Technical Comments

Comment on “Global pattern of nest predation is disrupted by climate change in shorebirds”

See allHide authors and affiliations

Science  14 Jun 2019:
Vol. 364, Issue 6445, eaaw8529
DOI: 10.1126/science.aaw8529

Abstract

Kubelka et al. (Reports, 9 November 2018, p. 680) claim that climate change has disrupted patterns of nest predation in shorebirds. They report that predation rates have increased since the 1950s, especially in the Arctic. We describe methodological problems with their analyses and argue that there is no solid statistical support for their claims.

Climate change affects organisms in a variety of ways (14), including through changes in interactions between species. Kubelka et al. (5) reported that a specific type of trophic interaction, namely depredation of shorebird nests, has increased globally over the past 70 years. The authors state that their results are “consistent with climate-induced shifts in predator-prey relationships.” They also claim that the historical perception of a latitudinal gradient in nest predation, with the highest rates in the tropics, “has been recently reversed in the Northern Hemisphere, most notably in the Arctic.” They conclude that “the Arctic now represents an extensive ecological trap for migrating birds, with a predicted negative impact on their global population dynamics.” These conclusions have far-reaching implications for evolutionary and population ecology, as well as for shorebird conservation and related policy decisions (6). Therefore, such claims require robust evidence, strongly supported by the data. Here, we dispute this evidence.

First, Kubelka et al. graphically show nonlinear, spatiotemporal variation in predation rates (their figure 2, A and B, and figure 3) and suggest that in recent years, predation has strongly increased in North temperate and especially Arctic regions, but less so in other areas. However, they only statistically test for linear changes in predation rates over time for all regions combined, and for each geographical region (their table S2) or period (before and after 2000; their table S6) separately. To substantiate their conclusions, they should have presented statistical evidence for an interaction between region/latitude and year/period on predation rate. Moreover, their analyses control for spatial autocorrelation but fail to model non-independence of data from the same site (pseudo-replication).

Using the data of Kubelka et al., we ran a set of mixed-effect models, structurally reflecting their results depicted in their figure 2, A and B, and figure 3, but including location as a random factor (Table 1) (7). These analyses show (i) that much of the variation in nest predation rate is explained by study site (>60%, compared to species: <5%), implying a reduced effective sample size; (ii) that all regions—except the South temperate—show similar predation rates; and (iii) that nest predation rates increase over time similarly across all geographical areas (Fig. 1, A to F). Linear models without interaction terms are much better supported than nonlinear models with interactions (Table 1), indicating that predation rates in the Arctic are not increasing any faster than elsewhere (Fig. 1, B, C, E, and F). Thus, these results provide no evidence that the rate at which nest predation increased over time varies geographically.

Table 1 Comparison of models explaining spatiotemporal variation in daily nest predation rate using the original Kubelka et al. data.

Letters and results in bold refer to panels in Fig. 1; A and D are the models reflecting figures 2A and 3A in (5). Each model is fitted with maximum likelihood and controlled for number of nests in a given population (ln-transformed) and for multiple populations at a given site or for a given species, using site and species as random intercepts. Daily predation rate (dependent variable) was ln-transformed after adding 0.01 [following (5)]. Predictors are Year (mean year of the study), Hemisphere (Northern versus Southern), Latitude (degrees), Geographical Area (Arctic, North temperate, North tropics, South tropics, South temperate), and Period [historic (1944–1999) versus recent (2000–2016)]. Models that include Period (instead of Year) are not supported by the data [less likely than the best model by factors of 69 to 320, as indicated by the evidence ratio (model weight of the first-ranked model relative to that of the given model, i.e., how many times the first-ranked model is more likely than the given model)]. Models including the interaction between time and geographical region/latitude do not improve the model fit or are much less supported by the data than are models without the interaction. See (7) for model output and analyses of total predation rates. Note that we used quadratic or third-order polynomial terms to mimic the relationships depicted in Kubelka et al.’s figures (5). Number of parameters denotes number of model parameters without the random effects. ΔAIC is the difference in Akaike information criterion between the first-ranked model (AIC = 349.8) and the given model. Model probability refers to Akaike weight (wi), the weight of evidence (probability) that a given model is the best-approximating model.

View this table:
Fig. 1 Spatiotemporal variation in daily nest predation rates of shorebirds.

(A to C) Predation rate in relation to year for different geographical regions: with interaction and using all populations (A), without interaction and using all populations (B), with interaction and using only the 88 populations with known exposure from the Arctic and North temperate region (C). The model behind (A) is less supported by the data than the model behind (B) by a factor of ~18 (Table 1). (D to F) Predation rate in relation to latitude for different periods: with interaction (period as two-level factor) and using all populations (D), without interaction (year as continuous variable) and using all populations (E), with interaction and using only the 98 populations with known exposure (F). The model behind (D) is less supported than the model behind (E) by a factor of ~70 (Table 1). In (A) to (F), lines and shaded areas represent model predictions with 95% confidence interval (CI) based on posterior distribution of 5000 simulated values. Note the weak (P > 0.64) temporal increase in (C) [estimate = 0.08 (95% CI, –0.07 to 0.2) from a linear model without interaction] and (F) [estimate = 0.06 (95% CI, –0.09 to 0.17)]. See Table 1 for model description and comparison and (7) for details. (G) Temporal change in the percentage of populations in which exposure was estimated [following (10)] to calculate predation rate. Note the sharp decline in the Arctic relative to the other regions [see (7) for overall and region-specific changes]. Circles represent data for 5-year intervals. (H) Modeled changes in predation rate over time assuming different values of B (proportion of nesting period observed; higher values indicate nests found sooner after egg laying) for populations with unknown exposure and year <2000 (leaving the original estimates for all remaining populations). This exercise explores the sensitivity of the results to using older studies where the stage at which nests were found is less certain. (I) Relation between true and estimated predation rate for different values of B [N = 65 populations, as in (5)]. The dashed line indicates a slope of 1 (i.e., estimated values equaling true values). In (G) and (I), lines and shaded areas represent locally estimated scatterplot smoothing with 95% CI; in (H), lines and shaded areas represent model predictions with 95% CI based on posterior distribution of 5000 simulated values.

Second, for the period under study, not only the climate has changed, but also the research methods. Hence, it remains unclear whether nest predation rates have indeed increased over time and if so, why. Kubelka et al. used the Mayfield method (8, 9) to calculate daily nest predation rates as the number of depredated nests divided by “exposure” [the total time (in days) all nests were observed]. However, 59% of the 237 populations they used lacked information on exposure. They circumvented this problem by estimating exposure based on the description of nest search intensity in the respective studies (10). The key question is when nests were found. Kubelka et al. decided that in 114 populations, nests were found such that 60% of the nesting period (egg laying and incubation combined) was “observed” (B = 0.6; nests searched once or twice a week). For 14 populations they used B = 0.9 (nests searched daily or found just after laying), and for 11 populations they used B = 0.5 (assuming nests found midway during the nesting period). However, the choice of B value remains subjective (7), and for 38% of the 128 populations where Kubelka et al. used B > 0.5, we found no information in the reference to suggest that this was appropriate. This issue is not trivial, because using higher B values (i.e., assuming that nests were found earlier than they actually were) overestimates exposure and hence underestimates nest predation rates.

The proportion of populations with estimated exposure declines over time (7), particularly after 2000 and especially in the Arctic (Fig. 1G). The timing of the decline coincides with Kubelka et al.’s definition of historic and recent data and with the suggested exponential rise of predation in the Arctic (their figure 2, A and B, and figure 3, A and B). Indeed, the results are sensitive to variation in estimated exposure during the “historic period” (Fig. 1H). Although Kubelka et al. correctly state that the estimated and true predation rates are highly correlated [using studies with quantitative information on exposure; see supplementary materials of (5)], the true rate is typically underestimated for the higher B values they used (Fig. 1I). Given these issues, the main result—the apparent increase in daily nest predation rate over time, especially in the Arctic—may simply be an artifact. To further assess the robustness of the change in predation rate over time, we used only populations where nest predation rates were calculated on the basis of known exposure (N = 98). These analyses reduced the effect of year by ~50% (7) and resulted in weak, nonsignificant linear trends (Fig. 1, C and F), which suggests that there is little evidence for changing predation rates.

Finally, we note that nest searching effort and frequency of nest visits likely increased in recent years as researchers learned how best to obtain accurate estimates of nest survival (1113). Researchers also intensified their activities (e.g., capturing adults to band, tag, and collect samples and placing monitoring equipment near nests, which may increase the predation rate) (14, 15). Thus, an increase in the quality of data reporting as well as increased research activity around nests may have further induced a time-dependent bias in estimates with an underestimation of true predation rates in the historic data (see above), and perhaps an overestimation in the contemporary data.

In summary, reanalysis of the data of Kubelka et al., evaluation of the quality and interpretation of the published data used, and considerations about changes in research methods over the past 70 years lead us to conclude that there is no robust evidence for a global disruption of nest predation rates due to climate change. We argue that their claim that the Arctic has become an ecological trap for breeding shorebirds is untenable.

  • These authors formed the core team behind this work. The remaining authors are listed alphabetically according to their first name.

References and Notes

Acknowledgments: We thank all who contributed to the original studies on predation, shared data, helped to find the original data, or helped in preparing this comment [see (7) for details]. Funding: Supported by the Max Planck Society (B.K.), an EU Horizon 2020 Marie Curie individual fellowship (4231.1 SocialJetLag; M.B.), the Czech University of Life Sciences (CIGA: 2018421; M.B.), a Netherlands Polar Programme grant (NWO 866.15.207; J.R. and T.P.), and the U.S. Fish and Wildlife Service (R.B.L.). Author contributions: M.B., J.R., R.B.L., and B.K. initiated this work; M.B., J.R., E.L.W., D.V.S., R.B.L., and B.K. extracted and evaluated the data; most authors collected data contained in the primary sources; E.L.W. investigated the differences between newly extracted and original data; M.B. conducted the statistical analyses and drew the figure with help from M.V. and E.L.W. and input from J.R., R.B.L., and B.K.; B.K. and M.B. wrote the manuscript with input from J.R., E.L.W., and R.B.L. and finalized it with input from all authors. Competing interests: Authors declare no competing interests. Data and materials availability: All data, code, methods, results, figures, and tables associated with this comment are freely available from Open Science Framework (7).
View Abstract

Stay Connected to Science

Navigate This Article