factors and Risk Areas of West Nile Virus 449

Page 10: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

temperature, it appears that both years 2007 and 2008 re-ceived warmer than normal temperature in their early seasonin CDs 5–7, while the situation was opposite in 2009 duringthe same time frame. The slightly wet to incipient andmoderate drought conditions in 2006 and warmer than nor-mal temperature earlier in the season seems to encourage theamplification of virus across southern California, e.g., Kernand Los Angeles (Fig. 3) in later season of 2007. The severeto extreme drought conditions in 2007 and warm spring maycontribute to the outbreak of WNV in summer 2008 asshown in Fig. 3. However, a significant drop of WNVinfections in mosquitoes in the same year was observed inKern County, where WNV has been very active since 2004.This abnormal record may be linked with below averagetemperature recorded in CD 5 (Fig. 5) earlier in the season.Based on Kern County WNV Strategic Response Planadopted in May 2008, Kern Mosquito and Vector ControlDistrict and Kern County Department of Public Healthreceived dedicated funding in 2008 to support mosquitocontrol activities, which may play a significant role inreducing the infections [e.g., 5]. Cooler temperatures inearly 2009 may have retarded the multiplication of mosqui-toes in the following seasons.

Based on the results of environmental factor analysis, thepositive role of temperature detected in the environmentalfactor analysis is consistent with the finding reported byReisen et al. [41, 42]. Temperature must be above theminimum temperature (14.3°C) required for virus replica-tion in mosquitoes for WNV to be disseminated throughoutthe year, especially in the cold months [42]. A temperaturestudy in southern California showed that mean temperatures

in Coachella Valley in Riverside County and the hilly partsof Los Angeles County were mainly above the minimum inthe coldest months of years 2003–2005, while temperaturesin Long Beach, Los Angeles and Shafter, Kern Countyfloated around the minimum in the same period of time,and the differences increase during summer [41]. The resultsalso show that areas with lower elevations tended to be moresusceptible to WNV invasion. This finding may be explainedby the observation of mosquitos in Colorado: mosquitospecies (e.g., Culex tarsalis) appeared to be abundant in theplain habitats with lower elevations and warmer temperatures,and the population dropped in areas with higher elevation andslightly lower temperatures [13]. Higher landscape diversity isusually associated with multiple land cover types such asurban, grass, and water. Highly urbanized area with diversehabitats can contribute to bird abundance [48]. Certain amountof vegetation moisture in the neighborhood provides a favor-able habitat for mosquito survival [12].

Future research should be directed to improve the map-ping of WNV-risk areas by taking the following factors intoconsideration. Firstly, WNV surveillance data were collect-ed from CA arbovirus bulletins published by CaliforniaVectorborne Disease Surveillance System; thus, it is possi-ble that incidents existed outside the range of samplingeffort, e.g., central Imperial. Secondly, the two ecologicalzones (as shown in Fig. 1) were defined using clusteranalysis solely based on weather and terrain conditions.Although we carefully selected the variables for principalcomponent analysis and cluster analysis, additional factorsmay exist to influence the outcome of cluster analysis, e.g.,the number of water impoundments (e.g., green pools). The

Fig. 5 Palmer drought severityindex (PDSI) and departurefrom normal 2-month (Februaryand March) average tempera-ture (from 1895 to 2009) forCDs 4–7. X-axis represents yearand Y-axises represent PDSI(left) and departure temperature(right). The weather informa-tion per CD was obtained fromNOAA National Climatic DataCenter

450 H. Liu, Q. Weng

Page 11: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

identified risk areas in zone 1 may be changed accordingly.Also, we used the land cover dataset for 2001 to measure thelandscape complexity and diversity. Land cover datasets forindividual years of 2007–2009 may provide more up-to-datecharacteristics of landscape. More field and entomologicalinvestigations are needed to support these preliminaryresults. Other environmental factors, such as soil moistureand evaporation, can be derived from high-resolution remotelysensed imagery and serve as additional inputs for the risk areaassessment.

Acknowledgments The authors appreciate the constructive commentand suggestion by the reviewer and the editor, which helps to improvethe manuscript. Hua Liu acknowledges a Summer Research FellowshipProgram grant from Old Dominion University Office of Research in2010. Qihao Weng acknowledges a senior research fellowship fromNASA allowing him to work at Marshall Space Flight Center, 2008–2009, on urban climate and public health applications.

References

1. Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster analysis.Newbury Park: Sage.

2. Bertolotti, L., et al. (2008). Fine-scale genetic variation and evolutionof West Nile Virus in a transmission "hot spot" in suburban Chicago,USA. Virology, 374, 381–389.

3. Bouden, M., et al. (2008). The geosimulation of West Nile Viruspropagation: a multi-agent and climate sensitive tool for riskmanagement in public health. International Journal of HealthGeograhics, 7, 35. doi:10.1186/1476-072X-7-35.

4. Brown, H. E., et al. (2008). Ecological factors associated withWest Nile Virus transmission, northeastern United States. EmergingInfectious Diseases, 14, 1539–1545.

5. Carney, R. M., et al. (2008). Efficacy of aerial spraying of mosquitoadulticide in reducing incidence ofWest Nile Virus, California, 2005.Emerging Infectious Diseases, 14(5), 747–754.

6. Cavareski, C. A. (1976). Relation of park size and vegetation tourban bird populations in Seattle, Washington. Condor, 78, 375–382.

7. Ceccato, P., et al. (2001). Detecting vegetation water content usingreflectance in the optical domain. Remote Sensing of Environment,77, 22–33.

8. Chase, J. M., & Knight, T. M. (2003). Drought-introduced mosquitoout-breaks in wetlands. Ecology Letters, 6, 1017–1024.

9. Chevalier, V., et al. (2009). Predicting West Nile Virus seroprevalencein wild birds in Senegal. Vector Borne and Zoonotic Diseases, 9, 589–596.

10. Cooke, W. H., III, et al. (2006). Avian GIS models signal humanrisk for West Nile Virus in Mississippi. International Journal ofHealth Geographics, 5, 36. doi:10.1186/1476-072X-5-36.

11. Delbart, N., et al. (2005). Determination of phenological dates inboreal regions using normalized difference water index. RemoteSensing of Environment, 97, 26–38.

12. Dohm, D. J., et al. (2002). Effect of environmental temperature onthe ability of Culex pipiens (Diptera: Culicidae) to transmit WestNile Virus. Journal of Medical Entomology, 39(1), 221–225.

13. Eisen, L., et al. (2008). Mosquito species richness, composition,and abundance along habitat-climate-elevation gradients in thenorthern Colorado front range. Journal of Medical Entomology,45(4), 800–811.

14. Emlen, J. T. (1974). An urban bird community in Tucson, Arizona:Derivation, structure, regulation. Condor, 76, 184–197.

15. Epstein, P. R., & Defilippo, C. (2001). West Nile Virus anddrought. Global Change & Human Health, 2(2), 105–107.doi:10.1023/A:1015089901425.

16. Epstein, P. R., et al. (1998). Biological and physical signs ofclimate change: Focus on mosquito-borne diseases. Bulletin ofthe American Meteorological Society, 79(3), 409–417.

17. Gao, B. (1996). NDWI-A normalized difference water index forremote sensing of vegetation liquid water from space. RemoteSensing of Environment, 58, 257–266.

18. Gibbs, S. E. J., et al. (2006). Factors affecting the geographicdistribution of West Nile Virus in Georgia, USA: 2002–2004.Vector Borne and Zoonotic Diseases, 6(1), 73–82.

19. Goddard, L. B., et al. (2002). Vector competence of Californiamosquitoes for West Nile Virus. Emerging Infectious Diseases, 8(12), 1385–1391. doi:10.3201/eid0812.020536.

20. Gould, E. A., & Higgs, S. (2009). Impact of climate change andother factors on emerging arbovirus diseases. Transactions of theRoyal Society of Tropical Medicine and Hygiene, 103, 109–121.

21. Gu, Y., et al. (2007). A five-year analysis of MODIS NDVI andNDWI for grassland drought assessment over the central GreatPlains of the United States. Geophysical Research Letters, 34,L06407. doi:10.1029/2006GL029127.

22. Horvitz, D. G., & Thompson, D. J. (1952). A generalization ofsampling without replacement from a finite universe. Journal ofthe American Statistical Association, 47(260), 663–685.

23. Huang, C., et al. (2001). A strategy for estimating tree canopydensity using landsat & etm+ and high resolution images overlarge areas. The proceedings of the Third International Conferenceon Geospatial Information in Agriculture and Forestry held inDenver, Colorado, 5–7 November, 1 disk.

24. Jackson, T. J., et al. (2004). Vegetation water content mappingusing Landsat data derived normalized difference water index forcorn and soybeans. Remote Sensing of Environment, 92, 475–482.

25. Jolliffe, I. T. (1986). Principal component analysis. Berlin: Springer.26. Knick, S. T., & Dyer, D. L. (1997). Distribution of black-tailed

Jackrabbit habitat determined by GIS in Southwestern Idaho.Journal of Wildlife Management, 61(1), 75–85.

27. Koenig, W. D., et al. (2007). West Nile Virus and Californiabreeding bird declines. EcoHealth, 4(1), 18–24. doi:10.1007/s10393-007-0086-4.

28. Landesman, W. J., et al. (2007). Inter-annual associations betweenprecipitation and human incidence of West Nile Virus in the UnitedStates. Vector Borne and Zoonotic Diseases, 7(3), 337–343.

29. Lian, M., et al. (2007). Using geographic information systems andspatial and space-time scan statistics for apopulation-based riskanalysis of the 2002 equine West Nile epidemic in six contiguousregions of Texas. International Journal of Health Geographics, 6,42. doi:10.1186/1476-072X-6-42.

30. Liu, H., & Weng, Q. (2009). An examination of the effect oflandscape pattern, land surface temperature, and socioeconomicconditions on WNV dissemination in Chicago. EnvironmentalMonitoring and Assessment, 159, 143–161.

31. Liu, H., & Weng, Q. (2011). Enhancing temporal resolution ofsatellite imagery for public health studies: a case study of WestNile Virus outbreak in Los Angeles in 2007. Remote Sensing ofEnvironment. doi:10.1016/j.rse.2011.06.023.

32. Liu, H., et al. (2008). Spatio-temporal analysis of the relationshipbetween WNV dissemination and environmental variables in Indi-anapolis, USA. International Journal of Health Geographics, 7,66. doi:10.1186/1476-072X-7-66.

33. Mahalanobis, P. C. (1936). On the generalised distance in statistics.Proceedings of the National Institute of Sciences of India, 2, 49–55.

34. McGarigal, K., & Marks, B. J. (1995). FRAGSTATS: Spatialpattern analysis program for quantifying landscape structure.General

Environmental Factors and Risk Areas of West Nile Virus 451

Page 12: Environmental Factors and Risk Areas of West Nile Virus in …macaulay.cuny.edu/eportfolios/bird2012/files/2012/07/... · 2012-10-03 · Environmental Factors and Risk Areas of West

Technical Report PNW-GTR-351. USDA Forest Service. PacificNorthwest Research Station. Portland, OR.

35. O'Leary, D. R., et al. (2004). The epidemic of West Nile Virus inthe United States, 2002. Vector Borne and Zoonotic Diseases, 4,61–70.

36. Ozdenerol, E., et al. (2008). Locating suitable habitats for WestNile Virus-infected mosquitoes through association of environ-mental characteristics with infected mosquito locations: a casestudy in Shelby County, Tennessee. International Journal ofHealth Geographics, 7, 12.

37. Palmer, W. C. (1965). Meteorological drought. Research PaperNo. 45, U.S. Department of Commerce Weather Bureau,Washington, D.C.

38. Pradier, S., et al. (2008). Land cover, landscape structure, and WestNile Virus circulation in southern France. Vector Borne and ZoonoticDiseases, 8, 253–263.

39. Reisen, W. K., et al. (2006). Role of corvids in epidemiologyof West Nile Virus in Southern California. Journal of MedicalEntomology, 43(2), 356–367.

40. Reisen, W. K., et al. (2009). Repeated West Nile Virus epidemictransmission in Kern County, California, 2004–2007. Journal ofMedical Entomology, 46, 139–157.

41. Reisen, W. K., et al. (2006). Overwintering of West Nile Virus inSouthern California. Journal of Medical Entomology, 43(2), 344–355.

42. Reisen, W. K., et al. (2006). Effects of temperature on the trans-mission of West Nile Virus by Culex tarsalis (Diptera: Culicidae).Journal of Medical Entomology, 43(2), 309–317.

43. Reisen, W. K., et al. (2004). Invasion of California by West NileVirus. Emerging Infectious Disease, 10(8), 1369–1378.

44. Reisen, W. K., et al. (2005). Avian host and mosquito (Diptera:Culicidae) vector competence determine the efficiency of WestNile and St. Louis encephalitis virus transmission. Journal ofMedical Entomology, 42, 367–375.

45. Ruiz, M. O., et al. (2010). Local impact of temperature and precipita-tion on West Nile Virus infection in Culex species mosquitoes innortheast Illinois, USA. Parasites & Vectors, 3, 19.

46. Ruiz, M. O., et al. (2004). Environmental and social determinantsof human risk during a West Nile Virus outbreak in the greaterChicago area, 2002. International Journal of Health Geographics,3, 8. doi:10.1186/1476-072X-3-8.

47. Sanders, L. (1990). L’analyse statistique des données en géographie.Montpelliers: Alidade-Reclus.

48. Savard, J. P. L., et al. (2000). Biodiversity concepts and urbanecosystems. Landscape and Urban Planning, 48, 131–142.

49. Shannon, C. E. (1948). A mathematical theory of communication.Bell System Technical Journal, 27, 623–656.

50. Soverow, J. E., et al. (2009). Infectious disease in a warmingworld: How weather influenced West Nile Virus in the UnitedStates (2001–2005). Environmental Health Perspectives, 117(7),1049–1052.

51. Vogelmann, J. E., et al. (1998). Regional land cover characterizationusing Landsat Thematic Mapper data and ancillary data sources.Environmental Monitoring and Assessment, 51, 415–428.

52. Whitman, L. (1937). The multiplication of the virus of yellowfever in Aedes aegypti. The Journal of Experimental Medicine,66, 133–140.

53. Winters, A. M., et al. (2008). Predictive spatial models for risk ofWest Nile Virus exposure in eastern and western Colorado. TheAmerican Journal of Tropical Medicine and Hygiene, 79(4), 581–590.

54. Wittich, C. A., et al. (2008). Identification of hyperendemic foci ofhorses with West Nile Virus disease in Texas. American Journal ofVeterinary Research, 69, 378–384.

55. Yang, K., et al. (2008). An integrated approach to identify distribu-tion of Oncomelania hupensis, the intermediate host of Schistosomajaponicum, in a mountainous region in China. International Journalfor Parasitology, 38, 1007–1016.

452 H. Liu, Q. Weng