Introduction

Globalisation has increased the spread of invasive species to new ranges with significant impacts on local ecosystems (Bellard et al. 2016). To combat these effects conservation managers have initiated many attempts at population suppression or outright eradication with varying degrees of success (Pluess et al. 2012; Tobin et al. 2014). Social insects are overrepresented in biological invasions and have been shown to be especially difficult to remove once established (Howse et al. 2023). Understanding why attempts at population suppression and eradication fail can improve future management of these invasive species.

Social wasps are an example of a globally significant invader (Lester and Beggs 2019). A flexible reproductive strategy and high output has allowed social wasps to reach high densities in new ranges all over the world (Beggs et al. 2011; Hanna et al. 2014; Lester and Beggs 2019; Loope and Wilson Rankin 2021). Their generalist feeding behaviours and habitat preferences allow them to readily dominate available resources (D'Adamo and Lozada 2009; Lester and Beggs 2019; Howse et al. 2020; McGruddy et al. 2021a; Moreyra and Lozada 2021) altering both recipient animal and plant communities (Thomas et al. 1990; Beggs and Wilson 1991; Beggs and Rees 1999; Beggs 2001; McGruddy et al. 2021b). Invasive wasps also represent a very important threat to apicultural industries as they act as both a competitor and direct predator of honey bees as well as a reservoir for honey bee viruses (Brenton-Rule et al. 2018; Dobelmann et al. 2020; Buteler et al. 2021). Wasps were implicated in as much as 14.9% of overwintering hive losses in New Zealand per year since 2015 (Stahlmann-Brown et al. 2022). A study by MacIntyre and Hellstrom (2015) estimated that invasive Vespula wasps cost the New Zealand economy NZ$133 million per year for their impacts on agriculture and human health.

Reducing the population of these invasive wasp species has become a priority for conservation managers globally (Edwards et al. 2017; Lester and Beggs 2019; Wilson Rankin 2021). Nest destruction, trapping and biological control have all been employed to control populations of invasive wasps but with limited success, especially at large scales (Beggs et al. 2011). Currently the most effective methods for controlling invasive wasps are toxic baits. These baits contain an insecticide within an attractive matrix. The wasps collect the bait and pesticide, taking it back to the nest where it is spread to other colony members via trophallaxis (Beggs et al. 2011). The commercially available bait, Vespex® (Merchento, New Zealand) contains the active ingredient fipronil (0.1%) which is a neurotoxin and broad-spectrum insecticide, within a matrix that attracts foraging workers. This bait is typical of many wasp control approaches and is designed to target vespuline wasps, relying on the species’ scavenging behaviours to take the protein-based bait back to the nest to be distributed to the other nest members. Vespex® has been shown to be effective at controlling wasps at scales of over 2000 ha, reducing nest activity by over 97%, 20–38 days after baiting (Edwards et al. 2017). Similar fipronil-based baits have been shown to significantly reduce wasp activity across the globe (Sackmann et al. 2001; Hanna et al. 2012; Rust et al. 2017) but eradications at landscape scales have not been achieved.

Previous studies examining the impact of toxic baiting efficacy have measured wasp activity by calculating nest traffic rates (Sackmann et al. 2001; Hanna et al. 2012; Edwards et al. 2017). Monitoring changes in nest traffic provides a measure of changes in foraging pressure exerted on an ecosystem. Nest traffic rates also provide information on colony size (Malham et al. 1991). Colony size in social insects is linked to important processes related to survival such as reproduction (Melo et al. 2023), colony defence (London and Jeanne 2003) and nest hygiene (Maák et al. 2019) and has been associated with increased genetic diversity and reduced viral load (Dobelmann et al. 2017), which may alter susceptibility to toxic baits. Factors that govern bait discovery and uptake are also important to understand when addressing bait efficacy. Nest-to-bait distance has previously been shown to impact baiting efficacy (Harper et al. 2016a). Investigating the arrangement of nests relative to both baits and to each other may provide insights into opportunities for increasing baiting efficacy.

In this study, a current wasp control method using the fipronil-based bait, Vespex®, was used to highlight what factors are most associated with wasp nest survival during a baiting operation. Our goal was to identify where measures might be improved for more effective wasp suppression in the pursuit of higher levels of population control or even eventual eradication at large scales. The methods were implemented using the manufacturer’s recommended protocol, informed also by Edwards et al. (2017), to control the common wasp (Vespula vulgaris) in a beech forest habitat, in New Zealand. Logistic and beta regression were used to analyse how aspects such as nest size and spatial distribution of both baits and nests influenced the magnitude of the bait’s effects.

Methods

Field site

The experiment took place on a privately-owned site in Northern Canterbury, New Zealand (Fig. 1). The focal site of the experiment was a 64-ha area within a patch of remnant beech forest with a canopy of predominately mountain beech (Fuscospora cliffortioides) but also including pockets of tōtara (Podocarpus totara) and other native species. The understorey was primarily beech saplings F. cliffortioides as well as mingimingi (Coprosma propinqua). The patch of forest containing the focal experiment site is located on the southeast side of Mt Balfour (42.4679° S, 173.0107° E) ranging from 636 to 878 m in elevation. A second site used as an untreated reference location was approximately 4 km south from the treatment site (41.3987° S, 172.9908° E). This reference site was used to account for natural seasonal variation in wasp activity. The reference site was in a valley on the eastern side of the Hanmer River. The northern side of the valley was covered in a similar forest habitat as in the treatment area while the southern side contained more altered open habitat. The area searched here was approximately 16 ha in area and varied from 554 to 700 m in elevation.

Fig. 1
figure 1

Maps showing the position and arrangement of the treatment and reference sites. Panel a shows a map of New Zealand with the study location marked in red. Panel b shows where the treatment and reference sites were relative to each other with the point marked ‘T’ representing treatments site and point marked ‘R’ representing the reference site. Panels c and d show maps of the Treatment and Reference sites, respectively. Red triangles show the position of all wasp nests discovered prior to baiting while blue squares mark the position of bait stations. Baits were spaced at 50-m intervals with 300-m spaces between baitlines as per manufacturer guidelines. The distance between these sites is approximately 4.5 km. Maps made in QGIS version 3.16 (QGIS Development Team 2021) using base map data sourced from Land Information New Zealand (https://data.linz.govt.nz/)

Full size image

Wasp nest marking and monitoring

In January 2023 over the course of 10 days (10–20 January), both the 64-ha treatment area and the 16-ha reference area were systematically searched for wasp nests. The areas were visually surveyed by two researchers walking slowly through the forest block. When a nest was discovered, it was marked with flagging tape and the location and elevation recorded using a GPS (Garmin 65 s) accurate to 15 m. The traffic rate, or average number of wasps entering and exiting the nest over the course of 1 min, was recorded for each nest. The traffic rate of the nest is an established measurement of wasp activity and is linked to colony size (Malham et al. 1991). One researcher counted the number of wasps entering the nest while, at the same time, another researcher counted the number of wasps exiting the nest. The sum of these numbers was recorded as a traffic rate over 1 min and the process was repeated 3 times sequentially, to calculate an average traffic rate. All traffic rate monitoring occurred during daylight hours between 9:00 am and 5:00 pm. Monitoring was halted during rain.

In March 2023 (20–22 March), these nests were resurveyed prior to baiting. Earlier re-surveying and baiting was thwarted by extreme weather events in February which made the sites inaccessible until mid-March. Traffic rates and any mortalities were measured and recorded before any bait was applied. Additionally, 17 nests that were discovered had their traffic rates calculated and recorded. A nest was considered dead if no wasps were observed during the 3 min of observation which was usually confirmed with another 3-min observation period the following day. Given the life history of these wasps, we assumed that all nests found in March, prior to baiting, were alive in January and were used to estimate natural or pre-baiting mortality rates. Baiting commenced on 23 March 2023 (for the purpose of the analysis this date is considered day 0).

Baiting regime

Within the 64-ha plot of forest, 4 bait lines were established. Wasptek™ bait stations (Merchento, New Zealand) were screwed onto trees at approximately 1.2 m high. These bait stations were spaced at 50-m intervals along the line with each line being spaced 300 m apart as suggested by the manufacturer and as used by Edwards et al. (2017). These distances were measured using GPS. In total 52 bait stations were used across the 64-ha treatment area (Fig. 1). In each bait station, approximately 30 g of Vespex® bait (chicken-based paste with 1 g/kg fipronil) was applied on 23 March 2023. On 26 March 2023, any remaining bait was collected and removed from the site as per manufacturer and Environmental Protection Authority (EPA) guidelines. The duration we allowed the bait to remain in the environment was the minimum suggested by the manufacturer. It is recommended to bait during sunny, clear conditions to encourage the greatest bait uptake. As mentioned, extreme weather events forced the baiting operation to take place later in the season than previously planned and so there were constraints on time due to site accessibility and appropriate baiting weather. The baiting window was flanked by periods of wind and precipitation and so could not be extended. Due to this constraint the bait was left out only 3 days compared to other studies such as Edwards et al. (2017) who had bait available for between 7 and 13 days.

Post-bait monitoring

Nests were checked and monitored again immediately after baiting. Due to the size of the field site, it took two days to check all nests across both reference and treatment sites. For this analysis, counts were clustered together and ascribed the later assessment date. All nests were checked by 27 March, the 5th day after baiting (hereafter, day 4). Traffic rates were calculated as described above, and any nest mortalities were recorded. Finally, nests were checked again in April, with all nests monitored by 11 April, the 20th day after baiting (hereafter, day 19). Similar fipronil-based baits have been shown to produce declines in nest traffic rate 24 h after application (Sackmann et al. 2001).

Nest extractions

Once all data collection had taken place, 15 of the surviving nests were exhumed for observation to visually assess the health of the surviving nests and assess the severity of sublethal effects of the bait. Twelve nests in the treatment site and three from the reference site were excavated and examined. Excavations occurred between 21 and 22 days after bait application. The size of the nest was visually inspected and evidence of new season queens as well as dead or dying workers was noted. These observations were purely qualitative, so no formal analysis was carried out.

Bait consumption

Thirteen bait stations were concurrently monitored to assess the level of bait uptake by wasps. Identifying rates of bait uptake can indicate how much bait may be required for an area or how long bait can be left out. Looking at the variation in bait uptake may tell us where efforts could be focussed. Baits were weighed using digital scales when they were placed out in bait stations on 23 March and reweighed daily until they were brought in on 26 March 2023. To account for potential evaporation, 3 baits were placed in a ventilated cabinet that roughly imitated the forest conditions but prevented wasps or other species from accessing the bait.

Camera data

During the baiting process, cameras were placed out at a subset of nests to assess the daily activity patterns of wasps at the nest. Cameras were set to record 1 minute of video footage every hour between the hours of 06:00 and 22:00. Cameras were placed in front of six nests, three at the treatment site and another three in the reference site. A4-sized pieces of fluted plastic board were placed near the entrance of the nest to allow a better view of wasps entering and exiting the nest, as the plastic gave a more visually contrasting background (Fig. 2). Recording began two days before baiting commenced and ran for a further four days after baits were deployed. Cameras were placed out again in April at the same nests for a further day. Footage was analysed and traffic rates calculated for each time stamp across the filming period. Unfortunately, only two cameras managed to capture footage over this entire period, one in the treatment site and one at the reference site, so this data is presented as an example of activity patterns from two nests.

Fig. 2
figure 2

The remote camera set up. The camera (to the right) was powered by an external power pack. The nest entrance is indicated by the red arrow. The yellow plastic flute board was used to create contrast, to more easily observe wasps entering and exiting the nest

Full size image

Statistical analysis

All statistical analysis was performed using R version 4.2.0 (R Core Team 2022). Mean proportional changes in traffic rate measured 19 days after baiting across the two sites were analysed using a Wilcoxon rank sum test with a continuity correction using the ‘wilcoxon.test’ function, from the ‘stats’ package. The test statistic (W) and p-value (p) are reported.

For the mortality analysis, logistic regression was used to evaluate how different variables influenced mortality. A binomial response variable was assigned to represent nest mortality (1 = alive, 0 = dead). Nest traffic rate has been shown to be a predictor of wasp nest size (Malham et al. 1991) so rates measured immediately prior to baiting were used as an estimate of nest size and denoted Size in the model. The distance between the nest and the nearest bait station in metres was also included as a term Bait distance. The terms Bait density 50 m, Bait density 100 m, and Bait density 200 m were used to denote the number of baits within a given radius of a wasp nest. Nest density 50 m, Nest density 100 m, and Nest density 200 m were three terms used to denote the number of other nests discovered within a given radius of a wasp nest. The full complement of terms was used to build an initial model using the ‘glm’ function, from the ‘stats’ package in R studio (R Core Team 2022). Backwards stepwise regression was used to select the variables in the final model using the ‘step’ function in the ‘stats’ package (R Core Team 2022). This process starts with the full complement of terms in a generalised linear model and works backwards, removing terms to produce the greatest decrease in the Akaike’s Information Criterion (AIC). When the removal of any further variables would produce an increase in AIC the process stops, and we are left with the final variables selected. The ‘Anova’ function from the package ‘car’ (Fox and Weisberg 2019) was used to produce a Type II analysis of deviance table to assess the relationship between the response and explanatory variables in the selected logistic regression model. The likelihood-ratio chi square statistic (G2), degrees of freedom (df) and associated p-values (p) are reported. Statistical significance was assumed at p < 0.05.

To explore potential drivers of sublethal effects, a subset of data was analysed using beta regression. Only nests that survived the baiting were used here. The proportional change in traffic rate was calculated from the traffic rate measured 19 days after baiting, divided by the traffic rate measured immediately prior to baiting for each nest. This proportion was used as the response variable Proportional change. A beta regression model was fitted using the ‘betareg’ function from the package ‘betareg’ (Cribari-Neto and Zeileis 2010). The model had the same full complement of terms listed above and similarly backwards stepwise regression was used to select terms for the final model using the ‘StepBeta’ function in the ‘betareg’ package. This selection process also used AIC to determine variable removal. A Type II analysis of deviance table was again produced using the ‘Anova’ function from the ‘car’ package (Fox and Weisberg 2019), producing likelihood-ratio chi square statistic (G2), degrees of freedom (df) and associated p-values (p).

To investigate how the degree of bait consumption impacts changes in wasp activity, the mean proportional change in nest traffic of nests within 50 m of the bait stations was plotted against the proportional change in bait mass at each station. To ensure non-negative fitted values the association was modelled using an exponential (semi-log) regression model in R (R Core Team 2022). A small positive constant (0.0001) was added to the mean proportional change in nest traffic, to avoid problems with logarithmic transformation of a response variable taking zero values. Note that as nest traffic could increase if any nests grew, no upper limit on proportional change in nest traffic exists in practice. The effect size and associated standard error, F-statistic (F), degrees of freedom (df), R2 value and associated p-value (p) are reported.

Seasonal autoregressive integrated moving average (ARIMA) models (Hyndman and Athanasopoulos 2021) were used to investigate the effect baiting and temperature had on the nest traffic daily patterns recorded by the trail cameras. Regularly spaced hourly traffic data from each nest were converted into a time series. Hourly temperature readings from the nearest weather station were retrieved from the CliFlo database (https://www.cliflo.niwa.co.nz/). An indicator variable was included to represent times following bait deployment (Online resource 1). ARIMA models were fitted to the observed nest traffic using the ‘auto.arima’ function in the R ‘forecast’ package (Hyndman and Khandakar 2008), and a best model was selected by minimising Akaike’s Information Criterion (AIC).

We analysed the spatial arrangement of wasp nests found in the treatment site to compare to the arrangement of baits using nearest neighbour index (NNI). These statistics were calculated using the ‘spatstat’ package (Baddeley and Turner 2005) in R. GPS points of nests were converted to point pattern datasets using the ‘ppp’ function in the ‘spatstat’ package. A polygonal boundary was drawn around the outermost GPS points to form the observation window. The NNI was then calculated as the observed mean nearest neighbour distance divided by the expected mean nearest neighbour distance (Clark and Evans 1954; Rogerson 2001). This expected value is derived from a hypothetical point pattern with the same number of points following a random distribution across the same sized area. A value less than one suggests an aggregated distribution and a value over 1 suggests a dispersed or uniform distribution. The observed mean nearest neighbour distance was calculated using the ‘nndist’ function. The expected mean nearest neighbour distance (\({\overline{r} }_{{\text{Expected}}}\)) was defined as one-half times the square root of the number of points (N) divided by the area of the observation window (A) (Clark and Evans 1954).

$${\overline{r} }_{{\text{Expected}}}=0.5\sqrt{(}N\div A)$$

A z-test was conducted, and significance was assumed at p < 0.05.

Results

In total 75 wasp nests were discovered and monitored prior to baiting. Fifty-nine wasp nests were discovered during the January monitoring period, 50 nests in the treatment site and 9 in the reference site. In March, a further 16 nests were discovered (13 in the treatment site, 3 in the reference site). All but 3 nests were confirmed to be Vespula vulgaris. These other nests belonged to the related Vespula germanica and were excluded from the data and analysis.

Mortality

Between January and March, prior to baiting, 3 out of 72 nests died. All 3 nests were from the treatment site, but no wasp control action had taken place over this time and so these mortalities could only be attributed to natural events such as flooding (a number of nests were in close proximity to waterways).

Across all nests in the treatment site, the bait application produced a median reduction in traffic rate of 96.5% by day 19 (Fig. 3a). By comparison, traffic rates in the reference site were more variable with a median decline of 38%, 19 days after baiting. The mean proportional change in traffic rate was significantly larger at the treatment site than at the reference site (W = 649, p-value < 0.001). Over the course of the study, no nests in the reference site were observed to die. In the treatment site one nest had failed 4 days after baiting. By day 19, the failed number was 17 out of 57 or 29.8% (Fig. 3b).

Fig. 3
figure 3

Plot a shows a pair of boxplots comparing the change in traffic rates measured before and after baiting at both the treatment and reference sites. The y-axis shows the proportion of the traffic rate measured prior to baiting that remained 19 days after baiting. The dashed line represents a value of 1 and values over or under 1 represent an increase or decrease in traffic rate, respectively. A value of zero indicates a nest that has died. The median value for nests in the reference site was 0.62, suggesting a drop in traffic rates of 38%. The box indicates the interquartile range while the bold black line shows the median value. The median decline in traffic rates in the treatment site was much more severe at 96.5%. Plot b shows binned traffic rate declines at nests in the treatment area. The numbers above the bars show the number of nests within each bin. All nests within the treatment area experienced declines in traffic rate after 19 days. Seventeen nests out of 57 experienced declines of 100% meaning that no wasp activity was observed at these nests during post-bait monitoring, so they were considered dead

Full size image

Logistic regression was used to investigate predictors of wasp nest mortality, with variables for the generalised linear model selected via backward stepwise regression. The variables included in the final model were Size, Bait distance, Nest density 50 m and Nest density 100 m. The size of the nest was found to have a positive association with nest survival suggesting that, given all else, larger wasp nests were more likely to survive baiting than smaller ones (βSize = 0.040 ± 0.020, G2 = 5.317, df = 1, p = 0.021) (Fig. 4a). A positive association was also found between mortality and Bait distance suggesting that, given all else, the odds of survival increased at nests with increasing distance to the nearest bait (βBait distance = 0.025 ± 0.010, G2 = 8.519, df = 1, p = 0.004) (Fig. 4b). Interestingly, the analysis showed a significant positive association between mortality and nest density. The term Nest density 50 m had a significant negative coefficient (βNest density 50 m = − 1.133 ± 0.392, G2 = 12.179, df = 1, p = 0.001) indicating that, given a constant value at all other variables, an increase in the number of other nests within 50 m of a given nest leads to a higher odds of nest mortality. The term Nest Density 100 m had a significant positive coefficient (βNest density 100 m = 0.467 ± 0.232, G2 = 5.720, df = 1, p = 0.017).

Fig. 4
figure 4

Bivariate plots of lethal and sublethal effects of toxic baiting. On plots a and b the y-axis shows nest survival, where a value of 1 or 0 represents a nest that survived or did not survive baiting, respectively. In plot a, traffic rate count prior to baiting was used as a proxy for nest size and was shown to have a positive association with nest survival. This trend suggests that smaller nests experienced higher chances of mortality. Plot b shows that the distance from the nest to the nearest bait is positively associated with nest survival, suggesting that nests closest to baits experienced higher chances of mortality. Plots a and b show fitted sigmoidal trendlines. Plots c and d include only data from nests that survived baiting. The y-axis of plots c and d shows the proportion of the traffic rate measured prior to baiting that remained 19 days after baiting. A value of 1 represents no change in traffic rate while values closer to 0 represent a larger decrease in traffic rate and therefore more significant sublethal effects. Plot c shows that with increasing nest size (measured as traffic rate prior to baiting), nests tended to experience greater declines in nest activity. Plot d shows that with increasing distances from the nest to the nearest bait, declines in nest activity reduce. Blue lines in plots c and d show fitted exponential trendlines

Full size image

Sublethal effects

Nests that survived baiting at day 19 often still exhibited important sublethal effects, most notably a decrease in traffic rate. Across the 40 surviving treatment nests, by day 19 traffic rates had declined by 83% on average, with all treatment nests experiencing a reduction in traffic rate by this time. Nests in the reference site, on average, also experienced an overall decline in traffic rate of 30% by day 19, but trends here were more varied with some nests increasing in activity. Cooler conditions almost one month later may have contributed to this decline.

Beta regression analysis was used to investigate the predictors of the severity of sublethal effects in the nests that survived baiting. In this analysis, nests that died were excluded. The proportional response variable Traffic rate proportion was calculated as the final traffic rate count divided by the traffic rate count immediately prior to baiting. Variable selection was carried out using AIC and backward stepwise regression. The predictor variables used in the final model were Size, Bait distance and Nest density 100 m. Based on this subset of nests, Size appeared to have a significant negative association with Traffic rate proportion (βSize = − 0.018 ± 0.006, G2 = 8.331, df = 1, p = 0.004) (Fig. 4c). This negative association implies that the larger of the remaining living nests, given all else, would be expected to experience a more dramatic decline in traffic rate. Bait distance was found to have a positive association with Traffic rate proportion (βBait distance = 0.009 ± 0.003, G2 = 6.597, df = 1, p = 0.010) (Fig. 4d). Like mortality, this association would suggest that given all else remains constant, nests further from bait stations would expect to experience less severe declines in traffic rate after baiting. Nest density 100 m was also included in the model using the AIC criterion and showed a negative association with Traffic rate proportion; however, this was found not to be statistically significant (βNest density 100 m = − 0.090 ± 0.050, G2 = 3.231, df = 1, p = 0.072).

Nest extractions

Nests in the treatment site reacted less aggressively to disturbance and extraction of the nest, with few workers typically observed defending the nest. Evidence of new season queens was present at all nests, suggesting the baiting did not achieve the ‘reproductive failure’ of the treatment nests sampled. Nests extracted in the treatment site also contained a large number of dead wasps at varying states of decay (Fig. 5). A strong, sour odour was emitted as we broke into the bottom layers of these treatment nests where dead workers appeared to accumulate. Mould was growing among the bodies of these wasps. Such accumulations were absent from the three nests excavated from the untreated reference site. Nests in the treatment site were smaller than those excavated from the reference site with, on average, 9.7 layers compared to 14.3 layers, respectively.

Fig. 5
figure 5

Pictures of a nest that was extracted from the treatment site in mid April 2023. Inside this nest, and in others from the treatment site, were large numbers of dead wasps that appeared to be decaying within the nest after baiting. There was a strong odour detected once the outer nest envelope was removed and large numbers of dead wasps were found mainly among the lower layers of the nest. These large deposits of dead wasps were not present in the nests extracted from the untreated reference site

Full size image

Bait consumption

Some wasps were observed to be consuming bait within an hour of it being placed in the bait stations, however, all stations contained bait remaining when checked on day 3 after baiting. Bait stations checked daily during the operation had dropped on average 3.26 g per day. The reference baits had lost an average of 1.62 g per day suggesting that, accounting for evaporation, an average of 1.64 g of bait per day was consumed at each station over the course of the baiting period. After 19 days, reference baits had decreased in mass by 16.7% on average, while baits out in the treatment site experienced a decrease of 33.1% on average (Fig. 6a). Bait consumption was variable across different stations in the treatment site. Bait station 8, located in a shaded, heavily forested valley, experienced less change in mass than even the three reference baits at 14.8%, while bait station 13, located on the bush edge lost 54.2% of its initial mass, by day 3 (Fig. 6a). This suggests bait placement in the environment may influence bait uptake. To investigate whether bait consumption was associated with changes in wasp activity, the proportion of bait remaining after the conclusion of the operations was plotted against the mean proportional change in traffic rate after 19 days, using nests within 50 m of each bait station (Fig. 6b). There was a statistically significant positive association (βlocal nest decline = 18.02 ± 6.59, F = 7.475 on 1 and 8 df, R2 = 0.483, p = 0.026).

Fig. 6
figure 6

Plot a shows the proportion of bait remaining at the 13 selected stations compared to 3 reference baits. Treatment bait stations (blue lines) were placed out on 23 March (day 0) and weighed daily until they were retrieved on 26 March (day 3). To account for evaporation, 3 reference baits (red lines) were placed in a ventilated cabinet to simulate outdoor forest conditions but prevent wasps from accessing the bait. Plot b shows the association between bait consumption and nest activity decline. Proportion of bait remaining is plotted on the x-axis. The y-axis shows the mean of the proportion of pre-baiting traffic rate calculated as the traffic rate of local nests measured 19 days after bait application divided by the traffic rate of those same nests measured prior to baiting. Only nests within 50 m of the bait stations were used in this analysis. The numbers shown on the plot correspond to the bait station. Some bait stations are excluded from the plot due to the lack of wasp nests within 50 m of the bait. The dark blue line of best fit shows a positive and statistically significant association between variables

Full size image

Daily activity patterns

Camera footage captured at the entrance of two wasp nests allowed observation of wasp activity patterns (Fig. 7). The observed nests tended to be active for between 13 and 14 h per day. Wasp activity began between 6:00 and 7:00 and ceased between 20:00 and 21:00, approximating to roughly 1 h before sunrise and 1 h after sunset, respectively. Nest activity was relatively variable and likely dependent on local weather conditions. Conditions on 21, 22 and to a lesser extent 23 March were comparatively cooler and wetter than the rest of the week (24–26 March 2023) and may have resulted in the slightly lower traffic rates seen in Fig. 7. Often two main spikes in activity occurred in the late morning and afternoon. The treatment nest shows a decline in traffic rate after baiting, which would be expected if the bait was impacting the nest. The reference nest from the untreated reference site does not show the same declining pattern, instead traffic rate increases in the 3 days after baiting. For both nests, various seasonal ARIMA models were constructed with a seasonal period of 24 h, using the regularly spaced observations on nest activity as the response variable, with temperature and bait presence as possible predictors. Neither temperature nor bait presence could significantly improve the model fit when evaluated by AIC. The time series of nest activity in the treatment site was best explained by a seasonal ARIMA(1,0,0)(1,1,0)24 model, which used activity patterns from the previous day as an important predictor and also included a negative drift term to explain the long-term, hour-to-hour decline observed in mean nest activity (Online resource 2). The time series of nest activity in the reference site was best explained by a seasonal ARIMA(1,1,1)(0,1,1)24 model, which again used activity patterns from the previous day as an important predictor. The reference site model also used the activity from the previous hour as a predictor and hence did not need a drift term to explain long-term patterns within the data (Online resource 2).

Fig. 7
figure 7

Plot of wasp nest traffic at two nests captured using automated camera traps. Nest traffic rates were counted using the footage captured at 1-h intervals from 6:00 to 22:00 daily. Filming began on 21 March 2023. Baiting occurred on 23 March, indicated by the dashed line and blue shading. Cameras were retrieved on 26 March, providing 3 full days of footage after baiting. Cameras were placed out again at the same nests on 11 April, providing footage from 19 days after baiting, although this latter data was not fitted using the time series models. Red squares show the activity measured at the nest in the reference site while blue circles show the activity patterns at the nest in the treatment site

Full size image

Spatial analysis of nests

Visual assessment of the spatial arrangement of both the discovered wasp nests and bait stations within the treatment site suggest that the two follow different patterns (Fig. 1c). While the bait stations were systematically deployed in space, wasp nests identified and monitored over the course of the baiting program were found to be spatially aggregated. The observed mean nearest neighbour distance was calculated to be 36.9 m. Based on the Clark and Evans (1954) formula, the expected nearest neighbour distance for our searched area was 178.5 m producing a nearest neighbour index (NNI) value of 0.209 (z = − 3.394, p = 0.001). This value is less than 1, indicating a significantly aggregated or clustered distribution.

Discussion

The aim of this research was to identify the factors that might be responsible for the survival of wasp nests during population control using toxic baits. The fipronil based bait, Vespex®, was shown to produce a knockdown in wasp numbers and killed many nests. A median decline in traffic rate of 96.5% across all treated nests was observed with a mortality rate of 29.8%, 19 days after bait application. All nests in the treatment area experienced a decline in traffic rate. Nests at the reference site showed a more variable pattern with an overall median decrease in activity of 38%. Some nests at the reference site showed declines in activity while for others activity increased over time. Logistic regression analysis suggested that all else remaining equal, larger nests, nests further from baits, and nests more isolated from their neighbours appeared to be associated with higher odds of survival after baiting. Similarly, beta regression suggested that given all else remaining equal, less severe sublethal effects of baiting were associated with smaller nests and those nests further from bait stations.

Some of these associations may be explained by patterns of forager recruitment to bait stations. Recruitment is the process by which individuals can signal nestmates to gather at a location, usually to exploit a resource (Czaczkes 2021). Social wasps including Vespula spp. exhibit some level of recruitment behaviours where foragers exchange information to promote nestmate discovery of the same resources (Schueller et al. 2010; Wilson-Rankin 2014; Santoro et al. 2015; Lozada et al. 2016). When competition for food resources is high wasps have been shown to quickly capitalise on resources close to the nest (Wilson-Rankin 2014), potentially explaining why we observed that both nest mortality and severity of sublethal effects are associated with distance to nearest bait. A link between recruitment ability and nest size was shown by Wilson-Rankin (2014), where smaller annual nests exhibited higher bait visitation in response to successful nestmates returning, compared to larger perennial nests. The author goes on to hypothesise that the wasps from the larger perennial nests may be used to foraging at greater distances from the nest and may overshoot baits that other nestmates have discovered. This foraging pattern may suggest that if a bait is discovered, small nests will more readily exploit it leading to larger doses of insecticide and increased chance of mortality. Larger nests may, in turn, show a weaker response to these baits which could explain their increased survival.

Larger nests have both physiological and behavioural advantages that may explain their lower risk of mortality compared to smaller nests. Dobelmann et al. (2017) showed that larger V. vulgaris nests often contain higher genetic diversity and lower viral loads. Higher genetic diversity and lower viral loads have subsequently been linked to increased insecticide tolerance in social insects (Milone et al. 2020; Zhu et al. 2022). Additionally, it was shown that insecticide tolerance increased with group size in termites, likely due to an increase in stress relieving behaviours (DeSouza et al. 2001; Watanabe et al. 2023). These benefits may explain increased survival of our larger nests, in addition to larger nests needing higher doses of toxic bait to achieve mortality. In the context of population control, it seems likely that ensuring enough bait is taken up by wasps from larger nests is essential to produce nest failure. Allowing for longer baiting windows may facilitate higher bait uptake.

The apparent susceptibility of larger nests to sublethal effects is likely a case of survivorship bias. The beta regression identified a negative association between nest size (measured as the traffic rate count of nests prior to baiting) and the severity of sublethal effects (measured as the proportion of the traffic rate measured prior to baiting that remained 19 days after baiting). It is possible that small nests that survived baiting experienced reduced sublethal effects because they did not encounter bait. Larger nests rapidly deplete resources near to them and so tend to have to forage further from their nest than their smaller counterparts (Wilson-Rankin 2014). This increased foraging distance makes it more likely that foragers from larger nests encounter baits in the landscape. Smaller nests by comparison have a reduced need to forage as widely and are less likely to encounter baits further from them. Increasing bait density would be needed to ensure that these smaller nests encounter baits and perhaps increase the efficacy of this eradication treatment.

Our analysis showed that in scenarios where all else would remain constant, nests closer to baits would experience more significant declines in traffic rate while those further away would be less affected. This result is mirrored in a study by (Harper et al. 2016b) who similarly showed that nests located further from bait stations experienced less severe declines in traffic rate. Despite this trend, both the Harper et al. study and this current study identified nests that survived baiting near bait stations. Surviving nests excavated 3 weeks after bait application displayed various levels of activity but all showed evidence of queen survival. Three surviving nests identified by Harper et al. within 100 m of baits were active 6 weeks after bait application. Queen survival was not measured in that study. Optimistically, one might hope that despite nest activity after treatment, the absence of surviving queens would effectively mean the nest is reproductively dead (with the exception of workers potentially producing males). Further research on the long-term survival of nests post-treatment is needed to identify rates of new queen mortality and whether this can be increased by adjusting the timing of bait application.

Logistic regression identified an association between higher nest density and lower nest survival after baiting. With the higher nest density we expect competition for resources to increase and, as already noted, in such situations wasps will quickly capitalise on resources close to the nest (Wilson-Rankin 2014), leading to increased uptake of bait and hence increased mortality. Viral and bacterial loads have been shown to increase with nesting density in Vespula pensylvanica (Loope and Wilson Rankin 2021), which as discussed above can predict susceptibility to pesticides (Zhu et al. 2022). Additionally, nest drift has been observed in social wasps including Vespula vulgaris, where individuals may temporarily inhabit and even forage for a nest that is not their own (Sumner et al. 2007; Santoro et al. 2019). This behaviour increases with nesting density and may increase a nest’s exposure to toxic baits.

In this study bait was available to wasps for only 3 days due to constraints from time and weather. This bait availability may have not been sufficient and may explain the lower level of mortality compared to Edwards et al. (2017). Their study used a similar 50 m by 300 m baiting strategy and was able to produce mean declines in traffic rate of 93% or more by 20 days post-baiting. Mean decline in traffic rate in our study only reached 88% by this same time. Baits were available to wasps in the Edwards et al. study for between 7 and 13 days, significantly longer than was possible in this current study, which may have allowed greater uptake of bait and more significant declines in wasp activity. Harris and Etheridge (2001) observed approximately 40% of bait was removed after 5 days. By comparison, we estimate less than 20% of bait placed out in our study was consumed after 3 days. Harris and Etheridge used a higher bait density than used in this study. Additionally, both Harris and Ethridge (2001) and Edwards et al. (2017) commenced baiting earlier in the year than was possible in our study, which could have influenced bait uptake. An earlier baiting date and longer duration of bait access may have reduced the number of new season queens that were observed to survive in our study. Nevertheless, despite weather constraints leading to reduced baiting duration, this study highlighted variation in baiting efficacy over a landscape. It will be important to address this source of variation when designing future baiting programs.

Though not identified in the present analysis, temperature is known to be a predictor of wasp foraging activity (Kasper et al. 2008). Camera footage of the two wasp nests shows evidence of a bimodal pattern of activity throughout the day. The results of our own analysis are based on only two nests and so we must be careful not to overstate our findings. It is possible, however, to see patterns of high daily wasp activity in the morning, followed by another spike in the afternoon. This pattern is not perfect and likely heavily influenced by local climate but has been observed in other wasp species such as by Santos et al. (2010). These authors observed social wasp foraging rates spike twice throughout the day, once in the morning and once in the later afternoon. This bimodal foraging pattern has also been described in social insect pollinators including bees (Apis cerana and Bombus spp.) as well as Lepidopterans and Dipterans, likely to avoid overheating in the midday sun (Xu et al. 2021). The pattern appears weakest in the data collected in our study, on days that were overcast and raining (21 and 22 March) (Fig. 7). Wasps were still observed to be foraging during these wetter days though. During one particularly heavy downpour, two nests were observed to have wasps entering and exiting with surprisingly high frequency. Kasper et al. (2008) showed that in Vespula germanica foraging activity drops by around 30% during rain but recovers rapidly once precipitation stops. Their data also showed foraging began and ended slightly before and after daylight, respectively. This pattern is similar to what we observed in the two V. vulgaris nests observed in this study.

The spatial arrangement of the bait stations was predetermined based on a regular 50 m × 300 m grid superimposed on the landscape. The placement of these baits was not influenced by the surrounding environment except in the cases of researchers being unable to reach the specific point (due to a steep gully or ravine). While convenient for the design of baiting programs, it is inconceivable that wasps choose their nesting location in the same way. The nests discovered in this study were shown to be spatially aggregated. Nest site selection in other hymenopteran groups has been shown to be driven by factors such as correct substrate type, optimal thermal conditions, resource distribution and patterns of dispersal (Strassmann 1991; Thomas 2002; Antoine and Forrest 2021; Veldtman et al. 2021). A lack of optimum nesting conditions can enhance this aggregated pattern of nesting (Potts and Wilmer 1997). It is likely that some aspect of the environment is dictating the distribution of wasp nests in our treatment site. Given we found that distance to bait is a predictor of nest mortality, placing baits close to nests would be ideal in order to increase baiting efficacy. Baiting would be optimised by targeting the pesticide to where wasp nests are located though currently, we lack the sufficient knowledge to be able to successfully predict nest locations. More research should be conducted on what informs wasp nesting behaviours, to focus baiting efforts where nests are most aggregated.

To achieve eradication of invasive social insects like wasps, control methods must be highly effective (Phillips et al. 2019; Lester et al. 2020; Howse et al. 2023). It is important to understand and learn from both successful and failed control programmes, especially identifying why some individual colonies survive while others die. Ongoing research is needed to achieve goals of large-scale eradication programs such as the ambitious ‘Predator Free New Zealand’ initiative highlighting this sentiment (Russell et al. 2015; Kopf et al. 2017). While our study area encompassed a combined area of approximately 80 hectares, it was limited to a single treatment site. Despite the difficulties of research at large scales we highlight its importance if we are to achieve landscape-scale results.

Analysis identified that bait efficacy is heavily influenced by both the size and distribution of wasp nests in the environment. Bait uptake was shown to be variable and the spatial distribution of the wasp nests themselves exhibit significant clustering. Investigating the drivers of nest site selection and bait uptake in different environments may allow for a more advanced, flexible baiting arrangement that could maximise bait discovery and efficacy. In the meantime, maximising baiting duration may increase the likelihood of all nests encountering bait, which would target those smaller nests that appear to more effectively avoid baiting, as well as larger nests which appear to better tolerate bait exposure.

Author contributions

M W F Howse and P J Lester contributed to the study conception and design. Data collection was performed by M W F Howse and A Reason. Analyses were performed by M W F Howse and J Haywood. The first draft of the manuscript was prepared by M W F Howse, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.