The association between meteorological variables and the abundance of Aedes taeniorhynchus in the Florida Keys

Authors


ABSTRACT:

The black salt marsh mosquito, Aedes taeniorhynchus, is a serious nuisance pest and a potential vector of a number of arboviruses. This study examined the effect of wind direction, wind speed, temperature, and time of year on the abundance of Ae. taeniorhynchus collected in CO2-baited light traps at 12 sites in the Florida Keys during 2004. The dependent variable analyzed was the natural log of weekly mosquito abundance. The previous week's wind speed and wind direction, and the current week's temperature were used as independent variables. Simple and multiple linear regression models were used to assess the significance and nature of association between the meteorological variables and the natural log of mosquito abundance, and to determine whether the meteorological variables had significant associations with mosquito abundance after also controlling for time of year. Week of year was treated as a circular independent variable in the regression models, using the sine and cosine of week in radians to model the periodic seasonal fluctuation in mosquito abundance. Mosquito abundance was significantly associated with all meteorological variables and with week of year. Individually, previous week's wind speed and wind direction, and current week's temperature were able to explain respectively 24.5%, 24.5%, and 52.1% of the variation in mosquito abundance observed over the year. Week of year had the strongest individual association with mosquito abundance, explaining 65.7% of the variation in mosquito abundance. The meteorological variables were still significantly associated with mosquito abundance, after controlling for week of year. Week and the meteorological variables together explained 79.2% of the variation in mosquito abundance. The regression models fit to the data from this study suggest a strong periodic seasonal variation in mosquito abundance, with meteorological conditions explaining a significant portion of the variation beyond the seasonal trend.

INTRODUCTION

The black salt marsh mosquito, Aedes taeniorhynchus Wiedemann, is the most abundant mosquito on Key Largo (DeMay and Hribar 2008). It is a serious nuisance pest and it has been found naturally infected with Everglades virus, St. Louis Encephalitis virus, and West Nile virus; recently it has been implicated as a potential vector of Rift Valley Fever virus (Hodapp et al. 1966, Chamberlain et al. 1969, Sudia et al. 1969, Hribar et al. 2003, 2004, Turell et al. 2008).

It is an occupational hazard of mosquito control that mosquitoes sometimes appear in areas that have been subjected to thorough adult and larval control measures. As early as 1903, Smith (1903) cautioned that mosquito control problems might be caused by mosquitoes that had moved into an area from elsewhere, what he called “migration.” According to Service (1997), “dispersal” is the proper term for this type of movement. Aedes taeniorhynchus often flies for long distances, up to 32 miles in some instances (Harden and Chubb 1960, Provost 1952). Vlach et al. (2006) documented interisland dispersal of Ae. taeniorhynchus among islands near Big Pine Key in the Florida Keys. Wind direction and wind speed can influence flight of this species, although the mosquitoes have no control over their movements when subjected to windborne transport (Provost 1952, Bidlingmayer 1985b). Wind direction and speed are among the meteorological factors that deserve more study in regard to their impact on dispersal of disease vectors (Pecoraro et al. 2007). We will investigate the association between the abundance of Ae. taeniorhynchus in the Florida Keys and wind direction, wind speed, and temperature, while controlling for the seasonal variation over the year.

MATERIALS AND METHODS

Mosquito collections

Weekly collections of mosquitoes were made at 12 sites on Cross Key and Key Largo, FL, from January to December, 2004 (Figure 1). No mosquito counts were obtained during week 9 (early March). Weekly trapping protocol was identical to that described by DeMay and Hribar (2008). Eight of the trap sites utilized in this study were the same sites used in the previously cited study: Cross Key, Crocodile Lake NWR, Key Largo Dump, County Road 905, Key Largo Gun Club, Bayberry Lane, Gulfstream Trailer Park, and the Wild Bird Center. Site descriptions may be found in DeMay and Hribar (2008). The other four sites are described below.

Figure 1.

Trap sites, no-spray zones, and weather buoy.

The trap site on the ocean side of the intersection of Card Sound Road and County Route 905 consists of a large tract of red mangrove (Rhizophora mangle L). The site is behind a shoreline ridge on a flat area extending a tenth of a mile wide and some three miles long called Dispatch Slough. The capstone Key Largo Limestone is about a foot below normal sea level; protected from heavy surf action, this area changes significantly from hypersaline in the dry season to hyposaline in the wet season, conditions that favor the development of red mangroves. Traps were set not far from Route 905 on the edge of a high hardwood hammock forest patch and the adjacent red mangrove forest.

Ocean Reef Gate is located at the entrance of the Ocean Reef Club, a large, upscale gated community on the northeastern end of Key Largo. Vegetation consists of well-developed high hardwood hammock forests typical of higher elevations. Solution holes are less common than at lower elevations.

Garden Cove is located about 12 miles south of the northeastern tip of Key Largo. Vegetation consists of high hardwood hammock species. The trap site is surrounded on three sides by well-developed stands of red mangrove.

Port Bougainville is an abandoned residential development. The State of Florida is completely leveling this area in the hope of returning it to its natural state. Three-story skeletal concrete buildings, artificial berms, debris, and construction equipment are found virtually everywhere. A considerable amount of the substrate was dynamited to create artificial canals. The area is transitioning gradually back to the original high hardwood hammock forest.

Key Largo is the largest of the Florida Keys and the only one directly connected to mainland Florida. Much of Key Largo, especially the northeastern end, is owned or managed by the State of Florida's Department of Environmental Protection or the U.S. Fish and Wildlife Service. State lands include John Pennekamp Coral Reef State Park and Key Largo Hammocks Botanical State Park; Federal lands include Crocodile Lake National Wildlife Refuge. Most of the smaller islands between Key Largo and the mainland of Florida are within Everglades National Park. Most of this land is designated a “no-spray” zone (Figure 1). No-spray zones are not subject to aerial adulticiding; ground adulticiding with permethrin and aerial larviciding with Bacillus thuringiensis israelensis do occur but are greatly restricted, adulticiding generally occurring only in close proximity to houses and larviciding done only in specified larval habitats. The result of these restrictions is that a large area of Key Largo and adjacent islands, including much optimal mosquito habitat, is not subject to control operations.

Weather data

Hourly wind direction (degrees clockwise from true north), wind speed (MPH), and temperature (o F) data were obtained from the Weather Warehouse (http://weather-warehouse.com/). Data were collected from the NOAA Molasses Reef weather buoy, the closest weather collection station to the study sites (see Figure 1), from midnight, January 1 to 11:59 PM, December 31, 2004. There were some gaps in the data, when one or more of the variables were not measured. However, every day in 2004 had a minimum of 12 h of data for each variable, for a total of 8,542 hourly observations. The hourly measurements were used to compute weekly averages for wind direction, wind speed, and temperature, to merge with the weekly mosquito count data. Because traps were deployed on Mondays and retrieved on Tuesdays, the previous week's wind speed and wind direction were used in the models. Temperature data were converted to degrees Celsius prior to analyses.

Statistical analyses

Simple and multiple linear regression models were used to investigate the association between Ae. taeniorhynchus abundance and the meteorological variables and time of year. As measurements for the meteorological variables were not site specific, the total number of mosquitoes collected over all sites was used in the regression analyses. Specifically, the dependent variable in the regression models was the natural log of the weekly mosquito abundance over all 12 sites. Candidate independent variables were the previous week's average wind speed and average wind direction, and the current week's average temperature. Bivariate regression models were estimated first, using only week or only one of the meteorological variables as an independent variable. Next, regression models were fit using the meteorological variables as independent variables, while also controlling for week. Finally, multiple linear regression models using all three meteorological variables, with and without adjusting for week, were considered.

Wind direction and week of year are circular variables. A circular week variable measured in degrees was obtained by the transformation Week × (360/52). The equivalence of 0 and 360 degrees for a circular variable can lead to nonsensical results if treated as a linear variable in statistical analyses. Hourly wind direction measurements were summarized weekly by computing the circular mean direction. Standard errors for the weekly mean directions were computed using a circular standard error. Several references for circular statistics are Fisher (1993), Jammalamadaka and SenGupta (2001), and Mardia and Jupp (2000).

To include wind direction and week of year in the regression models, the sine and cosine of these variables were used as independent variables, rather than inserting them directly into the models as covariates. For example, regressing the natural log of mosquito abundance (y) on the circular week variable (y), the linear model

image

was used. This model can be reparameterized as

image

showing that the regression model is a periodic function with overall mean level (a0), amplitude (b0), and phase shift (θ0). It is well-known and documented that mosquito abundance in the Florida Keys has seasonal variation (Hribar 2002). Using the sine and cosine of week in the regression models yielded a smooth, periodic estimated mosquito abundance with respect to time of year.

To investigate the nature of the association between the candidate independent variables and mosquito abundance, higher order terms of the linear variables (wind speed and temperature) and higher order trigonometric polynomials of the circular variables (wind direction and week), i.e.,

image

were explored as possible independent variables. Interactions between the explanatory variables were also explored.

Statistical analyses were done using SAS version 9.1 (SAS Institute Inc., Cary, NC), the R statistical software package (R Development Core Team 2007) and the circular statistics package for R (Lund and Agostinelli 2007).

RESULTS

Trap sites varied in terms of the weekly number of Ae. taeniorhynchus collected (Table 1). All sites had weeks for which no mosquitoes were collected, while peak abundance varied dramatically. For example, the Wild Bird Center site had a peak abundance of 910 mosquitoes in week 30 at the end of July, whereas the Key Largo Dump site had a peak abundance of 84,568 in week 32 at the beginning of August. Over the course of the study period the trap at the Key Largo Dump site collected the most mosquitoes, with Port Bougainville, Cross Key, Crocodile Lake, and Intersection comprising the other sites with a relatively large number of collected mosquitoes. The sites with relatively smaller abundance were Bayberry Lane, Garden Cove, Gun Club, Gulfstream Trailer Park, Ocean Reef Gate, Route 905, and Wild Bird Center.

Table 1.  Descriptive statistics for the weekly number of Ae. taeniorhynchus, by site (n = 51 for each site).
Trap SiteMinMaxSumMeanSD
Bayberry04,44611,816232688
Port Bougainville025,02556,3471,1053,807
Cross Key027,495109,6932,1515,460
Crocodile Lake056,519144,7782,8398,753
Key Largo Dump084,568196,3543,85013,167
Garden Cove01,7037,330144369
Gun Club07,83924,4094791,413
Gulfstream01,4562,98358223
Intersection039,19568,4891,3435,646
Ocean Reef Gate09,00938,4377541,866
Route 90507,83934,7726821,585
Wild Bird Center09102,94558169
All Sites20156,307698,35313,69333,671

There were two peaks in the weekly total number of mosquitoes collected over all sites, one in August and another in September (Figure 2). The two large spikes in abundance obscured the overall annual cycle and motivated the use of the natural log of mosquito abundance for the subsequent analyses. The seasonality of mosquito abundance was accompanied by seasonal variation in the weekly averages of the three meteorological variables considered (Figure 2). June, July, and August saw the highest average temperatures as well as the lowest average wind speeds. Temperatures were more consistent (lower standard errors) in the summer months as well. Winds generally came from the east or deviations off of east. Wind direction was most consistent during May and June.

Figure 2.

Weekly total number of mosquitoes collected over all sites; weekly averages (± 2 standard errors) of hourly temperature, wind speed, and wind direction measurements taken at Molasses Reef weather buoy, 2004.

The natural log of the total number of mosquitoes collected over all sites showed a clear, periodic trend by week of year (Figure 3). The regression model using the sine and cosine of the circular week variable was able to explain almost 66% of the variability in observed mosquito abundance over the year (Table 2). The meteorological variables, each on their own, explained less variation in mosquito abundance than week. However, each meteorological variable had a statistically significant association with mosquito abundance (left-hand side of Table 2). Wind speed had a moderately quadratic association with mosquito abundance, where lower and higher wind speeds were associated with greater mosquito abundance (Figure 3). Higher temperatures were associated with greater mosquito abundance. Using all three meteorological independent variables, about 65% of the variation in mosquito abundance was explained by the fitted regression model.

Figure 3.

Fitted regression models with natural log mosquito abundance as the independent variable, week and meteorological variables as individual dependent variables. Wind speed and wind direction measurements were previous week's measurements. Temperature was current week's measurement.

Table 2.  Summary of regression models, with and without adjusting for week of year. Dependent variable was natural log of mosquito abundance. Meteorological variablesa were the candidate independent variables. Thumbnail image of

The three meteorological variables were all still significant covariates of mosquito abundance when controlling for week of year (right-hand side of Table 2). There were significant interactions between week and wind direction and between week and temperature, meaning that the nature of the association between these meteorological variables and mosquito abundance depended on the time of year. Allowing all three meteorological variables in the regression model, while controlling for week, led to a model that could explain about 80% of the variation in the observed mosquito abundance, a 15 percentage point improvement in explained variation over the model with just week of year. Wind direction was no longer a statistically significant predictor in the model when already controlling for wind speed, temperature, and week of year. The above results suggest that mosquito abundance is significantly associated with meteorological conditions, above and beyond any seasonal variation.

DISCUSSION

The data collected during this study suggest clearly that wind direction and speed have an effect on number of mosquitoes collected in carbon dioxide-baited light traps. An examination of Figure 1 may help to understand why wind characteristics may have an effect on mosquito numbers in traps. The sites with the largest mean numbers of mosquitoes (Table 1) all are directly west of large no-spray areas (Figure 1). Since prevailing winds are from the east or deviations off of east, it is likely that mosquitoes collected by traps came from these no-spray areas. Of those sites where the smallest mean numbers of mosquitoes were collected, Gulfstream Trailer Park and Garden Cove are near the eastern coast of Key Largo, where there are few larval habitats or harborage areas for adult mosquitoes. The Wild Bird Center is west of an area subject to mosquito control operations, and mosquito control operations can affect studies of mosquito population dynamics (Shaman et al. 2002). Bayberry Lane and the Gun Club are west of no-spray areas, and higher numbers of mosquitoes might be expected at these sites. However, both sites are very near the coast; again, there may be few larval habitats or harborage sites for adults.

The relationships observed between meteorological variables and mosquito numbers are not surprising (Table 2). More mosquitoes are collected during warmer periods of the year and generally speaking, at lower wind speeds. Wind speed is higher during colder parts of the year, and interaction between temperature and wind speed may be the likely explanation for low trap catches during the colder months (Williams 1940). Wind is the meteorological variable most variable between sites (Bidlingmayer 1985a). Vegetation near traps may serve to protect the trap from high wind speeds and thus mitigate effects on mosquito trap catch (Bidlingmayer 1967). Unfortunately the weather data were collected at one site, the weather buoy, and so we were unable to determine whether there were site-specific effects of wind speed and wind direction on trap catch.

Wind is known or suspected to affect dispersal of mosquitoes. Several authors have reported such observations. Hamlyn-Harris (1933) observed transport of Aedes vigilax (Skuse) on winds. Horsfall (1954) observed Aedes vexans (Meigen) transported long distances on a strong cold front. Ming et al. (1989) noted sudden appearances of Culex tritaeniorhynchus Giles that were associated with southwesterly winds and passing cold fronts. Later, Ming et al. (1993) detected Cx. tritaeniorhynchus moving within an air mass directly behind a cold front. The mosquitoes were estimated to have been transported anywhere from 35 km to 200 km. Reynolds et al. (1996) reported on long-distance wind-borne transport of mosquitoes in India. They felt that even short-term flights would result in dispersion of individuals over several kilometers. Hribar (2007) speculated that Culiseta inornata (Williston) may have been blown into the Florida Keys by strong winds. Bogojević et al. (2007) in Croatia correlated movement of Ae. vexans (Meigen), Ae. caspius (Pallas), and Ae. sticticus (Meigen) with wind speed and direction. Mass invasion of Buenos Aires by Ae. albifasciatus (Macquart) was observed following wind gusts accompanying a storm (Bejarán et al. 2008). Whether this downwind dispersal is active or passive can be debated. According to Service (1980), some species of mosquitoes are adapted to disperse during periods of high winds. On the other hand, if wind speed is high then some mosquito species will fly downwind (Bidlingmayer 1985a).

It is interesting that two peaks of maximum mosquito collection were observed, one at lower wind speed and another at higher speed. Although it might be expected that there would be an inverse relationship between wind speed and mosquito numbers, this was not observed in the present study. Bidlingmayer and Evans (1987) and Bogojević et al. (2007) also did not find such a relationship in their studies. However, Bidlingmayer (1974) found a significant negative relationship between nightly wind speed and number of Ae. taeniorhynchus collected by various types of traps. Moreover, Hoffman and Miller (2003) found a significant inverse relationship between wind speed and mosquitoes captured in CDC light traps baited with dry ice. Although baiting light traps with dry ice usually results in a greatly increased catch of Ae. taeniorhynchus (Newhouse et al. 1966), Hoffman and Miller (2003) stated the effect of wind was to dilute the stimulus plume (CO2) downwind. They attributed the lower numbers in traps to dilution effect rather than to limitations imposed by wind on mosquito flight.

The dispersal of Diptera via wind has potentially serious implications for disease transmission. Sellers (1980) discussed the implications of such windborne transport of mosquitoes on the spread of viral diseases of animals, stating that viral diseases could be repeatedly introduced into an area due to annual movements of mosquitoes. Ritchie and Rochester (2001) and Johansen et al. (2003) determined, via analyses of wind patterns and disease outbreaks, that it was possible for wind-blown mosquitoes to have introduced Japanese Encephalitis into Papua New Guinea. Lindsay et al. (1995) related distribution of Anopheles gambiae s.l., and thus malaria, to nighttime wind patterns.

Nearly thirty years ago Service (1980) wrote that not much was known about wind-borne transport of mosquitoes. A dozen years ago Service (1997) reiterated that our knowledge of mosquito flight was still at an “embarrassing” state. Studies of impact of wind direction and speed on vector-borne diseases are few, but with the technology available today, cooperative studies among entomologists and climatologists should generate information that will prove invaluable in vector management (Paz and Broza 2007).

Acknowledgments

We thank the staffs of Crocodile Lake National Wildlife Refuge and Florida Department of Environmental Protection, Division of Parks and Recreation, for permission to collect mosquitoes on lands under their management. The following employees of the Florida Keys Mosquito Control District assisted in this study: Amy Sargent and Steve Bradshaw provided detailed information on no-spray zones; Kristin Bird prepared the map.

Ancillary