Skip to main content

Trap success data for capturing hemlock woolly adelgid

Cite this dataset

Sanders, Meg et al. (2022). Trap success data for capturing hemlock woolly adelgid [Dataset]. Dryad.


Hemlock woolly adelgid (HWA), Adelges tsugae Annand, threatens hemlock forests throughout eastern North America. Management efforts focus on early detection of HWA to ensure rapid management responses to control and stop the spread of this pest. This study’s goal was to identify an affordable, efficient trap to aid with airborne environmental DNA (eDNA) sampling approaches as an early monitoring tool for HWA. We initially compared HWA detection success between a standard sticky trap, commonly used for HWA monitoring, and trap designs potentially compatible with eDNA protocols (i.e., passive trap, funnel trap, and motorized trap). Passive, funnel, and motorized traps’ estimated capture success probabilities compared to sticky traps were 0.87, 0.8, and 0.4, respectively. A secondary evaluation of a modified version of the motorized trap further assessed trap performance and determined the number of traps needed in a set area to efficiently detect HWA. By modifying the original motorized trap design, its estimated capture success probability increased to 0.67 compared to a sticky trap. Overall, the cumulative capture success over the 16-week sampling period for the motorized trap was 94%  and 99% for the sticky trap. The number of traps did impact capture success, and trap elevation and distance to infested hemlocks influenced the number of adelgids captured per trap. As environmental DNA (eDNA)-based monitoring approaches continue to become incorporated into invasive species surveying, further refinement with these types of traps can be useful as an additional tool in the manager’s toolbox.


Trap Design Efficiency

The trap design testing took place at Pioneer Park (PIPK), Muskegon, Michigan, USA (Fig. 2; lat. 43.284127°, long. -86.368278°) a site with confirmed HWA infestations. Pioneer Park is 58.7 ha (145 ac) of county park and campground property along Lake Michigan. We designated the HWA infestation level as high based on a sistens count assessment outlined by Evans and Gregoire (2007). All traps were deployed in areas with known infested hemlock trees to test our trap designs. All four trap designs (motorized, passive, funnel, and sticky traps) were deployed for four weeks in the month of July 2020, which is during the sistens crawler stage. We organized our experiment in a randomized block design with five blocks. Each block comprised 36 cells for a total area of 625 m2. Trap contents were collected on a weekly basis for a total of four collection periods.

We assessed differences in HWA capture success for each of the four trap designs within each block and evaluated HWA distribution between blocks to account for potential effects of spatial variation in HWA across the study site. To assess adelgid capture success of the motorized and passive traps, we examined the petroleum-jelly-coated microscope slides under a Nikon SMZ645 dissecting microscope and counted the total number of HWA crawlers from the four slides of each trap. To assess adelgid capture success for the funnel traps, we counted crawlers in funnel traps by placing each trap’s contents into an individual petri dish and examining the contents underneath a dissecting microscope. To obtain adelgid counts for the sticky traps, we counted adelgids on each sticky trap using methods previously described by Dreistadt et al. (1998). Adelgids were counted on a 2.5-cm-wide vertical column down the center of each sticky insect card using a dissecting microscope. We used this technique on each of the five cards that made up every sticky trap.

To determine if spatial variation in HWA prevalence across our sampling site might impact our capture results, we evaluated HWA presence within each designated block at Pioneer Park by counting the number of ovisacs on hemlock branches using a method from the Pennsylvania Department of Conservation and Natural Resources (Johnson, 2020). This was quantified at the block level since differing amounts of HWA between blocks could impact trap success in catching HWA. We randomly selected 10 trees within every block and numbered the lower crown branches within 7.5 m of the ground starting on the north side and moving clockwise around the tree. We used a random number generator to select five branches around each tree and counted the number of ovisacs within a 25 cm length of the distal part of each branch.

All analyses were conducted using the program R v 4.0.3 (R Core Team, 2020). HWA estimates within each block and adelgid capture assessment data were non-normal despite transformations, thus we chose non-parametric analyses. To determine whether there were differences in HWA prevalence between blocks, we assessed differences between the average number of ovisacs counted from each block with a Kruskal-Wallis test using the package stats v 3.6.2. We estimated the probability that a non-sticky trap would capture HWA when a corresponding sticky trap (same block and same collection date) also captures HWA with a Wilson score interval (Wilson, 1927) using the package binom v 1.1-1. We also assessed differences in capture success between the different trap types using a generalized linear mixed model (GLMM), with trap type as the fixed effect and block as a random effect; the sticky trap was used as the reference. This was performed in the R package lme4 v 1.1-27.1 (Bates et al., 2015). Tukey’s post-hoc test was performed with the package multcomp v 1.4-20 (Hothorn et al., 2008) to evaluate differences in capture success across trap types. All statistical analyses used an alpha value of 0.05 to determine statistical differences.

Evaluation of Capture Success Related to Number of Traps and Landscape Features

The second part of our study took place at North Ottawa Dunes (Fig. 2; lat. 43.090484°, long. -86.247998°), a 240.2-ha (593-ac) Ottawa County Parks property of wooded sand dunes bordering Lake Michigan. This is a site with a known HWA infestation, and we designated the infestation level as low based on a sistens count assessment outlined by Evans and Gregoire (2007) (Sanders, 2021). We obtained Ottawa County Parks survey data (January–October 2020) with GPS locations of all hemlock trees within the park, as well as the locations of hemlock trees where visual surveys previously detected the presence of HWA ovisacs. 

Within North Ottawa Dunes, we established a 36.5-ha (90-ac) circle over our study area and sectioned it into 30 equal parts. The 30 equal sections (3 acres each) were divided into five replicate groups (A-E), with six sections per group. Each of these six sections hosted a different number of paired motorized and sticky traps. Section one contained one pair of motorized and sticky traps, section two contained two pairs of traps, so on and so forth up to the sixth section containing six trap pairs. This resulted in a total of 105 motorized and 105 sticky traps for the entire 36.5-ha (90-ac) area, and the density of the traps within each section ranged from 1 trap per 0.2 ha (0.5 ac) to 1 trap per 1.2 ha (3 ac). In every replicate group, the number of trap pairs and trap placement within each section was randomly assigned. Traps were attached to a 1.5 m pole, and the motorized and sticky traps were placed 2 m apart at each trap location. Traps were deployed for 16 weeks from April 7 through July 28, 2021, during both annual HWA egg-hatching events. Petroleum-jelly-coated slides from the motorized traps were collected biweekly and placed in 50 mL vials, and sticky traps were collected biweekly in clear, plastic storage bags. 

After each biweekly collection, we counted the number of adelgids observed on each trap. For the motorized traps, the number adelgids present on the four petroleum-jelly-coated slides were observed using a Nikon SMZ645 dissecting microscope, counted, and recorded. We assessed the number of adelgids collected on each sticky trap using the same method previously described for our trap design assessment (Dreistadt et al., 1998). For both the motorized and sticky traps, 20% of traps per collection period were recounted for quality assurance (R2 = 0.99). 

We created maps predicting distribution of HWA with the count data for each motorized trap by means of the inverse distance weighted (IDW) spatial interpolation method using ArcMap v 10.4.1 (ESRI, 2016) to visualize how adelgid counts varied in our study area throughout the summer. The IDW method predicts likely HWA numbers based on a linear-weighted combination of count data for sample locations. This method is appropriate for clustered data. IDW predicts values for unsampled locations by assuming those values are related more to closer data points than to those that are farther away. We used a power of 2 and a nearest-neighborhood search of 8 points in the analysis, so more localized trap counts influenced predictions of the nearby unsampled locations and accounted for all cardinal directions surrounding a location.

All statistical analyses performed in R used v 4.0.3 (R Core Team, 2020). We estimated the probability that a motorized trap would detect HWA when the corresponding sticky trap detected HWA with a Wilson score interval (Wilson, 1927) using the package binom v 1.1-1 to evaluate how our modifications to the motorized trap improved capture success compared to our initial trap design. We also used a GLMM to evaluate if the number of capture successes and failures differed between the sticky and motorized traps where trap type was considered a fixed effect, and the collection week and group ID (A-E) were included as random effects. This was performed in the R package lme4 v 1.1-27.1 (Bates et al., 2015).

To assess the level of spatial autocorrelation in the number of adelgids captured across our traps, we calculated Moran’s I using the program GeoDa (  Euclidean distances were calculated between each trap point. The bandwidth was set to 0.001 so that the median number of neighbors for each point (i.e., trap) was five (min neighbors = 1; max neighbors = 8). We performed the same analysis for each two-week collection period when crawlers were present to test for significant spatial autocorrelation with 999 permutations.

To determine if the number of traps deployed within each 1.2-ha (3-ac) section influenced whether an adelgid was captured. We used a GLMM to evaluate if a capture success within a section was correlated with the number of traps within each section. This analysis focused on data collected from April 21 – July 28, when adelgid crawlers were present. In the full model, the fixed effect included the number of traps per section. The collection week and replicate group ID (groups A-E) were included as random effects; sections with one trap were used as the reference. The null model included the random effects collection date and group ID (A-E). We then used an ANOVA to determine if the addition of the fixed effect significantly improved the model. This analysis was run using the lme4 package v 1.1-27.1 (Bates et al., 2015). We used the R package multcomp v 1.4-20 (Hothorn et al., 2008) for post-hoc analyses to evaluate significant differences in capture success between each number of traps per section using a Tukey’s post-hoc test. We also used a generalized linear model (GLM) to predict the number of traps that would need to be deployed within a 1.2-ha (3-ac) section to have a catch probability of 0.9 or greater. This analysis was performed for the active crawler period (April 21–July 28) and again with a subset of that data that represented the peak crawler period (May 19–June 16).  

We assessed if trap elevation, slope, aspect, and Euclidean distance to the nearest HWA-infested hemlock impacted the number of adelgids caught in a motorized trap. The adelgid count data were non-normal and over-dispersed. Because of this, we used a GLM with a negative binomial distribution using the package MASS v 7.3-53.1. The full model consisted of adelgid counts as the dependent variable and Euclidean distance, elevation, slope, and aspect as the independent variables. A reduced GLM model was also run after removing the non-significant terms, and the optimal model was selected using the lowest Akaike’s Information Criterion (AIC). All analyses used an alpha value of 0.05 to determine statistical differences.

Usage notes

Most files are in csv format and can be open with any type of spreadsheet software. The GIS related files can be opened with ArcGIS. Some open-source alternatives to ArcGIS include QGIS or specific R packages. 


US Forest Service, Award: STDP-R9-2019-01-FR