Tải bản đầy đủ - 0 (trang)
2 Case Study: WNV Dissemination in Indianapolis, 2002–2007

2 Case Study: WNV Dissemination in Indianapolis, 2002–2007

Tải bản đầy đủ - 0trang

366



Chapter Thirteen

land-cover (LULC) information was derived from three Advanced

Spaceborne Thermal Emission and Reflection Radiometer (ASTER)

images acquired on June 16, 2001, April 5, 2004, and October 13, 2006,

respectively. An unsupervised image classification method was chosen to classify each image into six categories: urban, agriculture, forest, grasslands, water, and barren lands. Overall accuracy of image

classification reached 88.33, 92.0, and 89.0 percent individually. The

resulting LULC maps are shown in Fig. 7.1b (for the image of June 16,

2001) and Fig. 7.1c (for the image of April 5, 2004). Figure 13.1 shows

the classified map for the image of October 13, 2006. It can be seen in

the image that urban, forest, and grassland were dominant habitats,

where urban fell in the central part of the city and forest was located

mainly in the north mixed with grassland. Agriculture mainly lay in

the southeastern and southwestern parts of the city. There were two

major reservoirs, the Geist Reservoir and the Eagle Creek Reservoir,

located in northern part. The White River runs north-south through

the city in the center.

Other environmental data were collected from the Indiana

Geological Survey Web site (http://igs.indiana.edu/). These data

included estimated percentages of impervious surfaces in Indiana in

2001, digital elevation model with 1.5-m resolution, hydrogeologic

terrains and settings, national wetland inventory data, pipe locations

in the National Pollutant Discharge Elimination System (NPDES),

facilities in NPDES, and industrial waste sites in Indiana. The outlines



Urban

Forest

Grasslands

Agriculture

Water

Barren lands

0 5 10



20



30



kilometers

N

W



E

S



FIGURE 13.1 LULC map of Indianapolis, Indiana, on October 13, 2006. See also

color insert.



40



Public Health Applications

of the U.S. Census block groups also were downloaded from the same

Web site. There were a 658 Census block groups in this study area.

This study was based on an assumption that selected environmental

variables in a certain Census block groups contributed the most to

WNV dissemination in that block group. As a result, all factors were

summarized and analyzed at the Census block-group level. A centroid of each block group was calculated by using ArcGIS. The shortest distances from each centroid to the pollutants and industrial waste

sites were calculated. The total length of streams and the area of wetlands in each block group also were computed in ArcGIS. Human

population density in each block group was selected to examine the

possible influence of human behavior on the spread of WNV. Table 13.1

lists all the variables used in the study.



Variable



Description



Mean for All 658 Block

Groups in Indianapolis



Land use land

cover (LULC)



Area percentage of each

LULC category



Varied by seasons and

years



Impervious

surface



Average percentage of

impervious surface



38.8 percent



Elevation



Elevation variation



12.9 m (42.3 ft)



Slope



Mean slope



1.6 degrees



Stream



Total stream length and

stream density



466.7 m (1531.2 ft)



Wetland



Total size of wetlands



51,437.4 m2

(553,667.6 ft2)



Pollutant



Distance from centroid

of each block group to

the closest pollutant in

National Pollutant Discharge

Elimination System (NPDES)



3201.8 m (10,504.6 ft)



Pipe



Distance from centroid of

each block group to the

closest pipe in NPDES



7831.3 m (25,693.2 ft)



Waste industry



Distance from centroid of

each block group to the

closest waste industry site



804.9 m (2640.7 ft)



Population

density



Human population density



1587/km2



Note: Study unit: U.S. Census block groups.



TABLE 13.1 Environmental Variables Selected for the Study in Indiana, 2001



367



368



Chapter Thirteen



13.2.2



Plotting Epidemic Curves



WNV was first identified in Indiana in 2001, and its transmission was

enhanced during the summer of 2002. C. pipiens was the main vector

mosquito in the state. Six cumulative epidemic curves were created to

show the peaks and temporal trends of mosquito WNV outbreaks in

Indianapolis in years 2002 through 2007. Epidemic patterns could be

identified based on monthly and annual comparisons. The spatial

outbreaks of WNV were tracked to indicate the movement of WNV

dissemination in the last 6 years.



13.2.3



Risk Area Estimation



It was significant to identify the risk areas of WNV mosquito outbreaks so that special care could be taken. A retrospective spacetime permutation model in SaTScan software was selected to

identify nonrandom WNV clusters in years 2002 through 2007.

This space-time model was based on a null hypothesis that there

was not any spatial, temporal, or spatio-temporal clusters in the

study area. The p value for the most likely cluster would be larger

than 0.05 with a 95-percent probability of change. Monte Carlo

hypothesis testing was used to calculate the test statistic for all possible clusters and 999 random replications. If a possible cluster

were among the 5 percent highest, then the significance level of the

test would be 0.05 (Dwass, 1957). The centroids and radii of the

most likely clusters with p values of less than 0.01 were recorded in

a DBF (.dbf file format, a major feature of dBase) table and then

visualized by using ArcGIS.

A K-means cluster analysis was developed to identify the highrisk areas for each month of July, August, September, and October. In

order to increase the sampling size in the statistical analysis, the

records collected in the same month but from different years were

combined. Three variables were selected from each location: the coordinates of the location, the number of positive mosquito pools in each

month, and the total number of mosquitoes in those pools. The spatial variations of high-risk areas in different months were identified

based on the results of cluster analysis.



13.2.4



Discriminant Analysis



Some environmental variables are considered to have more influence

than others on the spread of WNV. In order to identify which environmental factors played significant roles in WNV dissemination in

the study area, a stepwise Wilks’ lambda discriminant analysis was

applied to compare the block groups with WNV cases with those

without any cases relative to the set of environmental factors in

Table 13.1. There were 658 block groups used in this study. The analysis was done by year.



Public Health Applications



13.2.5



Results



Seasonal Outbreaks of WNV

Figure 13.2 presents epidemic curves of the WNV mosquito from

2002 through 2007. In the figure, the y axis presents mosquito counts

of WNV-positive tests. It becomes clear that year 2002 had positive

mosquito records from July to October; year 2003, from June to October;

year 2004, from May to October; and year 2005, from April to October;

whereas the records started from June to September in 2006 and from

May to August in 2007. The epidemic curve reaches a peak in August

in each year of 2002 through 2006 and in July in 2007. The variations

in the curves are consistent with the ecological observation that WNV

erupted in the summer and continued into the fall. We can conclude

that the outbreak of WNV was in August or July in Indianapolis in

last 6 years. As for the spatial patterns of WNV outbreaks, WNV dissemination always started from central longitudinal corridor and

spread out to the west and east. This observation indicates that some

environmental conditions may have significantly affected the spatial

patterns of the outbreaks. The results of discriminant analysis below

provide important clues to explain this observation.



Risk Areas

Spatial-temporal WNV clusters were detected in each year of 2002

through 2007 using a space-time permutation model in the SaTScan

program. Figure 13.3 shows that 2002 had six spatio-temporal clusters,



3,000



WNV mosquito cases



2,500

2,000



Y2002

Y2003

Y2004

Y2005

Y2006

Y2007



1,500

1,000

500

0

Apr



May



Jun



Jul



Aug



Sep



Oct



Month



FIGURE 13.2



Epidemic curves of mosquito-borne WNV in 2002 through 2007.



369



WNV mosquito cases



Chapter Thirteen



540



WNV mosquito cases



8/28–10/11, 6.82



Expected WNV cases



360



9/12–10/11, 7.36



8/28–9/11, 14.27



180



7/14–8/27, 2.36

8/28–10/11, 2.47



0

39.92



7/4–8/12, 10.25



39.84



La



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32



e



gitud



Lon



WNV mosquito cases



(a) Year 2002



540



WNV mosquito cases

Expected WNV cases



360

7/18–8/16, 12.43



180



8/2–8/16, 5.71

6/18–8/16, 12.83



0

39.92

39.84



La



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32



e



gitud



Lon



(b) Year 2003

WNV mosquito cases



370



WNV mosquito cases



540



Expected WNV cases



360

8/25–10/8, 5.36



180



6/26–7/25, 15.96



0

39.92



6/26–8/24, 10.21



39.84



La



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32



e



gitud



Lon



(c) Year 2004



FIGURE 13.3 WNV mosquito cases and their spatial-temporal clusters in the

years 2002 through 2007.



WNV mosquito cases



Public Health Applications



WNV mosquito cases



540



Expected WNV cases



360

8/23–9/6, 7.89

8/23–9/21, 10.62



180

0

39.92



7/24–8/22, 8.54



6/9–8/22, 7.47



39.84



La



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32



e



gitud



Lon



WNV mosquito cases



(d) Year 2005



WNV mosquito cases



540



Expected WNV cases



360

8/13–9/11, 9.36



180



6/29–7/28, 3.48



6/14–7/28, 13.11



0

39.92

6/14–6/28, 4.29



39.84



La



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32



e



gitud



Lon



WNV mosquito cases



(e) Year 2006



WNV mosquito cases



540



Expected WNV cases



360

180

5/18–7/1, 4.08



0

39.92



5/18–6/1, 9.72



La



39.84



tit



–86.00

–86.08



39.76



ud



e



–86.16



39.68



–86.24

–86.32

(f) Year 2007



FIGURE 13.3



(Continued)



e



gitud



Lon



371



372



Chapter Thirteen

whereas both 2003 and 2004 had three clusters. There were four clusters in each of 2005 and 2006 and two clusters in 2007. Each cluster

had a certain time period of outbreak and individual radius. The

p value for rejecting the null hypothesis of no clustering is 0.001.

Although these spatial-temporal clusters appeared randomly in

space, multiple clusters can be found on the southeast corner of the

study area in 2002 through 2006 with large radii (more than 8 km)

and a time period ranging from June to September. Two clusters from

two different years (2002 and 2004) even shared the same centroid

(39°69’N, 85°95’W) but with different radii (14.27 and 15.96 km) and

periods of outbreak (8/28 through 9/11 and 6/26 through 7/25,

respectively). The southeastern city was dominated by agriculture

and grassland. A high concentration of clusters with large radii indicates that agriculture and grassland provided favorable and extensive habitats for mosquito breeding during the summer and early fall.

Three clusters in 2002, 2004, and 2005 were found to share the same

centroid on the vegetation land in the central west of the city. This

finding supports the conclusion that mosquitos were attracted by

grassland in the study area.

The result of K-means cluster analysis on individual months of

July, August, September, and October shows that different months

had different high-risk areas. There were two clusters in Julys, four

clusters in Augusts, four clusters in Septembers, and two clusters in

Octobers in the study period. Figure 13.4 shows the high-risk areas in

the four months and their radii, which conform to the explosive transmission of WNV in Indianapolis during the summer. A cluster can be

found in the southeastern corner of the city in each month of August

and September. This finding supports the result of spatio-temporal

analysis in which five WNV clusters were observed in the same area

in 2002 through 2006. As noted in the preceding paragraph, this area

was dominated by crops (mainly corn and soybean) and grass. In

Indiana, corn usually is planted in April and May, and it becomes

mature in September and October, whereas soybeans usually are

planted in May and June and their leaves are shed mainly in September

and October. Dense corn and soybean fields in August and September

provide suitable temperature and sufficient moisture for mosquito

breeding, which might not happen in July when small leaves are still

growing, and in October, when crops are being harvested.

Two clusters appeared in the northwestern part of the city in

August and September. The August cluster was close to the Eagle

Creek Reservoir, and the September one included the reservoir.

According to the records of a U.S. Geological Survey (USGS) streamflow gauging station, mean stream flow in the Eagle Creek watershed

was highest in March and lowest in September. Wetlands were

expected to appear when the water level decreased from July to August

and September, which provided favorable habitats for mosquito



Public Health Applications



Radius: 6.14 km



Radius: 5.27 km



Radius: 5.37 km



Radius: 5.11 km

Radius: 4.57 km



Radius: 6.5 km



(a) July



(b) August



Radius: 5.62 km



Radius: 6.19 km



Radius: 5.61 km



Radius: 7.13 km

Radius: 5.50 km



Radius: 4.36 km



(c) September



(d) October

N



WNV case

W



Risk area



E

S



0 1.5 3



6



9



12



kilometers



FIGURE 13.4 Risk areas in July, August, September, and October, in the years 2002

through 2007. See also color insert.



breeding combined with the warm temperatures in August and

September. There was a cluster consisting of dense ponds and part of

the White River in the southeastern city in each month. A possible

explanation for this observation is that the ponds and the river provided plenty of moisture and still water for mosquito breeding.



373



374



Chapter Thirteen



Environmental Factors of WNV Dissemination

Table 13.2 shows the environmental variables that remained after the

discriminant analysis; their coefficients, eigenvalues, and grouping

accuracies; and the numbers of block groups with mosquito WNV

records and without mosquito cases. According to the results of the

stepwise Wilks’ lambda discriminant analysis, some environmental

variables seem to be more effective than others in differentiating block

groups with mosquito WNV cases from those without WNV records.

Area percentage of agriculture was one of the most effective variables

in the discriminant analysis in all years except 2006. A higher proportion of agriculture land was associated with more WNV cases. This

finding conforms to the risk-area analysis discussed earlier suggesting

that agriculture lands in the southeastern city were always within

high-risk clusters in different years. It could be explained that agriculture contributed to a cooler temperature in the surrounding areas in

the summer (Liu and Weng, 2008; Weng et al., 2004) and that farming

irrigation maintained a relatively constant moisture, which provided

a favorable environment for mosquito breeding.

Total size of wetland was another important variable for WNV

dissemination in the years 2002, 2005, and 2006. Larger wetlands

were linked to more mosquito WNV records. According to regulations of U.S. Environmental Protection Agency (EPA), wetlands are

those areas saturated by surface or ground water at a frequency and

duration sufficient to support vegetation prevalence. A possible

explanation of the observation is that high temperature and low to

median precipitation in July and August in these three years created

ideal conditions for the wetlands for mosquito breeding.

Total length of streams plays an important role in mosquito WNV

dissemination in the years 2002, 2003, and 2007. Longer streams were

associated with more WNV cases. This observation is in agreement

with a WNV study in Mississippi that suggested that stream density

contributed to WNV risk based on dead-bird occurrences (Cook et al.,

2006). Both stream density and total length of streams were originally

input in the discriminant analysis, but only the total length of streams

remained after the analysis. This result suggests that curvy streams

with still water provided a favorable environment for mosquitoes

breeding. Area percentage of water remained after the discriminant

analysis for the years 2004 and 2007. Water information derived

from ASTER imagery included big rivers, lakes, and dams. Higher

percentage of water sources certainly contributed to more mosquito

WNV cases.

In addition to agriculture, streams, wetlands, and some other

water sources, certain other variables also showed a positive contribution to WNV dissemination in various years, including mean slope

and elevation change for 2004, human population density in 2006,

and distance to the closest pollutant and distance to the closest waste



Classification

Function

Coefficient



Year



Variable



2002



Percentage of agriculture



0.143



Total length of streams



0.001



Total size of wetlands



0.001



Constant

2003



2004



0.126



Total length of streams



0.001



2005



0.154



Percentage of water



0.037



Mean slope



1.908



DEM variation



0.049



TABLE 13.2



46/612



86.90%



0.202



35/623



84.80%



0.201



25/633



84.80%



0.24



36/622



90.30%



–4.261



Percentage of agriculture



0.232



Total size of wetlands



0.001



Constant



0.226



–1.336



Percentage of agriculture



Constant



Grouping

Accuracy



–2.124



Percentage of agriculture



Constant



Eigenvalue



Block Group

with WNV/

without WNV



–2.499



375



Environmental Factors That Remained after Discriminant Analysis; Their Coefficients, Eigenvalues, and

Numbers of Block Groups with and without WNV Cases; and Grouping Accuracies



376

Year



Variable



2006



Total size of wetlands

Human population density

Constant



2007



Classification

Function

Coefficient

0.001



Grouping

Accuracy



0.157



29/629



85.30%



0.195



16/642



88.90%



0.001

–2.141



Percentage of agriculture



0.051



Total length of streams



0.001



Percentage of water



0.205



Distance to the closest pollutant



0.001



Distance to the closest waste industry



0.003



Constant



Eigenvalue



Block Group

with WNV/

without WNV



–4.709



TABLE 13.2 Environmental Factors that Remained after Discriminant Analysis; Their Coefficients, Eigenvalues, and

Numbers of Block Groups with and without WNV Cases; and Grouping Accuracies (Continued)



Tài liệu bạn tìm kiếm đã sẵn sàng tải về

2 Case Study: WNV Dissemination in Indianapolis, 2002–2007

Tải bản đầy đủ ngay(0 tr)

×