Critical review Lamarre et al

MOD003372

Critical review of Lamarre et al. (2017a)

Introduction

Summary

In the paper by Lamarre et al(Lamarre et al., 2017b), the authors set out to investigate the impact of greater snow geese, Anser caerulescens, on nesting shorebirds in the Canadian High Arctic (Bylot Island). They hypothesised that geese impact nesting shorebirds by increasing nest predation. To test this hypothesis, the authors modelled the relationship between proximity to the goose colony and the spatial distribution of predators (Arctic foxes Vulpes lagopus, skuas and gulls), nesting American golden plovers (AGP) Pluvialis dominica, and predation of artificial nests as modified by lemming populations.

They found that both distance and lemming density influenced the likelihood of observing nesting plovers and predators, and the predation of artificial nests. These relationships were complex but AGP were more likely to be observed outside the goose colony especially in high lemming years. Foxes and avian predators were more likely to be observed close to the colony centre, with lemmings influencing fox sightings which were more frequent in low lemming years. Lemming density did not affect avian predator sightings inside the colony, but sightings were slightly higher outside the colony in high lemming years.

Context

There is growing concern about the ecological impacts of growing numbers of geese in the Arctic. Large numbers of geese breed in the Canadian High Arctic, and numbers have been growing partly driven by climate change(Dickey, Gauthier and Cadieu, 2008; Gilg et al., 2012; Lefebvre et al., 2017), and changes in agricultural practices. Overabundance of geese can have deleterious consequences through overgrazing altering plant assemblages and by disrupting predator-prey relationships, particularly when they migrate to regions where they provide a short-term increase in prey opportunity for local predators (predator pulse).

As goose populations increase, the number of Arctic shorebirds (sandpipers and plovers) have been observed to be declining and it has been hypothesized that goose populations may be indirectly influencing this by increasing the abundance of local mutual predators (arctic foxes, gulls and skuas) which predate goose eggs and chicks, leading to increased predation of nesting shorebird eggs(Flemming et al., 2019b). This interaction is complicated by cyclical booms (every 3-4 years) in lemming populations(McKinnon, Berteaux and Bêty, 2014). The ‘bird-lemming’ relationship has been noted between sandpipers and lemmings in Sweden(Blomqvist et al., 2002). Artcic fox breeding success appears highest inside goose colonies in high lemming years(Chevallier et al., 2020).

Understanding how geese impact Arctic ecosystems is therefore of conservation concern and an issue for wildlife managers in North America and this paper is intended to contribute to this understanding.

Detailed discussion

Study design

Specifically, Lamarre et al examined the spatial variation in predator occurrence, nest predation and shorebird breeding success in relation to proximity to the goose colony. The authors estimated the occurrence of arctic foxes, predatory seabirds and nesting plovers through a series of transect surveys at varying distances from the centroid of the snow goose colony over 6 years.

The authors use a number of proxy measures as dependent variables. American Golden Plovers (AGP) was used an indicator of overall shorebirds. This was justified on the basis that the diversity and abundance of shorebirds is correlated with the occurrence of AGP (Supplement 1). Skuas (jaegers) and glaucous gulls were as an indicator of avian predators - ravens are also predators but were infrequently seen.

Transect distance to the centroid of the goose colony was used as an indicator of the influence of the geese. Finally predation of artificial nests (prepared quail’s eggs) within 48 hours as a proxy for shorebird nest predation, although the authors state that this can only be used as a measure of relative nest predation risk rather than as a direct measure of predation of real nests(Weidinger, 2001; McKinnon et al., 2010).

In addition they estimated the abundance of lemmings through trapping, and classified each year as a high or low according to estimated lemming density.

In all 1242 surveys were conducted during the breeding seaons on 267 transects, with the majority of transects being surveyed in at least 4 years. Artificial nests were deployed on 253 transects, with 138 being surveyed for at least 4 years. Data on nesting plovers was obtained for 233 transects¹.

Analysis

The unit of analysis was the transect. The authors modelled the likelihood of observing an AMGP, a predator, or evidence of nest predation within 48 hours on transect surveys, as a function of transect distance from the colony centroid and lemming status. They used a generalized logistic mixed model with transect and standardised (Julian) date as random effects. Random effects models allow for variation in the relationship between dependent (fixed) variables and responses of interest rather than assuming that the relationship is the same. In this case a random intercept model was used allowing the probability of the response varaible to vary between transects. This is an appropriate analytical method for this kind of grouped data(Bolker et al., 2009).

Models were fitted without parameters (null model), and with each combination of distance, quadratic distance based on the segmentation analysis,and lemming status. Interactions between distance and lemming prevalence were also included.

Model goodness of fit was compared using the AICc criterion and the model with the lowest AICc was deemed the best fit. The outputs of the logistic regressions were expressed as log odds and were calculated as a weighted average of the coefficients of each model with the best fitting model contributing the greatest weighting.

Log-odds are hard to intepret and need to be exponentiated to convert back to odds ratio² which are more intepretable. They tells us how much more (or less) likely the response is to occur per unit increase in distance from the colony centroid.

Results

Four main sets of results are presented in the paper:

• summaries of separate logistic random effects models relating distance from the colony (linear and quadratic) lemming-status, with each of the four response variables: probability of occurrence of arctic foxes; probability of occurrence of avian predators (sum of skuas and glaucous gull); probability of occurrence of breeding american golden plovers; and probability of artificial nest predation

• model outputs as log odds (these should be converted back to odds) which are the model predicted probabilities of the response - the authors split this by lemming frequency

• the distance of change points (where the distance - occurrence relationship changed, calculated by the Lavielle method³) from the centroid of the goose colony

• The change in occurrence per Km up to and beyond the change points

These results are presented in tabular and graphical format but some (model outputs and distance relationships) are presented in the text. Tabulating these would have improved clarity of reporting. I have extracted these to (Table1?) with log-odds transformed back to odds.

(Table1?): Summary of distance-occurrence relationships

Occurrence	High lemming (change in odds per km)	Low lemming (change in odds per km)	Change per km inside the colony (%)	Change per km outside the colony (%)	Distance of change point from centroid (km)
Arctic fox	0.27 (0.15-0.47)	0.69 (0.39-1.22)	-2	-0.29	8.5
Avian predators	0.32 (0.23-0.43)	0.48 (0.36-0.62)	-6.8	-1.1	8.5
Nest predation	0.51 (0.32-0.84)	0.70 (0.52-0.93)	-0.9	-0.5	-
Plover	0.50 (0.35-0.71)	2.36 (1.72-2.89)	3	1.3	9

Figure 4 in the paper summarises the relationship between occurences / predation and distances as set of bubble plots (scatter plots with varying point sizes) stratifying between high and low lemming years. It is a complex, multivariate chart made difficult to read partly because of the lack of contrast between high and low lemming years (shades of grey) the sizing of the ‘bubbles’, and the inclusion of change points and mean values for high and low lemming years which means there are several overlapping lines.

Use of contrasting colours to represent high and low lemming years would have improved clarity as would have reducing the size of the bubbles or using colour to represent logN. Use of colour for lines would have helped, or simplifying the charts by presenting some of the information elsewhere. The algorithm for variation in bubble size is unclear - the figure caption says variation is “proportional to log(N)”, but what N is is unclear.⁴

Figure 4 shows that the probability of arctic fox occurrence is much higher close to the centre of the colony in low as opposed to high lemming years and has a non-linear relationship with distance. This is consistent with arctic foxes switching to goose predation when lemmings are scarce.

There seems to be little difference in occurrence likelihood of predatory birds between high and low lemming years within the colony, but is much higher closer to the colony than further away. Outside the colony observations of avian predators are slightly higher in high lemming years.

Nest predation appears to show no relationship with lemming status and a weak linear relationship with distance.

The likelihood of observing nesting golden plovers is much higher outside the colony and is more likely in high lemming years. This is consistent with increased predation when lemmings are scarce.

Strengths

The authors design and methods seem thorough, based on a long history of studies on Bylot Island(Bety et al., 2002; Nolet et al., 2013; Flemming et al., 2019a). For example, quail’s eggs used for nest predation studies were well prepared to avoid suspicion by arctic foxes; golden plovers were chosen as the indicator for shorebirds because they are more reactive to human presence, making them easier to spot, and exhibit behavioural traits which readily indicate nesting status; they reported and used a well understood method for trapping lemmings. They sought to maximise detection rates of plovers by transect surveys at the peak breeding season and provide evidence to support detection by showing that AGMPs flushed well within the 150m limits of the transects.

Mixed modelling is an appropriate analytical method and seems to have been properly conducted by the authors.

Areas of weakness

Areas of potential weakness are:

• no description of how transects were selected other than habitat location (for example where they randomly sampled grid squares, chosen for practical reasons, or existing transects). To some extent this is tackled in the analysis by treating transect as a random variable in the models.

• not including climate variables - temperature, precipitation and date of snow melt for example - in the analysis. These are all measured extensively on Bylot Island[ref] and can influence trophic interactions in the Arctic(Juhasz et al., 2020).

• the authors could have modelled golden plovers as a function, not just of distance and lemming abundance, but also predator detection and also an metric of snow goose abundance. The data presented gives estimated pre transect counts by year of snow geese(Lamarre et al., 2018b). This would have given clearer insight into the impact of the colony and predators on nesting success of plovers.

• presentation could have been improved - model outputs could have been tabulated and Figure 4 simplified, further faceted and use made of colour or stronger contrast to clarify the distinction between high end low lemming years.

Conclusion

The paper successfully describes the spatial relationship between distance from the goose colony and observation of plovers and predators and the goose colony and shows that plovers do better outside the colony than inside, especially when lemmings are more abundant. This is consistent with the hypothesis that main predators, especially arctic foxes, prefer lemmings in years of plenty and switch to geese and plover predation when lemming populations crash. Given that both foxes and avian predators are much more prevalent close the centre of the colony, they either extensively predate any nesting plovers, or discourage nesting. The high density of geese may reduce nesting habitat, or alter food availability (although there is evidence that high density of goose faeces can increase invertebrate biomass and benefit breeding shorebirds and chick growth (Flemming et al., 2022)).

References

Bety, J., Gauthier, G., Korpimaki, E. and Giroux, J.-F., 2002. Shared predators and indirect trophic interactions: Lemming cycles and arctic-nesting geese.

Blomqvist, S., Holmgren, N., Åkesson, S., Hedenström, A. and Pettersson, J., 2002. Indirect effects of lemming cycles on sandpiper dynamics: 50 years of counts from southern sweden. Oecologia, 133(2), p.146158.

Bolker, B.M., Brooks, M.E., Clark, C.J., Geange, S.W., Poulsen, J.R., Stevens, M.H.H. and White, J.-S.S., 2009. Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution, 24(3), p.127135.

Calenge, C., 2006. The package adehabitat for the r software: Tool for the analysis of space and habitat use by animals. 197, p.1035.

Chevallier, C., Gauthier, G., Lai, S. and Berteaux, D., 2020. Pulsed food resources affect reproduction but not adult apparent survival in arctic foxes. Oecologia, [online] 193(3), pp.557–569. https://doi.org/10.1007/s00442-020-04696-8.

Dickey, M.-H., Gauthier, G. and Cadieu, M.-C., 2008. Climatic effects on the breeding phenology and reproductive success of an arctic-nesting goose species. Global Change Biology, [online] 14(9), pp.1973–1985. https://doi.org/10.1111/j.1365-2486.2008.01622.x.

Flemming, S.A., Nol, E., Kennedy, L.V., Bédard, A., Giroux, M.-A. and Smith, P.A., 2019a. Spatio-temporal responses of predators to hyperabundant geese affect risk of predation for sympatric-nesting species. PLoS One, 14(8), p.e0221727.

Flemming, S.A., Smith, P.A., Kennedy, L.V., Anderson, A.M. and Nol, E., 2022. Habitat alteration and fecal deposition by geese alter tundra invertebrate communities: Implications for diets of sympatric birds. PLOS ONE, [online] 17(7), p.e0269938. https://doi.org/10.1371/journal.pone.0269938.

Flemming, S.A., Smith, P.A., Rausch, J. and Nol, E., 2019b. Broad-scale changes in tundra-nesting bird abundance in response to hyperabundant geese. Ecosphere, [online] 10(7). https://doi.org/10.1002/ecs2.2785.

Gilg, O., Kovacs, K.M., Aars, J., Fort, J., Gauthier, G., Grémillet, D., Ims, R.A., Meltofte, H., Moreau, J., Post, E., Schmidt, N.M., Yannic, G. and Bollache, L., 2012. Climate change and the ecology and evolution of Arctic vertebrates. Annals of the New York Academy of Sciences, [online] 1249(1), pp.166–190. https://doi.org/10.1111/j.1749-6632.2011.06412.x.

Juhasz, C.-C., Shipley, B., Gauthier, G., Berteaux, D. and Lecomte, N., 2020. Direct and indirect effects of regional and local climatic factors on trophic interactions in the Arctic tundra. Journal of Animal Ecology, 89(3), p.704715.

Lamarre, J.-F., Legagneux, P., Gauthier, G., Reed, E.T. and Bêty, J., 2017a. Predator-mediated negative effects of overabundant snow geese on arctic-nesting shorebirds. Ecosphere, [online] 8(5), p.e01788. https://doi.org/10.1002/ecs2.1788.

Lamarre, J.-F., Legagneux, P., Gauthier, G., Reed, E.T. and Bêty, J., 2017b. Predator-mediated negative effects of overabundant snow geese on arctic-nesting shorebirds. Ecosphere (Washington, D.C), 8(5), p.e01788.

Lamarre, J.-F., Legagneux, P., Gauthier, G., Reed, E.T. and Bêty, J., 2018b. Data from: Predator-mediated negative effects of overabundant snow geese on arctic-nesting shorebirds. https://doi.org/10.5061/DRYAD.796T8.

Lamarre, J.-F., Legagneux, P., Gauthier, G., Reed, E.T. and Bêty, J., 2018a. Data from: Predator-mediated negative effects of overabundant snow geese on arctic-nesting shorebirds. [online] https://doi.org/10.5061/DRYAD.796T8.

Lefebvre, J., Gauthier, G., Giroux, J.-F., Reed, A., Reed, E.T. and Bélanger, L., 2017. The greater snow goose anser caerulescens atlanticus: Managing an overabundant population. Ambio, 46(2), p.262274.

McKinnon, L., Berteaux, D. and Bêty, J., 2014. Predator-mediated interactions between lemmings and shorebirds: A test of the alternative prey hypothesis. The Auk, [online] 131(4), pp.619–628. https://doi.org/10.1642/auk-13-154.1.

McKinnon, L., Smith, P.A., Nol, E., Martin, J.L., Doyle, F.I., Abraham, K.F., Gilchrist, H.G., Morrison, R.I.G. and Bêty, J., 2010. Suitability of Artificial NestsResponse. Science, [online] 328(5974), pp.46–47. https://doi.org/10.1126/science.328.5974.46-b.

Nolet, B.A., Bauer, S., Feige, N., Kokorev, Y.I., Popov, I.Y. and Ebbinge, B.S., 2013. Faltering lemming cycles reduce productivity and population size of a migratory Arctic goose species. Journal of Animal Ecology, 82(4), p.804813.

Weidinger, K., 2001. How well do predation rates on artificial nests estimate predation on natural passerine nests? Ibis, [online] 143(3), pp.632–641. https://doi.org/10.1111/j.1474-919x.2001.tb04891.x.

Footnotes

1. Personal analysis of supplemental data(Lamarre et al., 2018a)

2. \[ p = e^l/(1+e^l) where: l = log-odds \]

3. The Lavielle method segments a time series or distance into a number of segments and sequentially calculates the change between the first segment and subsequent segment until a significant difference is found or the end of the series is reached. The method is incorporated into the adehabitatL R package.(Calenge, 2006)

4. Possibilities are number of observations, number of transects at each distance or number of samples per transect. It is hard to tell but each of the high-low bubbles seem to lie at the same distances suggesting that N is the number of transects at each distance. There are 21 points - so it would seem that the points represent transects per km between 0 and 20 km