Migratory strategy drives species-level variation in bird sensitivity to vegetation green-up

Abstract

Animals and plants are shifting the timing of key life events in response to climate change, yet despite recent documentation of escalating phenological change, scientists lack a full understanding of how and why phenological responses vary across space and among species. Here, we used over 7 million community-contributed bird observations to derive species-specific, spatially explicit estimates of annual spring migration phenology for 56 bird species across eastern North America. We show that changes in the spring arrival of migratory birds are coarsely synchronized with fluctuations in vegetation green-up and that the sensitivity of birds to plant phenology varied extensively. Bird arrival responded more synchronously with vegetation green-up at higher latitudes, where phenological shifts over time are also greater. Critically, species’ migratory traits explained variation in sensitivity to green-up, with species that migrate more slowly, arrive earlier and overwinter further north showing greater responsiveness to earlier springs. Identifying how and why species vary in their ability to shift phenological events is fundamental to predicting species’ vulnerability to climate change. Such variation in sensitivity across taxa, with long-distance neotropical migrants exhibiting reduced synchrony, may help to explain substantial declines in these species over the last several decades.

Extended data

Extended Data Fig. 1 Study area of interest over North America.

Data were aggregated within each cell to calculate phenological measures and to characterize phenological change and sensitivity. Yellow cells represent the full extent of the study area. Cells were selected based on data density for both bird and green-up phenology (see Methods). Cell centres ranged from approximately 95° W to 54° W longitude and 26° N to 59° N latitude.

Extended Data Fig. 2 Derivation of the half-maximum from GAM results for each species–cell-year.

Circles at the top of each plot represent checklists where the species of interest was recorded, while circles at the bottom of each plot represent checklists where the species of interest was not recorded. Left panel: the green line represents the first detection of a given species in a given cell-year; the red line represents the first local maximum for the modelled probability of occurrence in an eBird checklist to come after the first detection; the gold line represents the probability of occurrence at that local maximum; the purple line represents Δ_p, the difference between the minimum modelled probability of occurrence prior to the first local maximum and the probability of occurrence at the local maximum (the minimum reporting probability here is 0); the dark blue line represents the probability of occurrence at $\frac{1}{2}{{{\Delta}}}_p$, half the difference between the maximum and minimum probabilities plus the minimum reporting probability; the light blue line represents the half-maximum date, the ordinal date (day-of-year) at which the modelled probability of occurrence equals $\frac{1}{2}{{{\Delta}}}_p$. Right panel: black lines represent posterior realizations of the GAM model fit for a single species–cell-year (500 realizations shown for clarity). The red lines represent the derived half-maximum date at each realization of the GAM model fit and were used to calculate the mean and 95% credible intervals for this metric.

Extended Data Fig. 3 Data processing pipeline using red-eyed vireo (Vireo olivaceus) as an example.

Estimates of arrival (half-maximum) were derived from generalized additive models (GAMs), which were then used as input for the intrinsic autoregressive (IAR) model to produce spatially smoothed estimates of arrival. The plot at the far left shows GAM results for a single cell-year for this species. Circles at the top of the plot represent eBird checklists in which red-eyed vireo was recorded, while circles at the bottom of the plot represent eBird checklists in which red-eyed vireo was not recorded. The black line represents the mean GAM fit, while the dashed red lines represent the 95% credible intervals. The solid blue and dashed blue lines represent the mean estimate and 95% credible intervals for the half-maximum, respectively. The plots in the centre column of the figure represent the estimated arrival date of this species over the study area for 2006. The plot at top centre represents the GAM-derived arrival estimates, while the plot at bottom centre represents the IAR-derived arrival estimates. Blue hues represent later arrival dates while pink hues represent earlier arrival dates for a given cell. The plots at far right represent a subset (the region bounded by the black box) of the maps in the centre column of the figure. Numbers in black represent the posterior mean of the arrival day (ordinal date), while the white numbers represent the posterior standard deviation of the arrival day.

Extended Data Fig. 4 Number of cells across the study area that met data requirements for each species and year.

Red hues represent more cells while white hues represent fewer cells. Only species that met minimum data requirements are shown. Since species-years with fewer than 3 valid cells were not run as a part of the IAR model (see Methods), each species–year has either 0 or 3 or greater valid cells.

Extended Data Fig. 5 Posterior estimates for a) ξ_APG (the species-specific phenological sensitivities; equation 11) and b) γ_APG (the species-specific effect of latitude on phenological sensitivities; equation 11).

Points represent posterior medians, thick lines represent 50% credible intervals, thin lines represent 95% credible intervals. The dashed grey line represents zero in each case.

Extended Data Fig. 6 Rate of change in green-up from 2002–2017 over the study area for (a) forest land cover types, and (b) all land cover types.

Colours for (a) and (b) represent the cell-specific posterior mean estimates of the rate of change in green-up over time (days change per year) with red hues representing more negative trends over time (earlier green-up) and yellow hues representing no trend over time. c, Posterior estimates for cell-specific rate of change in green-up for forest land cover types (black) and all land cover types (red). Points represent the posterior median estimates for the rate of change of each cell (ordered by latitude), thick lines represent 50% credible intervals, thin lines represent 95% credible intervals. The dashed grey line represents zero.

Extended Data Fig. 7 Directed acyclic graphs (DAGs) outlining the hierarchical models used in this study.

Boxes represent variables that were provided to the model, while ovals represent parameters estimated by the model. Corresponding equation numbers for each DAG given in lower right of each bounded box. Lettering corresponds to that shown in Supplementary Table 2, which provides descriptions of each variable represented in the DAGs.

Extended Data Fig. 8 Density plots for observed response variable data (y; corresponding to IAR-derived arrival dates for (a) and green-up dates for (b) and (c)) and response variable data simulated from the posterior predictive distribution (yrep).

These plots were used for graphical posterior predictive checks, to ensure that data simulated from the model were similar to the observed data for models examining (a) the sensitivity of bird arrival to vegetation phenology (Eqs. 9–14), (b) trends in green-up over time for forest land cover types (Eqs. 19–21), and (c) trends in green-up over time for all land cover types (Eqs. 19–21). Curves in red are a representation of the density of all response data used to fit each model. Curves in black are a representation of the density of data simulated from the posterior predictive distribution. Each iteration of the posterior chain yields a simulated dataset. Here 250 datasets simulated from the posterior predictive distribution are displayed (250 separate black lines). The general similarities between the red lines and black lines demonstrate that the models simulate data similar to the observed data.