International Journal of Health Geographics Open Access Research the Relationship between Mosquito Abundance and Rice Field Density in the Republic of Korea

Background: Japanese encephalitis virus (JEV), the causative agent of Japanese encephalitis (JE), is endemic to the Republic of Korea (ROK) where unvaccinated United States (U.S.) military Service members, civilians and family members are stationed. The primary vector of the JEV in the ROK is Culex tritaeniorhynchus. The ecological relationship between Culex spp. and rice fields has been studied extensively; rice fields have been shown to increase the prevalence of Cx. tritaeniorhynchus. This research was conducted to determine if the quantification of rice field land cover surrounding U.S. military installations in the ROK should be used as a parameter in a larger risk model that predicts the abundance of Cx. tritaeniorhynchus populations. Mosquito data from the U.S. Forces Korea (USFK) mosquito surveillance program were used in this project. The average number of female Cx. tritaeniorhynchus collected per trap night for the months of August and September, 2002-2008, was calculated. Rice fields were manually digitized inside 1.5 km buffer zones surrounding U.S. military installations on high-resolution satellite images, and the proportion of rice fields was calculated for each buffer zone. Results: Mosquito data collected from seventeen sample sites were analyzed for an association with the proportion of rice field land cover. Results demonstrated that the linear relationship between the proportion of rice fields and mosquito abundance was statistically significant (R 2 = 0.62, r = .79, F = 22.72, p < 0.001). Conclusions: The analysis presented shows a statistically significant linear relationship between the two parameters, proportion of rice field land cover and log 10 of the average number of Cx. tritaeniorhynchus collected per trap night. The findings confirm that agricultural land cover should be included in future studies to develop JE risk prediction models for non-indigenous personnel living at military installations in the ROK. Background Japanese encephalitis virus has been identified by the U.S. National Center for Medical Intelligence as operationally important to the U.S. military. Similarly, the pathogen is important to unvaccinated personnel from other militar-ies around the globe, expatriates, and travelers [1]. JE is a potentially debilitating and deadly flavivirus that is endemic in rural areas in east, south, and southwest Asia. Successful vaccination programs have diminished the number of illnesses in many countries, causing many to consider the disease to be rare and exotic [2]. However, the JEV is maintained in populations of wading birds and


Background
Japanese encephalitis virus has been identified by the U.S. National Center for Medical Intelligence as operationally important to the U.S. military. Similarly, the pathogen is important to unvaccinated personnel from other militaries around the globe, expatriates, and travelers [1]. JE is a potentially debilitating and deadly flavivirus that is endemic in rural areas in east, south, and southwest Asia. Successful vaccination programs have diminished the number of illnesses in many countries, causing many to consider the disease to be rare and exotic [2]. However, the JEV is maintained in populations of wading birds and its primary vector Culex tritaeniorhynchus is plentiful in many Asian countries. It has been shown that the number of JEV vector mosquitoes collected in an area is highly correlated with the number of JE patients in that area [3]. While a reliable JE vaccination exists, this disease still affects many Southeast Asian countries [4,5]. Recent outbreaks in the last five years (2005 through 2010) have occurred in the indigenous population and unvaccinated expatriates and travellers in India, Nepal, Taiwan, Malaysia, China, and Vietnam [6][7][8][9][10][11][12]. Recent maps showing the geographical distribution of JE have been published in several different papers [2,12,13]. The JE distribution map in Figure 1 is based on the U.S. Centers for Disease Control and Prevention (CDC) 2009 version of the map currently distributed to international travellers [13].
While JE clinical manifestations occur only in a small number that become infected, the case fatality rate can be as high as 40% for those demonstrating symptoms. Permanent neurologic sequelae are observed in 40 to 50% of those who survive severe disease, resulting in minor to severe complications and early death [14,15]. Furthermore, JE is often misdiagnosed and lacks effective treatment beyond supportive care [16,17].
U.S. military personnel have been stationed in the ROK since World War II. Currently, programmed extended tour lengths have greatly increased the number of accompanying family, placing family members and young unvaccinated children at risk of infection. While an established Korean pediatric JEV vaccination program has greatly reduced the number of cases occurring in the Korean population, JEV continues to be of military public health concern in the ROK because U.S. personnel are not vaccinated for JE. Because of this disparity, human disease rates in the indigenous population are unreliable for estimating transmission and disease risks to U.S. military, civilian, and family members stationed there.
Culex tritaeniorhynchus is exophagic, prefers zoonotic and domestic hosts, and tends to feed interspecifically [18]. Primary vertebrate reservoirs for JEV are large water birds of the family Ardeidae, while wild and domestic pigs are amplifying hosts [17,19]. Cx. tritaeniorhynchus is highly susceptible to viral infection; studies suggest that over 30% become infected after feeding on a viremic pig [20]. Mosquito populations are dependent on environmental and seasonal factors for their propagation. In the ROK, seasonal occurrence of Cx. tritaeniorhynchus are reported between mid-June and early October, with adult peak populations from late July through early September [21][22][23][24][25][26]. JEV infections within vector populations peak in late July and early August, with human disease rates peaking in late August and September [15,27]. Culex tritaeniorhynchus breed in open sunlit temporary and permanent water habitats with vegetation, and have an average flight range of 1.5 km [4,28]. The relative abundance of Cx. tritaeniorhynchus is closely related to local rice agricultural practices, where wetland rice fields and similar irrigation systems provide an optimal habitat for  [13]. Guam and Saipan (not shown on this map) have also had cases of Japanese Encephalitis. larval development [18]. The ecological relationship between mosquitoes and rice fields has been studied extensively [29][30][31]; results show that these man-made breeding sites result in a greater prevalence of Cx. tritaeniorhynchus than natural breeding sites [32,33].
Remote sensing and Geographical Information Systems (GIS) are increasingly used in disease epidemiology to map vector and disease distributions so that vector control efforts and disease intervention strategies are administered efficiently and effectively [29]. Past research has demonstrated that remote sensing and GIS capabilities can be used successfully to identify breeding habitats of Cx. tritaeniorhynchus in Japan, the ROK, India, and Australia [18].
Past studies on relative seasonal abundance and geographical distribution of Cx. tritaeniorhynchus in the ROK have contributed to the success and development of a JE disease surveillance systems for local populations [34]. These surveillance systems have been based on monitoring vector populations and seroprevalence in swine from selected slaughterhouses; while such surveillance systems have been effective, they have a negative impact on the limited human resources and man-power available to public health. For this reason, the Uniformed Services University, the Armed Forces Health Surveillance Center, Global Emerging Infections Surveillance and Response System (AFHSC-GEIS), the National Aeronautics and Space Administration (NASA) Goddard Space Flight Center (GSFC), and the 65 th Medical Brigade-Korea, began working together to develop a JEV risk model in the ROK that incorporates remote sensing, land cover, elevation, and historical climate data.
The goal of this project was to determine if quantification of wetland rice field land cover via remote sensing and GIS techniques could feasibly be used as a parameter in a larger risk model to predict the abundance of Cx. tritaeniorhynchus at U.S. military installations in the ROK. While past studies have shown a correlation between Cx. tritaeniorhynchus collected at cow and pig sheds and surrounding rice field acreage [35,36], this study examines the number of Cx. tritaeniorhynchus collected on primarily urban military bases and compares the number of mosquitoes collected to the amount of rice field surrounding those bases. Quantifying the relationship between the amount of rice fields surrounding U.S. military installations in the ROK and the abundance of Cx. tritaeniorhynchus captured at these sites will enhance the U.S. military's capability to predict the risk of JEV transmission to military, civilians, and family members assigned to the ROK. This proposed risk model would be used in conjunction with a predictive model for seasonal JEV transmission using real-time data and would serve as a disease risk monitoring system for U.S. military and civilians living in the ROK. The development of an accu-rate risk model based on remotely sensed land and climate data will allow for more efficient detection of increased JE risk to U.S. military populations in the ROK. Additionally, the risk model would diminish reliance on man-power intensive surveillance systems, and may provide rationale for the implementation of mandatory JEV vaccination to USFK military, civilian, and family members assigned to the ROK.

Study Area
The study area comprised seventeen U.S.

Mosquito Data
Retrospective mosquito data were provided by the USFK, which conducts routine mosquito surveillance on U.S. military installations to monitor all mosquito species. Mosquitoes were collected weekly using un-baited New Jersey (NJ) light traps in accordance with USFK preventive medicine protocol from May through October, 2002 to 2008. One to eight traps were established at each mili-  tary installation. Collections were sent to the Entomology Section, 5 th Medical Detachment, for identification where data on adult female mosquitoes (i.e. species, trap night, month, and year) were prepared. A summary of the total number of mosquitoes (males plus females) collected at each location from 2002 to 2008 is listed in Table 1. Although NJ light traps are relatively inefficient for collecting mosquitoes when compared to newer technologies, they are easy to operate and data can be compared with collections made over more than 20 years of surveillance in the ROK.
The final mosquito count at each military installation used in the main analysis of this study is an average from August to September from 2002 to 2008. August and September were selected for analysis to capture the peak Cx. tritaeniorhynchus breeding cycle [21]. To reduce the effects of sampling variation between years and installations, data were standardized as the number of female mosquitoes collected per trap night per month [mosquitoes per trap night = total female mosquitoes collected/ (number of traps * number of nights)]. The final mosquito capture rate for each military installation was calculated by adding the number of mosquitoes per trap night for months August and September from 2002 to 2008, then dividing by the number of nights mosquito sampling was conducted during that time period.

High Resolution Satellite Data
One IKONOS and ten QuickBird high resolution images, encompassing visible and near infrared data, taken between 2004 and 2008 were used for land cover classification. QuickBird is a commercial earth observation satellite operated by DigitalGlobe http:// www.digitalglobe.com that records multispectral imagery at 2.6 meter resolution at nadir. IKONOS is a similar commercial earth observation satellite operated by Geo-Eye http://www.geoeye.com that provides multispectral images at four meter resolution. Although the QuickBird images have a higher resolution than the IKONOS image, comparison studies have demonstrated that land classification outcomes are analogous for both image types [37]. Within each of the ten QuickBird images and one IKO-NOS image, military installations were located using latitude and longitude coordinates. Processing of image data involved two steps: digitizing and area computation.
Rice fields were identified using photointerpretation and manual digitization via ArcGIS ® 9.3 software (Environmental Systems Research Institute, Redlands, CA, USA, http://www.esri.com). The border of each military installation was digitized using publically available installation maps as aids. Buffers were used to designate the study inclusion area. The buffer area surrounding each military installation was based on the average flight range of Cx. tritaeniorhynchus [4], and was set at 1.5 km from the digitized military installation border. Rice fields within the buffer area of each military installation were then manually digitized as polygons. Manual digitization was chosen because the majority of images available for this project were acquired during the winter months when rice fields were bare and spectrally similar to other land cover, making automated classification difficult. Rice field land cover was digitized as large polygons consisting of many individual rice fields. Other non-rice features such as irrigation ditches and small roads may have been included in the rice field land cover. To minimize variation, all digitization was completed by one researcher in a consistent manner.
To obtain area measurements in kilometers, the digitized military installations buffer area and rice field shape files were reprojected from a Geographic Coordinate System (GCS) to a Universal Transverse Mercator (UTM) map projection. The buffer area at each military installation was calculated and reported in square kilometers. The digitized rice field polygons within the buffer area of each military installation was calculated and reported in km 2 . To account for variation in installation size and thus buffer area size, rice field densities were calculated as the proportion of rice fields within the buffer area at each installation (rice fields proportion = rice field area/military installation buffer area). An example of digitized rice field data is presented in Figure 3.

Rice Field Analysis
Simple linear regression analysis, using the SPSS ® statistical program (SPSS Inc., Chicago, IL, USA) was conducted on the data using two continuous variables, rice field density and log 10 (Table 1). An observational retrospective study of mosquito prevalence at U.S. military installations in the ROK found that Cx. tritaeniorhynchus was the most common mosquito species collected in 2002 [38]. In 2002, due to a high number of mosquitoes collected at Gunsan Air Base, the overall average of Cx. tritaeniorhynchus populations in the ROK increased by 27.2%, while populations of other mosquito species remained relatively constant during the same time period [38]. The increased number of Cx. tritaeniorhynchus caught at Gunsan Air Base in 2002 and 2004 could be due to the use of more efficient collection methods during these two years. At Gunsan, NJ light traps were baited with dry ice during collections in 2002 and the Mosquito Magnet ® was used for collections in 2004 [38]. The Mosquito Magnet is the most effective trap for collecting Cx. tritaeniorhynchus; capturing up to 650 times as many mosquitoes as the NJ light trap [23]. Primarily because trapping methods were substantially different at Gunsan than those at all other installations in the study sample, Gunsan Air Base was excluded from the final analysis.
A priori statistical power analysis stated that a sample of sixteen sites carried 80% power to detect the proportion of variability between mosquito abundance and rice field density in the model with a coefficient of determination (R 2 ) of 0.65. For our purposes, an R 2 of 0.65 was desired. The correlation coefficient (r) was also used in this analysis to elucidate the relationship between mosquito abundance and the rice field density seen in the data.

Results
Average mosquito counts per trap night and the corresponding rice field density statistics are presented for each installation in Table 2. The area surrounding half of the installations in this study were comprised of less than 10% rice fields. This is in sharp contrast to installations that were surrounded by the most rice fields; coverage ranged between 30-56% of the land surrounding five installations. Mosquito data were log 10 transformed prior to regression analysis. Analysis conducted using log 10 transformed mosquito data from 2002-2008 demonstrated a statistically significant linear relationship between rice field density and mosquito abundance (R 2 = 0.62, r = .79, F = 22.72, p < 0.001). A scatter plot of the data with the regression line is presented in Figure 4.
The resulting linear model [y = 2.71 × -0.28; where y equals the log10 of the average mosquitoes per trap night, and × equals the proportion of rice field land cover] was used to predict the abundance of post-construction Cx. tritaeniorhynchus at Camp Humphreys using the 2008 image. The data reveals that for every 10% increase of rice field land cover surrounding an installation, an 86% increase of female Cx. tritaeniorhynchus per trap night can be expected.
In an exploratory analysis the regression model was used to predict the abundance of Cx. tritaeniorhynchus at Camp Humphreys in 2008. Despite major changes to the Camp Humphreys area, the proportion of rice fields surrounding the installation increased by only 1%. The number of predicted mosquitoes per trap night was 18.2 for the 2008 Camp Humphreys image; the number of observed mosquitoes per trap night was 22.2.

Discussion
This research addresses the issues of mosquito abundance and spatial patterns of agricultural land use. Emphasis has been put on determining the relationship between average Cx. tritaeniorhynchus collected per trap night and density of rice fields surrounding U.S. military installations in the ROK; as well as assessing the value of that relationship for inclusion in a larger disease risk model. The future application of rice field density as a parameter in models used to predict JEV disease risks can be based on the strength of linear association, as it is presented here.
Results revealed that 62% of the variation in the log 10 of Cx. tritaeniorhynchus collected per trap night on base was accounted for by the density of rice fields in the area surrounding the military base. The remaining 38% can be explained by inherent variability or other variables, such as species breeding cycle, proximity to blood source, or regional climatic conditions [17,[19][20][21]31]. Our data reaffirms past JE and Cx. tritaeniorhynchus research and suggests that rice fields account for a large portion of the mosquito's abundance but that populations cannot be predicted by rice fields alone [29][30][31]35,36].
A statistically significant linear association was demonstrated between rice field density and mosquito population. Given the strong relationship demonstrated between rice field density and mosquito populations, this variable should be considered when modelling Cx. tritaeniorhynchus propagation or JEV transmission risks.
Specific limitations to this study with regard to image data include the inability to automatically extract rice fields due to the lack of ground vegetation in many of the images used. Due to cloud cover and other limiting factors, the remotely sensed imagery used in analysis were not all acquired during the rice growing season, making automated land cover extraction software unable to accurately identify rice fields. While rice fields were extracted manually, the potential for misclassification of land cover remained. Field shape, image interpretation, and area photographs guided rice field identification. This study was also limited by the lack of images at each installation for each year or month within the study period. Thus, the authors were unable to explore variation in the correla-tion between rice field density and Cx. tritaeniorhynchus throughout time. This study, and potentially all studies conducted on military bases in the ROK, will be limited by the number of installations and number of installations within the USFK mosquito surveillance program.
This limitation restricts analysis and does not allow the geographical variability among installations to be explored.
Despite limitations, this study has a number of strengths. It utilized USFK mosquito surveillance data collected consistently for over 8 years in the ROK. This data continues to be collected and can be similarly used in the future. It also uses high resolution remotely sensed imagery to determine rice field land cover. Using remotely sensed data to map rice field land cover is less time consuming than conventional ground survey methods used in the past to calculate agricultural land use.

Conclusions
Using the data presented here, an overall statistically significant linear relationship between rice field density and Cx. tritaeniorhynchus abundance was found. Use of the regression model to predict mosquito abundance was inconsistent. Future studies to develop JE risk prediction models for non-indigenous personnel living at military installations in the ROK should include agriculture land use and irrigation patterns as variables.
There are a number of implications in this study for future research in JE risk prediction and Cx. tritaeniorhynchus population modelling in the vicinity of U.S. military installations in the ROK. First, automated risk