Mid-latitude climate extremes are projected to increase in frequency under global climate change. How this may affect migratory bird populations is not well understood. The mid-latitudes of North America experienced an extreme warming event during March 2012 that advanced the spring phenology of ecological productivity, resulting in lower levels of productivity during the summer. Here, we test the predictions that: (1) short-distance migratory birds, due to geographic proximity and more flexible migratory behavior, should advance their spring migration phenology; and (2) breeding populations, due to lower summer productivity, should have reduced occurrences. We used occurrence data for 353 bird species from the eBird database to calculate weekly occurrence anomalies for 2012 relative to the 2010–2014 average. We identified species having unusually large positive occurrence anomalies during March 2012 and species having unusually large negative occurrence anomalies during July–August 2012. For each category, we summarized migration strategies, geographic distributions, and annual associations with temperature and ecological productivity. Short-distance migrants whose winter and breeding ranges intersect the mid-latitudes advanced their spring migration phenology during March (n = 21). Long-distance migrants whose winter and breeding distributions were weakly associated with the mid-latitudes had lower occurrences during the summer (n = 32). Five species were shared between the two categories. Within species’ winter ranges, temperature and ecological productivity were higher than expected during March; within species’ breeding ranges, ecological productivity was lower than expected during the summer. These differences were strongest for the 21 short-distance migrants. Following our expectations, mid-latitude climate extremes and associated ecological consequences broadly affected avian migration and breeding activities within the region. Our findings suggest short-distance migrants are more flexible and resilient, whereas populations of long-distance migrants are at a distinct disadvantage, which may intensify if the frequency of these events increases.
Migratory birds are considered to be at greater risk of population loss under global climate change (IPCC 2013) due in large part to their complex life-history phenologies (Walther et al. 2002, Gordo 2007, Knudsen et al. 2011). As climate warming progresses, the phenological synchronization that exists between migratory birds and other trophic levels during migration and the breeding season may become increasingly disrupted (Thackeray et al. 2010). For migratory species that have a limited capacity to buffer against phenological mismatches, these disruptions may result in increased rates of mortality during migration (Klaassen et al. 2012) or declines in breeding success (Both et al. 2006, 2010, Møller et al. 2008, Saino et al. 2011). Phenological disruptions may also occur through the influence of climate extremes, events that result in conditions that abruptly and substantially exceed the long-term norm (Gaines and Denny 1993). For migratory birds, climate extremes that coincide with migration can reduce physiological performance (Bauchinger et al. 2011), increase metabolic costs (Senner et al. 2015), and increase the chance of mortality (Newton 2007, Møller 2011). Over the past several decades, the frequency of climate extremes has increased worldwide with the greatest concentration occurring at mid-latitudes in the Northern Hemisphere (Easterling et al. 2000, Rahmstorf and Coumou 2011, Coumou and Rahmstorf 2012, Horton et al. 2015). This same geographic region also hosts the majority of world's species of migratory birds during migration and the breeding season (Somveille et al. 2015). This spatial correspondence suggests that mid-latitude climate extremes may pose a significant threat to populations of migratory bird species. However, the implications of mid-latitude climate extremes for migratory birds have not been fully considered.
The effect of global warming on avian migration phenology is considered an important metric for estimating the ecological consequences of global climate change (Gordo 2007, Knudsen et al. 2011). Much of this work has focused on determining how closely trends in migration phenology follow expectations under global warming, and has often identified migration distance as an important correlate. Current evidence suggests, with some exceptions (Jonzén et al. 2006), that shifts in migration phenology under global warming are more evident for short-distance migrants whose winter ranges are located at higher latitudes (Butler 2003, Rubolini et al. 2007, Hurlbert and Liang 2012, Van Buskirk et al. 2012). This outcome is likely related to differences in the mechanistic cues that regulate migration behavior, which are largely dictated by the latitude of species’ breeding or winter range (Hagan et al. 1991). During spring migration, photoperiod and endogenous circannual rhythms trigger migration onset for species that winter at lower latitudes (Gwinner 2003), whereas species that winter at higher latitudes may rely more on external environmental cues to initiate migration, such as changes in ecological productivity (La Sorte et al. 2014). As global warming progresses, the more rigid endogenous control and low phenotypic variation in migratory behavior displayed by long-distance migrants has placed these species at greater risk of population declines (Both et al. 2006, Møller et al. 2008, Saino et al. 2011). In contrast, the higher phenotypic plasticity in migratory behavior displayed by short-distance migrants, possibly in combination with microevolutionary change (Pulido and Berthold 2004, 2010, Gienapp et al. 2007), has allowed these species to more effectively adjust to changing environmental conditions under global warming (Calvert et al. 2012, Van Buskirk et al. 2012, Charmantier and Gienapp 2014). Although the response to global warming by short-distance migrants may be adaptive, these expectations are based on birds’ responses to average changes in climate. To date, the response of migratory bird populations to mid-latitude climate extremes have not been thoroughly explored, and it is unclear if the generalizations developed under global warming can be similarly applied within the context of mid-latitude climate extremes.
An extreme warming event occurred within the mid-latitudes of North America during March 2012 (Fig. 1; Appendix A: Fig. A1). The magnitude of this event within the mid-latitudes was exceeded only three times during the 20th century (Fig. 1). The extreme warming event also altered the spring phenology of ecological productivity creating a breeding season resource bottleneck (Maron et al. 2015); specifically, ecological productivity first increased during the warming event and then declined during subsequent months, with the most substantial declines occurring during the summer (July–August; Appendix A: Figs. A2 and A3). The extreme warming event occurred at the beginning of spring migration in a region containing a large number of migratory species during both migration and the breeding season (Mac Arthur 1959, Rabenold 1993). As such, this event provides a unique opportunity to document how an extreme warming event and associated disruptions in ecological productivity affect migratory bird populations within the region.
Here, we use occurrence information from the eBird citizen-science database (Sullivan et al. 2014) to document the response of 353 North American bird species to the March 2012 extreme warming event and the lower summer ecological productivity. We consider weekly patterns of occurrence during the period 2010 to 2014 within the geographic region where the warming event was strongest (Appendix A: Fig. A4). Our objective is to document, relative to the 2010–2014 average, how patterns of occurrence within the mid-latitudes for these species were affected by the extreme warming event, and to test our predictions for short- and long-distance migrants. Specifically, due to geographic proximity and more flexible migratory behavior, we expect short-distance migrants that winter in close proximity to the mid-latitudes to advance their spring migration phenology. Conversely, due to their more southerly winter distributions and more rigid migratory behavior, we expect no change in spring migration phenology for long-distance migrants. We also expect species that breed within the mid-latitudes to display evidence of lower levels of occurrence during the summer of 2012. Here, we expect the lower summer ecological productivity to result in lower population sizes through a combination of factors including greater adult mortality, lower breeding success, and early autumn migration. By testing these predictions, our goal was to improve our understanding on how mid-latitude climate extremes may affect North American migratory bird populations now and into the future.
Data acquisition and preparation
We compiled daily avian occurrence information for North American birds for the period 2010 to 2014 from the eBird citizen-science database (Sullivan et al. 2014). eBird checklists contain species’ observations and the time, geographic location, and protocol of the sampling event. We selected checklists that contained all species seen or heard during the survey (complete checklists) that used stationary, traveling, or area sampling protocols. We organized the checklists by week and year within equal-area hexagons (49 811 km2) of a global icosahedron (Appendix A: Fig. A4) (Sahr et al. 2003, Sahr 2011).
The study area encompassed nonmarine hexagon cells located between 130° to 60° W longitude, and 35° to 53° N latitude (Appendix A: Fig. A4). We selected this region because it contained the strongest temperature anomalies during March 2012 (Appendix A: Fig. A1) and was well surveyed by eBird (La Sorte et al. 2013). This region is also north of the winter distributions of most migratory species in North America and encompasses the Great Lakes of North America, a region that contains an exceptionally large number of migratory species during the breeding season (Mac Arthur 1959, Rabenold 1993).
To estimate geographic characteristics of North American migratory bird distributions within the Western Hemisphere, we used range maps of species’ breeding and winter distributions from NatureServe (Ridgely et al. 2007). For analysis, we converted range map polygons to collections of equal-area hexagons of a global icosahedron having a cell size of 12 452 km2 (Sahr et al. 2003, Sahr 2011). We estimated total migration distance between each species’ breeding and winter ranges using the great circle (orthodromic) distance between the geographic centroids of the breeding and winter ranges, which were estimated by averaging the geographic locations of the hexagon cell-centers occurring within each species’ breeding and winter ranges. We log-transformed total migration distance to improve its distributional properties, and we used robust ANOVA to test for differences in total migration distance among groups of species.
Our environmental data consisted of gridded time-series of temperature, vegetation greenness, and net primary productivity (NPP) for the Western Hemisphere. We acquired average monthly daily mean temperature from the global CRU TS data set version 3.22 (Harris et al. 2014), which covers the period from January 1962 to December 2012 at a monthly temporal resolution and 0.5° spatial resolution. We estimated vegetation greenness using the Enhanced Vegetation Index (EVI) (Huete et al. 1994) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite (Justice et al. 1998). EVI measures canopy greenness, a composite property of canopy structure, leaf area, and canopy chlorophyll content (Myneni et al. 1995). We used EVI values calculated at 1-km spatial resolution and 16-day composite periods (MOD13A2V.005) for the period 2000 to 2012. We estimated NPP using the MODIS derived Net Photosynthesis (PSN) index, defined as gross primary productivity minus maintenance respiration (Running et al. 1999, Heinsch et al. 2003). We used PSN estimates produced at 1-km resolution and 8-day composite periods (MOD17A2V.055) for the period 2000 to 2012. PSN is in units of kilograms of carbon produced per square meter.
Occurrence and environmental anomalies
We selected North American bird species for analysis using the following procedure. We first identified regularly occurring North American species, introduced species, and accidental species using the American Ornithologists’ Union checklist of North American birds, Seventh edition (Chesser et al. 2013). We then excluded accidental species and species that were associated primarily with marine environments. We then selected species whose breeding ranges were located in North America and that occurred within the study area at some point during the annual cycle and had total migration distances >0. This procedure resulted in a total of 405 species for analysis.
We estimated weekly occurrence anomalies for each species from 2010 to 2014 based on the proportion of eBird checklists by week and hexagon cell that contained the species. To account for the annual growth in survey effort (number of checklists) from 2010 to 2014 and the potential for spatiotemporal biases in our anomaly estimates (see Appendix A: Fig. A5), we scaled weekly effort for 2011 to 2014 based on the weekly effort that occurred during 2010. Here, we randomly selected checklists for each week during the period 2011 to 2014 based on the number of checklists submitted during that same week in 2010. We repeated this procedure 100 times for each week during the period 2011 to 2014.
To generate occurrence anomalies, for each hexagon cell, week, and species, we first computed the average proportion of all checklists for that cell and week that contained the species. For the years 2011 to 2014, we averaged the proportions for each cell, week, and species across the 100 eBird checklist permutations. We then generated weekly baseline proportions for each species and hexagon cell by averaging these values across the period 2010 to 2014. We then calculated the 2012 weekly occurrence anomalies for each species, week, and hexagon cell by subtracting the weekly baseline proportion from the observed weekly proportion for 2012 (see Appendix A: Fig. A6 for an example). We only considered hexagons cells in these calculations that contained proportions of occurrence >0. We then summarized the weekly occurrence anomalies across hexagon cells for each species and year using the average weekly anomaly. To minimize the effect of small sample sizes on the quality of our estimates, we retained for analysis average weekly occurrence anomalies that were derived from >6 hexagon cells.
We selected for analysis 353 species (Appendix B: Table B1) from the original 405 whose weekly occurrence anomalies were located within a minimum of 10 unique hexagon cells within the study area for all weeks of 2012 combined. This procedure removed species that had small sample sizes across the annual cycle within the study area, which would interfere with the quality of our estimates. We used a distribution based outlier detection method (van der Loo 2010) to classify species into two categories. The first category identified species that presented evidence of having unusually large positive occurrence anomalies during March 2012. This category allowed us to identify species whose populations responded to the mid-latitude extreme warming event by advancing their spring migration phenology. The second category identified species that presented evidence of having unusually large negative occurrence anomalies during the summer of 2012 (July–August). This outlier category allowed us to identify species whose populations duirng the breeding season had lower levels of occurrence in association with the lower levels of summer ecological productivity within the region (see Appendix A: Figs. A2 and A3).
We described the 2012 environmental conditions within species’ winter and breeding ranges across the annual cycle using temperature, EVI, and PSN anomalies. The baseline for the temperature anomalies was the 1901–2013 average (CRU TS 3.22), which was calculated at a monthly temporal resolution. The baseline for the 2012 EVI anomalies and PSN anomalies was the 2000–2011 average. EVI and PSN anomalies were calculated at 16-day and 8-day temporal resolution, respectively. Because of missing data, PSN anomalies were not calculated for the first four 8-day periods of 2012.
All analysis was conducted in R, version 3.1.2 (R Development Core Team 2015). Outliers were identified using the getOutliers function with the Method II procedure in the extreme values library with α = 0.05 (van der Loo 2014). Robust ANOVA was implemented using the lmRob function in the robust library (Wang et al. 2014).
From the 353 migratory bird species we considered, we identified 21 species having unusually large positive occurrence anomalies during March 2012 (Fig. 2; Appendix B: Table B1). Weekly patterns of occurrence during 2012 for these species displayed, on average, a strong positive response during the month of March and patterns of occurrence that did not differ from zero for the remaining months (Fig. 2). We identified 32 species having unusually large negative occurrence anomalies during the summer of 2012 (July–August; Fig. 2; Appendix B: Table B1). Weekly patterns of occurrence during 2012 for these species displayed, on average, a moderately positive response during the month of March and a moderately negative response during July–August (Fig. 2). Five species were shared between the March and summer categories (Appendix B: Table B1). The two responses identified for the 32 summer species differed on average from zero but did not differ from the response observed during the same periods for the 21 March species (Fig. 2). When considering the distribution of migration distances for species in the two categories, the March species had on average shorter migration distances (median = 1077 km) relative to the summer species (median = 2086 km; robust ANOVA, F = 11.02, P < 0.001).
Winter and breeding distributions were distributed differently for the two categories of outlier species (Fig. 3). The winter ranges for the March outlier species were highly concentrated within the southeastern United States, whereas the winter ranges for the summer outlier species had a limited concentration in the Caribbean and Central America with distributions extended throughout South America (Fig. 3). The breeding ranges for the March outlier species were concentrated at mid-latitudes in the eastern portion of North America, whereas the breeding ranges for the summer outlier species were concentrated primarily north of the mid-latitudes (Fig. 3).
Temperature, EVI, and PSN anomalies were defined in a different fashion across the winter and breeding ranges for the two categories of outlier species (Fig. 4). As expected under global warming, monthly temperature anomalies on both the winter and breeding ranges for both categories of species were on average significantly higher than expected during the majority of 2012 (Fig. 4a,b). During the beginning of 2012 and especially during the month of March, monthly temperature anomalies on both the winter and breeding ranges reached their highest levels for both categories of species (Fig. 4a,b). However, only on the winter ranges did the temperature anomalies differ between the two categories of species, and this was most pronounced during the month of March for the March outlier species (Fig. 4a,b).
Enhanced Vegetation Index anomalies on both the winter and especially breeding ranges were higher on average during the beginning of the year with the largest anomalies occurring during the month of March for both categories of species (Fig. 4c,d). During subsequent months, EVI anomalies declined on average, with the largest negative anomalies occurring during the period July–August on both the winter and breeding ranges for both categories of species (Fig. 4c,d). These differences were on average more pronounced for the March outlier species, especially on the winter ranges during the month of March (Fig. 4c,d).
Similar to EVI, PSN anomalies on both the winter and breeding ranges were on average significantly greater than expected during the month of March, but were only significantly lower than expected during the beginning of July (Fig. 4e,f). Similar to EVI, PSN anomalies differed significantly between the two categories species only on the winter ranges, with the March outlier species having higher than expected PSN in March and lower than expected PSN during the beginning of July (Fig. 4e,f).
Following our expectations, short-distance migrants whose winter and breeding ranges intersect the mid-latitudes had the strongest phenological response to the extreme warming event, and species whose breeding ranges intersected the mid-latitudes had lower occurrences in association with the lower levels of summer ecological productivity. Interestingly, few species displayed both responses, and unlike the species that displayed a migration response, the species that displayed a summer response were long-distance migrants that wintered primarily south of the mid-latitudes with breeding ranges centered just north of the mid-latitudes. Thus, species that displayed a summer response had weaker winter associations with the extreme warming event and weaker breeding season associations with the lower levels of ecological productivity. These findings suggest short-distance migrants that have a year-round presence within the mid-latitudes have a greater level of flexible and resilience to climate extremes and associated consequences for ecological resources, whereas populations of long-distance migrants that have a more limited year-round presence within the mid-latitudes and more rigid migratory behavior are at a distinct disadvantage, especially when climate extremes result in degraded ecological conditions on the breeding grounds. However, the majority of bird species examined in the study did not display a strong occurrence-based response, suggesting that the most significant implications may by constrained to a few migratory species.
From an ecological perspective, extreme events are defined as episodes during which the acclimatory capacities of an organism are abruptly and substantially exceeded (Gutschick and BassiriRad 2003), which can result in reproductive failure and reduced adult survival (Moreno and Møller 2011). Our findings suggest that this may have occurred for some of the species we examined, which were characterized primarily as long-distance migrants, whose populations are generally considered to be at greater risk under climate change (Both et al. 2006, Møller et al. 2008, Saino et al. 2011). More specifically, our findings suggest mid-latitude climate extremes occurring during the early spring had a negligible effect on the migration phenology of long-distance migrants that winter south of the mid-latitudes even though temperatures and ecological productivity within their winter ranges during March 2012 were higher than normal. In contrast, short-distance migrants that winter in closer proximity to the mid-latitudes responded more readily to the strong increases in temperatures and ecological productivity on their winter ranges. However, the lower patterns of occurrence during the breeding season were associated most strongly with the long-distance migrants, even though declines in ecological productivity were more pronounced on the breeding ranges of the short-distance migrants.
Because of the extensive geographic separation between breeding and wintering grounds, long-distance migrants are unlikely to have the capacity to detect and then adjust their migration phenologies to account for the effects of mid-latitude climate extremes. This geographic separation may therefore buffer long-distance migrants from the consequences of early spring mid-latitude climate extremes. Events that occur later in the spring, however, may influence en-route migration decisions of long-distance emigrants as they track the “green wave” to their breeding ground (La Sorte et al. 2014). For example, while our findings indicate that short-distance migrants had the strongest phenological response to the extreme warming event, there was one notable exception, the American Golden-Plover (Pluvialis dominica). The breeding and wintering grounds for the American Golden-Plover are widely separated within the Western Hemisphere, and this species only occurs at mid-latitudes during migration, with arrivals north of the Gulf of Mexico occurring in late February to early March (Johnson and Connors 2010). Therefore, as individuals approached mid-latitudes, which is the traditional stopover location for long-distance migratory shorebirds that breed in the Arctic (Skagen and Knopf 1993), our findings suggest that the extreme warming event resulted in phenological shifts in en-route migration timing, resulting in more rapid northward advancement and unusually strong mid-latitude occurrences.
Our findings for species that showed a strong phenological response suggest migration timing was cued by unusually high March temperatures and associated unusually high levels of ecological productivity on the wintering grounds. Greater plasticity in migration behavior, in possible combination with recent microevolutionary change in the level and form of migratory behavior (Pulido and Berthold 2004, 2010) and recent poleward shifts in winter distributions (La Sorte and Thompson 2007), appear to have allowed these species to follow the advanced phenology of ecological productivity. Moreover, this flexibility may have allowed these species to take advantage of the early pulse of resources in the absence of competitors (i.e., other migratory species still on the wintering grounds or en-route to the breeding grounds), which may have supported their continued occurrence within the mid-latitudes, even under lower levels of summer ecological productivity. However, a number of these short-distance migrants did not maintain occurrence levels during the summer, suggesting not all species benefited in a similar fashion to the early pulse of resources. Exploring the source of this variation may help clarify the ecological mechanism behind the occurrence-based responses observed in this study.
The increasing frequency of mid-latitude climate extremes under global climate change is thought be to be due to the phenomenon of Arctic amplification (Francis and Vavrus 2012). Here, the Arctic is warming more rapidly compared to the lower latitudes, a response that is being intensified as Arctic sea ice and snow cover diminish (Serreze et al. 2009, Screen and Simmonds 2010). Artic amplification also affects the seasonal location and intensity of high-altitude jet streams, including the polar-front jet stream in the Northern Hemisphere. Jet streams are closely associated with the development and propagation of storms, and alterations in the polar-front jet stream may result in changes in the frequency and intensity of mid-latitude climate extremes in the Northern Hemisphere (Archer and Caldeira 2008, Pena-Ortiz et al. 2013, Horton et al. 2015). Although determining the exact causes of climate extremes remains challenging (Barnes 2013, Screen and Simmonds 2013, Barnes et al. 2014, Cohen et al. 2014), the location and strength of the polar-front jet stream and its effects on high-latitude winds in North America has been identified as a key correlate of migration intensity (La Sorte et al. 2015). The findings from this study therefore add to the relevance of the polar-front jet stream for North American migratory birds in that we now must consider how the jet stream influences mid-latitude climate extremes.
Ecological productivity and its seasonal phenology have been broadly affected by climate change (Richardson et al. 2013, Keenan et al. 2014), and in this study we show that mid-latitude climate extremes can affect the spring phenology of ecological productivity, which may lead to resource bottlenecks (Maron et al. 2015) later in the season that can impact populations of long-distance migrants as they arrive on their breeding grounds. Further work is needed to determine how these responses affect long-term fitness and how these affects differ across migration strategies and across different classes of climate extremes (e.g., temperature, precipitation, wind) occurring during different phases of the annual cycle.
By improving our understanding of the full spectrum of environmental and ecological perturbations migratory birds are likely to encounter under global climate change, ecologists are in a better position to build more accurate predictive models and develop more effective mitigation and conservation strategies (La Sorte and Jetz 2010). If the frequency of mid-latitude climate extremes increases under climate change, the chances of migration or breeding activities being disrupted are likely to increase, with long-distance migrants potentially facing a greater risk of population declines.
We thank S. Kelling, M. Iliff, and J. Ward for valuable discussions, an anonymous reviewer for their suggestions, the eBird team for their support, and the many eBird participants for their contributions. This work was funded by the Wolf Creek Foundation, the National Science Foundation (CDI-1125228), and NASA (NNH12ZDA001N-ECOF).