Abstract
Detection of newly established populations of Agrilus planipennis Fairmaire, the most destructive forest insect to invade the United States, remains challenging. Regulatory agencies currently rely on artificial traps, consisting of baited three-sided panels suspended in the canopy of ash (Fraxinus spp.) trees. Detection trees represent another survey option. Ash trees are girdled in spring to attract ovipositing A. planipennis females then debarked in fall to assess larval presence and density. From 2008-2010, systematic grids of detection trees and artificial traps were established across a 390-km2 area for the SLow Ash Mortality pilot project. We compared probabilities of detection associated with detection trees and artificial traps along varying A. planipennis density proxies estimated as distance-weighted averages of larval counts (detection trees) or adult captures (traps) within 800 m of each detection tree or trap. Detection trees were consistently more likely to be positive, that is, detect A. planipennis, than traps in all three years, even when traps were placed in canopies of detection trees. Probability of detection with a single detection tree was >50% when density proxies for the area were <5 larvae per detection tree, while the probability of detection with an artificial trap placed in the same area was <35%, even when density proxies exceeded 25 larvae per detection tree. At very low densities of <5 larvae per detection tree, using three detection trees would increase detection probabilities to 90%, while five artificial traps would increase the detection probability only to 40%.
More than 450 species of non-native forest insects are known to be established in the United States (Aukema et al. 2010). Approximately 15% of these alien species cause significant damage to trees in forest or urban forest settings in the United States (Aukema et al. 2010), costing landowners, municipalities, and plant-based industries billions of dollars annually (Kovacs et al. 2010, 2011; Aukema et al. 2011). Nearly all non-native phytophagous insects arrive in new habitats through accidental transportation with commodities, baggage, or as hitchhikers. Many efforts, including pathway analyses, preshipping treatment, quarantines, and inspections, have been implemented to prevent new introductions of non-native organisms into the United States and other countries. These efforts are largely offset, however, by burgeoning rates of international trade and travel (Aukema et al. 2010), making it likely that introductions of non-native, potentially invasive insects will continue (National Research Council 2002, Work et al. 2005, Brockerhoff et al. 2006, McCullough et al. 2006).
Developing effective detection and monitoring tools is a high priority when a new and potentially invasive insect pest is identified. Early detection allows implementation of management actions during the initial stages of establishment that can slow growth and spread of the pest population, potentially decreasing damage and possibly providing an opportunity for eradication (Shigesada and Kawasaki 1997, Myers et al. 2000, Liebhold and Tobin 2008, Suckling et al. 2012, Tobin et al. 2013). Tools capable of detecting low pest densities are also necessary for accurately delimiting the extent of an infestation and measuring effects of management activities. Ideally, a detection and monitoring tool should enable program managers to reduce uncertainty regarding the presence and density of a pest at an acceptable cost. Accuracy and precision of different detection tools can vary with pest density, however, affecting the cost–benefit relationship. In addition, not all detection tools may be available or appropriate for use under all circumstances.
For insects that produce long-range sex or aggregation pheromones, identification of active compounds can facilitate production of highly attractive lures. Pheromone lures for numerous Lepidopteran foliage feeders and some scolytine bark beetles, for example, are commercially available, enabling baited traps to be used locally or distributed across large geographic regions to detect or track infestations (e.g., Borden 1989, El-Sayed 2011).
Detection of invasive insects that do not produce long-range pheromones presents a greater challenge. This situation is exemplified by the emerald ash borer, Agrilus planipennis Fairmaire, an invasive pest native to Asia first identified in Detroit, MI, and Windsor, ON, Canada, in 2002 (Cappaert et al. 2005). In its native range, A. planipennis is considered a secondary pest, colonizing stressed trees (Yu 1992, Poland and McCullough 2006). North American ash species, however, lack a co-evolutionary history with A. planipennis and even healthy trees are colonized and killed (Cappaert et al. 2005, Wei et al. 2007, Anulewicz et al. 2008, Rebek et al. 2008). To date, A. planipennis populations are established in 20 states and two Canadian provinces (EAB.info 2013), and tens of millions of ash (Fraxinus spp.) trees in forests and urban settings have been killed (EAB.info 2013). Recent analyses indicate A. plannipennis has become the most costly forest insect to ever invade the United States (Kovacs et al. 2010, Aukema et al. 2011).
Like its North American congeners, A. planipennis does not produce long-range sex or aggregation pheromones, although short-range or contact pheromones may be involved in mating behavior (Lelito et al. 2009; Silk et al. 2009, 2011; Ryall et al. 2012). Adult A. planipennis rely on visual cues and host volatiles to locate ash trees and potential mates (Rodriguez–Saona et al. 2007, Crook et al. 2009, Crook and Mastro 2010). Visual surveys to identify infested ash trees, used extensively after the initial identification of A. planipennis, are problematic. Infested trees typically exhibit few external symptoms for at least 4 yr, until larval densities build to levels high enough to disrupt nutrient and water transport (Cappaert et al. 2005, Siegert et al. 2010, Tluczek et al. 2011).
Currently, the tools used most commonly for A. planipennis detection and monitoring include girdled ash trees and artificial traps baited with ash volatiles. Adult A. planipennis are attracted to and preferentially oviposit on ash trees stressed by girdling (McCullough et al. 2009a, b), likely in response to stress-induced changes in volatile emissions (Rodriguez–Saona et al. 2006) and perhaps light reflectance (Bartels et al. 2008). Grids of ash trees girdled in spring, then debarked in fall or winter to assess A. planipennis larval presence, for example, “detection trees,” have been used for surveys in Michigan, Ohio, and other states since 2005 (Rauscher 2006, Hunt 2007, Poland and McCullough 2010, SLAMEAB.info 2012). In previous studies, A. planipennis larval densities were consistently higher in girdled trees than in trees stressed by wounding or herbicides, or baited with volatile attractants (McCullough et al. 2009a, b; Tluczek et al. 2011). Although girdled trees are effective for A. planipennis detection or monitoring, debarking girdled trees can be labor intensive and trees suitable for girdling may not always be available or accessible.
Beginning in 2008, regulatory agencies have relied on artificial traps baited with lures containing volatile compounds associated with ash foliage or bark and wood for A. planipennis surveys (U.S. Department of Agriculture–Animal and Plant Health Inspection Service [USDA–APHIS] 2008, 2010). Numerous field studies have compared trap designs, colors, and lures to see which combination captured the greatest number of A. planipennis beetles (Francese et al. 2005, 2010; Marshall et al. 2009, 2010a; Crook and Mastro 2010; Poland et al. 2011). Traps currently used by the national A. planipennis program in the United States consist of three-sided plastic prisms in specific shades of purple or green attractive to A. planipennis (Crook et al. 2009, Francese et al. 2010). Traps are assembled on-site, coated with a clear sticky substance to capture beetles, then placed in the canopy of ash trees.
Relatively few trials have attempted to compare the efficacy of both girdled ash trees and artificial traps. In a 16-ha forested site with a very low A. planipennis density, McCullough et al. (2011) found larvae in 100% of the girdled ash trees when they were debarked in fall, while only 25% of the purple prism traps baited with Manuka oil and suspended in ash trees captured a single A. planipennis adult. Other studies recorded captures of adult A. planipennis beetles on sticky bands wrapped around the circumference of girdled or ungirdled ash trees and purple or green prism traps (Marshall et al. 2009, 2010b). Results were affected by various factors including tree size or exposure to sun, A. planipennis population levels, trap height, and compounds used as lures. When girdled trees were debarked, however, larvae were consistently present even in trees that captured no adult beetles on sticky bands (Marshall et al. 2009, 2010b).
Much is still unknown about the relative effectiveness of artificial traps and girdled detection trees at varying A. planipennis densities or whether these tools can accurately estimate A. planipennis population densities. Data collected from 2008 to 2010 as part of the SLow Ash Mortality (SLAM) pilot project (SLAMEAB.info 2012) provided a unique opportunity to measure detection trees and artificial traps across a wide range of A. planipennis population levels. The SLAM pilot project was a large-scale operational program that involved personnel from universities, and state and federal agencies. Objectives included development, implementation, and measurement of an integrated management strategy to determine whether EAB population growth or the progression of ash mortality could be delayed in a localized outlier infestation. A major aspect of the SLAM operation involved intensively monitoring A. planipennis annually using girdled ash detection trees and artificial traps deployed systematically across the project area that encompassed nearly 400 km2. Our specific goals in this article were to use data from the SLAM pilot project to compare the ability of the detection trees and artificial traps to 1) detect A. planipennis and 2) estimate local population densities across the SLAM project area.
Materials and Methods
In September 2007, an isolated A. planipennis infestation was identified near Moran, Mackinac Co., in the Upper Peninsula of Michigan, when larvae were found on a girdled ash tree debarked by state surveyors. Additional ash trees in the immediate area were subsequently felled and debarked, resulting in identification of 10 more infested trees. Ash trees in the area appeared healthy and nonsymptomatic, but examination of cross-sections from infested trees indicated the A. planipennis infestation likely originated at least 4 yr earlier. This relatively recent and localized infestation was subsequently selected as the focus of the SLAM pilot project to measure the effectiveness of integrating management options to reduce the progression of ash mortality in the area. Activities undertaken as part of the SLAM pilot project included extensive ash inventories and establishment of long-term plots to monitor ash condition, along with intensive surveys of A. planipennis (Poland and McCullough 2010, SLAMEAB.info 2012). Cooperators in the SLAM pilot project established an irregularly shaped project area encompassing 390 km2, which extended 20–29 km north from the Mackinac Bridge and 15–22 km west of Lake Huron in 2008. The SLAM project area encompassed a mixture of private and federal forest lands, and rural, residential, and municipal areas.
Detection Trees.
In spring 2008, systematic grids of girdled ash detection trees and artificial traps (see below) were established at three densities (Table 1) to survey A. planipennis distribution across the project area. The dense survey grid (16 detection trees per 2.6 km2) was centered on the area where infested trees were identified in 2007 and was surrounded by the coarse survey grid (four detection trees per 2.6 km2) (Table 1). The standard survey grid (one detection tree per 2.6 km2) (Table 1) extended from the coarse grid to the perimeter of the project area. Numerous grid cells were excluded from the A. planipennis survey because no ash trees were present or, in a few cases, because survey crews were unable to secure landowner permission for access.
Area and number of grid cells, girdled detection trees, and artificial traps, in the dense, coarse, and standard systematic survey grids used to monitor A. planipennis in the SLAM pilot project area from 2008 to 2010
Area and number of grid cells, girdled detection trees, and artificial traps, in the dense, coarse, and standard systematic survey grids used to monitor A. planipennis in the SLAM pilot project area from 2008 to 2010
From 12 May to 30 May 2008, in total, 383 girdled detection trees were established; 56% were in the dense grid, 26% were in the coarse grid, and 18% were in the standard grid (Table 1). Crews were directed to select trees that were 10–15 cm diameter at breast height (DBH; measured 1.3 m above ground). This size was considered optimal to eliminate the risk of smaller trees breaking during high winds and to avoid the excessive amount of time required to debark and examine larger trees. Because A. planipennis beetle activity is consistently higher in sunny areas than in shade (Yu 1992; McCullough et al. 2009a, b), crews were encouraged to preferentially select open-grown trees, or trees growing along roadways, the edge of woodlands, or in canopy gaps. Trees were girdled using drawknives to remove the outer bark and phloem from a 15- to 30-cm-wide band around the circumference of the trunk, ≈1.1–1.3 m above ground (McCullough and Siegert 2008). Global positioning system coordinates of each tree were recorded. Detection trees could not be established in some grid cells located in areas designated as protected wilderness or as critical habitat for a federally listed endangered dragonfly.
Detection trees were felled and examined from 15 September to 19 December 2008. Surveyors carefully removed outer bark on the trunk and any branches >5 cm in diameter with drawknives. Thin layers of phloem were gently shaved away until sapwood was exposed. Larval counts were recorded for each detection tree. Representative samples of larvae were collected from infested trees and placed into vials with ethanol. If an intact larva could not be recovered, at least one suspect gallery was collected, bagged, and saved for subsequent inspection. Debarked trees and limbs in forested areas or along secondary roads were bucked into small sections to speed decomposition. In developed or residential areas, logs and tops were collected and transported to a disposal yard.
Similar methods were used to establish girdled detection trees in 2009 and 2010, although the portion of the project area surveyed intensively (e.g., the dense grid) increased. From 11 May to 20 May 2009, in total, 579 girdled detection trees were established in the SLAM project area; 70% were in the dense survey grid, 18% were in the coarse grid, and 12% were in the standard survey grid (Table 1). In 2010, crews established a total of 748 detection trees from 3 May to 4 June, with 65, 30, and 5% in the dense, coarse grid, and standard grids, respectively (Table 1). Quality control measures implemented in 2009 and 2010 included inspection of up to 10% of the detection trees by supervisors, to ensure tree selection, girdling, and data collection were completed appropriately. Detection trees were debarked to assess presence of A. planipennis larvae from 15 September to 19 December 2009 and from 13 September to 29 October 2010.
Artificial Traps.
Artificial traps and lures used in the SLAM pilot project each year were provided by the USDA–APHIS. Design, lures, and placement of the artificial traps followed guidelines issued by USDA–APHIS for the national A. planipennis detection program (USDA–APHIS 2008). Each artificial trap consisted of a three-sided prism constructed from purple corrugated plastic (4 mm in thickness). The three panels were each ≈60 cm tall by 40 cm wide and the surface of the panels was coated with clear Tanglefoot. A lure with Manuka oil, released from a pouch at 50 mg/d (Synergy Semiochemicals Corp., Burnaby, B.C., Canada), was attached to each trap. Compounds in Manuka oil, derived from the New Zealand tea tree, Leptospermum scoparium J. R. and C. Forst (Myrtaceae), are similar to compounds associated with ash wood and bark. Traps baited with Manuka oil lures have been show to be attractive to A. planipennis (Crook et al. 2008; Grant et al. 2010, 2011; Poland et al. 2011). Baited traps were suspended in the canopy, at least 1.5 m aboveground, on the sunny or exposed side of ash trees.
Survey crews established a total of 229 artificial traps from 2 June to 20 June 2008 in the standard, coarse, and dense trapping grids (1, 4, or 16 traps per 2.6 km2, respectively) (Table 1). Crews placed 223 of the artificial traps in the dense survey grid (97%); 171 of those traps (75%) were placed in the canopy of girdled detection trees (Table 1). In the coarse and standard grids, artificial traps were established only in grid cells where ash trees were present but could not be used as girdled detection trees (e.g., protected wilderness). Global positioning system coordinates of all artificial traps were recorded. Traps were collected from 18 August to 12 September 2008. Suspect A. planipennis beetles were removed from traps, placed into individual vials with ethanol, labeled, and returned to Lansing, MI, for identification by the Survey Program Manager, MI Department of Agriculture. A few traps were blown out of trees and could not be recovered.
Survey crews installed 331 artificial traps from 1 June to 19 June 2009 and 475 artificial traps between 7 June and 25 June 2010 (Table 1). The same three-sided prism trap design was used as in 2008, but in 2009 and 2010, traps were baited with a lure consisting of an 80:20 ratio of Manuka and Phoebe oils, emitted from pouches at 50 mg per day. Like Manuka oil, Phoebe oil, derived from the Brazilian walnut tree, Phoebe porosa Mez (Lauraceae), is attractive to A. planipennis beetles (Grant et al. 2011, Poland et al. 2011). In 2009, 81% of the traps were placed in the dense survey grid while 8 and 11% of the traps were placed in the coarse and standard grids, respectively. Thirty percent of the traps in the dense grid were placed in the canopy of detection trees. When artificial traps were not placed in the canopy of a detection tree, surveyors preferentially selected ash trees ≥20 cm DBH, growing in open areas or along edges. In 2010, no artificial traps were placed on detection trees and when possible, traps were placed in ash trees at least 50 m from a detection tree. In 2010, 88% of the artificial traps were placed in the dense survey grids, 9% in the coarse grid, and 3% in the standard grid (Table 1). As in 2008, artificial traps were the only survey tool used in grid cells that fell within designated wilderness or protected areas. Artificial traps and all suspect beetles captured on traps were collected from 17 August to 12 September 2009 and 16 August to 10 September 2010.
Analysis of Paired Detection Trees and Artificial Traps.
Using data from the 177 artificial traps placed on detection trees in 2008 and the 99 artificial traps placed on detection trees in 2009, we contrasted the probabilities that an artificial trap placed on a detection tree and the detection tree itself would be detected as infested (i.e., “positive”) using a generalized linear mixed model with a logit link in the glmmPQL function in the MASS package for the statistical package R (R Development Core Team 2010). For these analyses, the paired detection tree and artificial trap were considered the experimental unit and the paired group was treated as a random effect. This enabled us to avoid potentially confounded effects between beetles attracted to detection trees and beetles captured on artificial traps. We also included an estimate of the mean local A. planipennis population, as described below, using larval counts from detection trees. The total number of larvae counted on each detection tree was used for this and subsequent analyses because of the uniform size of the detection trees. We similarly used the total number of adults captured on each artificial trap because of the uniform size of the traps. On average (±SEM), DBH of the detection trees was 12.1 ± 0.11 cm, corresponding to ≈2.8 m2 of phloem surface area (McCullough and Siegert 2007), while surface area of each trap was 0.72 m2.
Probability of Detection Estimates.
In 2009, 232 artificial traps were placed in ash trees that were not girdled and in 2010, all of the artificial traps were placed on nongirdled ash trees. Larval counts and number of adult beetles from unpaired detection trees and artificial traps, respectively, were used to estimate the mean local A. planipennis population present within 800 m of each artificial trap and detection tree separately for 2009 and 2010. To derive these estimates, we first calculated the Euclidean distance between all detection trees and artificial traps. Subsequently, the detection trees or artificial traps located within an 800 m radius of each artificial trap or detection tree were used to calculate a weighted estimate of the mean number of larvae (detection trees) or adult beetles (artificial traps) in the area surrounding a given detection tree or artificial trap. Mean density estimates were weighted using the distance of the detection trees or artificial traps to the point of interest. Specifically, the weight of each detection tree or artificial trap was estimated as 1 − (distance of detection tree or artificial trap/distance of all detection trees or traps). For all estimates, the value of the focal detection tree or artificial trap (e.g., the detection tree or artificial trap for which the neighborhood larval density or adult density was being estimated) was excluded from the analysis. Only detection trees and artificial traps with at least 10 detection trees or artificial traps located within 800 m were used for analyses.
Logistic regressions were performed in R (R Development Core Team 2010) to determine the relationship between A. planipenis density and the likelihood of A. planipenis detection by a detection tree or artificial trap (e.g., at least one larva recorded from the detection tree or one beetle recovered from the artificial trap). Subsequently, odds ratios determined through logistic regressions from the 2010 results were used to calculate the probability that a detection tree or artificial trap was detected as positive following the relationship probability=odds/(odds+1). To illustrate the effect of increasing or decreasing the number of detection trees or artificial traps used, we adjusted the detection probabilities from 2010 across a range of A. planipennis larval densities. The probability was adjusted following the simple relationship: adjusted probability=1 − (1 − probability of detection)number of DTs or ATs.
In 2009, 30% of the artificial traps in the dense trap grid (delineated by a 2.4-km radius around the epicenter of the infestation) were suspended in the canopies of detection trees, while detection trees were distributed across the project area. These conditions resulted in an underrepresentation of artificial traps that met our minimum requirement of having at least 10 detection trees or artificial traps located within 800 m in areas distant to the core of the infestation where A. planipennis densities were generally low (Tables 1 and 2). Given this situation, the 2009 data were excluded from the analysis described below.
Total number of unpaired detection trees and artificial traps and percentage detected as positive for A. planipennis by distance class from the epicenter of the infestation in 2008 to 2010
Total number of unpaired detection trees and artificial traps and percentage detected as positive for A. planipennis by distance class from the epicenter of the infestation in 2008 to 2010
Proportion of Positive Detection Trees and Artificial Traps.
A grid of 9,752 equidistant points was overlaid on the 2010 SLAM pilot project area, effectively dividing the project area into 200 by 200 m cells. Mean counts of larvae from detection trees and adult beetles from artificial traps were estimated for each grid point using all detection trees and artificial traps within 800 m of each point in the grid. The proportion of detection trees and artificial traps detected as positive within 800 m of each grid point was also estimated. Because of the extensive size of the project area and the patchy nature of ash distribution, some grid points fell within areas where ash trees did not occur, so no detection trees or artificial traps were established in these areas. To ensure our estimates were reasonable, we analyzed 561 grid points, all of which had at least 10 detection trees and 10 artificial traps within 800 m. We then contrasted the proportion of detection trees and artificial traps detected as positive using local adult and larval A. planipennis counts, as well as direct comparisons within grid points. In addition, the standard deviation in larval and adult beetle counts among detection trees and artificial traps was estimated for each grid point. These estimates were used to determine the coefficient of variation for larval and adult beetle counts to provide comparable estimates of variation for artificial traps and detection trees. Coefficients of variation were calculated as the standard deviation of the number of larvae or adults detected recorded per grid point divided by the mean number of larvae or adults detected across the project area.
Results
Girdled detection trees were more likely to be detected as positive for A. planipennis than artificial traps placed on the same trees in 2008 and again in 2009. In 2008, only six (3.5%) of the 171 artificial traps placed on girdled detection trees captured an A. planipennis beetle, but 18 (10.5%) of those trees had one or more A. planipennis larvae. Similarly, in 2009, larvae were present on 17 of the 99 detection trees (17.2%) that also had an artificial trap, but only four (4.0%) of the 99 traps placed on those detection trees captured a beetle. Results from a GLMM confirmed that the probability of detection was significantly higher for girdled detection trees than artificial traps in 2008 (t170=−6.49; P < 0.001) and 2009 (t97=3.78; P < 0.001). In 2009, there was a positive and significant interaction between the estimated local A. planipennis density and the detection tool (detection tree or artificial trap) (t97=2.78; P=0.007). This result indicates that differences between the detection trees and artificial traps were more pronounced when local A. planipennis densities were higher, likely because of both detection trees and artificial traps having a relatively low probability of detection in areas where A. planipennis densities were extremely low.
Probability of Detection Estimates.
Mean larval estimates for the 800-m radius surrounding each artificial trap or detection tree ranged from 0 to 42.5 larvae per detection tree in 2009 and 0–52.6 larvae per detection tree in 2010, with average counts (±SEM) of 4.84 ± 0.01 larvae and 9.15 ± 0.51 larvae per detection tree in 2009 and 2010, respectively. Mean adult beetle estimates for the 800-m radius surrounding each artificial trap or detection tree ranged from 0 to 11.4 beetles per trap, with an average (±SEM) of 2.86 ± 0.02 beetles in 2009 and 0–2.70 beetles per trap, averaging 0.23 ± 0.002 beetles in 2010. When data from 2009 and 2010 were analyzed together (including only unpaired detection trees and artificial traps), there was no interaction between the local A. planipennis density and the year in which sampling was undertaken (z=−1.36; P=0.17). However, an interaction was observed between the detection tool and the year in which sampling took place (z=3.4; P < 0.001). In 2009, fewer unpaired artificial traps and detection trees were used for density estimates and analysis than in 2010. Even in 2009, however, we observed significant effects of local A. planipennis density (z=4.013; P < 0.001), detection tool (z=−2.11; P=0.0348), and their interaction (z=1.997; P=0.0458) (Table 2). The interaction likely resulted from a difference between detection trees and artificial traps present only at higher A. planipennis densities, reflecting the concentration of detection trees and artificial traps in the epicenter of the A. planipennis infestation and the underrepresentation, particularly of artificial traps, near the edge of the project area (Table 2).
In 2010, when artificial traps were placed only on ungirdled ash trees, a larger number of traps had at least 10 detection trees within 800 m (295 and 232, respectively), increasing the power of our analyses compared with 2009. In 2010, the effect of local A. planipennis density on the probability of detection was once again significant (z=3.83; P < 0.001), as was the effect of detection tool (z=2.925; P=0.003). However, unlike 2009, there was no significant interaction between local A. planipennis density and the detection tool (z=1.168; P=0.242).
Results from 2010 showed that the probability of detection observed for detection trees and artificial traps improved as the local A. planipennis density increased, particularly for detection trees (Fig. 1). For example, the probability of detection with detection trees was >50% in areas where an average of only five larvae per tree were found when the local detection trees were debarked. In contrast, even when average larval counts exceeded 25 larvae per tree, the probability of detection with artificial traps was <35%. An important consideration when interpreting these results is that local A. planipennis larval density was estimated from girdled trees, not healthy trees. Therefore, the A. planipennis larval counts on the X-axis of Fig. 1 and other figures likely overestimate the local population and should be interpreted as a relative density estimates, rather than absolute densities.
Relationship between average A. planipennis larval counts from local detection trees (within 800 m) and the probability of a detection tree or artificial panel trap being detected as positive based on 2010 survey data.
Figure 2 illustrates the increased probability of detection if one to five detection trees or artificial traps are used in a given locale. When the local A. planipennis population is very low, for example, <5 larvae per tree, using three detection trees increases the probability of detection from 57 to 90% and if five detection trees are used, the likelihood of detection approaches 100% (Fig. 2 a). Once larval counts average 15 larvae per tree, the probability of detection with a single detection tree exceeds 95% and using additional detection trees would have little benefit (Fig. 2 a). For the artificial traps, when larval counts in the area average five larvae per tree, the probability of detection goes from 10% with a single artificial trap up to 40% if five artificial traps are used (Fig. 2 b), which is still lower than the detection probability associated with a single detection tree (Fig. 2 a). At an average of 15 larvae per tree, using five artificial traps instead of a single artificial trap increases the probability of detection from ≈20 to 62% (Fig. 2 b).
Relationship between average A. planipennis larval counts estimated from detection trees within 800 m, and the probability of detecting an A. planipennis infestation for (a) detection trees, and (b) artificial traps, when one to five detection trees or artificial traps are used, based on 2010 survey data.
Proportion of Positive Detection Trees and Artificial Traps.
Figure 3 presents a direct contrast of the proportion of positive detection trees to artificial traps in 2010. More than half the points fell below the 1 to 1 line, indicating a higher proportion of detection trees were detected as positive compared with artificial traps. This difference was also reflected in the standardized mean difference between the proportion of positive detection trees and artificial traps, which was fairly high (0.78). The standardized mean difference was calculated as: (Mean value for detection trees − Mean value for artificial traps)/Combined standard deviation.
Comparison between the proportion of detection trees and artificial traps detected as positive for A. planipennis per grid point in the 2010 survey. Black line represents the 1 to 1 line. (n=560 grid points).
The relationship between the average larval counts on local detection trees and the proportion of detection trees and artificial traps that detected A. planipennis in 2010 is shown in Fig. 4 a. Estimates of the local A. planipennis population was strongly related to the likelihood that a detection tree in the area would be infested (R2=0.69). Figure 4 b presents the relationship between the average number of adult A. planipennis beetles captured on artificial traps in the local area and the proportion of detection trees and artificial traps that were positive. On average, less than two beetles were captured per trap, and the highest number of beetles on a single trap in 2010 was 18. The relationship between A. planipennis adult counts on traps and the proportion of positive detection trees was fairly weak (Fig. 4 b), which reflects the high incidence of positive detection trees in areas where traps did not capture a single adult beetle (Fig. 4 b). A similarly weak relationship was observed between the mean number of larvae recorded per detection tree and the mean number of adults captured per artificial trap per grid point (Fig. 5).
Relationship between the proportion of detection trees and artificial traps detected as positive for A. planipennis per grid point as a function of (a) average larval counts, or (b) adult beetle captures in the 2010 survey. (n=560 grid points).
Relationship between the mean number of A. planipennis larvae in detection trees and the mean number of adults captured on artificial traps per grid point in 2010. (n=560 grid points).
Coefficients of variation, which represent the standard deviation scaled by the mean, indicated a high level of variation for both detection trees and artificial traps when A. planipennis larval counts were relatively low (Fig. 6). Coefficients of variation can be roughly interpreted as noise to signal ratios, indicating that at low population densities, the noise reached or approached four times the value of the signal. These results indicate that within a relatively small area bounded by an 800-m radius, densities recorded from detection trees or artificial traps were highly variable where A. planipennis densities were low. Attempting to extrapolate local population levels based on results from one or a few detection trees or artificial traps at this spatial scale, therefore, is not justified when A. planipennis densities are low. There was no discernible relationship between the coefficients of variation between detection trees and artificial traps (Pearson correlation=−0.01; P=0.88) (Fig. 6 c).
Coefficients of variation for A. planipennis larval counts or adult beetles captured in artificial traps per grid point in 2010. These are represented as (a) the coefficient of variation of the larvae counted per detection tree, (b) the coefficient of variation of adult captures per artificial trap, or (c) a regression of the coefficient of variation values for the larval counts per detection tree and adult captures per artificial trap. (n=389, 226, and 226 grid points, respectively).
Discussion
The success of efforts to reduce the spread of invasive species through the use of quarantines or active management relies on the ability to effectively detect, delimit, and monitor infestations. Early detection can greatly increase the success of efforts to eradicate, contain, or manage invasive pests (Suckling et al. 2012, Tobin et al. 2013), including actions to reduce the spread and population growth of A. planipennis (Mercader et al. 2011a, b). Furthermore, early detection provides municipal foresters with time to secure funds for insecticide treatment or staged removals of ash trees on public land, rather than reactively removing dead and dying trees. For example, in a recent simulation encompassing a 10-yr period, costs of treating up to 50% of urban ash trees annually with a highly effective systemic insecticide were substantially lower than removing trees as they declined (McCullough and Mercader 2012). Unfortunately, budgets usually limit the efforts that can be expended in surveys, and managers must carefully consider the relative cost-effectiveness of survey tools.
Placing a baited artificial trap into the canopy of an ash tree would presumably require less labor than debarking a girdled ash tree, translating to a lower price per unit. In practice, however, the time and labor required to set and recover an artificial trap and collect suspect beetles is often relatively similar to the time needed for an experienced surveyor to girdle and subsequently debark a detection tree, assuming the detection tree is ≤20 cm in diameter (McCullough et al. 2011). In this operational project, detection trees were substantially more likely to detect A. planipennis than artificial traps in every analysis we performed, especially when the larval A. planipennis population was relatively low. Of particular relevance is the higher likelihood of detection when the use of multiple detection trees or artificial traps was considered or when the proportion of positive detections was considered. These results show that even if more artificial traps than detection trees are used in an area, the detection trees will be more likely to detect an A. planipennis infestation than the artificial traps. For example, the probability of detection associated with a single detection tree was 80% if larval counts in detection trees within an 800-m radius averaged 10 larvae (≈3.5 larvae/m2 of phloem surface). At this same density, a single artificial trap corresponded to a detection probability of only 15%. Using five artificial traps would increase the detection probability to 50%, which is still substantially lower than the probability associated with a single detection tree.
The patchy distribution of A. planipennis within an infested site reflects the tendency of A. planipennis to disperse short distances when ash trees are immediately available (Mercader et al. 2009, Siegert et al. 2010), the typically clumped distribution of ash trees in forested or rural settings, and the attraction of A. planipennis females to stressed ash trees (McCullough et al. 2009a, b; Siegert et al. 2010; Tluczek et al. 2011). Not surprisingly, in a relatively new infestation, the likelihood that any one ash tree will be infested is low (Mercader et al. 2012), and in the absence of a strong radius of attraction, any detection tool can be expected to yield variable results. This was observed for both artificial traps and detection trees in terms of the proportion of positive detections and in the variability of estimates of the local population density based on larval counts from detection trees or adult beetle captures on artificial traps. This high variability indicates efforts to estimate the A. planipennis infestation level across a 2-km2 area (e.g., bounded by an 800-m radius) based on counts from only a few detection trees or artificial traps placed within that area should be made cautiously. Moreover, the A. planipennis density estimates in this study were generated from the detection tools themselves. Given the markedly greater attraction of egg-laying A. planipennis females to ash trees stressed by girdling than to healthy trees (McCullough et al. 2009a, b), the larval counts from the detection trees almost certainly overestimate the true population density in the local area. Therefore, detection trees as well as artificial traps are both probably more sensitive than the results presented here indicate.
Despite the likely underestimation of the sensitivity of detection trees, the estimated probability of detection with detection trees presented here was considerably higher than would be expected if nongirdled ash were used for surveys (Mercader et al. 2012). In contrast, the probability of A. planipennis detection with artificial traps in our project area was substantially lower than the probability of detecting A. planipennis by simply debarking an ungirdled ash tree (Mercader et al. 2012). As noted above, the sensitivity of artificial traps was probably underestimated in this study, and these results should not be interpreted as evidence that artificial traps are always less sensitive than debarking ungirdled trees. These results do, however, stress the relatively high sensitivity of detection trees compared with other currently available detection tools. Owing to the comparatively low probability of detection with artificial traps, increasing the number of artificial traps up to five did not increase the probability that an infestation would be detected to the level of a detection tree, even in areas with moderately high A. planipennis populations. Differences between detection trees and artificial traps were less pronounced at very low A. planipennis densities, reflecting the generally poor detection ability of both survey tools. These results highlight the fact that although detection trees are considerably more likely to detect infestations than ungirdled trees or artificial traps, detecting new infestations of A. planipennis where densities are very low remains difficult.
The use of detection trees for A. planipennis detection or survey can be limited by the availability of trees suitable for girdling. The size of detection trees used in the SLAM project (10–20 cm in DBH) reflects the need to select a tree that can be efficiently girdled, felled, and debarked, but also of sufficient size to remain attractive to egg-laying A. planipennis females during the summer (McCullough and Siegert 2008). Trees meeting these criteria may be unavailable or inaccessible for use as detection trees, particularly in urban areas and when multi-year surveys are necessary. Sentinel ash trees, for example, trees in large pots or planted specifically for A. planipennis detection, have been used occasionally in operational programs (Sargent et al. 2010). Artificial traps, in contrast, are not destructive and can be used in sensitive areas or in locales where ash trees are scarce. The traps may generate interest among residents or visitors, providing an opportunity for outreach efforts to increase awareness of A. planipennis.
Results from artificial traps, however, should be interpreted cautiously. While a positive result clearly indicates the presence of an A. planipennis infestation, a negative result is likely to occur even when moderately high infestations are present. Any measurement of detection and survey tools must include consideration of the opportunity costs associated with false negatives. Numerous field trials have been conducted to measure trap designs and lures for A. planipennis, but many of these occurred in sites with a moderate or even heavy infestation (e.g., Crook et al. 2008, 2012; Grant et al. 2010, 2011; Ryall et al. 2012). Multiple studies in many sites have shown an alternative artificial trap design, termed the double-decker trap (McCullough and Poland 2009), is consistently more effective for A. planipennis detection in areas with low or ultra-low population densities than baited canopy traps (Marshall et al. 2010a, b; McCullough et al. 2011; Poland et al. 2011). A sampling method that involves debarking two branches to assess A. planipennis larval presence was measured on open-grown urban ash trees (24–34 cm in DBH) and may provide a useful tool for municipal foresters (Ryall et al. 2011). Efforts to further assess A. planipennis attraction to other host volatiles or short-range pheromones (Silk et al. 2009, 2011; Ryall 2012) are continuing and may eventually lead to improved lures. Thorough analyses that incorporate costs of failing to detect new A. planipennis infestations are needed, however, to more fully measure the efficacy not only of individual detection tools but also the relative cost-effectiveness of combining girdled detection trees with various traps and lures.
We thank our colleagues and members of the SLAM Management Board, particularly Steven Katovich, Therese Poland, and Noel Schneeberger (USDA Forest Service), Andrew Storer (Michigan Technological University), Brenda Owen (Michigan Association of Timbermen), Robert Heyd (Michigan Department of Natural Resources), and Ken Rauscher (Michigan Department of Agriculture). Poland and Katovich provided helpful comments on an earlier draft of this manuscript. We also thank Gabriel Carballo, Travis Perkins, and Amos Ziegler (Michigan State University [MSU]) for their assistance with data collected for the SLAM pilot project. We appreciate the efforts of all the surveyors who collected the SLAM data, often in challenging and difficult conditions. Funding for the SLAM project and our work was provided by the USDA Forest Service, the American Recovery and Reinvestment Act, and USDA–APHIS.
References
