Modeling Culex tarsalis Abundance on the Northern Colorado Front Range Using a Landscape-Level Approach

Transcription

1 Modeling Culex tarsalis Abundance on the Northern Colorado Front Range Using a Landscape-Level Approach Author(s): Jessica A. Schurich ; Sunil Kumar ; Lars Eisen ; and Chester G. Moore Source: Journal of the American Mosquito Control Association, 30(1): Published By: The American Mosquito Control Association DOI: URL: BioOne ( is a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses. Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne s Terms of Use, available at terms_of_use. Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder. BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research.

2 Journal of the American Mosquito Control Association, 30(1):7 20, 2014 Copyright E 2014 by The American Mosquito Control Association, Inc. MODELING CULEX TARSALIS ABUNDANCE ON THE NORTHERN COLORADO FRONT RANGE USING A LANDSCAPE-LEVEL APPROACH JESSICA A. SCHURICH, 1,2 SUNIL KUMAR, 3 LARS EISEN 4 AND CHESTER G. MOORE 4 ABSTRACT. Remote sensing and Geographic Information System (GIS) data can be used to identify larval mosquito habitats and predict species distribution and abundance across a landscape. An understanding of the landscape features that impact abundance and dispersal can then be applied operationally in mosquito control efforts to reduce the transmission of mosquito-borne pathogens. In an effort to better understand the effects of landscape heterogeneity on the abundance of the West Nile virus (WNV) vector Culex tarsalis, we determined associations between GIS-based environmental data at multiple spatial extents and monthly abundance of adult Cx. tarsalis in Larimer County and Weld County, CO. Mosquito data were collected from Centers for Disease Control and Prevention miniature light traps operated as part of local WNV surveillance efforts. Multiple regression models were developed for prediction of monthly Cx. tarsalis abundance for June, July, and August using 4 years of data collected over The models explained monthly adult mosquito abundance with accuracies ranging from 51 61% in Fort Collins and 57 88% in Loveland Johnstown. Models derived using landscape-level predictors indicated that adult Cx. tarsalis abundance is negatively correlated with elevation. In this case, lowelevation areas likely more abundantly include habitats for Cx. tarsalis. Model output indicated that the perimeter of larval sites is a significant predictor of Cx. tarsalis abundance at a spatial extent of 500 m in Loveland Johnstown in all months examined. The contribution of irrigated crops at a spatial extent of 500 m improved model fit in August in both Fort Collins and Loveland Johnstown. These results emphasize the significance of irrigation and the manual control of water across the landscape to provide viable larval habitats for Cx. tarsalis in the study area. Results from multiple regression models can be applied operationally to identify areas of larval Cx. tarsalis production (irrigated crops lands and standing water) and assign priority in larval treatments to areas with a high density of larval sites at relevant spatial extents around urban locations. KEY WORDS Colorado, Culex tarsalis, Geographic Information System, landscape, West Nile virus INTRODUCTION A Geographic Information System (GIS) has the capacity to integrate environmental and ecological factors with entomological and virological data for improved techniques in vector management (Beck et al. 1994, Dale et al. 1998). Further, the usefulness of models based on remote sensing and/or GIS-derived data to predict vector presence or abundance, or the level of risk for vector-borne diseases, has been well documented (Hay et al. 1998, Eisen and Eisen 2011). Landscape-level models, which consider both the diversity of the landscape and the impact of specific landscape features on the abundance of a species, can provide an improved understanding of the spatial patterns of species abundance with strong linkages to specific environments (Aukema et al. 2006, Gottschalk et al. 2007, Otte et al. 2007, Chuang et al. 2012). 1 Colorado Mosquito Control, Inc., Brighton, CO Department of Bioagricultural Sciences and Pest Management, Colorado State University, Fort Collins, CO Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO Department of Microbiology, Immunology and Pathology, Colorado State University, Fort Collins, CO Indeed, the behaviors and habitat preferences of a vector species of interest are critical considerations when modeling species distribution or abundance (Ruiz et al. 2010). Culex tarsalis Coquillett is recognized as the primary species of concern in the transmission of West Nile virus (WNV) to humans in Colorado (Bolling et al. 2007, Gujral et al. 2007). The 3- county area including Boulder, Larimer, and Weld counties in Colorado accounted for 46.2% (1,369) of the 2,965 human cases reported in Colorado during 2003, 7.9% (287) of the 3,630 human cases reported in Colorado during 2007, and 57.2% (181) of the 316 human cases reported in Colorado in 2013 (data still being collected for 2013; US Geological Survey). This 3-county area on the northern Colorado Front Range should be considered a high-risk setting for future WNV infections in humans. We anticipate that, as with any biological system, there is a combination of the interactions between climatological variables, the epizootic cycle, and landscape heterogeneity that drive vector abundance and WNV amplification within this region. Studies that have included data for weekly and monthly climatological conditions have demonstrated positive associations between temperature and Cx. tarsalis abundance, likely in part due to a shortened gonotrophic cycle and faster development of immatures at higher temperatures (Reisen 7

3 8 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30,NO. 1 Fig. 1. Mosquito trap locations within the spatial modeling area of the northern Colorado Front Range. The municipalities of Fort Collins and Loveland are located in eastern Larimer County and Johnstown in western Weld County. et al. 2010, Chuang et al. 2011). On the northern Colorado Front Range, a weak correlation was described between summer precipitation and abundance of Cx. tarsalis, while inclusion of mean weekly temperature improved abundance models (Brown et al. 2011). With regard to snow water content and rainfall, Wegbreit and Reisen (2000) found a strong relationship in the San Joaquin Valley of California between adult Cx. tarsalis abundance and snow depth, snow water content, and river runoff, while abundance was not strongly correlated with rainfall. It seems plausible that a similar relationship may exist on the northern Colorado Front Range, where much of the water used for irrigation is dependent on snowpack. This water supply is diverted variably throughout the summer months via a network of irrigation canals and reservoirs, and demand can vary with drought conditions and rainfall on a seasonal basis. Climatological conditions can differ across a landscape dependent on the spatial heterogeneity and topography of the environment. Improved predictive power for models of risk for exposure to vectors or vector-borne pathogens can also be achieved through consideration of landscapelevel heterogeneity and the impacts of vegetative differences across a landscape (Beck et al. 1994, Diuk-Wasser et al. 2006, Barker et al. 2009b, Ruiz et al. 2010). Previous work has identified spatial factors, including proximity to larval habitats and human population density, which modify the risk of WNV infection in humans (Winters et al. 2008, Eisen et al. 2010). Culex tarsalis adults are associated with specific habitats, for example riparian ecotones in rural environments of the San Joaquin Valley of California (Reisen et al. 1992) or areas with the presence of grass or hay (Chuang et al. 2011). In the northern Colorado Front Range, Cx. tarsalis abundance was lower in traps surrounded by 2 to 3 land cover classes than for traps surrounded by 1 or by 4 to 5 land cover classes measured at a 50- m scale around traps (Barker et al. 2009b). The work by Barker et al. (2009b) describes differences in fine-scale landscape use by Cx. tarsalis and a possible preference for vegetation. Further, the strongest predictive power for Cx. tarsalis abundance near the Cache la Poudre River on the northern Colorado Front Range resulted when a 100-m buffer was applied to data from the 2001 National Land Cover Dataset (Maki 2005). The spatial distribution of larval habitats also likely impacts the dispersal of adult mosquitoes. In the northern Colorado Front Range, distances to

4 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 9 Table 1. Compilation of landscape variables used in the spatial analysis and modeling of Culex tarsalis abundance. Data layer resolutions for independent variables are listed in parentheses. The buffer distance around the trap locations provides a comparison of spatial extent for landscape predictors. Buffer distance around trap location Source Continuous variables Slope in degrees (10 m) Mean 500 m Aspect in degrees (10 m) Majority 500 m Elevation (10 m) Mean 500 m and 250 m Topographic exposure (10 m) Mean 500 m Impervious surface (30 m) Mean 500 m and 250 m Distance to larval sites (1 m) Mean at 1,000 m, 500 m, 250 m, 175 m Colorado Mosquito Control, Inc. Perimeter of larval sites (1 m) Sum at 1,000 m, 500 m, 250 m, 175 m Colorado Mosquito Control, Inc. Distance to irrigated lands (1 m) Mean at 1,000 m and 500 m Northern Colorado Water Conservancy District Area of irrigated lands (1 m) Sum within 1,000 m and 500 m Northern Colorado Water Conservancy District Normalized difference vegetation index (250 m) Mean and standard deviation at 500 m and 250 m projects/weld Categorical variables LANDFIRE vegetation class (30 m) Majority 500 m and 250 m Water source of larval Majority at 1,000 m, 500 m, Colorado Mosquito Control, Inc. habitats (1 m) 250 m, 175 m Larval habitat type (1 m) Majority at 1,000 m, 500 m, 250 m, 175 m Colorado Mosquito Control, Inc. larval habitats derived from the Colorado Gap Analysis Project were not correlated with Cx. tarsalis abundance within a 400-m range, while Aedes vexans (Meigen) abundance decreased significantly with increasing distance from larval habitats (Barker et al. 2009b). These findings suggest that Cx. tarsalis commonly disperses.400 m from larval habitats, possibly along specific dispersal corridors related to fine-scale habitat heterogeneity and permanency of water. In California, Cx. tarsalis have been found to be very dispersive and hunt along riparian corridors or vegetative transitions (Bailey et al. 1965). We selected spatial predictors for this study based on the previous findings for environmental variables, as well as the biological relevance of these predictors to the distribution and larval habitat preference of Cx. tarsalis. Immatures of this mosquito species are associated with clear or foul water in irrigation systems, corrals, or slaughter yards, and with pools in streambeds or fresh or saline riparian wetlands (Carpenter and LaCasse 1955, Reisen 1993). Spatial predictors were also chosen from previous work performed in regions with similar topographic and land-use patterns as the northern Colorado Front Range. Water use and availability in the study area is highly dependent on snowmelt runoff and storage in reservoirs. The depth of water which fills many irrigated lands and reservoirs can vary on a weekly, monthly, and yearly basis depending on demand and average daily temperature. An improved understanding of the factors that drive vector mosquito abundance at the landscape level is critical in managing risk of human exposure to WNV-infected mosquitoes. It also will aid in creating effective Integrated Pest Management programs, including programs that minimize pesticide applications. Our primary objective in this study was to utilize raster-based environmental covariates in a GIS to determine the effects of landscape heterogeneity on mean monthly abundance of Cx. tarsalis adults on the northern Colorado Front Range (i.e., the municipalities of Fort Collins and Loveland Johnstown). Our objective was to make inferences about the importance of landscape variables that contribute to the population buildup of Cx. tarsalis, using observed monthly abundance data collected from , to provide future predictions of vector abundance. MATERIALS AND METHODS Study area and data collection Our study area encompassed the cities of Fort Collins and Loveland in Larimer County, CO, and the town of Johnstown in Weld County (Fig. 1). The elevation across these municipalities ranges from 1,448 1,562 m above sea level. The confluence of the Cache la Poudre River and Big Thompson River, which emerge from the Rocky Mountains in western Larimer County, drain into the South Platte River in neighboring Weld County. The river corridors are lined with cottonwood (Populus spp.) and willow (Salix spp.), and often include oxbows that are filled with sediment and contain dense patches of cattails (Typha spp.).

5 10 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30,NO. 1 The landscape in the study area is heterogeneous, with urban to suburban irrigated residential vegetation dominant within city limits and irrigated agriculture comprising much of the lands just east of the municipal boundaries (Brown et al. 2011). Notably, the water surface area in Loveland is approximately 2.5 times that of neighboring Fort Collins (Gujral et al. 2007), as a result of an extensive reservoir storage system in the northeastern portion of Loveland and associated irrigated lands on the outer limits of the city. The climate of the study area is characterized by cold winters and hot summers with low humidity (Winters et al. 2008). Conditions are semiarid and precipitation is variable. The average annual rainfall in Fort Collins from 1971 to 2000 was 393 mm (Mountain States Weather Services, Fort Collins, CO). Adult mosquitoes were collected using Centers for Disease Control and Prevention miniature light traps (John W. Hock Company, Gainesville, FL) that were suspended m above the ground and operated from afternoon ( h) until morning ( h). Each trap was baited with approximately 1.8 kg of dry ice, and was operated with a 6-V battery and 2,180- rpm motor fan. Weekly trapping was conducted at 41 fixed locations in Fort Collins, 37 fixed locations in Loveland, and 4 fixed locations in Johnstown during June through August The surveillance monitoring network for the municipalities of Fort Collins and Loveland includes traps located approximately 1.3 km apart across city limits in areas that provide suitable harborage and proximity to larval mosquito habitats. Mosquitoes were recovered from the traps each morning, immobilized with dry ice, and transported to the laboratory for identification to species using a published key (Darsie and Ward 2005). Monthly averages for Cx. tarsalis adults were calculated from weekly data collected during June, July, and August of Environmental GIS data The spatial modeling area was digitized to extend 8.05 km (5.0 mi) beyond the larval control boundaries for the targeted municipalities. The locations of these boundaries were provided by the mosquito control contractor, Colorado Mosquito Control, Inc. (CMC), Brighton, CO. The spatial modeling area encompasses 367,195 ha of land and water within Larimer and Weld counties. All data layers were geoprocessed and clipped to the spatial modeling area using ArcGIS version 10 (ESRI, Redlands, CA). All values for geoprocessed GIS data layers with associated buffer radii (Table 1) were extracted in the Universal Transverse Mercator projection (Zone 13N, NAD 1983 datum). LANDFIRE land cover classification data: LANDFIRE uses land cover classifications defined by NatureServes s ecological systems classifications (Ryan and Opperman 2013). The 30-m-resolution LANDFIRE raster was obtained and the categorical values were reclassified from 205 attribute classes to 8 classes: open water, developed open space, nonsuitable habitat, barren land, pasture and hay, cultivated crops, woody wetland, and herbaceous/introduced wetland following the classification methods described by Eisen et al. (2010). Majority land cover class was extracted at 250-m and 500-m buffer radii using the Zonal Statistics tool in ArcGIS (Table 1). Topographic variables: The Digital Elevation Model (DEM; 10-m resolution) raster (Gesch 2007) was used to extract topographic variables, including mean elevation, majority aspect class, and mean slope (Table 1). We used the Surface Tool in ArcGIS to generate rasters for slope and aspect (in degrees). The Focal Statistics tool in ArcGIS was used to derive mean elevation and slope at 250-m and 500-m extents using the DEM. For the analysis of aspect, we reclassified the majority aspect into 8 equal intervals (45u each). Topographic exposure was calculated as the difference between the elevations from 10-m DEM and 500-m buffer radius raster. We used the Raster Calculator in ArcGIS to create the topographic exposure raster from which the mean value for topographic exposure was extracted for all trap locations. Topographic exposure can identify areas of exposure or protection, oftentimes from wind, across the landscape structure (Mikita and Klimanek 2010). Impervious surface: The 2006 National Land Cover Dataset (NLCD) Percent Developed Impervious Surface was downloaded and converted to a 30-m raster (Table 1). The Statistics tool in ArcGIS was used to calculate mean value within a 500-m radius buffer around trap locations using the 30-m NLCD raster. Impervious surface can identify areas of urban development and more densely populated settings. Perimeter of and distance to larval mosquito sites: To assess the impact of perimeter of larval mosquito sites on abundance of Cx. tarsalis adults, the polygon shape file of larval mosquito sites (Table 1) was converted to a polyline file using the Features tool in ArcGIS. The polyline to raster conversion tool was then used to generate a surface raster from which the total perimeter of larval mosquito sites was extracted at 4 spatial extents. The area of land included in each buffer radii was 9.6 ha (175 m), 19.6 ha (250 m), 78.5 ha (500 m), and ha (1,000 m). Pixel counts were summed for the total perimeter of larval sites included within the buffer radii, and the values were extracted for each trap location using Zonal Statistics in ArcGIS. The rationale

6 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 11 behind the selection of perimeter, rather than total water surface, was that open surface water such as large reservoirs with little vegetation contribute minimally to larval mosquito abundance (with the exception of the shallow perimeter, which often holds vegetation) when compared to a line of vegetation, such as cattails, along a ditch that may impede flow and serve as a viable habitat for larval Cx. tarsalis. The mean distance to the perimeters of larval sites for the 4 abovementioned spatial extents around trap locations were obtained by converting the polygon shape file for larval mosquito sites, obtained from CMC, to a 1.0-m raster. The Euclidean Distance tool in ArcGIS was used to calculate the mean distance to perimeters of larval mosquito sites from each trap location. Habitat type and water source for larval mosquito habitats: The habitat type and water source categories for each larval habitat, which was classified by CMC, was joined with the larval mosquito site polygon shape file obtained from CMC (Table 1). Colorado Mosquito Control, Inc. uses the following categorical assignment to designate the type of larval habitats to which a numerical value was assigned for extraction of majority type: Temporary standing water (1), Irrigated field (2), Lake (3), Marsh (4), Swamp (5), Riparian (6), Ditches (7), Depressions (8), and Retention ponds (11). Majority water source categories as described by CMC were designated as follows: Flooding (1), Groundwater seepage (2), Irrigation (3), Manually controlled waters (4), Other types (5), Rain (6), Seepage (7), and Residential watering (8). The shape files for water source and habitat types were converted to a 1.0- m raster. The majority habitat and water source within the 4 buffer radii were extracted for each trap location. The larval control programs for the municipalities of Fort Collins, Loveland, and Johnstown have been in place for multiple years and many of the larval sites have been identified around the traps. It is important to note, however, that the size of these habitats may change with water use and rainfall on a weekly basis. The habitat type and water source layers for larval mosquito sites were not continuous data layers and were applied as categorical layers to identify those habitats that are important contributions to Cx. tarsalis abundance. Colorado Mosquito Control, Inc. did not specifically measure the size of larval habitats at the time of this study. Irrigated lands: Data for irrigated croplands were obtained for Division 1 from the Northern Colorado Water Conservancy District, Berthoud, CO, for 2007 (Table 1). The polyline file of irrigated croplands was converted to a 1.0-m raster. Distance to and area of irrigated croplands was obtained by calculating the Euclidean distance from each trap location to the nearest irrigated lands at buffer radii of 500 m and 1,000 m. Data for the irrigated lands were static and did not provide any information about water use on a seasonal basis. At the time of this modeling approach, dynamic data for area of irrigated lands were not available. Abundance modeling: Four-year ( ) adult mosquito abundance data were averaged for June, July, and August months and abundance models were run separately for Fort Collins (n 5 41) and Loveland Johnstown (n 5 41). Separate models were developed because of the differences in the integrated mosquito management programs used by these municipalities. Each of the communities performs larval mosquito control within municipal boundaries to reduce larval mosquito populations. The municipalities of Loveland and Johnstown also perform targeted mosquito adulticiding around surveillance trap locations when.50 adult Culex spp. mosquitoes (typically composed of Cx. tarsalis, Cx. salinarius Coquillett, and Cx. pipiens L.) are collected in 1 night. The city of Fort Collins does not perform mosquito adulticiding unless elevated levels of WNV activity in mosquitoes, humans, or birds is considered to warrant a public health response. Additionally, Loveland and Johnstown are located along the drainage of the Big Thompson River, whereas the Cache la Poudre River flows through Fort Collins. Landscape-level predictors were log transformed after geoprocessing in ArcGIS as determined by the distribution, using log 10 transformation, to achieve close to normal distributions. Pearson s correlation coefficient was used to evaluate the relationships between landscape predictors and mean monthly abundance of mosquitoes in Systat (Version 12; SYSTAT Software Inc., Chicago, IL). Spearman rank correlation was used to identify the strength of correlations between categorical landscape predictors and mean monthly abundance of mosquitoes. Multicollinearity was examined using Systat and only 1 variable from a set of highly correlated variables (r $ 0.75) was used in the multiple regression models. We used lm and stepaic functions in the MASS package in R (R Core Team 2013) for model development and model selection. We used stepaic to identify the relevant spatial extent within the heterogeneous landscape that had the greatest impact on abundance of Cx. tarsalis adults. Correlation coefficients were not evaluated for categorical variables; rather, we selected stepaic to evaluate the significance of land-use patterns, larval habitat type, and water source data on Cx. tarsalis abundance. The continuous or categorical landscape variables with the most predictive power based on Akaike s Information Criterion and multiple R 2 were used to build monthly multiple regression models. These multiple regression models assessed the variation in Cx.

7 12 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30,NO. 1 Fig. 2. Yearly comparison for mean Culex tarsalis abundance (x axis) per month. Weekly data were collected from fixed surveillance trap locations to obtain monthly averages for the years (a) 2007, (b) 2008, (c) 2009, and (d) 2010 in Fort Collins (FC) and Loveland Johnstown (LV-JT), CO.

8 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 13 Table 2. Multiple regression models for June, July, and August months describing variation in Culex tarsalis adult abundance with significant spatial extents of landscape-level predictors in Fort Collins, CO. Best models for all 3 months were selected using Akaike s Information Criterion. Variables in different models are arranged in order of their decreasing relative importance based on partial R 2 values. Model/Variable Parameter estimate Partial R 2 P-value June (adjusted R ; P, ) (Intercept) 31, Majority water type (manually controlled) (1,000 m) , Mean elevation (500 m) Mean elevation 2 (500 m) Majority water type (irrigation) (1,000 m) Perimeter of larval sites (1,000 m) July (adjusted R ; P, ) (Intercept) 1, , Majority water type (manually controlled) (1,000 m) , Mean elevation (500 m) Majority habitat type (depressions) (1,000 m) Impervious surface (250 m) Majority water type (groundwater seepage) (1,000 m) Distance to larval sites (1,000 m) August (adjusted R ; P, ) (Intercept) , Majority water type (manually controlled) (1,000 m) , Distance to irrigated lands (500 m) Majority habitat type (depressions) (1,000 m) Majority land cover type (developed open space) (500 m) Table 3. Multiple regression models for June, July, and August months describing variation in Culex tarsalis adult abundance with significant spatial extents of landscape-level predictors in Loveland Johnstown, CO. Best models for all 3 months were selected using Akaike s Information Criterion. Variables in different models are arranged in order of their decreasing relative importance based on partial R 2 values. Model/Variable Parameter estimate Partial R 2 P-value June (adjusted R ; P, ) (Intercept) Area of irrigated lands (500 m) , Perimeter of larval sites (500 m) Impervious surface (250 m) Majority land cover type (developed low intensity) (500 m) July (adjusted R ; P, ) (Intercept) Majority habitat type (retention ponds) (1,000 m) , Area of irrigated lands (500 m) , Perimeter of larval sites (500 m) Mean elevation (500 m) Majority land cover type (developed low intensity) (500 m) August (adjusted R ; P, ) (Intercept) , Majority habitat type (retention ponds) (1,000 m) , Area of irrigated lands (500 m) , Mean elevation (500 m) , Perimeter of larval sites (500 m) Majority land cover type (emergent herbaceous wetlands) (500 m) Distance to irrigated lands (500 m)

9 14 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30,NO. 1 Table 4. Pearson correlation coefficients for static predictors from univariate analysis in Systat for Fort Collins, CO. Coefficients were obtained for the relationship between landscape predictors and average monthly adult abundance for Culex tarsalis during Correlation coefficients that were statistically significant (P, 0.05) are highlighted in bold. Landscape predictor June abundance July abundance August abundance Elev 500 m Elev 250 m Elev 500 m STD Topo Exp Impervious 500 m Impervious 250 m Slope Log slope Majority aspect Dist larval 1,000 m Dist larval 500 m Log dist larval 500 m Dist larval 250 m Log dist larval 250 m Dist larval 175 m Log dist larval 175 m Perim larval 1,000 m Log perim larval 1,000 m Perim larval 500 m Log perim larval 500 m Perim larval 250 m Log perim larval 250 m Perim larval 175 m Log perim larval 175 m Dist irr 500 m Area irr 500 m Log area irr 500 m Dist irr 1,000 m Area irr 1,000 m Log area irr 1,000 m tarsalis abundance and how landscape variables, at multiple spatial extents, may affect the seasonal abundance pattern of this vector species across the municipalities. RESULTS Adult mosquito abundance data obtained from CMC for indicates that July contributes the largest proportion of adult mosquitoes on average over a season (i.e., June August) in all years included in this study. Of the total 15,607 Cx. tarsalis mosquitoes collected in Fort Collins during , 64.3% were collected during the month of July. A similar pattern was seen in Loveland Johnstown, with 58.1% of the total 16,312 Cx. tarsalis mosquitoes collected during July. Additionally, seasonal variation in mean monthly abundance of Cx. tarsalis occurred in each of the examined years (Fig. 2). While a comparable overall number of Cx. tarsalis mosquitoes were collected in Fort Collins and Loveland Johnstown, there were differences in mean monthly abundance across these municipalities. The observed variation in abundance data highlights the importance of spatial and temporal patterns, likely associated with climate and landscape heterogeneity within the study area. Model performance In general, the models for Cx. tarsalis abundance derived from static landscape predictors performed well and provided information about the directionality of the relationships between predictors and vector abundance. Furthermore, the results indicated that the spatial extent and significance of landscape predictors varied between Fort Collins and Loveland Johnstown. In Fort Collins, the June model explained 61.0% (P, ) of the variation in abundance of Cx. tarsalis adults, compared with 51.0% (P, ) of the variation in July and 58.0% (P, ) of the variation in August (Table 2). In Loveland Johnstown, the June model explained 57.0% (P, ) of the variation in abundance of Cx. tarsalis adults, compared with 83.0% (P, ) of the variation in July and 88.0% (P, ) of the variation in August (Table 3).

10 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 15 Fig. 3. Relationships between decreasing elevation and distance to irrigated lands and mean monthly Culex tarsalis abundance in Fort Collins, CO. Effects of topography on Cx. tarsalis abundance The mean monthly abundance of Cx. tarsalis adults was consistently negatively correlated with elevation within 250-m and 500-m buffers around trap locations in Fort Collins (Table 4 and Fig. 3) and Loveland Johnstown (Table 5 and Fig. 4) for all months examined from univariate analysis. A stronger correlation was observed between Cx. tarsalis abundance and elevation at a spatial extent of 500 m versus 250 m from StepAIC. No significant relationship was detected between mean monthly abundance of Cx. tarsalis adults and majority aspect or slope in either Fort Collins or Loveland Johnstown in any month. We found a positive correlation between mean monthly abundance of Cx. tarsalis adults and topographic exposure for buffer distances of 500 m around trap locations in Fort Collins (r , P ) (Table 4) and Loveland Johnstown (r , P ) during August (Table 5). Impervious surfaces at a spatial extent of 500 m (r , P ) and 250 m (r , P ) were significantly negatively correlated with Cx. tarsalis abundance in June in Fort Collins (Table 4). Differences in model predictions Variable results were obtained for Fort Collins versus Loveland Johnstown in the analyses of impact of larval habitat type, water source, or the mean distance to or perimeter of larval sites on mean monthly abundance of Cx. tarsalis adults. For example, we found that the perimeter of larval mosquito sites within a 500-m buffer around trap locations was a significant predictor of Cx. tarsalis abundance in June, July, and August in Loveland Johnstown (P, 0.05) (Table 3). The relationship between the perimeter of larval sites and mean monthly abundance was positive in June, July, and August (Table 3). The negative relationship between distance to irrigated lands and mean monthly abundance of Cx. tarsalis adults at a spatial extent of 500 m was significant in August in Fort Collins (P, 0.05) (Table 2). The distance to irrigated lands within a 500-m buffer around trap locations was a significant predictor of Cx. tarsalis abundance in August in Loveland Johnstown (P, 0.05) (Table 3). In both Fort Collins and Loveland Johnstown we found increases in mosquito abundance with decreasing distance to irrigated croplands (Figs. 3 and 5). Of the 9 categorical variables for larval mosquito habitat types evaluated at 175 m, 250 m, 500 m, and 1,000 m from trap locations, depressions in July (P ) and August (P ) at a spatial extent of 1,000 m were notable predictors in mean monthly abundance models for Cx. tarsalis adults in Fort Collins. In Loveland Johnstown, retention ponds improved the predictive power of abundance models in July (P, ) and August (P, ) at a spatial extent of 1,000 m. Of the 8 categorical variables for majority of water source, the following were

11 16 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30,NO. 1 Table 5. Pearson correlation coefficients for static predictors from univariate analysis in Systat for Loveland Johnstown, CO. Coefficients were obtained for the relationship between landscape predictors and mean monthly adult abundance for Culex tarsalis. Correlation coefficients that were statistically significant (P, 0.05) are highlighted in bold. Landscape predictor June abundance July abundance August abundance Elev 500 m Elev 250 m Elev 500 m STD Topo Exp Impervious 500 m Impervious 250 m Slope Log slope Majority aspect Dist larval 1,000 m Log dist larval 1,000 m Dist larval 500 m Log dist larval 500 m Dist larval 250 m Log dist larval 250 m Dist larval 175 m Log dist larval 175 m Perim larval 1,000 m Log perim larval 1,000 m Perim larval 500 m Log perim larval 500 m Perim larval 250 m Log perim larval 250 m Perim larval 175 m Log perim larval 175 m Dist irr 500 m Area irr 500 m Log area irr 500 m Dist irr 1,000 m Area irr 1,000 m Log area irr 1,000 m notable predictors in mean monthly abundance models for Cx. tarsalis adults in Fort Collins: manually controlled water in June (P, ), July (P, ), and August (P, ). Further, the majority of land cover type from LANDFIRE vegetation class, within a 500-m buffer of trap locations, contributed to the predictive model for Loveland Johnstown for the following classes: emergent herbaceous wetlands in August (Table 3). DISCUSSION Our study underscores the importance of landscape variables at different spatial extents to impact Cx. tarsalis abundance in both space and time. While we recognize that weather events can impact the population buildup of Cx. tarsalis, we did not include such variables in our modeling efforts due to the lack of GIS-based weather data of a sufficiently fine scale to provide comparisons across the trapping network of the study area. Important landscape-level variables were identified based on mosquito abundance training data through the use of multiple regression modeling. The relationship between these variables and their impact on monthly mosquito abundance can be applied operationally to target control efforts based on the landscape dynamics. The results from observed abundance data indicate that the month of July contributes significantly to the population buildup of Cx. tarsalis and suggests that it is the spike in vector populations during July that drive WNV transmission in July and August. Bolling et al. (2007) similarly found that Cx. tarsalis peaks in early July along the plains of northeastern Colorado. Understanding this population increase can allow for the selection of long-term larvicides at larval sites where water levels have dropped following peak snowmelt runoff during June, pretreatments in areas prone to irrigation, or application of adulticides in areas where adult abundance is higher during July. Efforts to model the abundance of adult Cx. tarsalis on a monthly basis using landscape-level predictors in Larimer County indicated that adult Cx. tarsalis abundance is negatively correlated with elevation. This suggests, and is not a novel idea, that Cx. tarsalis abundance increases with decreasing elevation within our study area, and is

12 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 17 Fig. 4. Relationships between decreasing elevation and distance to irrigated lands and mean monthly Culex tarsalis abundance in Loveland Johnstown, CO. consistent with spatial patterns for increasing WNV disease incidence with decreasing elevation (Winters et al. 2008). Elevation may work in combination with slope of the local landscape to produce flat environments at lower elevations well suited for development of larval Cx. tarsalis because of poor drainage and the clustering of larval habitats. This is in accordance with the results from Barker et al. (2009a), where higher abundance of Cx. tarsalis in the northern Colorado Front Range was recorded between elevations of 1,200 1,450 m, compared with elevations above 1,450 m. Increasing elevations are associated with increased slope along the foothills and in canyons that lead to the Continental Divide. Mosquito species richness in northern Colorado has also been found to be higher for elevations below 1,600 m (Eisen et al. 2008), which is likely in part a result of the presence of suitable habitats with favorable, warmer microclimates at lower elevations. While the elevation variation within our study area, based on 500-m buffers around trap locations, was relatively small 1,487 1,563 m for Fort Collins and 1,456 1,561 m for Loveland Johnstown the variation in mosquito abundance data across the landscape indicates that there are areas that pose greater potential for the presence of larval mosquito sites related to changes in elevation and slope. Furthermore, the inclusion of elevation at a spatial extent of 500 m around mosquito surveillance trap locations in abundance models suggests that elevation change is a proxy for the distance to major Cx. tarsalis habitats. The inclusion of elevation in June and July models in Fort Collins, but not August suggests that mountain runoff and irrigation early season may impact larval production by increasing the surface area of aquatic sites. We postulate that migration by adult Cx. tarsalis across rural areas closely associated with irrigated crops and a dispersal corridor into more developed areas occurs along the periphery of the urban rural interface because of the viable larval habitats, cooler environments, and better harborage for adult Cx. tarsalis these settings provide. Our work provides additional support for a positive association between the area of irrigated lands and mean monthly abundance of Cx. tarsalis adults, which is a plausible finding given that increases in surface water are likely to create variable larval mosquito habitats on a temporal scale. In Loveland Johnstown the association between the area of irrigated lands and Cx. tarsalis abundance was positive at a spatial extent of 500 m during June, July, and August, suggesting that increases in surface water is a proxy for increased Cx. tarsalis abundance. The negative correlation between the distance of irrigated lands and adult Cx. tarsalis populations in Fort Collins during August provides a biological understanding for the importance of irrigated tail waters to serve as significant aquatic larval habitats to Cx. tarsalis mosquitoes. It is important to note that differences in the parameter estimates and strength of predictive power for irrigated lands may be a result of gaps in the temporal data for irrigated lands or may be a

13 18 JOURNAL OF THE AMERICAN MOSQUITO CONTROL ASSOCIATION VOL. 30, NO. 1 Fig. 5. Predicted mean monthly abundance of Culex tarsalis adults for August from multiple regression modeling. The abundance index is based on a predicted mean monthly abundance (i.e., number of adult mosquitoes per trap); minimal adults, low adults, moderate adults, or elevated adult mosquitoes. product of variation in the size of irrigated lands being reported in the data layer. The irrigated croplands layer applied in this study did not account for real-time water usage and highlights a possible problem in the static nature of the layers applied in these modeling efforts. Our work considered 4 spatial extents (175 m, 250 m, 500 m, and 1,000 m) that are significant for the perimeter of and distance to larval sites surrounding urban environments. In general, the perimeter of larval sites was most important in Loveland Johnstown at a spatial extent of 500 m, while larval habitat types and water sources improved predictive power of monthly models at a spatial extent of 1,000 m in Fort Collins. In summary, the contribution of irrigated lands and select larval habitat types to improve the fit of our multivariate mosquito abundance models highlights the importance for production of Cx. tarsalis in the naturally semiarid Larimer County. These findings are consistent with previously demonstrated positive associations between irrigated lands and elevated incidence of reported WNV disease in humans (Eisen et al. 2010, DeGroote and Sugumaran 2012). The expression of the LANDFIRE vegetative classes in abundance models provides empirical support for the importance of emergent wetlands to contribute to local Cx. tarsalis populations. The significance of manually controlled water in abundance models

14 MARCH 2014 EFFECTS OF LANDSCAPE HETEROGENEITY ON CULEX TARSALIS 19 underscores the relationship between water diversion across the region and mosquito abundance and the less important influence of rainfall. Additionally, the significance of retention ponds in Loveland Johnstown during July and August models suggests that water may move differently across Loveland than Fort Collins. The variability in the observed importance of the land-use patterns underscores the effects of landscape heterogeneity on Cx. tarsalis populations. As a more practical output of the study, we generated abundance maps for Cx. tarsalis adults that used model predictions to describe areas where mosquito abundance is likely to be elevated. Output maps revealed that abundance of adult Cx. tarsalis is predicted to be highest at the eastern edges of the municipal boundaries of Fort Collins and Loveland (Fig. 5). Barker et al. (2009b) recorded high abundances of Cx. tarsalis and Cx. pipiens for trap locations extending several hundred meters away from major larval habitats into dry and less hospitable plains landscapes with scattered trees. A spatial risk study conducted by Eisen et al. (2010) observed patterns of elevated WNV disease incidence in less densely populated census tracts in Boulder, Larimer, and Weld counties of Colorado. The results from abundance models with spatial extent considerations thus stress the importance of adequate larval control boundaries that encompass suitable larval Cx. tarsalis sites. Our study highlights the importance for proximity to larval Cx. tarsalis sites at a spatial extent of 500 m and stresses the need for operational consideration of larval control boundaries to improve the efficacy of mosquito management efforts and limit the dispersal of vector mosquitoes into municipal boundaries. ACKNOWLEDGMENTS We thank Boris Kondratieff for suggestions and considerations pertaining to mosquito biology, which improved this work. We also thank the cities of Fort Collins and Loveland, the town of Johnstown, the Northern Colorado Water Conservancy District, and Colorado Mosquito Control, Inc. for the use of data on mosquito abundance, irrigated lands, local weather, and larval mosquito sites. Sunil Kumar was supported by the US Geological Survey. REFERENCES CITED Aukema BH, Carroll AL, Zhu J, Raffa KF, Sickley TA, Taylor SW Landscape level analysis of mountain pine beetle in British Columbia, Canada: spatiotemporal development and spatial synchrony within the present outbreak. Ecography 29: Bailey SF, Eliason DA, Hoffmann BL Flight and dispersal of the mosquito Culex tarsalis Coquillett in the Sacramento Valley of California. Hilgardia 37: Barker CM, Bolling BG, Black WC IV, Moore CG, Eisen L. 2009a. Mosquitoes and West Nile virus along a river corridor from prairie to montane habitats in Eastern Colorado. J Vector Ecol 34: Barker CM, Bolling BG, Moore CG, Eisen L. 2009b. Relationship between distance from major larval habitats and abundance of adult mosquitoes in semiarid plains landscapes in Colorado. J Med Entomol 46: Beck LR, Rodriguez MH, Dister SW, Rodriguez AD, Rejmankova E, Ulloa A, Meza RA, Roberts DR, Paris JF, Spanner MA, Washino RK, Hacker C, Legters LJ Remote sensing as a landscape epidemiologic tool to identify villages at high risk for malaria transmission. Am J Trop Med Hyg 51: Bolling BG, Moore CG, Andersen SL, Blair CD, Beaty BJ Entomological studies along the Colorado Front Range during a period of intense West Nile virus activity. J Am Mosq Control Assoc 23: Brown HE, Doyle MS, Cox J, Eisen RJ, Nasci RS The effect of spatial and temporal subsetting on Culex tarsalis abundance models a design for sensible reduction of vector surveillance. JAmMosq Control Assoc 27: Carpenter SJ, LaCasse WJ Mosquitoes of North America, north of Mexico. Berkeley, CA: Univ. Calif. Press. p Chuang TW, Hildreth MB, Vanroekel DL, Wimberly MC Weather and land cover influences mosquito populations in Sioux Falls, South Dakota. J Med Entomol 48: Chuang TW, Hockett CW, Kightlinger L, Wimberly MC Landscape-level spatial patterns of West Nile virus risk in the Northern Great Plains. Am J Trop Med Hyg 86: Dale PER, Ritchie SA, Territo BM, Morris CD, Muhar A, Kay BH An overview of remote sensing and GIS for surveillance of mosquito vector habitats and risk assessment. J Vector Ecol 23: Darsie RF Jr, Ward RA Identification and geographical distribution of the mosquitoes of North America, north of Mexico. Gainesville, FL: Univ. Press of Florida. p DeGroote JP, Sugumaran R National and regional associations between human West Nile virus incidence and demographic, landscape, and land use conditions in the coterminous United States. Vector- Borne Zoonotic Dis 12: Diuk-Wasser MA, Brown HE, Andreadis TG, Fish D Modeling the spatial distribution of mosquito vectors for West Nile virus in Connecticut, USA. Vector-Borne Zoonotic Dis 6: Eisen L, Barker CM, Moore CG, Pape WJ, Winters AM, Cheronis N Irrigated agriculture is an important risk factor for West Nile virus disease in the hyperendemic Larimer-Boulder-Weld Area of North Central Colorado. J Med Entomol 47: Eisen L, Bolling BG, Blair CD, Beaty BJ, Moore CG Mosquito species richness, composition, and abundance along habitat-climate-elevation gradients in the Northern Colorado Front Range. J Med Entomol 45: Eisen L, Eisen RJ Using geographic information systems and decision support systems for the