Research Article |
Corresponding author: Arno Thomaes ( arno.thomaes@inbo.be ) Academic editor: Alessandro Campanaro
© 2017 Arno Thomaes, Pieter Verschelde, Detlef Mader, Eva Sprecher-Uebersax, Maria Fremlin, Thierry Onkelinx, Marcos Méndez.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
Thomaes A, Verschelde P, Mader D, Sprecher-Uebersax E, Fremlin M, Onkelinx T, Méndez M (2017) Can we successfully monitor a population density decline of elusive invertebrates? A statistical power analysis on Lucanus cervus. In: Campanaro A, Hardersen S, Sabbatini Peverieri G, Carpaneto GМ (Eds) Monitoring of saproxylic beetles and other insects protected in the European Union. Nature Conservation 19: 1-18. https://doi.org/10.3897/natureconservation.19.11761
|
Monitoring global biodiversity is essential for understanding and countering its current loss. However, monitoring of many species is hindered by their difficult detection due to crepuscular activity, hidden phases of the life cycle, short activity period and low population density. Few statistical power analyses of declining trends have been published for terrestrial invertebrates. Consequently, no knowledge exists of the success rate of monitoring elusive invertebrates. Here data from monitoring transects of the European stag beetle, Lucanus cervus, is used to investigate whether the population trend of this elusive species can be adequately monitored. Data from studies in UK, Switzerland and Germany were compiled to parameterize a simulation model explaining the stag beetle abundance as a function of temperature and seasonality. A Monte-Carlo simulation was used to evaluate the effort needed to detect a population abundance decline of 1%/year over a period of 12 years. To reveal such a decline, at least 240 1-hour transect walks on 40 to 100 transects need to be implemented in weekly intervals during warm evenings. It is concluded that monitoring of stag beetles is feasible and the effort is not greater than that which has been found for other invertebrates. Based on this example, it is assumed that many other elusive species with similar life history traits can be monitored with moderate efforts. As saproxylic invertebrates account for a large share of the forest biodiversity, although many are elusive, it is proposed that at least some flagship species are included in monitoring programmes.
Lucanus cervus, Natura 2000 monitoring, elusive saproxylic beetles, Monte-Carlo simulation, population decline
Monitoring global biodiversity is essential for nature conservation in order to understand and counter its current loss due to anthropogenic disturbances (
The European stag beetle, Lucanus cervus (further called the stag beetle), is a good model species to investigate whether the population trend of a strongly elusive terrestrial invertebrate can be adequately monitored. This saproxylic species is often considered as an umbrella species, representing the large saproxylic diversity inhabiting forests and half open landscapes (
Different monitoring protocols have been evaluated for the stag beetle: acoustic larval detection (
Here, data have been used from three transects in north-western Europe which have been monitored nearly daily for seven up to ten years to parameterize a simulation model that estimates the stag beetles’ relative abundance. This model is then adjusted to include a population decline of 1%/year and used for a Monte-Carlo simulation. This decline was derived from European guidelines (
C-Season: Centred measurement for the day of the season which is equal in each year calculated as: (Julian date (1 to 365) – 170) /30
T-Season: Centred measurement for the day of the season but shifted based on the temperature of that specific year to accommodate a season that was triggered by a certain temperature calculated as: (Julian date (1 to 365) – first day with 18°C or more + 1) /30
MAB: Median absolute bias on the trend estimation calculated as the median value of the absolute difference between the trend introduced in the simulation and the trend estimated by the validation model.
The data were compiled from three published studies on transects that have been monitored daily during the activity period of the stag beetle, i.e. mid-May till early July, for several years (Table
Metadata of the stag beetle transect walks including location, longitude, latitude, altitude, start, end year, habitat, duration and reference to the protocol and number of transect walks (#).
Location | Long (°E) | Lat (°N) | Alt (m) | start year | end year | Habitat | Duration of transect walk | Reference for protocol | # |
---|---|---|---|---|---|---|---|---|---|
Tairnbach (Germany) | 8.75 | 49.25 | 191 | 2008 | 2014 | Forest edge | 1h |
|
681 |
Basel (Switzerland) | 7.58 | 47.57 | 262 | 1991 | 2000 | Park | 1.25h |
|
510 |
Colchester (UK) | 0.88 | 51.88 | 28 | 2005 | 2011 | Urban | 0.5–1h | Fremlin and Fremlin ( |
459 |
It is assumed that the number of individuals observed along the transects mirrors the local population density as there is no population density function available for this species. This situation is common for many monitored species (
As mentioned in the introduction, the stag beetle abundance shows a strong seasonal and temperature-dependent pattern. Therefore, the second part estimates the temperature effect using a spline smoother. This smoother was included to gain insight into the relationship between temperature and stag beetle abundance, being linear or multi-polynomial. The latter option is based on the observation that stag beetles’ activity increases until they are fully active from 18°C onwards (
# stag beetles ~ Intercept + offset (log(transect duration)) + transect + year:transect + s(temperature) + s(C-Season)
or
# stag beetles ~ Intercept + offset (log(transect duration)) + transect + year:transect + s(temperature) + s(T-Season)
The two variants of the explanatory model were, in the first place, evaluated based on the analysis of the model residuals in relation to model variables. As both variants performed equally well, the best model was finally selected by the AIC criterion (see results). Model residuals were plotted against weather variables that were not used in the model (see section 2.1) and moon cycle data to detect any remaining variability and correlation coefficients were calculated to decide whether to include these variables in the explanatory model.
New data sets were created consisting of 10 up to 100 transects, with each transect simulated for 12 years (cfr.
The observed temperature data was modelled with a generalised linear mixed model (GLMM) including season as a second degree polynomial as explanatory variable and transect and year as crossed random effects:
Temperature ~ Intercept-temperature + C-Season + C-Season² + (1|Transect) + (1|Year)
To generate simulated temperature data, the GLMM with the original coefficients was converted to a GLM by changing the random effects in normally distributed fixed effects with zero as mean and their sigma as variance. The auto-correlation was set to 0.7, based on visual and empirical interpretation (Suppl. material
The number of stag beetles observed per transect walk was simulated based on the selected explanatory model from the model selection. To facilitate the simulation, this GAM model was simplified to a GLM simulation model including a second degree polynomial of temperature and third degree polynomial of C-Season, based on the degrees of freedom in the original model. The third degree of temperature was not used in the simulation model as its significance in the explanatory model depended on the year:transect interaction and is therefore not an overall population characteristic but a statistical compensation for this interaction. Moreover, the temperature effect of a second or third degree polynomial on the number of stag beetles observed remains very similar (Suppl. material
# stag beetles ~ rpois(Expected count /transect duration)
Log (Expected count /transect duration) = Intercept + transecti + yearj + poly(temperature, 2)
+ poly(C-Season, 3) + ((1–0.12)1/12) * year
transecti ~ Normal(0, sdtransect)
yearj ~Normal(0, sdyear)
Four monitoring scenarios (Weekly, Warmest of 7d, Peak temperature and Daily) were applied to the simulated data to comply with different monitoring protocols. In the Weekly scenario, each transect was monitored weekly during one up to eight weeks centred around the period with peak abundance. In the Warmest of 7d scenario, the transect was monitored on the warmest day of each week representing a monitoring protocol that depends on the weather forecast for the coming seven days. This is a simplification of the method proposed by
On the subsets of data sampled with the different scenarios, a GLM validation model was fitted similar to the simulation model, but without year and transect effects to improve the processing time (they could be left out as these were centred around zero). If the subset included data of less than four weeks, then C-Season was left out of the validation model as the period is too short to fit the season effect properly. When modelling data from the Peak temperature scenario, both temperature and C-Season were left out of the validation model for the same reason. From each validation model, the parameter estimate and p-value for year were extracted.
Simulations were run 1000 times for each of the different simulation options, i.e. 10 to 100 transects (sample size), one to eight days/weeks of monitoring per year and transect (frequency) and for each of the four scenarios. Power (1 - type II error) was calculated as the percentages of p<0.05 (type I error) with parameter estimate <1 (i.e. prediction of a declining trend) for each of the simulation options. Based on these results, the minimum effort (= frequency * sample size) needed to reach a power > 90% was assessed. A threshold of 90% has been repeatedly suggested for reliable trend detection (e.g.
Finally, the power of three existing monitoring protocols was calculated: two in Flanders and one in Slovenia. In Flanders (Northern Belgium) a monitoring protocol for this species was designed including 36 transects and eight weeks of monitoring during the presumed warmest day of the week (
The model with C-Season had a lower AIC (7971) than the model variant with T-season (AIC = 8743) meaning that the hypotheses presented in
Coefficients of the explanatory GAM model variant with lowest AIC which explains the stag beetle abundance. The table includes coefficients and their significance, estimated degrees of freedom for the smoothers of C-season and temperature and percentage deviation explained by the model (%dev. expl.: % deviation explained).
Coefficients | edf | %dev. expl. | ||||||
---|---|---|---|---|---|---|---|---|
Transect | Year x Transect | s(Temp) | s(C-Season) | |||||
Basel | Colchester | Tairnbach | Basel | Colchester | Tairnbach | |||
-1.13*** | 0.24*** | -0.11 | -0.12*** | 0.11*** | 0.26*** | 2.898*** | 2.991*** | 47.4 |
Only with three of the four scenarios, a power of 90% was achievable but the effort and number of transects needed differed (Figure
The MAB criteria provided very similar results compared to the power (Suppl. material
The original monitoring for Flanders, Flanders scenario, yielded a power of 95% within 12y. The Flanders start-up scenario still had a power of 79%. The scenario of Slovenia, with only ten transects, yielded a power of 23%. This was quite low but, for a period of 24y the power increased to 81%.
Statistical power for different scenarios, number of transects and frequency as number of days (for Peak temperature, Weekly and Warmest of 7d) or weeks (for Daily) per year and transect for monitoring the stag beetle (Lucanus cervus).
With the statistical power analysis presented, it was shown that it is at least feasible to monitor population density changes of the stag beetle with an effort of 240 days/y. This effort can be applied successfully with different combinations of scenarios, number of transects (between 30 and 100) and frequency. Before concluding which monitoring strategy and effort is most advisable to employ, the impact and alternatives for the missing density function, the limitations of the data used and the consequences of methods used for the results of the power analysis will be discussed first.
One of the main methodological problems for population trend analysis is the use of a relative abundance measure (here number of stag beetles found along a transect) to estimate the absolute population size (e.g.
The between-site variation on the number of the stag beetles observed is difficult to assess as only three sites have been monitored.
The most efficient way of monitoring the stag beetle seems to involve a scenario with weekly transects walks during the warmest evening. The scenario with transect walks concentrated after a first evening with 18°C or higher seems to have missed the period with abundance peak resulting in a very low power. Possibly, the stag beetle emergence in this region is triggered by lower temperatures and this causes the mismatch. However, if this peak period can be predicted, then the power of such a monitoring scenario might be much higher. When T-season is used in the simulation model instead of C-season, the Peak temperature scenario has the lowest effort needed to reach a power of 90% (results not shown). This is due to the fact that the simulation model and data sampling are then ideally tuned as both are based on the same hypothesis i.e. the period with abundance peak starts on the first evening with a temperature of 18°C. In reality, the start of this peak might be more complex and therefore more difficult to predict. Especially in different regions, stag beetle emergence might be expected to respond differently and thus different monitoring protocols might be needed for each region if this were to be applied. Consequently, it might be difficult to organise a large network of transects and instruct volunteers if the monitoring differs at each transect depending on the local temperature or climate zone. In that case, it might be easier to have transects that need to be walked weekly on the warmest day or even on a fixed day.
The Warmest of 7d scenario is simulated with the simplifications of a perfect weather forecast (i.e. the warmest evening is known at the beginning of the week) and so, in reality, the power might decrease slightly due to an imperfect weather forecast. However, as the power of the Weekly scenario is quite similar, this effect is expected to be limited. For more southern locations, this effect might be even smaller as days with unsuitable weather become rare.
An advantage of the Warmest of 7d above the Peak temperature scenario is that the effect of season remains evaluated. By this, changes in seasonality can also be detected. For example, climate change is expected to negatively affect the activity period (
Daily sampling clearly results in oversampling of a site in terms of population trend detection and is therefore not advised when trying to optimise the monitoring effort. However, this sampling technique might be very useful when only a limited number of transects is available or to study other population parameters, e.g. gaining insight into the period with peak abundance.
When comparing different options with the same effort, it seems that, in the presented simulation, the number of transects and frequency has little additional impact on the power. Thus, different combinations can be used to bring this monitoring into practice. Due to some simplifications that were included in the simulation, e.g. constant seasonal effects at all locations and equal decline at all sites, it is not advisable to use the lowest sufficient effort calculated but rather select a more robust estimate of the effort needed. Therefore, it is concluded that any combination with the Warmest of 7d scenario and an effort of minimal 240 days per year and between 40 and 100 transects can be used to realise the monitoring of this species to detect the given trend. A higher number of transects only slightly improves the power (cfr.
When comparing our results with other studies, it is concluded that the effort needed to monitor this elusive stag beetle (240 surveys/y) is not higher than for other invertebrates.
Based on the current study, it is assumed that many other elusive species with similar life history traits can likely be monitored with a similar magnitude of effort. Many other stag beetles species share the short activity period, crepuscular activity and temperature dependence (e.g.
It is concluded that it is possible to monitor a rather small population density decline of 1% per year for the elusive stag beetle with a moderate monitoring cost of 240 transect walks per year. Based on this example, it is assumed that many other elusive species with similar life history traits can be monitored with moderate efforts. This finding is especially important as saproxylic insects represent a large share of the total forest biodiversity (e.g.
This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors. Thanks to two anonymous reviewers for improving an earlier version of this paper.
Special issue published with the contribution of the LIFE financial instrument of the European Union.