SEARCH

SEARCH BY CITATION

Keywords:

  • Aedes aegypti;
  • Culex quinquefasciatus;
  • land cover;
  • mosquito abundance;
  • seasonal presence;
  • Arizona

ABSTRACT:

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

It is currently unclear what role microhabitat land cover plays in determining the seasonal spatial distribution of Aedes aegypti and Culex quinquefasciatus, disease vectors of dengue and West Nile Virus, respectively, in Tucson, AZ. We compared mosquito abundance to sixteen land cover variables derived from 2010 NAIP multispectral data and 2008 LiDAR height data. Mosquitoes were trapped with 30–9 traps from May to October of 2010 and 2011. Variables were extracted for five buffer zones (10–50 m radii at 10 m intervals) around trapping sites. Stepwise regression was performed to determine the best scale for observation and the influential land cover variables. The 30 m radius buffer was determined to be the best for observing the land cover-mosquito abundance relationship. Ae. aegypti presence was positively associated with structure and medium height trees and negatively associated with bare earth; Cx. quinquefasciatus presence was positively associated with pavement and medium height trees and negatively associated with shrubs. These findings emphasize vegetation, impervious surfaces, and soil influences on mosquito presence in an urban setting. Lastly, the land cover-mosquito abundance relationships were used to produce risk maps of seasonal presence that highlight high risk areas in Tucson, which may be useful for focusing mosquito control program actions.


INTRODUCTION

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

Understanding the spatial dynamics of mosquito populations and how environmental factors affect mosquito abundance and distribution is central to preventing disease outbreaks by controlling mosquito populations. The mosquitoes Aedes aegypti and Culex quinquefasciatus, disease vectors for dengue fever and West Nile Virus (WNV), respectively, are commonly found throughout the Southwest. While neither species is native to the Southwest, both species have thrived in the human-modified desert environment. Because the mosquito's presence is highly dictated by the environment, understanding the spatial land cover-mosquito abundance relationship is essential for species distribution modeling and mapping the risk of seasonal Ae. aegypti and Cx. quinquefasciatus presence.

Mosquito-borne diseases have had an increasing impact on human health in Arizona in the past decade, including active outbreaks of WNV and a rising potential for a dengue outbreak. The rapidly growing cities of AZ are providing an ever-expanding habitat for mosquitoes, increasing the potential human health impact the existence of these vectors present. In Arizona, climate factors have a known influence on the mosquito population; mosquito presence is highest during the summer months with the mosquito population increasing at the start of the summer monsoon season and decreasing as temperatures fall in the late summer months (Hoeck et al. 2003). Land cover factors have also been shown to play a controlling role in the distribution and abundance of Ae. aegypti (Focks et al. 1993) and Cx. quinquefasciatus (Reiter and LaPointe 2007). However, the role of the land cover in defining the spatial variability of mosquito presence in the urban desert environment is less clear.

In response to potential outbreaks of vector-borne diseases, scientists and public health organizations have sought to use land cover-mosquito abundance relationships to develop place-based models to predict mosquito abundance and associated disease risk. A spatial epidemiology approach has produced many place-based models for malaria vectors, but fewer place-based models have been developed for other important disease vectors, such as dengue (Kolivras 2006, Fuller et al. 2010) and WNV (Liu et al. 2008, Ghosh 2011). Additionally, while general climate, land use, and mosquito abundance relationships are well documented for the Southwest (Hayden et al. 2010, Kolivras and Comrie 2004, Morin and Comrie 2010), no spatial models exist for dengue and WNV vectors that incorporate land cover variables or that highlight the risk of mosquito presence in Tucson, AZ.

Mosquito-borne disease-vector-land cover relationships have shown that (1) land cover greatly influences mosquito presence and (2) disease-vector distribution is strongly governed by environmental factors that create suitable habitats (Kolivras 2006). However, a traditional ground-based approach to monitoring mosquitoes is labor intensive and cost prohibitive. To improve monitoring, a large scale approach to identify the areas at highest risk of mosquito presence would be the best method to allocate limited resources. Geospatial mapping using remote sensing and Geographic Information Systems (GIS) are highly useful tools for assessing the spatial epidemiology of vector-borne diseases as they allow for accurate identification and mapping of the spatial distribution of high risk mosquito breeding habitat on a large scale. Until recently, technological limitations (i.e., low spatial resolution of satellite imagery, need for large storage capacity, etc.) limited epidemiological applications. Existing mosquito habitat and risk models often used imagery with too coarse a spatial resolution to adequately define relationships between microhabitat land cover and mosquito abundance in urban environments. Remote sensing technology advancements and increased access to high spatial resolution (∼1m) multispectral aerial imagery data and three-dimensional (3-D) LiDAR (Light Detection and Ranging) data will allow for improved identification of the mosquito's habitat preferences as defined by land cover.

Species distribution models (SDMs) provide the easiest way to incorporate remotely sensed data, such as land cover characteristics, into predictive models. SDM techniques can be applied to the modeling of many infectious diseases, beginning by predicting where the vector will be present. Modeling disease occurrence based on environmental conditions has been shown to be an effective way of developing a predictive spatial model (Kolivras 2006, Fuller et al. 2010). Many studies have used remotely-sensed data to identify mosquito habitat areas and to develop regression-based predictive models. Developing such models would identify areas of high risk mosquito presence in Tucson, AZ.

This study takes a spatial epidemiological approach to improve our understanding of mosquito habitat requirements and the spatial distribution of mosquito abundance. This study seeks to (1) use high spatial resolution image data to characterize urban mosquito microhabitat through land cover variables and (2) determine the best scale at which to monitor land cover variables to predict the risk of mosquito presence at the neighborhood level. Understanding the land cover-mosquito abundance relationships will assist in improving the identification of mosquito habitats in the urban Southwest and will allow for focusing of mosquito control actions.

MATERIALS AND METHODS

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

Study area

The study area is located in the urban center of Tucson, AZ, covering approximately 76 km2 between coordinates 32°12′25″N–32°15′28″N latitude and 110°54′35″W–110°54′35″W longitude (Figure 1). The study area was determined by the location of mosquito trapping sites. Located in the Sonoran Desert, Tucson experiences moderate winters (average January low temperature = 3.8° C), prolonged hot summers (average May–September high temperature = 35.7° C), and limited precipitation (average annual precipitation = 311.1 mm) with high spatial variability in rainfall totals (Western Regional Climate Center 2010. Arizona climate summaries. Reno.)

image

Figure 1. Study area and mosquito trap locations for 2010 and 2011. Near-infrared, red, and green false color multispectral NAIP (National Agricultural Imaging Program) image of the urban study area are a composite of a mosaic of the area in Tucson (AZ, USA) acquired June 2010.

Download figure to PowerPoint

Some rainfall occurs during the winter, but the majority of the rainfall occurs during the summer monsoon season, from July through September. 2010 was a dry hot year with above average temperatures and below average rainfall (National Weather Service Forecast Office. 2010 climate report for Tucson. http://www.wrh.noaa.gov/twc/climate/monthly/2010.php.). The year 2011 had record low temperatures in January and February and record high temperatures in the summer, with one of the driest years on record up to June and one of the wettest years on record in the second half of the year (National Weather Service Forecast Office. 2011 climate report for Tucson. http://www.wrh.noaa.gov/twc/climate/monthly/2011.php). The study area provides a representative snapshot of the urban desert environment with a high presence of impervious surfaces and native and introduced vegetation in natural and irrigated landscapes. The study area includes four land use categories: residential, managed green spaces (golf courses and parks), mixed use, and washes.

Adult mosquito collection

Ae. aegypti and Cx. quinquefasciatus presence and abundance were measured at trapping sites throughout the study area. Adult mosquito trapping was carried out weekly from June to October (from now on referred to as the mosquito season) of 2010 and 2011. Trapping occurred at 49 sites during the 2010 mosquito season and continued at 30 of the sites during the 2011 mosquito season (Figure 1). Carbon dioxide (CO2)-baited suction traps were placed at trapping sites of various land use types (residential, mixed use, washes, and managed green spaces which includes parks and golf courses) (Table 1). Traps were retrieved approximately 14 h after placement. Trapping sites were documented with photographs, GPS points, and site descriptions. Mosquitoes were identified by species and sex. Analysis of mosquito trapping data was based on seasonal total counts of adult female Ae. aegypti and Cx. quinquefasciatus for predicting risk of seasonal mosquito presence.

Table 1.  Distribution of mosquito traps in the various land use classes and characterization of land use classes for the study area. The percentage of mosquito traps for the four land use types and the percentage of each land use type for the study area were calculated to determine how well represented each land use type was by the mosquito trap coverage.
Land Use TypeNumber of Traps% of Traps% of Study Area
Managed Green Spaces102011
Mixed Use132733
Residential204152
Wash6124

Land cover classification and quantification

To quantify land cover around sites, a land cover map was created from National Imagery Agriculture Program (NAIP) multispectral (blue, green, red, and near-infrared) 1 m spatial resolution aerial imagery and LiDAR elevation data. The NAIP imagery was collected in June, 2010 by the United States Department of Agriculture. A Normalized Difference Vegetation Index (NDVI) was derived from the NAIP imagery. NDVI, derived from the red and near-infrared bands of the multispectral data, was used as a proxy for vegetation greenness (Tucker 1979). A canopy height model (CHM) was derived from February, 2008 LiDAR height and ground elevation data. The LiDAR data point sample density was one point/m or better with a horizontal accuracy of one m and a vertical accuracy of 37 cm (STGS 2009).

The combination of LiDAR and multispectral image data has been shown to improve land cover classification, specifically identification of vegetation classes (Hartfield et al. 2011, Secord and Zakhor 2007). NAIP, NDVI, and CHM were used as inputs to Classification and Regression Tree (CART) classification to produce a land cover map (Rulequest Research 2008). Training points for the classification were selected through expert knowledge. The land cover map was comprised of eleven classes: bare earth, pavement, structure, pool, water, shadow, herbaceous (0 m), shrub (0.3–0.9 m), and three tree height classes, low height tree (1.2–3.0 m), medium height tree (3.3–9.1 m), and high height tree (9.4+ m) (Figure 2). The water class is composed of all non-chlorinated water bodies such as ponds and lakes (natural or constructed), while the pool class refers to chlorinated water bodies. A distinction is made between chlorinated and non-chlorinated water bodies as chlorinated water bodies were determined to be unsuitable for larval development. Latitude and longitude coordinates of the trapping sites were recorded with a global positioning system for GIS processing using ArcGIS 10 (ESRI, Redlands CA, U.S.A). Vector layers were generated for trapping site locations and for presence and abundance of both species.

image

Figure 2. Land cover map based on classification and regression tree (CART) classification using 2010 4-band multispectral NAIP data, NDVI data, and 2008 LiDAR CHM data.

Download figure to PowerPoint

To assess the micro-scale land cover factors, buffer zones were created around each trapping site. Five buffer zones were defined by radii (10–50 m at 10 m intervals). Sixteen land cover variables were extracted for each buffer zone to characterize the trapping sites. Extracted land cover variables included the percent coverage of the 11 land cover classes and three additional “composite” vegetation classes, and two additional landscape descriptors – NDVI and distance to waterways. NDVI was used to characterize the sites separately from the land cover classification as it is calculated directly from the multispectral data and is without bias regarding land cover classes and plant physiology. Distance to waterways was calculated using Euclidean distance to the nearest water feature and was included in the site characterization as waterways may provide larval breeding habitat. Additional information of the land cover variables and extraction techniques is in Table 2. Land cover variable extraction was repeated for the five buffer zones to determine the best spatial scale at which to identify the significant land cover variables in the land cover-mosquito abundance relationship.

Table 2.  Land cover variables were extracted for the five buffer zones (10–50 m radii at 10 m intervals) around each trapping site. Land cover variables were used to characterize the mosquito habitat. Land cover was used as independent variables in the stepwise regression.
VariableDescription
Land coverComposed of 11 classes: bare earth, pavement, structure, pool, water, shadow, and five vegetation classes – herbaceous (0 m), shrub (1.2–0.91 m), low height tree (2.1–3.0 m), medium height tree (3.3–9.1 m), and high height tree (9.4+ m). Percent coverage of each class was calculated
All VegetationPercent coverage of a composite of the five vegetation classes
All Vegetation with heightPercent coverage of a composite of the four vegetation classes defined by height (shrub, low tree, medium tree, high tree)
All TreesPercent coverage of a composite of the three tree classes (low tree, medium tree, and high tree)
Distance to WaterEuclidean distance in meters from trapping site GPS point to nearest waterway (washes and man-made ponds). Pools were not included
Average NDVINDVI was used as proxy of vegetated cover. Average NDVI for buffer zone was extracted from NDVI raster

Land cover accuracy assessment

A GPS receiver was used to select accuracy points throughout the entire study area. Accuracy points were not selected within the classification's training site pixels. At least 25 accuracy points were chosen for each of the 11 classes, except for the water class due to the lack of water bodies in the study area for both sampling and assessment. Points were selected in the middle of objects to avoid confusion that could be caused by mixed pixels possibly representing multiple classes. Forty-nine points were collected in proximity of mosquito trapping sites.

The ERDAS Imagine 9.3 accuracy assessment tool was used to examine what the accuracy points actually represent. Assessment points were compared to the CART classification of those same points. Data from the accuracy assessment were used in an error matrix which provides producer's and user's accuracies along with kappa values for each of the 11 classes and overall accuracy and kappa values calculated for the entire classification product (Table 3) (Congalton 1991).

Table 3.  Error matrix results for the land cover classification performed with 2010 4-band multispectral NAIP data, NDVI data, and 2008 LiDAR CHM data. Thumbnail image of

Statistical analysis

Relationships between seasonal Ae. aegypti and Cx. quinquefasciatus abundance and land cover variables at trapping sites (n=49 for both species) were determined using stepwise regression. The mosquito counts and land cover variables were log transformed and square-root transformed, respectively, to improve assumptions of normality and homogeneity of variance (Gomez and Gomez 1984). Stepwise regression was used to select the significant independent variables (P ≤ 0.15 thresholds for variable inclusion) with a manual backward elimination to remove “noise” variables. Manual backward elimination consisted of iteratively removing non-significant (P > 0.05) independent variables and observing the effect on the error rate of the model. Models for the five buffer zones were compared based on their R2 values to determine the best scale for observing land cover-mosquito abundance relationships for both species. Prior to the stepwise regression, the multicollinearity of the explanatory variables was examined based on the variance inflation factor (VIF). We found high multicollienearity (VIF > 10) for several land cover variables, so multiple models were necessary to analyze them separately. JMP 9 (SAS Institute Inc.) was used for all statistical analysis.

Species presence risk maps

Based on the linear equation derived from the regression analysis, presence risk maps of seasonal Ae. aegypti and Cx. quinquefasciatus were produced using the “raster calculator” tool in ArcGIS 10. Binary raster layers were created for each land cover variable. The presence of each land cover variable within a 60 × 60 m moving window was calculated. Using the land cover layers and the land cover-mosquito abundance relationships, predictive SDMs were created to determine the risk of seasonal Ae. aegypti and Cx. quinquefasciatus presence. The presence risk maps highlight the spatial variability of seasonal mosquito abundance in Tucson. The presence risk maps of seasonal Ae. aegypti and Cx. quinquefasciatus presence were produced at a spatial resolution of 60 m.

Residual plots (estimated mosquito counts were subtracted from the observed mosquito counts) and cross-validation were used to assess the performance of the models. The residual plots provide information on whether the model's error is random and if the model can be improved by the inclusion of additional predictor variables. The models were evaluated by examining model residuals and the relationship between the modeled residual and observed data for both mosquito species. Cross-validation provides information on the generality of the risk maps and involved running ten iterations of the model that omitted a number of random trapping sites (5, 10, 15, 20, 25, 30 traps) and observing the effect on the average R2 value for comparison purposes.

RESULTS

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

Vector abundance

A total of 2,207 adult female Ae. aegypti and 3,944 adult female Cx. quinquefasciatus were collected from 49 trapping sites during the mosquito season. In both years, Cx. quinquefasciatus had an overall higher trapped presence than Ae. aegypti, however only a few sites were more productive in terms of the number of Cx. quinquefasciatus trapped. Individual trap totals varied considerably, even at trapping sites in close proximity (Figures 3a, 3b). Ae. aegypti and Cx. quinquefasciatus were both observed at 47 sites, and only one species was observed at two of the sites. However, at least one species was observed at each trapping site. The spatial distribution of mosquito abundance and trapping sites was not even, with a high percentage of mosquito trapping sites located in the residential neighborhoods near the center of the study area.

image

Figure 3. Distribution of mosquitoes. (a) Distribution of seasonal Ae. aegypti abundance per site; (b) distribution of seasonal Cx. quinquefasciatus abundance per site. The size of the circle represents the number of mosquitoes trapped at each site. High Ae. aegypti counts were observed in residential neighborhoods while high Cx. quinquefasciatus counts were observed in the parks, managed green areas around the university campus, and washes in the residential neighborhoods.

Download figure to PowerPoint

Landscape scale

To determine the best scale at which to define the landscape-abundance relationship, the stepwise regression models were compared based on the models' R2 values (Table 4). Land cover variables account for some of the variation in seasonal Ae. aegypti abundance for all buffer zone radii. Land cover accounts for some of the variation of seasonal Cx. quinquefasciatus abundance at the 30 m, 40 m, and 50 m radii buffer zones but does not account for any variation in seasonal Cx. quinquefasciatus abundance at the 10 m and 20 m radii buffer zones. The 30 m radius buffer zone had the highest R2 values for both Ae. aegypti (0.66) and Cx. quinquefasciatus (0.21). Ae. aegypti abundance showed a stronger relationship (higher R2 values) to land cover variables than Cx. quinquefasciatus abundance at all buffer sizes. The landscape scale results indicate that 60 m (30 m radius) is the best scale at which to observe the land cover-mosquito abundance relationship. The Ae. aegypti model's higher R2 values indicate that land cover plays a more significant role in defining Ae. aegypti habitat and seasonal abundance than it does in defining Cx. quinquefasciatus habitat and abundance.

Table 4.  Overview of the significant land cover variables used to construct the mosquito abundance models for Ae. aegypti (Aedes) and Cx. quinquefasciatus (Culex) at the five buffer radii. Quantities in the spaces are the coefficients assigned to the variables that were significant (P<0.15). This table also shows the model goodness of fit represented by the R2 value for each model.
Land Cover Variables 10m 20m 30m 40m 50m
Aedes Culex Aedes Culex Aedes Culex Aedes Culex Aedes Culex
Structure2.661.782.061.792.32
Pavement–1.192.10
Herbaceous–1.50–2.10
Shrub–5.65
Pool
Water2.26
Bare Earth–1.11
Shadow
Low Tree–0.96
Medium Tree1.323.135.002.854.71 4.60
High Tree1.481.71
All Tree1.27
All Vegetation with Height
All Vegetation
Water Dist (m)
Buffer Avg NDVI0.0003
R2 Value0.580.000.510.000.660.210.560.080.620.18

Land cover-mosquito abundance relationships

The results of the multicollinearity analysis did not identify a collinear relationship between the explanatory land cover variables. The results of the stepwise regressions (Table 4) reveal the relationship between land cover variables and log-transformed seasonal Ae. aegypti and Cx. quinquefasciatus counts at the five buffer sizes. The 30 m radius buffer models indicate that land cover variables explain 66% of the variation in seasonal Ae. aegypti abundance and 21% of the variation in seasonal Cx. quinquefasciatus abundance. The regression models indicate that at the best scale for observing the land cover-mosquito abundance relationship (30 m radius) three land cover variables define the habitat of both species. Structure and medium height tree presence have positive associations with Ae. aegypti abundance, while bare earth presence had negative associations with Ae. aegypti abundance. Pavement and medium height tree presence have positive associations with Cx. quinquefasciatus abundance, while shrub presence had negative associations with Cx. quinquefasciatus abundance.

Predictive mosquito habitat maps

Static maps were created using the land cover-mosquito abundance relationships for Ae. aegypti and Cx. quinquefasciatus to highlight the risk of seasonal mosquito presence in Tucson. The Ae. aegypti presence risk map (Figure 4 a) was created based on the significant predictive land cover variables percent structure, bare earth, and medium tree:

image

Figure 4. Presence risk map of (a) seasonal Ae. aegypti and (b) seasonal Cx. quinquefasciatus presence. Presence risk maps have a spatial resolution of 60m. Risk of presence was determined from the predicted mosquito seasonal abundance. Five risk classes were then defined by equal intervals of the percent seasonal mosquito abundance.

Download figure to PowerPoint

  • image

While the Cx. quinquefasciatus presence risk map (Figure 4b) was created using significant land cover variables pavement, shrub, and medium tree:

  • image

Because the 30 m radius (60 m diameter) buffer zone was identified as the best scale to observe the land cover-mosquito abundance relationship, the presence risk maps were produced at 60 m spatial resolution.

The Ae. aegypti presence risk map shows that Ae. aegypti very high risk areas are isolated to small sections of residential neighborhoods. The majority of residential neighborhoods in the center and eastern portion of the study have moderate risk of seasonal Ae. aegypti presence, while the less developed portions of the study area, along the interstate and the southwestern section, shows very low risk of seasonal Ae. aegypti presence. In comparison to the relatively low risk of seasonal Ae. aegypti presence, the entire study area sees a significantly higher risk of seasonal Cx. quinquefasciatus presence. The Cx. quinquefasciatus presence risk map shows that Cx. quinquefasciatus' very high risk areas are larger and less defined by specific land use designations (e.g., residential neighborhoods). The map indicates that a majority of the study area has at least a moderate risk of seasonal Cx. quinquefasciatus presence, including the less developed southwestern portion of the study area where very low Cx. quinquefasciatus presence would be expected.

Accuracy assessment and model evaluation

Accuracy assessment of the supervised classification of the land cover map found that the classification had an accuracy of 86.3% and a kappa value of 0.85 (Table 3).

Performance of the land cover-mosquito abundance model was assessed examining the pattern of the residual plots (Figures 5a, 5b). Model accuracy was determined by the R2 value of the residual plots. The residuals indicate that the Ae. aegypti model (residual plot R2=0.58) is performing better (smaller residuals) than the Cx. quinquefasciatus model (residual plot R2=0.84; larger residuals indicating an overestimation at lower range of the observed counts). The non-random pattern of the residual plots of both models indicates that the predictor variables are not capturing all of the explanatory information necessary for the most accurate prediction of seasonal mosquito abundance and that additional factors could play a significant role in determining seasonal Ae. aegypti and Cx. quinquefasciatus abundance. Excluding climatic variables in the model accounts for some of models' poor performance; both Ae. aegypti and Cx. quinquefasciatus have known population responses to climate (Hayden et al. 2010, Kolivras and Comrie 2004). Additionally, due the small spatial scale or area in which the mosquito exists, spatial resolution of the image data causes a limitation in pinpointing mosquito presence. The model uses high spatial resolution image data using a large scale approach to the region to identify areas at high and low risk for mosquito presence. However, the model cannot account for landscape variables that are too fine for the image data to pick up. Thus, the model's modest fit can be partially explained by the lack of ground based social factors – such as artificial container coverage that is likely to be present in human modified environments.

image

Figure 5. Model residual plots as a function of (a) Ae. aegypti seasonal abundance and model residuals (b) Cx. quinquefasciatus seasonal abundance and model residuals. Ae. aegypti residual plot has a R2 value of 0.57; Cx. quinquefasciatus residual plot has a R2 value of 0.84. For visualization purposes, the Cx. quinquefasciatus plot x-axis ends at 300, there are points that are not displayed on the plot, however they are included in the calculation of the R2 value.

Download figure to PowerPoint

Cross-validation was performed to test the robustness of the land cover-mosquito abundance model by running ten iterations of the model that omitted a number of random trapping sites (5, 10, 15, 20, 25, 30 traps). The resulting R2 values (Table 5) were compared and the results of all the trapping sites produced the most accurate model of Cx. quinquefasciatus abundance, while the Ae. aegypti model would be improved by using a sample of the trapping data. The absence of an observable drop in the Ae. aegypti R2 values highlights the importance of selecting trapping site locations that are well representative of the potential Ae. aegypti habitat.

Table 5.  Cross-validation of the risk maps involved removing a number of randomly selected mosquito traps. Each iteration of the model was run 10 times and the average R2 value was calculated for comparison. Additionally, the number of times the model was run where an R2 value of 0.00 was calculated is recorded.
# Traps RemovedR2 Value# With No Significant Variables
Aedes Culex Aedes Culex
50.710.1404
100.710.0907
150.730.1305
200.710.0907
250.690.0209
300.740.0907

DISCUSSION

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

This study analyzed the spatial distribution of seasonal Ae. aegypti and Cx. quinquefasciatus abundance and their relationships with land cover in Tucson for the mosquito season. This study demonstrated that (1) urban micro-habitat land cover variables play an important role in the spatial distribution of mosquitoes and (2) the spatial scale of the analysis affects our ability to identify and quantify the land cover-mosquito abundance relationship. Furthermore, this study used the land cover-mosquito abundance relationships to produce presence risk maps of seasonal Ae. aegypti and Cx. quinquefasciatus that highlight the land cover influences on the spatial distribution of mosquitoes in Tucson. Because this study focuses on land cover influences, climatic factors such as temperature and precipitation were not included. However, it is possible that microclimate differences between trapping sites could be significant in explaining some of the observed variance in Ae. aegypti and Cx. quinquefasciatus presence and abundance.

Previous studies have reported that mosquito presence is positively associated with higher vegetation coverage (Zhou et al. 2007, Vezzani et al. 2005). In Tucson, high mosquito activity has been recorded in residential neighborhoods (Fink 1998), areas typically characterized by higher vegetation coverage. By incorporating vegetation height data, this study was able to move beyond general vegetation presence and determine that both species have a distinct preference for the presence of medium height trees (3–9 m).

The results of this study show that land cover partially determines Ae. aegypti and Cx. quinquefasciatus habitat suitability in urban environments. Presence of both Ae. aegypti and Cx. quinquefasciatus at 47 of the 49 sites indicate that the mosquitoes have common habitat requirements, with similar associations to human built environmental variables (structure and pavement) and vegetation cover (medium height trees and/or shrubs). While both species were found at a majority of the trapping sites, the differences in Ae. aegypti and Cx. quinquefasciatus seasonal abundance emphasizes the species' different habitat requirements. The models show that Cx. quinquefasciatus has an overall higher predicted risk of presence than Ae. aegypti throughout the study area. This supports previous findings that while Cx. quinquefasciatus share habitat with Ae. aegypti, Cx. quinquefasciatus typically has a wider range of available habitats (Almiron and Brewer1 1996, Pires and Gleiser 2007).

Comparison of the models indicates that there is a significantly stronger relationship between land cover-Ae. aegypti abundance than land cover-Cx. quinquefasciatus abundance. The stronger performance of the Ae. aegypti model indicates that when not accounting for climatic influences, land cover may play a more significant role in defining Ae. aegypti seasonal abundance than land cover plays in defining it for Cx. quinquefasciatus. This suggests that Ae. aegypti potentially has more stringent habitat requirements that could explain why we have yet to see a dengue outbreak in the Tucson region. With less available potential habitat, Ae. aegypti has yet been unable to produce the large population for a continuous period of time, one environmental factor necessary for dengue to take hold. Similarly, Cx. quinquefasciatus' less stringent habitat requirements, as defined by land cover, means there is plenty of potential habitat and could explain why the mosquito and WNV have successfully established themselves in the Tucson region.

The influence of scale can be seen in the buffer analysis. The best model for both species used the land cover variables from the 30 m radius buffer zone. This implies that if the scale of the analysis is too fine then the heterogeneous nature of the urban landscape will not be adequately defined and the significance of the land cover in determining the spatial variability of seasonal mosquito presence will be underrepresented. Aggregating data to 60 m spatial resolution produces presence risk maps at a far courser scale than the original input data. However, this aggregation does not mean that using lower spatial resolution original data would produce an equivalent presence risk map. Accurate modeling of the mosquito habitat still requires the initial use of high spatial resolution image data in order to produce an accurate land cover classification which can explain the fine-scale spatial relationship between land cover and mosquito abundance.

Successful prediction of the spatial distribution of Ae. aegypti and Cx. quinquefasciatus seasonal abundance could allow vector control efforts to target areas at highest risk for mosquito presence, thereby focusing mosquito control actions to areas that would benefit the most from implementation of mosquito mitigation practices. Further research is needed to determine if the mechanisms behind the observed association between mosquitoes and land cover variables is limited to this particular region and climate. If land cover influences prove to be predictive at other scales and in other regions, this study's findings may be useful in guiding future disease-vector surveillance and mosquito control efforts. Additionally, because land cover does not fully explain the spatial distribution and risk of seasonal presence for either Ae. aegypti or Cx. quinquefasciatus, further research on the urban microclimate and land cover-microclimate interactions is needed to improve the predictive risk models.

Acknowledgments

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED

Mosquito collection and identification was headed by Elizabeth Willott with assistance in weekly trapping provided by Cory Morin. We thank the Pima Association of Governments for sharing the LiDAR data and to the many residents who allowed mosquito trapping on their properties. NAIP data were provided by the USDA's Farm Service Agency (FSA) through the Aerial Photography Field Office. Special thanks to Kyle Hartfield for his invaluable assistance with this work and Drs. Michael Crimmins and Stephen Yool for their comments on the manuscript. Support for this project was provided by the Arizona Remote Sensing Center and NSF grant DEB 0948334.

REFERENCES CITED

  1. Top of page
  2. ABSTRACT:
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgments
  8. REFERENCES CITED
  • Almirón, W.R. and M.E. Brewer. 1996. Classification of immature stage habitats of Culicidae (Diptera) collected in Córdoba, Argentina. Mem. Em. I. Oswaldo Cruz 91: 19.
  • Congalton, R. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens. Environ. 37: 3546.
  • Fink, T., B. Hau, B. Baird, S. Palmer, S. Kaplan, F. Ramberg, D. Mead, and H. Hagedorn. 1998. Aedes aegypti in Tucson, Arizona. Emerg. Infect. Dis. 4: 703704.
  • Focks, D.A., D.G. Haile, E. Daniels, and G.A. Mount. 1993. Dynamic life table model for Aedes aegypti (Diptera: Culicidae): Simulation results and validation. J. Med. Entomol. 30: 10031017.
  • Fuller, D.O., A. Troyo, O. Calderon-Arguedas, and J.C. Beier. 2010. Dengue vector (Aedes aegypti) larval habitats in an urban environment of Costa Rica analysed with ASTER and QuickBird imagery. Int. J. Remote Sens. 31: 311.
  • Ghosh, D. 2011. Geospatial analysis of West Nile Virus (WNV) incidences in a heterogeneous urban environment: A case study in the twin cities metropolitan area of Minnesota. In: J.A. Maantay and S. McLafferty (eds.) Geospatial Analysis of Environmental Health, pp. 153169. Springer Netherlands , New York .
  • Gomez, K.A. and A.A. Gomez. 1984. Data that violate some assumptions of the analysis of variance. In: Statistical Procedures for Agricultural Research, 2nd ed. pp. 294315. John Wiley & Sons, New York .
  • Hartfield, K.A., K.I. Landau and W.J.D. van Leeuwen. 2011. Fusion of high resolution aerial multispectral and LiDAR data: Land cover in the context of urban mosquito habitat. Remote Sens. 3: 23642383.
  • Hayden, M.H., C.K. Uejio, K. Walker, F. Ramberg, R. Moreno, C. Rosales, M. Gameros, L.O. Mearns, E. Zielinski-Gutierrez, and C.R. Janes. 2010. Microclimate and human factors in the divergent ecology of Aedes aegypti along the Arizona, US/Sonora, MX Border. EcoHealth 7: 6477.
  • Hoeck, P., F. Ramberg, S. Merrill, C. Moll, and H. Hagedorn. 2003. Population and parity levels of Aedes aegypti collected in Tucson. J. Vector Ecol. 28: 6573.
  • Kolivras, K.N. and A.C. Comrie. 2004. Climate and infectious disease in the southwestern United States. Prog. Phys. Geog. 28: 387398.
  • Kolivras, K.N. 2006. Mosquito habitat and Dengue risk potential in Hawaii: A conceptual framework and GIS application. Prof. Geogr. 58: 139154.
  • Liu, H., Q. Weng and D. Gaines. 2008. Spatio-temporal analysis of the relationship between WNV dissemination and environmental variables in Indianapolis, USA. Int. J. Hlth. Geogr. 7: 66.
  • Morin, C. and A.C. Comrie. 2010. Modeled response of the West Nile virus vectorCulex quinquefasciatus to changing climate using the dynamic mosquito simulation model. Int. J. Biometeorol. 54: 517529.
  • Pires, D.A. and R.M. Gleiser. 2007. Cuerpos de agua temporarios y permanentes como habitats larvales de mosquitos (Diptera: Culicidae) en la ciudad de Cordoba. Biol. Acuat. 23: 68.
  • Reiter, M.E. and D.A. Lapointe. 2007. Landscape factors influencing the spatial distribution and abundance of mosquito vector Culex quinquefasciatus (Diptera: Culicidae) in a mixed residential-agricultural community in Hawai'i. J. Med. Entomol. 44: 861868.
  • Secord, J. and A. Zakhor. 2007. Tree detection in urban regions using aerial Lidar and image data. IEEE Geosci. Remote S. 4: 196200.
  • STGS. 2009. Pima Association of Governments 2008 Digital Orthophotography and LiDAR Report. Sanborn Total Geospatial Solutions.
  • Tucker, C.J. 1979. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8: 127150.
  • Vezzani, D., A. Rubio, S.M. Velazquez, N. Schweigmann, and T. Wiegand. 2005. Detailed assessment of microhabitat suitability for Aedes aegypti(Diptera:Culicidae) in Buenos Aires, Argentina. Acta. Trop. 95: 123131.
  • Zhou, G., S. Munga, N. Minakawa, A.K. Githeko and G. Yan. 2007. Spatial relationship between adult malaria vector abundance and environmental factors in western Kenya highlands. Am. J. Trop. Med. Hyg. 77: 2935.