Summary statistics
To date, The Connecticut Agricultural Experiment Station (CAES) has collected and tested 4,602,240 female mosquitoes comprised of 47 species in 8 genera. Approximately 98% of these collections were obtained from 92 trapping sites in 73 towns throughout the state, while the remainder of collections were from an additional 365 supplemental sites sampled between 1996 and 2007. Eighty-eight percent of collections come from CDC Light Traps, CDC Gravid Traps and Biogents BG Sentinel Traps (beginning in 2012). There have been several other collection methods used throughout the years that account for 11.6% of the mosquitoes collected (S. Table 1). Overall, there was considerable variation in mosquito abundance, surveillance effort, species richness/evenness, and the proportion of single species detections across CT (Fig. 1). One clear trend was that surveillance effort was greatest in CT’s human population centers (predominately CT’s southwestern and central counties) where WNV is commonly detected and along the CT-Rhode Island border where EEEV is most commonly detected (Fig. 1A). Another noticeable visual trend was that species evenness tends to be higher in the eastern portion of CT (Fig. 1B).
Maps of total mosquito abundance (log10 transformed) (A), total number of trap nights (A), average annual mosquito species richness (B), average annual mosquito species evenness (B), and average annual prevalence of single species detections (C) across 87 mosquito surveillance sites throughout Connecticut, U.S. sampled with ground level CDC CO2-baited light traps from 2001 to 2019. (A) Point sizes represent abundance while colors represent trap-nights; (B) point sizes represent species richness while colors represent species evenness; (C) point sizes represent prevalence of single species detections. (A–C) Solid black lines represent county political boundaries. The figure was created in R V 3.6.3 using the following packages: ggplot2 and maps.
Objective 1: annual collections of mosquito populations among sites
Our first objective was to identify spatial and temporal linear and nonlinear trends in mosquito abundance among sites. We also examined coarse-scale correlations between statewide (i.e., annual) and site-wide abundance and weather and land classification variables. All regression results and tables are provided as supporting information in Supporting Information: Regression Tables.
Mosquito abundance
Temporal regressions
After accounting for trapping effort, regression parameters estimating the relationship between site-level mosquito abundance and year of collection were positive using generalized linear mixed effects models (GLMMs) (“Year”—Estimate 0.03, t-value 9.11) and generalized additive mixed effects models (GAMMs) (“Year”—Est. 0.77, t-value 2.7, p = 0.007), suggesting that site-level mosquito abundance has increased in CT since 2001 (Fig. 2A,B): this trend resulted in a predicted 60% increase in annual abundance from 2001 to 2019. While these regressions identified possible increasing trends in site-level abundance, they provided an overall poor-fit to the data: AIC scores from fixed effect GLMMs were higher than random effects-only models (ΔAIC 415.1). This poor model fit may be in part driven by directly modeling Year as a fixed continuous effect; Year as a random categorical effect may better capture variation in mosquito collections30. Despite large differences in AIC scores between fixed and random effects-only models, we detected a pattern of increasing intercept values when examining “Year” as a random effect (S. Fig. 1), providing further evidence of an increasing temporal trend in site-level mosquito abundance.
Average annual mosquito abundance (A), number of trap nights (B), mosquito species richness (C), mosquito species evenness (D), the annual correlation between mosquito species richness and evenness (E), and the prevalence of single mosquito species detections (F) across 87 mosquito surveillance sites throughout Connecticut, U.S. sampled with ground level CDC CO2-baited light traps from 2001 – 2019. For (A)–(D) and (F), points represent the average across all sites, solid lines represent the standard error of the average, and dashed lines are added to aid interpreting each plot as a time series. For (E), points represent the average across all sites while solid lines represent the 95% CI of the correlation point estimate. The figure was created in R V 3.6.3 using base functions.
Spatial regressions
After accounting for trapping effort, regression parameters estimating the relationship between site-level mosquito abundance and latitude/longitude were positive using a GLMM (“Latitude (centered)”—Est. 0.49, t-value 5.48; Longitude (centered)”—Est. 0.20, t-value 4.78), indicating that mosquito abundance tends to increase on a south to north and west to east gradient (which reflects the overall transition in land cover from developed to forested in CT). The best fitting fixed effect GAMM included Longitude by Latitude smoothing terms, which also predicted positive relationships between abundance and site coordinates (Smoothing term 1: Est. 0.24, p = 0.06; Smoothing term 2: Est. 0.05, p = 0.67). GAMM predictions of site-level mosquito abundance were considerably more complex than GLMM predictions, yet still supported the overall trend of increasing abundance from south to north and west to east (S. Fig. 2). Overall, the fixed effect GLMMs provided an extremely poor fit to the data compared to random effects-only GLMMs (Latitude—ΔAIC 1092.7; Longitude—ΔAIC 1099.8). These poor model fits may be in part driven by directly modeling coordinate (i.e., site) as a fixed continuous effect: GAMM predictions that account for nonlinear relationships between abundance and spatial location may provide a more appropriate fit to the data while site as a categorical random effect in the GLMMs may better capture variation in mosquito collections30.
Weather correlations
When comparing statewide annual mosquito abundance to weather variables, we found no correlations between summer temperatures, spring temperatures or precipitation. This was despite detecting a slight annual increase in temperatures across all three seasons examined (average daily temperature GLMM Est., Season/Summer: 0.05 °C, Prior Spring: 0.02 °C, Prior Winter: 0.07 °C) and a slight annual decline in within season and prior spring precipitation (total precipitation GLMM Est., Season/Summer: − 4.23 mm, Prior Spring: − 3.38 mm; Prior Winter: 2.22 mm) in CT since 2001. However, we did find a positive correlation between total summer precipitation and annual statewide mosquito abundance (r = 0.50, CI 0.07–0.78).
Land cover correlations
When comparing total site-wide abundance to land cover classifications, we found positive correlations between percent land cover categorized as barren (r = 0.22, CI 0.01–0.41), forested wetland (r = 0.34, 0.14–0.52), and non-forested wetland (r = 0.21, 0.004–0.41). We also found a negative association in total site-level abundance and percent land cover categorized as grass (r = − 0.35, − 0.52 to − 0.15).
Species richness
Temporal regressions
After accounting for trapping effort, regression parameters estimating the relationship between site-level species richness and year of collection were positive using both GLMMs (“Year (centered)”—Est. 0.10, t-value 9.46) and GAMMs (“Year”—Est. 1.78, t-value 1.93, p = 0.05) (Fig. 2C): this trend resulted in a predicted 10% increase in site-level species richness from 2001 to 2019. Overall, fixed effects GLMMs of species richness provided an overall poor fit to the data when compared to a random effects-only model (ΔAIC 319.37). However, we did observe a pattern of increasing intercept values when examining “Year” as a random effect (S. Fig. 3), further indicating that mosquito species richness has annually increased across sites in CT since 2001.
Spatial regressions
Similar to models of site-level mosquito abundance, GLMMs of species richness by coordinate predicted positive relationships (Latitude (centered): Est. 0.63, t-value = 2.11; Longitude (centered): Est. 1.26, t-value = 9.34), indicating the species richness tends to increase along a south to north, west to east gradient. The best fitting GAMM included Longitude by Latitude smoothing terms, which also predicted positive relationships between species richness and site coordinate (Smoothing term 1: Est. 1.45, p = 0.0001; Smoothing term 2: Est. 0.70, p = 0.05). The GAMM further predicted a complex relationship of species richness among sites, yet overall predicted richness was lowest in the southwest/central portions of CT (areas of greatest development) and highest along coastal/eastern portions of CT (areas of non-forested and forested wetlands) (S. Fig. 4). The fixed effect GLMMs provided very poor fits to the data compared with random effects-only models (Latitude: ΔAIC 953.01; Longitude: ΔAIC 871.93; see the above results for Site-level collections: spatial regressions for possible reasons for these poor fits).
Weather correlations
We found no correlations of note between mosquito species richness and seasonal temperatures and precipitation.
Land cover correlations
Positive correlations of note for site-level species richness included: coniferous forest (r = 0.25, 0.04–0.43), deciduous forest (r = 0.56, 0.40–0.69), and forested wetland (r = 0.43, 0.23–0.58). Negative correlations included: barren (r = − 0.30, − 0.48 to − 0.10), developed (r = − 0.66, − 0.77 to − 0.53), grass (r = − 0.24, − 0.43 to − 0.03), and open water (r = − 0.31, − 0.49 to − 0.11).
Species evenness
Temporal regressions
Trends in species evenness were negative using both GLMMs (“Year”—Est. − 0.01, t-value − 7.86) and GAMMs (“Year (centered)”—Est. − 0.04, t-value − 5.58, p = 0.000) (Fig. 2D): this trend resulted in a predicted 12% decrease in site-level species evenness from 2001 to 2019. Similar to fixed effects GLMMs of species richness, fixed effects GLMMs of species evenness were less informative than a random effects-only model (ΔAIC 66.5). Declining intercept values were evident when evaluating “Year” as a random effect (S. Fig. 5), further supporting an overall annual decline in species evenness estimates among sites.
Spatial regressions
Similar to spatial models of species richness, GLMMs predicted positive relationships between species evenness and coordinate (Latitude (centered): Est. 0.36, t-value = 7.63; Longitude (centered): Est. 0.18, t-value = 8.54); the best fitting GAMM, which included Longitude by Latitude smoothing terms, also predicted positive relationships (Smoothing term 1: Est. 0.12, p = 0.01; Smoothing term 2: Est. 0.16, p = 0.004). GAMM predictions of site-level species evenness were equally complex to predictions of abundance and richness, and predicted evenness to be highest in southcentral and eastern CT (S. Fig. 6). Fixed effect GLMMs provided very poor fits to the data compared with random effects-only models (Latitude: ΔAIC 502.6; Longitude: ΔAIC 488.4; see the above results for Site-level collections: spatial regressions for possible reasons for these poor fits).
Weather correlations
We did find a negative correlation between statewide prior spring minimum temperatures and mosquito species evenness (r = − 0.49, − 0.77 to − 0.04).
Land cover correlations
Positive correlations of note for species evenness included: deciduous forest (r = 0.46, 0.28–0.61) and forested wetland (r = 0.22, 0.01–0.41). Negative correlations included: barren (r = − 0.37, − 0.54 to − 0.18), developed (r = − 0.45, − 0.60 to − 0.26), and open water (r = − 0.32, − 0.50 to − 0.12).
Correlations between abundance, richness, and evenness
The relationships between abundance, richness, and evenness varied depending on the scale examined. Across all years of data at the site-level, the correlation between abundance and richness was positive (r = 0.53, 0.36–0.67), the correlation between abundance and evenness as negative (r = − 0.35, − 0.52 to − 0.15), and there was no correlation of note between richness and evenness. Across all sites at the year-level, there were no correlations of note between abundance, richness, and evenness. Annual statewide correlations between richness and evenness (RRE) were positive for all years yet there was no noticeable annual trend in these correlations (Fig. 2E). Spatially, the average site-level RRE was 0.15 (± 0.03 SE). Furthermore, the magnitude and direction of RRE tended to increase on a south to north gradient (r = 0.31, 0.11–0.49), yet there was no apparent relationship in RRE along a west to east gradient (S. Fig. 7). We did detect a positive correlation between RRE and average maximum spring temperatures (r = 0.46, 0.01–0.76) as well as a positive correlation between RRE and percent land cover classified as coniferous forest (r = 0.23, 0.02–0.42).
Single detection events
Single detection events were defined as the prevalence of single species detections at a site (i.e., number of species with a single pool divided by species richness). Changes in single species detections could indirectly indicate range expansion among species (i.e., the prevalence of single detections decreases with time) and/or areas of unique mosquito diversity (i.e., the prevalence of single detections changes across space).
Temporal regressions
We detected no overall pattern of increasing/decreasing annual prevalence of single-species detections among sites (GLMM, “Year”—Est. − 0.13, t-value = − 1.12, p = 0.22; GAMM, “Year”—Est. 0.02, t value = − 0.31, p = 0.75) (Fig. 2F). These models were considered equivalent to a random effects-only GLMM (ΔAIC < 2), and thus, there were no obvious temporal trend of increasing/decreasing frequency of single species detections among sites (S. Fig. 8). We did find a negative correlation between both collections and single-species detections among sites (r = − 0.81, − 0.92 to − 0.56), indicating that increases in collections are associated with increased species detections within the mosquito community.
Spatial regressions
Unlike all previous models, GLMM and GAMM spatial regressions of single species detections by coordinate all provided poor fits to the data and indicated no obvious linear and nonlinear trends in single species detections in CT (Fig. 1B, S. Fig. 9).
Objective 2: annual collections of mosquito populations among species
Our second objective was to identify species-level linear and nonlinear annual collection trends that would suggest growth or decline in mosquito community composition. Because data were aggregated across the state to the species-level in these analyses, we did not perform any spatial regressions with this data. All regression results and tables are provided as supporting information in Supporting Information: Regression Tables.
Total abundance
Temporal regressions
Annual trends in total abundance per mosquito species were positive (GLMM “Year” Est. 0.05, t-value 7.59; GAMM “Year” Est. 0.33, t-value 2.85, p = 0.0045), with a predicted doubling of per-species annual collections in CT from 2001 to 2019 (Fig. 3A). Fixed effects models of species-level collections were less informative than random effects-only models (ΔAIC 11.2); however, there was a clear pattern of increasing “Year” random intercept estimates in the null model (S. Fig. 10), further supporting an increase in abundance across species.
Average annual mosquito abundance (A) and the prevalence of single site detections (B) across 46 commonly captured mosquito species in Connecticut, U.S. All individuals were collected across 87 sites sampled with ground-level CDC CO2-baited light traps from 2001 to 2019. Points represent the average across all species, solid lines represent the standard error of the average, and dashed lines are added to aid interpreting each plot as a time series. The figure was created in R V 3.6.3 using base functions.
Single detection events
Among species, our regressions identified a linear (GLMM, “Year”—Est. − 0.03, t-value = − 8.47) and non-linear (GAMM, “Year”—Est. − 0.02, t-value = − 3.35, p = 0.0009) decline in single-site detections (Fig. 3B). In these models of single-site detections among species, fixed effects models provided a poor overall fit to the data compared to random effects-only models (ΔAIC 145.6). There was also a strong temporal pattern of decline in the intercept estimates for “Year” when “Year” was evaluated as a random effect (S. Fig. 11), further suggesting a pattern of spatial growth (i.e., a decline in single-site detections) among mosquito species in CT.
We did find a negative correlation between species-level collections and the proportion of single site detections among species (r = − 0.81, − 0.92 to − 0.56), indicating that increases in collections are associated with increased spatial detections (i.e., declines in single species detections).
Species-specific trends
Since 2005, five additional species have been documented in CT (Table 1), including Aedes atlanticus (Dyar and Knab), Aedes flavescens (Muller), Aedes infirmatis (Dyar and Knab), Aedes spencerii (Theobold), and Psorophora howardii (Coquillett). Initial detections of each species were along the southern border of CT (Fig. 4). We further identified nine species possibly undergoing population increases in CT: Ae. albopictus, Aedes taeniorhynchus (Wiedemann), Anopheles crucians (Wiedemann), Anopheles quadrimaculatus (Say), Anopheles walkeri (Theobald), Cx. erraticus, Culex territans (Walker), Psorophora columbiae (Dyar and Knab), and Ps. howardii (Table 1, Fig. 4). Many of these novel and expanding species tended to display a more southern distribution (Fig. 4), suggesting that many of the species possibly experiencing population expansions are moving from south to north. We found further evidence of a south to north expansion of species populations when examining correlations between total site-level species richness and latitude (r = − 0.29, − 0.47 to − 0.08).
The latitudinal distribution of forty-six mosquito species collected in light traps across 87 surveillance locations in Connecticut, U.S. sampled with ground-level CO2-baited light traps from 2001 to 2019. Species are ordered by their average location of detection across all light trap collections in the CAES database. Fill colors represent the time period of first detection during standardized surveillance; border colors indicate statistical evidence of growth or decline in collections throughout CT; NS not significant. Species listed in Table 1 which are not listed here include Aedes flavescens and Aedes spencerii as there are only one collection of each species across all trapping effort. The figure was created in R V 3.6.3 using the ggplot2 package.
We also detected two possibly declining species, Aedes trivittatus (Coquillett) and Aedes sticticus (Meigen) (Table 1, Fig. 4). Outside of the species listed in Table 1, eight species displayed statistical evidence of growth in either total annual collection or total spatial detections while eight species displayed statistical evidence of declines in collections or detections (see Supporting Information: Regression Tables). Due to the lack of evidence of growth/decline in both collections and detections for these 16 species, we did not make any conclusions as to whether these species were actually increasing or decreasing in the state.
Source: Ecology - nature.com
