Orkney Native Wildlife Project - Stoat Trapping Trial ... Trapping Trial... · Stoat Trapping Trial...
Transcript of Orkney Native Wildlife Project - Stoat Trapping Trial ... Trapping Trial... · Stoat Trapping Trial...
Stoat Trapping Trial Analysis Report
Iain Malzer, Bea Ayling, Carmen Biondo
Introduction:
Trapping trials have been undertaken in three phases for the Orkney Native Wildlife Project (ONWP) across three seasons: winter, summer and autumn. These trials were undertaken to assess the effectiveness of different combinations of trap type and bait, in key habitats, across different seasons in catching stoats on the Orkney Mainland as a prelude to a major eradication project planned for 2019.
The aims and results of the initial winter trial, conducted by contractors, SCL, can be found in a
separate report, however data collected in this trial was examined as part of this wider trapping trial
analysis report.
It was decided to undertake further trapping trials in summer and autumn to address some of the
shortcomings of the initial trapping trial carried out between December 2017 and February 2018,
and to collect trapping data during the summer season. This will serve two purposes: inform the HRA
with regards to disturbance to breeding birds, and to determine whether trapping rates differ when
natural food availability is greater, and females have dependant young.
The aims of both the summer and autumn trapping trials were to:
1. Test the attractiveness of alternative bait types to stoats: smoked fish, beef offcuts, dried
rabbit, tinned cat food and hens’ eggs – and synthetic lures.
2. Assess the effectiveness of trapping during the summer season when natural food is more
readily available
3. Assess the effect of season, trap type and habitat type on catch success.
Methods:
Stoats were trapped across three, roughly 5km2 zones, on the Orkney Mainland, west of Kirkwall:
Hobbister, Grimbister and Wideford. Within each of these zones, traps were set on a grid of 350m
resulting in 36 trapping locations per zone (in winter) and 18 (in summer and autumn). Traps were
not rigidly set at grid locations, but were placed along linear features such as walls or ditches to
maximise encounters with stoats. The traps were set almost equally across two major habitat types:
‘heathland’ and ‘fields’ (including both improved and unimproved grassland).
Trapping trial study areas: the number of trapping locations used during summer and autumn trials. In winter,
double the number of traps were used in each of these areas.
At each trapping location, effectively three traps were placed: two (double‐set) DOC150s in a double‐
entrance tunnel and one (single‐set) DOC200 in closed‐end tunnel.
The double‐set DOC150 double‐entrance tunnel set up was changed between trial periods. An internal
wire mesh barrier between the two traps (effectively creating two closed‐end tunnels within the same
box) was present during the winter and summer trials. This baffle was removed from all double‐set
traps in August 2018 following advice from Angela Newport (an expert trapper and stoat detection
dog handler) and the autumn trial was conducted without them present.
The single‐set DOC200 trap was placed in a shorter, closed‐end tunnel. The distance from the box
entrance to the trap was extended using a short section of drainage tube attached to the front baffle.
This design amendment was added to these closed‐end tunnels following an earlier incident prior to
the start of the trials when a cat caught its paw in a trap.
As mentioned above, three trapping trials were undertaken in total:
1. The winter trial took place between 7th December 2017 and 14th February 2018 and
involved 7 checks of the traps (undertaken by contractors SCL)
2. The summer trial took place between 4th June and 5th August 2018, involving 6
approximate weekly checks by RSPB/ONWP staff; and
3. The autumn trial took place between 11th September and 5th November 2018 and involved 6
approximate weekly checks by RSPB/ONWP staff.
The summer and autumn trials varied from winter trials in their use of baits and lures. During the
winter trial, a single type of bait was used, fresh skinned rabbit. For this reason, data on bait
preferences for winter trial are not available.
In the summer and autumn trials, a more systematic approach to baiting / luring was undertaken. Five
bait types, smoked fish, beef offcuts, dried rabbit, tinned cat food and hens’ eggs – and in the autumn
only, synthetic lures were used in rotation across all trapping areas.
Results and Analyses:
Trap encounters were low with 75 stoats caught over all three trials and thus the data are sparse.
Data can be analysed at the following levels:
The winter trial: this data can be tested for impacts of zone, habitat, trap type, and days since
trapping began.
The summer and autumn trials: this data can be tested for impacts of zone, habitat, trap type,
bait and days since trapping began.
Winter trial summary
In the winter trial there were 3 zones, 36 trap locations, effectively 3 traps (2xDOC150; 1xDOC200) at
each trap location and 7 trap checks. Each single trap in each trap round was recorded as a single line
of data, resulting in 2268 data points. Some traps were not set on a given round due to loss or
malfunction and so 27 lines of data which had 0 trap nights were dropped from the winter analysis.
In the winter trial, each of these had the related variables:
Zone: Hobbister, Grimbister, Wideford
Trap type: double‐set DOC150 in double‐entrance tunnel, single‐set DOC200 in closed‐end
tunnel
Habitat: heathland, fields
Lure peg used: Yes or No
Trap nights: integer, the number of days the trap was active in each round.
Time days: The number of days since trapping began in the given zone.
Stoat capture ‐ this was coded as a dummy variable: 1 for capture; 0 for no capture.
There was a total of 19,833 trapping nights in the winter trial, with each trap left out for a mean of 8.9
nights (min 3, max 21) per round. The total number of stoats captured in the winter analysis was 41.
However, it has only been possible to use 33 catches in the statistical analysis as the catch data from
the final round of trap checks were not provided by SCL.
Of these 33 stoats, 29 (88%) were caught in double‐set DOC150s in a double‐entrance tunnel (with
wire mesh partition between the two traps) and 4 (12%) were in the single‐set DOC200 trap in a
closed‐end tunnel. On three occasions, both DOC150 traps within the same box at a single location on
a single round had captured two stoats.
1.5% of checks to traps resulted in a stoat capture and the rate of capture in the winter trial trapping
night was 0.002. The rate of capture per trap night was 0.0006 in fields and 0.0026 in heathland. The
rate of capture per trap night declined with the trap round (see Figure 1)
Winter trial modelling approach
A logistic modelling approach within a generalised linear mixed model framework was used to model
the binary variable stoat capture as a function of the variables: zone, trap type, habitat, bait stick used
and time days. To model expected variation between the individual trapping locations, the GPS label
of the trapping location was included as a random effect. Time days coded as a categorical variable
was also fitted as a random effect in the initial models but did not explain much variation while tending
to prevent models converging and was therefore removed. To account for the fact that traps left out
for longer may have a higher success rate, the log of the trapping nights variable was included as an
offset variable.
A stepwise AIC approach was used to select the most parsimonious model from the various candidate
models as the random effect term did not change between models.
Table 1 shows the estimates and significance of the terms included in the model. The terms that
remained in the most parsimonious model are shown in bold.
The effect size of the categorical variables is shown relative to the base level. The time since the
trapping trial began in winter was a highly significant term showing a decrease in the probably of
capturing a stoat per day since trapping began (Figure 2). The trap housing type was also highly
significant and suggested an increased chance of catching stoats in the double‐set DOC150s in a
double‐entrance tunnel. The final model also showed an effect of habitat, where the traps in
heathland had a considerably higher chance of catching stoats but due to a wide standard error, this
term was not significant. The bait and zone terms were not useful in explaining the variation in the
probability of stoat captures.
Variable Estimate (SE) Z Value P Value
Time days ‐0.05 (0.01) ‐4.00 <0.001*
Trap type Double‐set DOC150 in
double‐entrance tunnel
1.58 (0.58)
2.71
<0.001*
Habitat: Heathland
1.65 (0.99)
1.66
0.095
Table 1: Winter trapping trial model outputs. Terms that remained within the final model are highlighted in
bold.
Figure 1: Model predictions of probability of capture per day since trapping began for double‐set DOC150 traps
in a double‐entrance tunnel (black) and a single‐set DOC200 in closed‐end tunnel (red).
Dotted lines are the mean + or ‐ SE
Summer trial summary
In the summer trial, the three same zones were trapped. However, for the summer trial, only half
the number of the winter locations and traps were used, resulting in 18 trap locations and again they
Zone Hobbister Wideford
‐0.43 (1.06) ‐0.43 (1.08)
‐0.41 ‐0.46
0.68 0.63
Lure peg use Y
0.86 (0.66)
1.3
0.20
were almost equally split between ‘heathland’ and ‘fields’. The raw stoat catch data collected can be
found in the Appendix.
In total, during the whole period of the trial from 04/06/2018 and 05/08/2018, there were 978
occurrences of traps being set. 12 of these lines of data had to be removed due to having 0 trap
nights.
In the summer trial, each of these had the related variables:
Zone: Hobbister, Grimbister, Wideford
Trap type: double‐set DOC150 in double‐entrance tunnel, single‐set DOC200 in closed‐end
tunnel
Habitat: heathland, fields
Bait: egg, beef, cat food, smoked fish, dried rabbit
Trap nights: integer, the number of days the trap was active in each round.
Time days: The number of days since trapping began in the given zone.
Stoat capture ‐ this was coded as a dummy variable: 1 for capture, 0 for no capture.
There was a total of 9454 trapping nights in the summer trial, with each trap left out for a mean of 9.7
nights (min 3, max 27) each round. The total number of stoats captured in the summer trial was 13.
10 (76%) of these were in the double‐set DOC150 traps in a double‐entrance tunnel, and 3 (23%) were
in single‐set DOC200 traps in closed‐end tunnels.
There were no instances in the summer trial of two stoats being caught in the double‐set DOC150
traps in a double‐entrance tunnel. 1.3% of trap checks resulted in a stoat capture and the rate of
capture per summer trial trapping night was 0.001.
Seven stoats were caught in heathland habitat and six in grassland (five of which were in rough,
unimproved grassland, near or adjacent to heathland). The one stoat caught in improved grassland
was on the coastal edge.
The rate of capture per trap night was similar in fields and heathland during the summer trapping trial
and the decline in captures per trap round was less marked than in the winter trial.
Summer trial modelling approach
The logistic mixed modelling approach was undertaken for the summer analysis was largely similar to
that undertaken for the winter trial. The major difference was the addition of the multi‐level factor
for bait types in the summer models. Initially, an interaction term was included between the trap type
and bait type but limited the ability of the models to converge. The models included a random effect
for trapping location which helped explain much of the variation that was not explained by the
modelled fixed effects.
Due to the lack of captures there was very little power with which to explain the variation in probability
of capture and standard error within the estimates was high. Thus, few terms were deemed
significant. The most effective model for the summer trial data suggested an influence of bait type,
where hen’s egg showed a higher chance of catching stoats than the other bait types and time days.
The trap type, habitat and zone were not useful in explaining variation during the summer trial and
dropped out of the final models.
Variable Estimate (SE) Z Value P Value
Time days ‐0.06 (0.03) ‐2.06 0.04*
Bait type Cat food
Dried Rabbit Hen’s Egg
Smoked Fish
0.08 (1.66) 0.99 (1.66) 2.87 (1.33) 0.84 (1.62)
0.05 0.60 2.15 0.52
0.96 0.54 0.03* 0.60
Habitat Heathland
‐0.37 (1.15)
‐0.33
0.74
Zone Hobbister Wideford
‐0.48 (1.07) ‐0.15 (1.13)
‐0.45 ‐0.14
0.65 0.89
Trap Type Double‐set
DOC150 in double‐entrance tunnel
0.75 (0.73)
1.01
0.31
Table 2: Summer trapping trial model outputs. Terms that remained within the final model are highlighted in
bold.
Autumn trial summary
The autumn trial was similar to the summer trial. The same zones and trap locations were used. The
raw stoat catch data collected can be found in the Appendix.
The same baits were used, though there was the addition of synthetic scent lures instead of a food
bait on 126 traps during the whole trapping trial period. The total autumn trapping period was from
14/09/2018 until 05/11/2018. In total there were 969 occurrences that traps were set. This is slightly
lower than the expected number (3 zones, 18 trap locations, 3 traps, 6 checks = 972) as three traps
were not checked on one round due them becoming temporally unavailable.
Each trap had the following related variables:
Zone: Hobbister, Grimbister, Wideford
Trap type: double‐set DOC150 in double‐entrance tunnel, single‐set DOC200 in closed‐end
tunnel
Habitat: heathland, fields
Bait: egg, beef, cat food, smoked fish, dried rabbit, lure.
Trap nights: integer, the number of days the trap was active in each round.
Timedays: The number of days since trapping began in the given zone during this trapping
trial.
Stoat capture. This was coded as a dummy variable: 1 for capture; 0 for no capture.
There was a total of 7812 trapping nights in the autumn trial, with each trap left out for a mean of
8.06 nights (min 2.5, max 22) per round. The total number of stoats captured in the autumn trial was
29. 25 (86%) of these were in double‐set DOC150 traps in double‐entrance tunnels and 4 (14%) were
in single‐set DOC200 traps in closed‐end tunnels. There were no occasions during the autumn trapping
trial that both DOC150 traps placed in the same double‐entrance tunnel box captured two stoats in
the same round.
3% of trap checks resulted in a stoat capture and the rate of capture per autumn trial trapping night
was 0.0037. Eight of the captured stoats were in fields, while 21 were in heathland (total capture rate
in fields: 0.0024; in heath: 0.0046). Of the eight caught in fields, four were in unimproved grassland
or adjacent to heathland, three were in improved grassland but on the coastal edge and one was in
improved grassland but along a linear farm track.
Figure 2 shows the trend in capture rates as the trapping rounds progressed. There was a decrease
over the first five trapping rounds but trapping rate increased to its peak on the 6th round.
Autumn trial modelling approach
A logistic mixed modelling approach was used again for the autumn trapping data. Bait types were
treated in the same way as the summer data, but an extra bait type ‘lure’ was added. There were a
number of different scent lures trialled but due to a low sample size these were all pooled into a single
‘lure’ level within the bait type variable. The models included a random effect for trapping location
which helped explain much of the variation that was not explained by the modelled fixed effects.
The most effective model for the autumn trial data suggested a strong influence of the smoked fish
bait, which was estimated to have around 1.5 times higher capture rate than the other bait types. The
raw data suggest 11 (37%) of the autumn stoats were captured with smoked fish. There was no
interaction between the zone and the bait. The trap type also had a significant term with the double‐
set DOC150 traps in double‐entrance tunnels having a higher capture rate. There was a weaker effect
of habitat, where trapping in the heathland habitats resulted in higher capture rates, but this was not
carried through into the final model based on AIC. Interestingly, in the autumn models there seemed
to be no impact of time days on capture rate suggesting that stoat capture rates did not significantly
reduce during over the trapping period (this is coherent with Figure 2 which shows a weaker decline
in capture rates over the trapping trials in the autumn period). Table 3 below shows the results from
the best model.
Variable Estimate (SE) Z Value P Value
Bait type Cat food Dry Rabbit Hen’s Egg
Smoked Fish Lure
0.00 (0.82) 0.96 (0.69) ‐0.41 (0.92) 1.41 (0.66) ‐0.14 (0.67)
0.10 1.39 ‐0.45 2.13 ‐0.15
0.99 0.17 0.66 0.03* 0.88
Trap Type DOC150 in double‐entrance tunnel
1.22 (0.55)
2.23
0.06*
Habitat Heathland
0.75 (0.43)
1.75
0.08
Zone Hobbister Wideford
0.12 (0.46) 0.07 (0.48)
0.26 0.15
0.80 0.88
Time days ‐0.01 (0.01) ‐0.84 0.40
Table 3: Autumn trapping trial model outputs. Terms that remained within the final model are highlighted in
bold.
Figure 2: Rate of capture per trap night over the seven trap rounds during the winter (black) and six trap
rounds in the summer (red) and six trap rounds in the autumn (green).
Discussion and limitation of trials:
There are more stoat captures in the winter and autumn trapping trials which allows more
useful modelling of the data from these seasons. Rates of capture in winter and autumn were
higher than in summer so this is not due to there being more trap nights in winter. Higher
food abundance in spring and summer could mean stoats are less likely to use traps. It could
also be that there is a preference for fresh rabbit bait, but it was only used during winter so
we cannot compare these data.
There is a trend of declining captures since trapping began in winter and (albeit less marked)
in summer. This suggests there is an impact of trapping on the local population. This is not
the case in autumn when the capture rates decline but then increase again on the final round.
This could be due to the possibility of dispersing animals and an increased number of animals
moving into the trapping area.
The analysis shows that the double‐set DOC150 traps in double‐entrance tunnels had a higher
rate of captures. The models control for the fact there are effectively two DOC150 traps in the
same housing and there is a significant effect in the winter and autumn models. These trap
types also caught more stoats in the summer, and probably there is still an effect, but the lack
of captures overall means there is too high a standard error and the term falls out of the final
model. In addition, during the winter trial, there were three occasions when two stoats were
captured in the two traps within a double‐set DOC150 in double‐entrance tunnel. This is likely
due to a preference of the box type, where the elongated box with a tunnel structure appeals
to the inquisitive nature of stoats. This shows that this trap box type has a higher chance of
catching stoats.
In the summer models, despite a real lack of power, hen’s egg as a bait type showed a
considerably higher capture rate, and was probably the most useful term in the summer
models. In autumn, the hen’s egg bait was not useful for capture, and in fact the models
suggest this had the lowest capture rate. Only 2 stoats were caught on hen’s egg bait during
the autumn. In autumn, we saw significantly higher rates of capture with the smoked fish
bait in comparison to all other baits. Therefore, seasonality could be an important factor in
bait use.
Heathland has higher capture rate in winter and autumn, though this was not the case in the
summer models. It would be useful to have had a better habitat variable than ‘fields’, for
example, distinguishing between coastal grassland, rough in‐bye land and pasture. Also, it
would be worth marking if the traps are along some kind of boundary or linear feature, and
what this is.
Stoat gender was not able to be tested for due to lack of data. Many stoats were too
decomposed to be able to sex and therefore the dataset was much reduced.
Combining all seasonal data into one model would be worth trying, but since baits were used
inconsistently we would not be able to ask the models the full suite of questions.
Recommendations for TAG:
We would like the TAG to agree that:
‐ The ONWP should mostly use double‐set traps in double entrance tunnels for the
eradication.
‐ A mix of baits and lures should continue to be used and trialled through adaptive
management. Egg as bait should be used in spring and smoked fish in autumn.
‐ Trap placement should focus on heathland, rough unimproved grassland and coastal edge.
‐ All trapping data collected to date and into the future should be reviewed for stoat capture
preferences and the trapping operation adapted accordingly.
Appendix – Raw Stoat Catch Data: Summer and Autumn trials
NB: In the ‘stoat gender’ column, 0 refers to the fact that it was not able to be recorded mainly due
to decomposition.