*Correspondence and present address: Alicia M. Ellis, Department of Geography and Earth Sciences, UNC-Charlotte, Charlotte, NC 28223, USA. E-mail: firstname.lastname@example.org
1Researchers often use the spatial distribution of insect offspring as a measure of adult oviposition preferences, and then make conclusions about the consequences of these preferences for population growth and the relationship between life-history traits (e.g. oviposition preference and offspring performance). However, several processes other than oviposition preference can generate spatial patterns of offspring density (e.g. dispersal limitations, spatially heterogeneous mortality rates). Incorrectly assuming that offspring distributions reflect oviposition preferences may therefore compromise our ability to understand the mechanisms determining population distributions and the relationship between life-history traits.
2The purpose of this study was to perform an empirical study at the whole-system scale to examine the movement and oviposition behaviours of the eastern tree hole mosquito Ochlerotatus triseriatus (Say) and test the importance of these behaviours in determining population distribution relative to other mechanisms.
3A mark–release–recapture experiment was performed to distinguish among the following alternative hypotheses that may explain a previously observed aggregated distribution of tree hole mosquito offspring: (H1) mosquitoes prefer habitats with particular vegetation characteristics and these preferences determine the distribution of their offspring; (H2) mosquitoes distribute their eggs randomly or evenly throughout their environment, but spatial differences in developmental success generate an aggregated pattern of larval density; (H3) mosquitoes randomly colonize habitats, but have limited dispersal capability causing them to distribute offspring where founder populations were established; (H4) wind or other environmental factors may lead to passive aggregation, or spatial heterogeneity in adult mortality (H5), rather than dispersal, generates clumped offspring distributions.
4Results indicate that the distribution of tree hole mosquito larvae is determined in part by adult habitat selection (H1), but do not exclude additional effects from passive aggregation (H4), or spatial patterns in adult mortality (H5).
5This research illustrates the importance of studying oviposition behaviour at the population scale to better evaluate its relative importance in determining population distribution and dynamics. Moreover, this study demonstrates the importance of linking behavioural and population dynamics for understanding evolutionary relationships among life-history traits (e.g. preference and offspring performance) and predicting when behaviour will be important in determining population phenomena.
Though researchers have advocated a more direct, mechanistic link between individual- and population-level processes (Wiens et al. 1993; Lima & Zollner 1996; Sutherland 1996; Lidicker 2002; Zollner & Lima 2005), translating the behaviour of individuals to the spatial distribution of populations has remained a major challenge in ecology (Morales & Ellner 2002; Biro, Post & Parkinson 2003). There remain few empirical studies attempting this integration (Levin et al. 2000), particularly studies that monitor the behaviour of individuals at the whole-system scale (Biro et al. 2003) and that test alternative mechanisms generating population phenomena (Jones 2001). Small-scale studies are useful for understanding the behaviour of individuals, but the scale at which they are conducted limits our understanding of the importance of behaviours in determining population processes (Lowe 2003; McMahon & Matter 2006). For example, studies have tested the effects of competitor and predator cues on the oviposition preferences of insects in a laboratory setting; however, it is unclear how important observed behaviours are in a natural setting where insects are exposed to a variety of different cues (Munga et al. 2006). Field studies, on the other hand, can be conducted at scales that are relevant to populations, but often they do not examine the actual behaviour of individuals (Jones 2001; Lidicker 2002) (e.g. field studies of the distribution of mosquito offspring: Lounibos 1981; Barker et al. 2003b). For this reason, field studies often cannot distinguish patterns generated by behaviour (e.g. habitat or oviposition preferences) from patterns generated by other processes (e.g. spatial differences in mortality) (Jones 2001).
One situation where the empirical link between behavioural and spatial population distribution is often assumed but rarely demonstrated is the influence of oviposition-site selection on the distribution of insect offspring. The link between oviposition and population distribution is presumably strong because the manner in which females distribute their offspring across a heterogeneous environment has direct consequences for population distribution, offspring survival and individual fitness (Reiskind & Wilson 2004; Rudolf & Rodel 2005). Empirical studies have examined oviposition and its consequences for offspring performance in several insects including galling sawflies, wasps, and aphids (reviewed in Jaenike 1990; Price 1994), grass and leaf miners (e.g. Scheirs, De Bruyn & Verhagen 2000), and butterflies (e.g. Bossart 2003; Ladner & Altizer 2005). However, most of these studies have been performed in controlled laboratory conditions and may not therefore be relevant to the behaviours and performance of organisms in natural settings. Other studies have examined the consequences of oviposition preference for spatial population distribution in the field by documenting the distribution of offspring among habitats (mosquitoes, Lounibos 1981; aquatic beetles, Binckley & Resetarits 2005; tree frogs, Rudolf & Rodel 2005). However, these field studies rarely study the pre-oviposition behaviour of adults and often assume that offspring distribution reflects adult oviposition preferences (e.g. Barker, Brewster & Paulson 2003a; Barker et al. 2003b; Rapley, Allen & Potts 2004; Heisswolf, Obermaier & Poethke 2005). This is despite the fact that several processes other than oviposition preference can generate spatial patterns of offspring density, such as dispersal limitations or spatial differences in mortality.
Previous work has demonstrated that the spatial distribution of offspring (eggs and larvae) of the eastern tree hole mosquito Ochlerotatus triseriatus (Say) is aggregated in the north-eastern USA (Ellis 2007). Within a 0·02 km2 area that included three adjacent habitats (open field, deciduous forest with dense understory vegetation, evergreen forest with little understory vegetation) of approximately equal size (c. 0·006 km2), more eggs and higher larval densities were found in the deciduous forest stands (Ellis 2007). The correlation between oviposition pattern and offspring distribution observed in the study suggests that aggregated distributions of offspring may result from (Hypothesis 1) preferences for habitats with particular vegetation characteristics by ovipositing females. However, several mechanisms in addition to habitat selection could generate the previously observed aggregated distribution of O. triseriatus offspring. First, mosquitoes may actually distribute their eggs randomly or evenly throughout their environment, but there may be spatial differences in egg mortality or hatching success (Hypothesis 2). Second, mosquitoes may randomly colonize particular areas but have limited dispersal capability (Hypothesis 3) causing them to distribute offspring in a relatively small area where founder populations were established. Third, mosquitoes may passively aggregate in particular habitats due to environmental factors such as wind turbulence (Hypothesis 4). Finally, spatial heterogeneity in adult mortality (Hypothesis 5) may generate clumped distributions of adults and, subsequently, their offspring. These hypotheses are not mutually exclusive, and certain aspects of them may operate simultaneously. However, incorrectly assuming that offspring distributions are generated primarily by oviposition preferences may not only compromise our ability to understand the mechanisms determining population dynamics, but may have important implications for our understanding of the evolutionary relationships among life-history traits (e.g. oviposition preference and offspring performance).
Accordingly, the purpose of this study was to evaluate at the whole-system scale the movement and oviposition behaviours of adult tree hole mosquitoes that may generate the previously observed aggregated distribution of their offspring. A mark–release–recapture experiment was performed to distinguish among the alternative hypotheses presented above. I predicted that: (1) the distributions of eggs and adults are clumped (not random or overdispersed) within a 0·01–0·02 km2 area; (2) dispersal is not passive or limiting within a 0·02 km2 area; and (3) the distribution of adults is associated with specific habitat characteristics. These results would provide support for the habitat selection hypothesis (Hypothesis 1).
The eastern tree hole mosquito Ochlerotatus triseriatus (formerly Aedes triseriatus) (Reinert 2000) is the most abundant tree hole mosquito in New Hampshire, where this study was conducted, and one of the most common tree hole mosquito species throughout the USA (Means 1979; Darsie & Ward 2005). Other mosquito species occasionally inhabit tree holes where this study was conducted (i.e. Ochlerotatus japonicus, Culex pipiens, Anopheles barberi), but only rarely and at extremely low densities (Ellis 2007). O. triseriatus oviposits primarily in water-filled tree holes and in artificial containers (e.g. tyres, buckets) (Means 1979), and tends to remain in or near forested regions (Sinsko & Craig 1979; Nasci 1982a; Ellis 2007). Adult females obtain blood meals from several different hosts including small mammals, birds, humans, and occasionally large mammals (Nasci 1982b, 1985). They lay their eggs just above the water line and have desiccation-resistant eggs (Means 1979). Larvae are filter-feeders and browsers of leaf detritus and microbes (Wallace & Merritt 1980; Merritt, Dadd & Walker 1992) and experience intense density dependence in the larval stage (Fish & Carpenter 1982; Livdahl 1982; Hard, Bradshaw & Malarkey 1989; Ellis 2007). O. triseriatus is primarily univoltine in New Hampshire, but exhibits instalment hatching (Livdahl 1982), which leads to a relatively long period of emergence (2–3 months) for each generation. O. triseriatus experiences little or no predation in the larval stage where this study was conducted (Barrera 1988).
sites and spatial trapping design
Releases of adult populations were performed in two isolated forest patches; one in Lebanon, NH (43°69′ N, 72°22′ W) and one in Piermont, NH (43°969′ N, 72°081′ W) (Fig. 1). Patches were selected because they were relatively isolated mixed evergreen and deciduous forests with similar flora, fauna and vertical forest structures. In addition, patches were selected based on size and the scale at which patterns of adult and larval densities were observed in previous studies (Service 1971b; Beier, Berry & Craig 1982; Ellis 2007). The Piermont site is a patch of forest (c. 0·02 km2) on a large rocky outcrop in the flood plain of the Connecticut River surrounded by corn fields on all sides (Fig. 1a). The Lebanon site is a patch of forest (c. 0·01 km2) located in the centre of a circular exit ramp off Interstate 89S. It is bordered primarily by roads, and residential and commercial buildings (Fig. 1b). The dominant tree species at each site included beech Fagus grandifolia, maple Acer saccharum, oak Quercus rubrum, pine Pinus strobus and hemlock Tsuga canadensis.
At each site, a grid with grid points located 15 m apart was established to designate sampling locations for collecting marked mosquitoes (101 grid points at the Piermont site, 47 grid points at the Lebanon site) (Fig. 1). The distance between grid points was selected to allow adequate characterization of forest structure using 5 m by 5 m plots at each grid point with a minimum of 5 m between each plot (see below). To decrease the probability of capturing most individuals directly next to release points, grid points were not placed less than 16 m from a release location. Two grid points at the Piermont site were not included because they were located on a very large cliff and were not accessible.
Vegetation type and amount were characterized in late June or early July within a 5 m by 5 m plot at each point in the grid (modified from Macarthur & Horn 1969; Erdelen 1984). Plot size was selected to permit adequate habitat characterization at a scale that was logistically feasible and hypothesized to be relevant to the scale at which mosquitoes perceive their environment. Elevation was recorded, the diameter at breast height (c. 1·5 m) for all trees (> 1 cm d.b.h.) was measured, and a spherical densitometer (concave) placed at breast height at the centre of the plot was used to determine percent canopy coverage. To characterize forest structure, a grid was established within each 5 m × 5 m plot with points located every 1 m (36 points total) and vertical forest structure was measured at each of these 36 points. At each point, a pole was used to determine the presence/absence of foliage vertically every 0·5 m up to 2·5 m (a total of 11 measurements for each point along the grid). This height is consistent with the height at which larvae of O. triseriatus have been found in oviposition traps placed in the field (Scholl & DeFoliart 1977; Sinsko & Grimstad 1977). The amount of foliage at each height (FOL1, FOL2 ... FOL11) was calculated as the proportion of all points in each plot where foliage was present at that height (Erdelen 1984). Vegetation analyses also included the identification and quantification of all trees, saplings, seedlings and shrubs. However, preliminary analyses revealed that the distribution and abundance of individual plant species were not important in determining the distribution of recaptured individuals. The number of trees, saplings and seedlings and the proportion of each of these that were hardwood were used instead, as these composite metrics better summarized differences among plots.
Released individuals were obtained from eggs collected in oviposition traps at both release sites. Two oviposition traps were placed c. 1·5 m off the ground on the closest tree to each specified location on the grid in late May 2005. Traps were 16-oz black plastic cups (Plum Party, Long Island City, NY, USA). Before the experiment, both traps were used to collect eggs that were later hatched and used in the experiment. During egg collection, both traps were lined with brown paper towels upon which females laid their eggs. Eggs were collected from late May until the middle of July and stored at 21 °C and a day : night cycle of 16 : 8 h. Eggs were hatched simultaneously by immersing paper towels in a hatching medium created by dissolving one protist pellet (Carolina Biological Supply, Co., Burlington, NC, USA) in 1 L of water. Larvae were reared at densities found in naturally colonized containers (c. 140 larvae L−1) (Ellis 2007) at 21 °C and a day : night cycle of 16 : 8 h. Once larvae began to pupate, pupae were removed every other day, counted and placed in marking cages (Bug Dorm-2, BioQuip, Rancho Dominguez, CA, USA). Twice as many pupae were designated for the Piermont site on each day because this site was approximately twice as large as the Lebanon site. Counting pupae in this way allowed estimation of the total number of marked individuals released at each site and ensured that both populations had similar age structures at the time of release. The number of individuals released was, however, an estimate due to some mortality between pupation and actual release. To determine species composition of the released populations, a sample of larvae was removed, preserved, and later identified to species. All larvae identified in these samples were O. triseriatus (280 for the first release, 84 for the second).
Individuals were marked with fluorescent DayGlo dust (Fire Orange for first release, Saturn Yellow for second release, DayGlo Color Corp., Cleveland, OH, USA) by placing c. 3–5 g of the dust in a cage and shaking (Hagler & Jackson 2001). Prior to the experiment, the effect of the dust on survivorship and the duration of the mark were tested. Two cages (Bug Dorm-2, BioQuip, Rancho Dominguez, CA, USA) containing 20 blood-fed adult females (c. 5–8 days old) were placed on the ground of a forest where they were exposed to natural environmental conditions. Mosquitoes were provided a site for oviposition, vegetation for resting and coverage, and a dilute sucrose solution. The number of individuals surviving in each cage was counted daily for 7 weeks. The marking lasted the entire duration of this experiment on surviving individuals. Because there was no replication, statistical tests examining differences in survival were not possible. However, there did not appear to be any difference in survival among treatments.
release and recapture
Two releases were performed at the Lebanon site and one release at the larger Piermont site in the fall of 2005. Releases were concentrated in the fall because the time needed to collect eggs and rear individuals to adulthood was too long to allow releases to be performed earlier. Moreover, evidence suggests that the oviposition behaviour of O. triseriatus changes throughout the season (Edgerly et al. 1998) and oviposition in the fall when eggs enter diapause is crucial for determining population distribution the following spring.
Approximately 2820 individuals (males + females) were released at the Piermont site on 23 August 2005 and c.1410 (males + females) were released at the Lebanon site on 24 August 2005. Because of low hatching rates, all 1930 individuals (males + females) collected for the second experiment were released at the smaller site in Lebanon, NH on 13 September 2005. A sample of the adult population from both sites was taken prior to the first release to determine the female to male ratio. The ratio was not significantly different from a 1 : 1 female to male ratio (Piermont: χ2 = 0·10, P = 0·75, Lebanon: χ2 = 0·31, P = 0·58). Individuals were 7–12 days old at the start of each release and females obtained a blood meal 1 day before being released from a guinea pig (housed according to standard Institutional Animal Care & Use Committee protocol). Blood feeding is necessary for egg development in O. triseriatus (Clements 1992) and ensured that females would begin searching for oviposition sites after being released.
For the experiment, one of the two oviposition traps at each grid point was randomly selected to monitor oviposition activity. This trap was lined with brown paper towels as above. The second oviposition trap at each grid point was used to capture marked and wild unmarked females. This trap was lined with a piece of brown paper bag (cut to the same size as the paper towel used in the first trap), which was covered with brushable Tangle-Trap insect trap coating (Biocontrol Network, Brentwood, TN, USA). Both traps were filled with c. 150 mL of water and maintained at this level throughout the experiment.
Traps were sampled weekly for 6 weeks. Tests of the effect of the dust on mosquitoes suggested that c. 50% mortality would occur 3–4 weeks after being marked and released. A 6-week sampling period was selected to encompass the period of time before mortality substantially limited the probability of recapturing marked individuals. Each week, the sticky paper was removed, unmarked and marked mosquitoes were counted and discarded, and a fresh sticky lining was inserted. Paper towels in oviposition traps were removed, and new ones put in place. The number of eggs per trap per week was counted using a dissecting microscope.
Distributions of eggs and adults
To determine if the distribution of eggs and recaptures was random, clumped, or overdispersed, spatial autocorrelation in the total number of eggs and the total number of recaptures per trap at each site and for each release at the Lebanon site was tested using Moran's I statistic (Sokal & Oden 1978) in ARCGIS (9·0 ESRI, 2004) (with inverse Euclidean distance and no standardization).
To confirm that the distribution of marked individuals matched that of wild unmarked individuals, regressions between the probability of recapturing marked individuals and the total number of unmarked individuals captured per trap were performed. Because the distribution of recaptures appeared to be spatially clumped (see ‘Results’), the total number of recaptures per trap was regressed against the number of unmarked individuals using models with and without spatial autocorrelation (i.e. models with no spatial covariance, Gaussian and exponential covariance structures and no nugget, and Gaussian and Exponential models with a nugget) (according to, Keitt et al. 2002) with the proc mixed option in SAS 9·1 (SAS Institute, 2002–03). Because the results for each regression were similar between all models (with- and without spatial covariance) for each response variable at each site, probit regressions, which assume independence among observations but are more appropriate for discrete response variables (i.e. number of recaptures), were used for all of the remaining analyses.
Probit regressions of the probability of recapturing marked individuals vs. the number of unmarked individuals were performed separately for each site and separately for the two releases at the Lebanon field site. The number of recaptures per trap ranged from zero to two for the first release in both sites. For the second release (Lebanon site only), the number of recaptures per trap ranged from zero to eight and the frequency distribution of these observations was highly skewed in the positive direction. Thus, for the second release at the Lebanon site, 4 bins were created: 0, 1, 2, and greater than 2 recaptures per trap. The probit option in proc logistic in SAS 9·1 (SAS Institute, 2002–03) was used for the regressions.
Patterns of dispersal
The location of recaptures was used to determine the distance and rate at which tree hole mosquitoes dispersed. To test if the direction of individual movement was passive via wind, the average orientation of recaptured individuals with respect to the centre of the grid was compared with the prevailing wind direction at each site. Hourly wind speed and wind direction data were obtained from the Lebanon Municipal Airport weather station, Lebanon, NH (National Climatic Data Center). This station is located c.10 km north-east of the Lebanon field site and c. 45 km south of the Piermont site and is the closest station to both sites from which hourly data could be obtained. Daily, weekly and total averages of wind speed and direction were calculated and the prevailing wind direction determined using circular statistics (Batschelet 1981). The directional bias in the movements of recaptured individuals away from the release point was examined by testing if the mean x- and y-coordinates of recapture locations were significantly different from zero. For the Piermont site, directionality and average orientation of recaptures were tested from a point directly between the two release points.
Probability of recapture vs. oviposition activity
Probit regressions were performed to determine the relationship between probability of recapturing marked individuals and the total number of eggs laid per trap. Regressions were performed separately for each site and separately for the two releases at the Lebanon field site. The same bins used in logistic regressions described above were used for recaptures during the second release (4 bins: 0, 1, 2, and > 2 recaptures per trap).
Probability of recapture vs. vegetation characteristics
To determine the relationship of the location of recaptures with vegetation characteristics, ordination and regression techniques were used. Principal components analysis (PCA) was performed using canoco 4·5 (ter Braak, Plant Research International, Wageningen, the Netherlands) to summarize all independent environmental variables (listed in Table 1). First, PC scores were calculated by combining data from the two sites. PCA was computed from the correlation matrix after centring and standardizing the environmental variables (for this and all remaining PCA analyses). Multivariate analysis of variance in proc glm in SAS 9·1 (SAS Institute, 2002–03) was used to test for the effect of forest patch (site) on the first three principal components. Univariate anovas were then performed on each variable to dissect the manova response.
Table 1. Principal component loadings for environmental variables. Abbreviations are described in the text. Large loadings (> |45|) are in bold
Percent canopy coverage
Number of trees
Proportion of trees hardwood
Number of saplings
Proportion of saplings hardwood
Number of seedlings
Proportion of seedlings hardwood
Number of shrubs
Percentage of variance explained
Because this initial principal component analysis demonstrated that forest structure differed substantially between sites (Fig. 3), PC scores were obtained for each site independently and these scores were used in the remaining analyses (see ‘Results’). For each site, the principal components that individually explained c. 10% or more of the variation among trap locations (the first four PCA scores for the Lebanon site, and the first three for the Piermont site) were used.
Regression analysis was then used to determine if the number of recaptures was related to any environmental variables (i.e. PCs). To account for possible spatial autocorrelation in the environmental (i.e. PCs) and response variables, the total number of recaptures per trap was regressed against the PC scores using models with and without spatial autocorrelation as described above (i.e. models with and without spatial covariance). Because the results for each regression were similar between all models, probit regressions (which assume independence among observations) were used. Data were analysed separately for each site and for each release, and four bins were created for the second release (0, 1, 2 and > 2 recaptures per trap).
Eighteen of c. 1410 marked females were recaptured at the Piermont site (c. 1·3% recapture rate), and 13 of c. 705 marked females were recaptured at the Lebanon site for the first release (c. 1·8% recapture rate) (Fig. 1). Shortly after the first release at both sites, there was a large storm with heavy winds and rain that may have led to substantial mortality or long-distance passive dispersal of marked individuals resulting in lower recapture rates for the first release. For the second release at the Lebanon site, 96 of c. 965 marked females were recaptured (c. 10% recapture rate) (Fig. 1). A total of 257 unmarked individuals were caught at the Piermont site and a total of 127 were caught at the Lebanon site.
distributions of eggs and adults
The spatial distributions of recaptures and eggs were highly clumped at the Piermont site (Moran's I = 0·07, 0·12, respectively, P < 0·01 for both variables). For the Lebanon site, the distribution of recaptures for the first release was random (Moran's I = –0·03, P > 0·15), while the number of recaptures for the second release was slightly clumped (Moran's I = 0·01, 0·05 < P < 0·10). The spatial distribution of eggs for the duration of the experiment at the Lebanon site was random (Moran's I = 00·05, P > 0·10).
The probability of at least one recapture per trap was positively related to the total number of unmarked individuals captured for both releases in both sites. This relationship was significant at both sites for the first release (P = 0·03 for both sites) ( = 5·3, 67·6% concordant with the model for the Piermont site, = 5·6, 66·3% concordant with the model for the Lebanon site) (Fig. 2a,b), but was only marginally significant for the second release (Lebanon site only: P = 0·09) ( = 3·03, 53·5% concordant with model) (Fig. 2c). Thus, at both sites, plots with more recaptures also tended to have more wild unmarked individuals suggesting that recaptures of marked individuals reflect processes operating in the wild populations.
patterns of dispersal
Marked individuals were captured at the edge of both forest patches within 2 weeks of each release. There was significant directional bias in the movement of individuals at the Piermont site in a south-east direction (x-coordinate P = 0·07, y-coordinate P < 0·0001), but no directional bias at the Lebanon site (x-coordinate P = 0·06, y-coordinate P = 0·15). For the first release at the Piermont site, the wind at the time of release was 2·2 kmph from the west, and the average wind speed for that day was 7·2 kmph ± 1·0 (1 SE) (direction highly variable). For the first release at the Lebanon site, there was no wind at the time of the release and the average wind speed that day was 5·0 kmph ± 1·0 (1 SE) (direction highly variable). There was also no wind during the second release at the Lebanon site (13 September 2005) and the average wind speed that day was 3·6 kmph ± 1·0 (1 SE) (direction highly variable). Average wind speed and direction for the entire duration of the experiment, was 7·2 kmph ± 0·3 (1 SE) from the west. The average orientation of recaptures with respect to the centre of the grid at each site was significantly different from a west to east wind direction (P < 0·0001 for both sites; t = 26·76, d.f. = 17 for Piermont; t = 15·20, d.f. = 107 for the Lebanon site).
probability of recapture vs. oviposition activity
The probability of at least one recapture per trap was positively related to the total number of eggs per trap for both releases in both sites. This relationship was significant for the first release at the Piermont site (P = 0·007) ( = 7·7, 75·0% concordant with model) (Fig. 2d) and for the second release at the Lebanon site (P = 0·001) ( = 8·01, 69·5% concordant with model) (Fig. 2f). The relationship was not significant for the first release at the Lebanon site where the recapture rate was low (P = 0·08) ( = 3·23, 64·0% concordant with model) (Fig. 2e).
probability of recapture vs. vegetation characteristics
Multivariate analysis of variance of principal component scores calculated from combined data of the two sites and individual anova tests indicated that the two sites differed significantly in vegetation structure (P < 0·001 for all tests). PC scores obtained for each trap location at the Piermont site exhibited a much larger range of variability than those obtained for the Lebanon site (Fig. 3). The Piermont site tended to have fewer seedlings, more shrubs, more vegetation > 1 m, and a higher variance in many measures of forest structure (A.M. Ellis, unpublished data). Because the vegetation structures of the two sites were significantly different, each site was analysed separately.
When PCs were computed for each site separately, the first three PC scores for the Piermont site cumulatively explained 56% of the variance and the first four PCA scores for the Lebanon site cumulatively explained 57% of the variance in environmental variables among trap locations. Table 1 lists the loadings for all environmental variables. The number of saplings and the amount of foliage above 1 m had large (i.e. > |0·45|) positive loadings on PC1 for both the Piermont and Lebanon field sites. Variables with large loadings on PC2 were also similar for both field sites. Traps with large PC2 values had lower percent canopy coverage, more shrubs, and more foliage at or less than 1·5 m. For the Piermont site, the proportion of trees that were hardwood, the proportion of saplings that were hardwood, and the number of seedlings had large positive loadings on PC3. For the Lebanon site, the number of trees, foliage at c. 0·5 m, and total d.b.h. were large and negative for PC3. Elevation, the number of seedlings, the proportion of seedlings that were hardwood, and foliage at c. 0·5 m had large loadings on PC4.
The probability of at least one recapture per trap for the first release at the Piermont site was negatively related to PC1 ( = 6·1, P = 0·01) (Fig. 4a), and was positively related to PC2 ( = 4·5, P = 0·03) (Fig. 4b) (P > 0·07 for PC3) (full model: = 10·8, P = 0·01, 80·4% concordant with model). For the Lebanon site, the probability of at least one recapture per trap for the first release at this site was not significantly related to any of the first four PCAs (all P > 0·40, full model: = 1·0, P = 0·91, 53·7% concordant with model). For the second release, however, the probability of at least one recapture was positively related to PC2, similar to the Piermont site ( = 4·6, P = 0·03) (Fig. 4c) and negatively related to PC4 ( = 7·2, P = 0·007; P > 0·70 for PC1 and PC3) (Fig. 4d) (full model: = 11·2, P = 0·03, 71·9% concordant with model). Thus, at both sites, more mosquitoes were recaptured in plots that had more foliage at or less than 1 m, more shrubs, and less canopy coverage. More mosquitoes were recaptured in sites with fewer saplings at the Piermont site, but in sites with fewer seedlings at the Lebanon site.
Theory predicts that females should distribute their offspring to maximize offspring performance (Rausher 1983; Schriber 1983; Thompson 1988; Valladares & Lawton 1991; Nufio & Papaj 2004). However, the distribution of offspring can be determined by a number of factors other than the oviposition preferences of adults (H1), such as spatial differences in egg mortality or hatching success (H2), dispersal limitations (H3), passive aggregation (H4), or spatial differences in adult mortality (H5). Here, a mark–release–recapture experiment was performed to distinguish among these alternative hypotheses.
Results were consistent with the hypothesis that tree hole mosquitoes exhibit some degree of habitat selection (H1), but did not rule out the possibility that observed distributions were influenced by passive physical aggregation (H4) or site-specific differences in adult mortality (H5). Clumped distributions of eggs and recaptures were not consistent with the hypothesis that aggregated offspring distributions are generated by random or overdispersed oviposition with spatial differences in egg hatch or success (H2). However, differences in the degree of clumping and differences in the amount of variability in environmental variables among the two sites suggest that differences in forest structure or the size of the forest patch may have been important in generating a more clumped distribution of adults and offspring in the Piermont site. This may have occurred because the Piermont site has more diverse microhabitats or because the forests differ in the spatial clustering of abiotic factors or the scale at which the forest structure changes.
The movement of tree hole mosquitoes also did not appear to be passive via prevailing wind or limiting in a 0·02 km2 patch of forest, which is not consistent with the hypothesis that aggregated offspring distributions are determined by random colonization followed by limited dispersal from founder populations (H3). Though the location of recaptures in this study did not appear to be related to the prevailing direction of wind, results cannot completely rule out the hypothesis of passive aggregation (H4). Vegetation within the forest can create turbulent wind flow (Clements 1999), which may have influenced movement on smaller spatial scales within each site. Mosquitoes may congregate in sites with more understory vegetation if, for example, sites with more vegetation tend to be the least windy (H4). Passive physical aggregation has been observed in other insects with weak flight capacity in sheltered areas (e.g. gall midges and moth flies, Lewis & Stephenson 1966); however, because vegetation also lowers wind speeds inside the forest, particularly at ground level (Clements 1999; Compton 2002) and because wind speeds tended to be low throughout the experiment (except for one short storm shortly after the first release), wind speeds were probably lower in the forest than the speed at which the flight of mosquitoes becomes depressed (c.3 m s−1 for mosquitoes, Clements 1999). Moreover, many studies indicate that small-bodied winged insects can control the degree to which movement is affected by wind (Johnson 1969; Isaacs & Byrne 1998; Loxdale & Lushai 1999). O. triseriatus and other mosquitoes, locusts, aphids, and other dipterans have been found to fly close to the ground where wind speeds are low (Taylor 1960, 1974; Isaacs & Byrne 1998), quickly stop flying when wind speeds increase (Johnson 1969; Clements 1999), and/or take flight only when winds are calm (Johnson 1969). Additionally, O. triseriatus and many other winged insects (e.g. mosquitoes, aphids, whiteflies and thrips) fly primarily at dusk and dawn when wind speeds are typically lower (Service 1971a; Clements 1999; Loxdale & Lushai 1999; Compton 2002), allowing them greater control over their flight (Southwood 1962; Taylor 1974; Loxdale & Lushai 1999). These behavioural processes greatly increase the opportunity for tree hole mosquitoes to have active and directed flight. In fact, when winds are low, it may be more likely that differences in wind speed are used as cues that mosquitoes use to select particular habitats. Future studies examining this issue in mosquitoes are clearly needed.
Results also demonstrated that more marked females were recaptured in sites with dense understory vegetation (< 1 m), little foliage between 1 and 2·75 m, and a more open canopy providing support for the hypothesis that mosquitoes exercise some degree of habitat selection (H1). Though some studies have examined the movement of adult mosquitoes (e.g. Lounibos 1981; Beier et al. 1982; Barker et al. 2003a,b), very few have focused on the searching behaviours of gravid females prior to oviposition (Clements 1999). In fact, very few studies in general have examined the day-to-day trivial flight of adult winged-insects (Loxdale & Lushai 1999). Results presented here are consistent with increasing evidence that many woodland mosquito species such as O. triseriatus can be found in dense ground vegetation and may exhibit highly patchy distributions associated with vegetation type (Service 1971b; Beier et al. 1982; Clements 1999). Though mosquitoes may select habitats in part based on the density or diversity of blood meal hosts within them (Clements 1999), individuals in the current study were released after obtaining a blood meal in the laboratory. Because females remain unresponsive to hosts for c. 1–3 days after blood-feeding (Bentley & Day 1989; Clements 1999), host cues were probably not important in determining movement behaviours during the initial days following a release, but may have influenced movement after this refractory period.
The association of adult densities with vegetation characteristics does not, however, completely rule out the possibility that distribution patterns are also influenced by differential adult mortality in different parts of the forest (H4). Though predation is one obvious mechanism that could generate site-specific mortality, very little is known about the impact that predators may have on the densities of adult mosquitoes other winged forest insects. Studies have documented the potential for predation of mosquitoes by spiders (Roitberg, Mondor & Tyerman 2003; Jackson, Nelson & Sune 2005), but their influence in natural settings is unknown. Insectivorous bats in New Hampshire can consume up to half their weight in insects nightly; however, mosquitoes are typically only a small portion of their total insect diet (Anthony & Kunz 1977). Moreover, bats may tend to exploit only local mosquito swarms that can be spatially and temporally variable (Anthony & Kunz 1977). Thus, the effect of predation on insect density is probably highly localized, temporally variable and minor at a landscape scale. Moreover, results presented here demonstrate that tree hole mosquitoes have the ability to disperse quickly such that local effects of predation would be quickly masked unless predation rates are exceptionally high (i.e. greater than dispersal rates) and consistent at small spatial scales. More studies are clearly needed to determine the influence of predation on the distribution and abundance of mosquitoes and other winged insects in natural systems.
Based on theory (e.g. the preference–performance hypothesis) (Rausher 1983; Schriber 1983; Thompson 1988; Valladares & Lawton 1991; Nufio & Papaj 2004), a weak oviposition preference suggested by results of the current study should be an evolutionary consequence of differences in offspring fitness among habitats. In accord with these expectations, a previous study with tree hole mosquitoes in New Hampshire demonstrated differences in offspring fitness that would be consistent with the evolution of weak habitat preferences for habitats with little understory vegetation (Ellis 2007). At low densities, offspring performance did not differ among deciduous forest habitats with dense understory vegetation and evergreen habitats with little understory vegetation suggesting that when overall densities are low, no fitness differences exist between habitats. However, offspring performance was higher in habitats with more understory vegetation when larval densities were high. Thus, preferences for habitats with dense understory vegetation would be favoured when regional abundances are high, but would not be favoured when regional abundances are low. Temporal variability in abundance could therefore select for weak habitat preferences such as those suggested by the results of the current study.
Few empirical studies have simultaneously examined the active movement of adults and the influence of movement on population processes. However, the studies that have been conducted illustrate the importance of behaviour in determining population and evolutionary processes. Lowe (2003) performed a mark–release–recapture experiment to estimate stream dispersal and population growth parameters of a stream salamander and found that upstream dispersal equalized population growth along the stream. Similarly, McMahon & Matter (2006) used mesocosm experiments to examine the influence of emigration on population parameters in desert pupfish. He found that emigration increased survival and reproduction as compared with populations where emigration was prevented. Though these and related studies are clearly an important step in linking behaviour to population processes, there remains a considerable lack of studies that monitor the behaviour of individuals at the whole-system scale (Biro et al. 2003) and that test the relative importance of movement in determining population phenomena. Studies examining the relative importance of behaviour, such as that presented here, are essential for our ability to predict in what systems and under what conditions the details of individual behaviour are necessary for understanding population processes and for our ability to understand relationships among life-history traits (e.g. oviposition preference and offspring performance).
This research was supported by NSF Grant DEB-0209736 to Mark A. McPeek and an endowment awarded by the R. Melville Cramer 1877 Foundation. Thanks to Mark McPeek, Kathryn Cottingham, Matthew Ayres, Doug Bolger, Toomas Tammaru, and an anonymous reviewer for their helpful comments on the manuscript. Thanks to Mr and Mrs Ward and Ellen Hanscome for allowing me to conduct research at the Piermont and Lebanon sites, respectively. Ashley Hetrick helped with field and laboratory work.