Community-science reveals delayed fall migration of waterfowl and spatiotemporal effects of a changing climate

Abstract

en

1. Climate change has well-documented, yet variable, influences on the annual movements of migratory birds. The effects of climate change on fall migration remains understudied compared with spring but appears to be less consistent among species, regions and years. Changes in the pattern and timing of waterfowl migration in particular may result in cascading effects on ecosystem function, and socio-economic and cultural outcomes.
2. We investigated changes in the migration of 15 waterfowl species along a major flyway corridor of continental importance in northeastern North America using 43 years of community-science data. We built spatially- and temporally explicit hierarchical generative additive models for each species and demonstrated that climate, specifically the interaction between minimum temperature and precipitation, significantly influences migration phenology for most species.
3. Certain species' migratory movements responded to specific temperature thresholds (climate migrants) and others reacted more to the interaction of temperature and precipitation (extreme event migrants). There are already significant changes in the fall migration phenology of common waterfowl species with high ecological and economic importance, which may simply increase in the context of a changing climate.
4. If not addressed, climate change could induce mismatches in management, regulations and population surveys which would negatively impact the hunting industry. Our findings highlight the importance of considering species-specific spatiotemporal scales of effect on climate on migration and our methods can be widely adapted to quantify and forecast climate-driven changes in wildlife migration.

Résumé

fr

1. Les changements climatiques ont des influences bien documentées, mais variables, sur les mouvements annuels des oiseaux migrateurs. Les effets des changements climatiques sur les migrations automnales demeurent peu étudiés par rapport aux migrations printanières, mais il semble qu'ils soient moins constants d'une espèce, d'une région et d'une année à l'autre. Les changements dans le patron et le calendrier de la migration de la sauvagine en particulier peuvent avoir des effets en chaîne sur la fonction des écosystèmes et des impacts socio-économiques et culturels.
2. Nous avons étudié les changements dans la migration de 15 espèces de sauvagine le long d'un corridor de migration d'importance continentale dans le nord-est de l'Amérique du Nord, en utilisant 43 ans de données scientifiques communautaires. Nous avons construit des modèles additifs généralisés hiérarchiques spatialement et temporellement explicites pour chaque espèce et avons démontré que le climat, en particulier l'interaction entre la température minimale et les précipitations, influence de manière significative la phénologie de la migration pour la plupart des espèces.
3. Les mouvements migratoires de certaines espèces répondent à des seuils de température spécifiques (migrateurs climatiques) et d'autres réagissent davantage à l'interaction entre la température et les précipitations (migrateurs d'événements extrêmes). La phénologie des migrations automnales d'espèces de sauvagine commune qui ont une grande importance écologique et économique connaît déjà des changements importants, qui pourraient simplement s'accentuer dans le cadre des changements climatiques.
4. S'ils ne sont pas pris en compte, les changements climatiques pourraient induire des décalages dans la gestion, les réglementations et les enquêtes de population, ce qui aurait un impact négatif sur l'industrie de la chasse. Nos résultats soulignent l'importance de prendre en compte les échelles spatio-temporelles spécifiques sur la migration et nos méthodes peuvent être largement adaptées pour quantifier et prévoir les changements induits par le climat dans la migration de la faune.

1 INTRODUCTION

Changes in the phenology of migrating birds, or lack-thereof, due to changing climate have distinct implications for the success, survival and management of migratory birds worldwide. In the northern hemisphere, where springs are arriving earlier and warmer temperatures lingering into the fall, climate-related changes include advanced pre-breeding migration and earlier breeding (Gallinat et al., 2015; Lehikoinen et al., 2019), and prolonged time spent on the breeding grounds prior to fall migration (Green & Krementz, 2008; Krementz et al., 2012; Schummer et al., 2010). While there is ample evidence on the influence of climate change on spring bird migration (Barton & Sandercock, 2018; Van Buskirk et al., 2009), fall migration remains understudied in climate change research (Gallinat et al., 2015), despite its ecological and socio-economic importance.

While less is known on the influence of climate on waterfowl migration compared with songbird research, work to date suggests that fall migration is occurring later in most species, both in North America (Andersson et al., 2022) and Europe (Lehikoinen & Jaatinen, 2012). Waterfowl hunters also report certain species are conspicuously delaying fall migration and remaining closer to their breeding grounds into the fall and winter months (Andersson-Sköld et al., 2015; Slagle & Dietsch, 2018). Shifts in the timing of fall migration in waterfowl are linked to warming temperatures, reduced snow cover and large-scale weather patterns, which prolong post-breeding food availability, causing postponed migration departure (Andersson et al., 2022; Schummer et al., 2010; Thurber et al., 2020; Xu & Si, 2019). Warmer fall temperatures are also associated with southerly headwinds, which may slow down migration speed (Ferrarese et al., 2009; Liechti, 2006). Interspecific or inter-population differences in fall migration phenology likely stem from a complex interplay of multiple ecological mechanisms (Chmura et al., 2019).

Understanding the phenological changes in waterfowl migration is an important component of full-life cycle migratory bird conservation and associated harvest management planning (Pavón-Jordán et al., 2015). Shifts in migration phenology could have extensive effects including: reduced reproductive success and survival, habitat overuse in key staging areas (Abraham et al., 2005), loss of socio-economic opportunities by waterfowl hunters and associated tourism, impacts to subsistence harvesting by indigenous communities and reduced efficacy of monitoring and management efforts by land managers and policy makers (Dickey et al., 2008; Meehan et al., 2021; Notaro et al., 2016; Schummer et al., 2017). Understanding changes in the timing and distribution of fall migration will help direct waterfowl harvest policies by ensuring that hunting seasons coincide with waterfowl occurrence (Andersson et al., 2022). Managers can open the hunting season to coincide with peak migration abundance for maximum harvest and hunting success but these decisions must often consider other economic and social values in addition to ecological information such as when peak migration occurs (Fuller et al., 2021; Stiller et al., 2022; Vrtiska, 2021). For waterfowl management, there is also often a keen desire to keep hunting regulations simple to help recruit and retain hunters (Johnson et al., 2015) but maintaining simple rules for the hunting season could be difficult if the species show divergent response to climate change. Species or populations breeding in biomes found to be the most perturbed by climate change, such as mountains, boreal and temperate-latitude biomes, may be unequally influenced by the changing climate (Jiguet et al., 2010; Thackeray et al., 2016).

Research assessing phenology changes of bird migration in the context of climate often faces methodological limitations, including: (1) uncertainty in where climate influences migratory phenology, which often leads to conclusions drawn from climate at the spatial scale of the migration observation (i.e., breeding or stopover sites), rather than at the scale of influence at the origin of migration initiation (Schmaljohann et al., 2017) and (2) large uncertainties in the exact timing or duration of the influence of climate on migration which often leads to conclusions drawn based on the same timing as observation or coarse monthly averages (Gordo, 2007; van de Pol et al., 2016). Both issues result in studies making, oftentimes unsupported, a priori decisions on when and where climate influences migration. Ensuring that the effect of climate is measured at the appropriate spatiotemporal scales is rarely empirically tested, and the findings of this kind of study are broadly applicable to all animal ecologists.

To increase our understanding of how and why waterfowl fall migration is changing, we investigated changes in long-term migration of waterfowl using 43 years of community science data in a major flyway corridor along the St. Lawrence River in Québec, Canada. We built and compared multiple models across various combinations of space and time to best explain migration for 15 waterfowl species. We then used these models to determine whether peak migration date and passage period length changed over time. We hypothesized that (1) for many waterfowl species environmental factors such as specific temperature thresholds (climate migrants) or the interaction between temperature and precipitation causing snow events (extreme events migrants) drive fall migration phenology (frost wave hypothesis; Xu & Si, 2019), (2) a changing climate in northeastern North America, particularly milder falls/winters are delaying migration of waterfowl species over time and (3) the species-specific scale of effects will differ between species and will reveal different effects of climate on migration phenology.

2 MATERIALS AND METHODS

2.1 Study area

We restricted our analysis to the St. Lawrence Valley in southern Québec (Figure 1), which is in the Lower Great Lakes/St. Lawrence Plain Bird Conservation Region. It contains the most important breeding area for the American Black Duck (Anas rubripes), a highly prized harvest species, on the L'Isle-Verte Plateau, and the vast wetlands of the Lake Saint-Pierre region is the second largest staging ground for aquatics birds in the entire St. Lawrence (Lehoux et al., 2003). The fluvial sections on the St. Lawrence contains open water in the winter, attracting large numbers of ducks (Lepage et al., 2015). This area is expected to have substantial environmental disruptions due to climate change, including decreased water levels and flow of the St. Lawrence River, a loss of freshwater wetlands fed by the river, rising sea levels modifying or flooding coastal wetlands, and delays in ice formation on the river and coastlines (Lepage et al., 2015).

FIGURE 1

Map of the study region. Inset map—Outline of Canada and the USA in light grey with the province of Québec in darker grey. Main map—Province of Québec with the 12 economic regions for observations of species/day of year (DOY)/year were aggregated in light blue, and the six spatial scales used for climate variable extractions of one of the economic regions in shades of blue-grey. Centering of spatial scales differed for each of the economic regions as the southernmost edge of the first scale began at the same latitude as the centroid of each economic region.

2.2 Fall migration observation data

Étude des populations d'oiseaux du Québec (ÉPOQ) is a database of bird observations managed by QuébecOiseaux (Larivée, 2001). It is North America's longest-running, and prior to eBird, largest checklist program. Data are collected according to guidelines designed to maximize scientific rigour (Dunn et al., 1996). ÉPOQ consists of more than six million observations that have been collected since the early 1900s by birders from all around Québec, as well as many retro-active contributions from other published works that date back to the 1700s. Although both professional and amateur birders are able to submit their observations to ÉPOQ, 90% of the data is collected by 10% of the observers, called ‘expert observers’ (Francoeur, 2012). ÉPOQ is an invaluable source of historical bird abundance information in Québec, particularly for data compiled from the late 1960s to 2012. After 2012, the growing popularity of eBird largely replaced ÉPOQ, and the reduction in the number of observations reduced our ability to draw inference for multiple species. Unlike standardized bird surveys, ÉPOQ observations are mostly opportunistic, variable in effort and there is no explicit indication of a species' absence. However, they are well suited to provide information on trajectory and timing trends (Dunn et al., 1996).

We compiled all ÉPOQ records from 1970 to 2012 for waterfowl species from 15 August to 31 January of the subsequent year to ensure we captured a buffer period before and after fall migration. To minimize potential biases with opportunistic data collection, we restricted our analyses to complete checklists (as indicated by the observers) that had an effort of ≥30 mins and ≤8 h. We restricted the analysis to a focal area within 100 km of the St. Lawrence River, starting near Cornwall, ON at the Québec border and extending eastward to Riviere-du-Loup, QC. After applying all the filters, we identified 15 focal species that had a minimum of 10 years in which a species was reported in ≥40 checklists per year, which we identified as our minimal threshold for inclusion (Table S1) and restricted our analyses to these years. To spatially aggregate the ÉPOQ records, we georeferenced each observation to one of the 12 economic regions in the study area (Figure 1) and aggregated all observations of species/day of year (hereafter DOY) per year for each of the economic regions. Because ÉPOQ does not explicitly indicate a species' absence, we assumed the species was absent if not included on a complete checklist. We did not include checklists with geese only for duck species, as they typically represent observations of large flocks flying overhead, and therefore do not necessarily represent effort in suitable waterfowl stopover habitat.

2.3 Climate data and spatio-temporal scales

To estimate the influence of climate to the north of each economic region, which broadly reflects conditions from where birds would be migrating from, we delineated six spatial scales. Starting at the mid-latitude for each economic region and spanning the width of the region, the spatial scales moved northward at the following distances: 0–50 km, 50–150 km, 150–250 km, 250–400 km, 400–600 km, and 600–1000 km. The spatial scales also widened beyond the width of the region at a logistic growth scale (0, 4, 45, 118, 496, 800 km) so that the scales would capture possible east–west migration of birds as they travelled southwards (Figure 1). As the initial width of the first (and smallest) spatial scale is set to match with the width of the economic region, which differ in size, there is some variation in spatial scales for each economic region. This variation is small compared to the size differences between the scales (Figure S1), with the largest difference occurring at the largest scales which, proportionally, would be the smallest comparative difference across the economic regions. Expert waterfowl researchers and biologists across the study region guided the decision for the spatial scale sizes and the funnel-like shape of the scales.

Natural Resources Canada provided spatiotemporal data of minimum temperature and precipitation from interpolated weather station data (McKenney et al., 2011). This included 5-day averages (pentads) for minimum temperature and precipitation. The only other variable available from this dataset (maximum temperature) was not used in the analysis due to its correlations with minimum temperature, and the consensus among waterfowl experts was that minimum temperature during the fall was the most ecologically relevant variable. We extracted minimum temperature and precipitation at six spatial scales for each of the 12 economic regions (Figure 1). To assess how climatic conditions had changed during migration, we identified the time-period during which 90% of the EPOQ observations were made and calculated the seasonal (fall migration period) mean minimum temperature and mean precipitation during this time-period for each spatial scale. We then used a linear model (LM) to test seasonal trends in minimum temperature and precipitation with effects for year, spatial scale and their interaction that differed across spatial scales.

2.4 Modelling changing phenology of waterfowl migration over time

Previous research has used time-window analysis to explore spatiotemporal climate trends on migration (Bailey & van de Pol, 2016; Jarjour et al., 2017) but this approach is limited to linear models with a priori functions (i.e., linear or quadratic). We instead used a hierarchical generalized additive models (HGAMs; Pedersen et al., 2019) since it offered more flexibility. The HGAM allowed us to model the abundance of migrating birds in function of the DOY with a global smooth term (i.e., a nonlinear function) and allowed for a second group-level smooth term for the different year. That is to say that the abundance of migrating birds as predicted by DOY, was allowed to have its own functional response for each year, while penalizing functions that are too far from the average (Pedersen et al., 2019).

We constructed the HGAMs with Tweedie distributions and log link using the mgcv package (Wood et al., 2016) in Program R (R Core Team, 2021) with the species abundance as the response variable and the following predictors as written in R:

$$\begin{array}{c}\log \left(\text{Count}\right)~\mathrm{s}\left(\mathrm{DOY},\mathrm{bs}=“\mathrm{tp}”\right)+\mathrm{s}\left(\text{MINT}\right)+\mathrm{s}\left(\text{logPCP}\right)\hfill \\ +\mathrm{ti}\left(\text{MINT},\text{logPCP}\right)+\text{Year}+\mathrm{s}\left(\mathrm{DOY},\text{Year},\mathrm{bs}=“\mathrm{fs}”\right)\hfill \\ +\mathrm{s}\left(\text{Region},\mathrm{bs}=“\mathrm{re}”\right)+\text{offset}\left(\log \_\text{Effort}\right)\hfill \end{array}$$

With (1) DOY as a smooth term, (2) minimum temperature (MINT) as a smooth term, (3) logged precipitation (PCP) as a smooth term, (4) tensor product smooth with an interaction between minimum temperature and precipitation, (5) Year as a fixed effect, (6) a factor smooth interaction of DOY and Year, (7) the economic region (Region) as a smoothed random effect, (8) and the logged hours of effort (EFFORT) as a cumulative offset. We log transformed precipitation and effort, as we assumed that both effects would saturate as they increased. Adding the effort offset to the model means that the aggregated species data are not absolute counts, but instead a measure of the # individuals/hour of effort. This, as well as adding the economic region as a random effect, mitigates possible biases of aggregating the species data on regions that differ in geographic size and human density.

For each species, we ran 24 models, one for each combination of the six spatial scales and four time-periods (present, 5-day lag, 10-day lag and 15-day lag from the observation) of the climate data. We also tested a null model which did not include any temperature or precipitation smooths. We checked for concurvity (i.e., the nonparametric version of multicollinearity), however since the mgcv estimation procedures have been developed with concurvity issues in mind (Wood, 2008), we considered <0.8 observed concurvity as acceptable for our model variables. Models were compared using second-order Akaike's Information Criterion adjusted for small sample sizes (AICc; Hurvich & Tsai, 1989). Models were ranked according to the strength of support for each model, using measures of the difference between each candidate model and the top model (Burnham & Anderson, 2002).

2.5 Assessing change of peak migration and passage period over time

Using the top model for each species from the set of 24 spatiotemporal scale candidate models, we predicted the annual abundance, as measured in # of individuals per hour of effort for the 15 focal species in two ways. First, we made predictions based solely on the DOY and year (i.e., excluding the scale-specific climatic conditions and region). These predictions represent the annual optimal migration of birds through the area (‘optimal’), regardless of climate variables we extracted at the species-specific scale of effect. The DOY and year effects do inherently incorporate broad-scale climate even though there are not climate variables from the species-specific spatiotemporal scales input in the model (i.e., there are broad climate associations with year and time of year even without including those variables in the model). Second, we made predictions for each economic region based on the climatic conditions calculated from the species-specific spatiotemporal scale (i.e., including all smooths). These predictions represent the “realized” times for each species to be present in the area, based on (1) the climatic conditions at the species-specific spatiotemporal scale, and (2) the DOY and year effects. Because weather conditions were highly correlated across economic regions, and the underlying DOY smooth was the same for all economic regions, we calculated the mean predicted abundance across regions for each DOY to explore changes in predicted peaks of migration over time. We identified the DOY for peak abundance and constructed LMs to model the peak abundance as a function of year and type of prediction. We quantified the length of time in days it took to pass between the 25th and 75th percentile of the predicted total abundance of individuals migrating each year and we constructed GLMs with a Poisson distribution to model the length of the passage period, in days, as a function of year, type of prediction (i.e., “optimal” (year and DOY only) or “realized” (year, DOY, and climate at the species-specific spatiotemporal scale)), and their interaction. A significant interaction effect would represent a situation in which the climate conditions at the species-specific scale were alone responsible for influencing migratory changes, while a significant main effect of year alone would suggest that migration was changing, but that these changes were a cumulative effect of climate at multiple scales or stopovers.

3 RESULTS

3.1 Warmer, wetter fall over time

For the 15 focal species, 90% of EPOQ observations occurred between September 17 and December 4. During the date range of fall migration, minimum temperature increased by (reported means and 95% CI): 0.55°C (0.44–0.66) per decade (year, F = 91.16, p < 0.001; Figure S2a) and mean precipitation increased by 0.01 mm (0.05–0.16) per decade (year, F = 13.92, p < 0.001; Figure S2b). Neither of these climatic trends differed across spatial scales, so we omitted the interaction of year and scale from the final models.

3.2 Species-specific spatiotemporal scales of climate effects on bird abundance

For all 15 species, the global climatic model, including the effects of minimum temperature, precipitation, and the interaction between minimum temperature and precipitation, was preferred over the null model. Inferring migration movements from the HGAMs that predict daily abundance during fall migration, many species appear to migrate in response to temperatures near freezing and are more likely to migrate when there is also precipitation within the five-day pentad (Table S2). However, the spatial and temporal scale of the top climatic predictors varied between species (Figure 2). Abundance of most species (American Black Duck, Canada Goose, Common Eider, Common Goldeneye, Hooded Merganser, Mallard, Northern Pintail (Anas acuta), Ring-necked Duck (Aythya collaris), Snow Goose (Anser caerulescens), and Surf Scoter) were best predicted by the prevailing local climatic conditions in the survey area (0–50 km). Blue-winged Teal was best predicted by climate conditions at 50–150 km from the survey area. Barrow's Goldeneye was best predicted at 250–400 km, Long-tailed Duck was best predicted at 400–600 km, and American Wigeon was best predicted at 600–1000 km.

FIGURE 2

Heatmaps of logged AICc of the 24 candidate spatiotemporal climate hierarchical generalized additive models (GAMs) of community-science daily count of 15 waterfowl species from 1970 to 2012 as a function of day of year, climatic variables, and year in the St. Lawrence Lowlands of Québec, Canada. The top model for each species is indicated in red.

The temporal scale of climate effects were more variable (Figure 2). Abundances of five species were best predicted by the present weather conditions (American Black Duck, Common Goldeneye, Long-tailed Duck, Ring-necked Duck, and Surf Scoter). Weather conditions with a 5-day lag best predicted abundance for four species (Barrow's Goldeneye, Green-winged Teal, Hooded Merganser, and Mallard), a 10-day lag best predicted for five species (Blue-winged Teal, American Wigeon, Canada Goose, Northern Pintail, and Snow Goose), and a 15-day lag best predicted Common Eider abundances.

3.3 Influence of climatic variables on waterfowl abundance during migration

Compared with a null model, models that included climatic variables explained more of the variation seen in the data for all 15 waterfowl species, with top model fit ranging from 26.7% deviance explained (American Black Duck) to 70.7% (Long-tailed Duck; Table S2). All 15 species responded to an interaction between minimum temperature and precipitation as an explanatory variable for bird abundance during fall migration. For some species (‘climate migrants’), a clear temperature band or range is correlated with a higher abundance of individuals during migration (e.g., Canada Goose, Mallard, Northern Pintail, Ring-necked Duck, Greater Snow Goose; Supplemental Figures). For example, temperatures near 0°C or freezing drives higher abundances of Canada Geese during fall migration (Figure 3a); yet there is also an interaction with precipitation where milder temperatures (~5°C) is similar to colder temperatures without precipitation in its effect of driving Canada Goose numbers. Other species, such as the American Wigeon (‘extreme event migrants’) show no obvious temperature band (Figure 3b), and instead migration movements for these species are more driven by equal interactions of minimum temperature and precipitation (extreme weather events).

FIGURE 3

(a) Predicted abundance of Canada Goose (Branta canadensis) based on the minimum temperature and precipitation conditions at the species-specific spatiotemporal scale; (b) Predicted abundance of American Wigeon (Mareca americana) based on the minimum temperature and precipitation conditions at the species-specific spatiotemporal scale.

3.4 Changes in fall migration phenology of waterfowl over time

Of the 15 focal species, the realized peaks of migration from the species-specific spatiotemporal models were significantly delayed over time for eight species (American Wigeon, Canada Goose Common Eider, Green-winged Teal, Hooded Merganser, Mallard, Ring-necked Duck, and Snow Goose), and significantly earlier for one species, the Common Goldeneye (Table S2; Figure 4). We did not observe significant differences between the shifts in peak migration from the realized predictions and the optimal predictions for any species (Table 1).

FIGURE 4

Modelled effects of realized peak abundance for 15 species of waterfowl during fall migration from 1970 to 2012 in Québec, Canada, based on realized effects for DOY and climatic conditions. Statistically significant trends (p < 0.05) are in blue.

TABLE 1. Model coefficients and p-value from linear regressions of model-derived peak migration date and passage period length as a function of year.

Species name	Peak date year effect p-value	Peak date Yr*Pred p-value	Passage year effect p-value	Passage Yr*Pred p-value	∆ peak date per decade	∆ passage period per decade
American Black Duck	0.602	0.764	0.362	0.325	NS	NS
American Wigeon	<0.001	0.972	0.967	0.839	12.49 (3.78–21.19)	NS
Barrow's Goldeneye	0.196	0.451	<0.001	0.921	NS	−8.29 (−12.37–4.21)
Blue-winged Teal	0.736	0.308	0.278	0.873	NS	NS
Canada Goose	0.002	0.646	<0.001	0.135	6.82 (1.62–12.02)	6.76 (5.2–8.32)
Common Eider	<0.001	0.763	0.836	0.408	14.2 (5.72–22.67)	NS
Common Goldeneye	<0.001	0.297	<0.001	0.659	−13.08 (−20.48–5.68)	1.97 (0.34–3.6)
Green-winged Teal	0.001	0.981	<0.001	0.66	5.95 (0.83–11.08)	2.43 (1.08–3.78)
Hooded Merganser	0.005	0.772	<0.001	0.718	6.66 (−0.58–13.9)	−3.54 (−5.53–1.55)
Long-tailed Duck	0.72	0.819	0.017	0.913	NS	−1.38 (−2.93–0.17)
Mallard	<0.001	0.96	<0.001	0.594	18.66 (5.00–32.31)	3.39 (1.38–5.39)
Northern Pintail	0.447	0.836	0.08	0.444	NS	NS
Ring-necked Duck	0.001	0.28	0.033	0.972	2.89 (1.06–4.72)	1.36 (−0.47–3.18)
Snow Goose	<0.001	0.576	<0.001	0.321	5.77 (3.24–8.3)	4.16 (2.83–5.49)
Surf Scoter	0.179	0.174	<0.001	0.991	NS	3.02 (1.35–4.69)

• Note: Changes in peak date and passage period (in # days per decade) are reported from the realized model prediction (i.e., including year, DOY, and climate covariates). Bolded values are those statistically significant (p < 0.05).

Changes in the migration phenology differed among the significantly delayed species. Most species generally show unimodal curves when plotting abundance across the season (e.g., Hooded Merganser, Figure S19; Ring-necked Duck, Figure S27). However, the predicted migration of Common Eiders radically changed over the 43-year period; in the 1970s–1980s, the highest daily abundance of Common Eider occurred in the late summer while by the late 1990s the curve shifts to a small daily abundance peak in the mid-season, which grows to a clear unimodal curve in the latter years (Figure S13). The predicted abundance vs. DOY for the two geese species (Canada and Snow Goose) show a delayed shift of the peak as well as a growing trend for bimodal curves over time, particularly for Snow Geese (Figures S11 and S29). The predicted change of migration for the Mallard is the least consistent over time, with a mix of unimodal and bimodal daily predicted abundance vs. DOY and years where daily abundance peaked late in winter (Figure S23).

The passage length, measured as length of time in days between the 25th and 75th percentile of the predicted total abundance of individuals migrating each year, significantly lengthened over time for 7/15 species (Canada Goose, Common Goldeneye, Green-winged Teal, Mallard, Ring-necked Duck, Snow Goose, and Surf Scoter; Table S2, Figure 5). Comparatively, the passage periods of three species (Barrow's Goldeneye, Hooded Merganser and Long-tailed Duck) have significantly shortened over time (Table S2, Figure 5). As for peak migration, there were no species with significant differences in the rate of change of passage periods from the realized predictions and the optimal predictions (Table 1).

FIGURE 5

Modelled effects of year the realized passage period (from Q25 to Q72) for 15 species of waterfowl during fall migration from 1970 to 2012 in Québec, Canada, based on combined effects for DOY and climatic conditions. Statistically significant trends (p < 0.05) are in blue.

4 DISCUSSION

Using 43 years of community-science data for 15 species, we found the interaction between minimum temperature and precipitation influences fall migration phenology of waterfowl in the northern part of the Atlantic flyway of eastern North America. For 12 of the 15 species, we found significant shifts in either peak migration, passage period, or both across the study period. While all species showed a response to the interaction between temperature and precipitation, some demonstrated responses at specific temperature thresholds (climate migrants) while others showed no distinct pattern for temperature and instead reacted more to the interaction of the two variables (extreme events migrants). Our findings support other recent work documenting changes in migration phenology or winter distribution of waterfowl due to milder falls and winters (Gordo, 2007; Meehan et al., 2021; Notaro et al., 2016; Thurber et al., 2020; Xu & Si, 2019), and provides the first evidence of species-specific temperature threshold and spatiotemporal scales of effect driving changes in waterfowl fall migration.

This study highlights the need to consider multiple spatial and temporal scales when assessing phenological changes to bird migration in response to climate. Using climate at the scale of the individual observation, rather than at the scale of influence (Schmaljohann et al., 2017), or making assumptions to the timing and duration of the influence of climate (Gordo, 2007), may lead to a spatiotemporal mismatch and false conclusions. Spatiotemporally explicit climate models also provided additional insight at which stage of a migratory species' annual cycle climate change may influence fall migration phenology. For example, the timing of migration of the American Wigeon, a species that breeds in the high tundra, had a lagged climate signal hundreds of kilometres north from observation data collected in the St. Lawrence Lowlands. We hypothesize that wigeon may be triggered to leave breeding grounds with both increased precipitation and cold, and these events in the far north are correlated with increased numbers of wigeons along their migration route weeks later. Attempting to correlate this species' migration phenology and abundance during migration to climate in the St. Lawrence Lowlands would create a spatiotemporal mismatch between the scale of influence and observation, likely resulting in a type II error of no or weak climate effects on migration, when in fact climate does affect migration but manifests at greater spatiotemporal scales.

Models including climatic variables at species-specific spatiotemporal scales (‘realized’ models) were best at predicting species' abundance in our study region. As there were no significant differences between shifts in peak migration or passage period between the ‘realized’ and ‘optimal’ predictions, it was not the changes of climate in these areas in isolation that were responsible for directional shifts in migration phenology of waterfowl. Instead, these trends were a cumulative effect of climate at multiple scales or stopovers. Trends in fall migration of waterfowl over space and time may include changes in the timing of migration as well as a shift in wintering grounds over time (Cox et al., 2023; Verheijen et al., 2023). Climate broadly impacts the annual life cycle of migratory birds, even if observations in a specific region can be tied to climate at species-specific spatiotemporal scales.

Migration decisions by waterfowl are believed to be primarily driven by energetic considerations; when the metabolic costs of increased thermoregulation increases in tandem with a decrease in food quality and availability at freezing temperatures, at some point it becomes less costly for an individual to relocate southward than remain in the north (Newton, 2007; Schummer et al., 2010). We found support that the interaction between temperature and precipitation significantly influences the abundance of individuals for all 15 species during fall migration. These findings corroborate other research suggesting waterfowl are driven south by cold or inclement weather (Schummer et al., 2010, 2017; Weller et al., 2022; Xu & Si, 2019). Species with specific temperature thresholds or ‘climate migrants’ such as Canada Goose, Mallard, Northern Pintail, Ring-necked Duck, and Greater Snow Goose may have more predictable waves of migration that coincide with their temperature bands at species-specific spatiotemporal scales. Those species whose response is driven more equally by the interaction of precipitation and temperature, such as American Wigeon, are less predictable ‘extreme event migrants’, who are pushed south by less predictable storm events. Other studies have demonstrated that both the more predictable and generally cooling trend in the fall/early winter, as well as singular extreme weather events are associated with fall migration in waterfowl. For Mallards fitted with satellite telemetry units, relocation probability during fall migration was primarily influenced by winter condition characteristics, specifically snow cover in previous days and current snow depth (Weller et al., 2022). Compared to other metrics, such as frost days, ice cover, and temperature, even a single snow day was more likely to trigger relocation (Weller et al., 2022). Comparatively, the onset of freezing temperatures drove White-fronted and Swan Geese in Northern Asia southward from stopover sites before snow conditions became a factor (Xu & Si, 2019).

The fall migration phenology of many waterfowl species in eastern North America is rapidly shifting. Twelve species in this study had significant shifts in the timing of their migration peak, passage period length, or both over the 42 years of our study period. For most of these species, the peak of migration is occurring later than it did a decade or more previously, except for the Common Goldeneye. These findings add support to the growing body of work documenting delayed fall migration in waterfowl and possible resulting changes in winter distribution (Lehikoinen & Jaatinen, 2012; Meehan et al., 2021; Notaro et al., 2016; Thurber et al., 2020). Over 30 years, 6/15 northern European waterfowl species have delayed their fall migration, likely due to climate change, although this was not explicitly included in the analyses (Lehikoinen & Jaatinen, 2012). In the Chesapeake Bay, 21 waterfowl species significantly delayed mean migration dates when comparing 2002–2013 with pre-1958 data (Canada Goose, Gadwall (Mareca streper), Greater Scaup (Aythya marila), Surf Scoter, White-winged Scoter (M. deglandi), and Black Scoter (M. americana)). Across 29–35 years in southern Ontario, 4/6 species of waterfowl species exhibit delayed migration (Mallard, American Black Duck, American Wigeon, and Gadwall). The two other species in this study, Blue-winged and Green-winged Teals did not show changes in migration phenology (Thurber et al., 2020), similar to our findings for Blue-winged Teals during a similar timeframe. Unlike the previous work, our research investigates the influence of climatic variables at multiple spatiotemporal scales on fall migration of waterfowl and then uses the modelled relationship to quantify shifts in both fall migration peak and passage period. To our knowledge, this provides the first evidence of species-specific effect sizes of temperature and precipitation interactions, at the appropriate spatiotemporal scales that climate influences species-specific migration phenology. We encourage other researchers to adopt ours or a similar approach that investigates the effect of climate change on waterfowl and other bird migration at the most appropriate spatiotemporal scales.

Notable exceptions for shifts in fall migration phenology in our study was the American Black Duck, Blue-winged Teal, and the Long-tailed Duck. The American Black Duck was the species with the most irregular abundance distributions in fall migration over time, changing from unimodal to multimodal erratically. As such, there was no distinct shift in migration peak or passage period over time. As short-distance migrants and opportunistic foragers that can switch to waste agricultural grains or other anthropogenic food sources when wetlands food sources become unavailable at freezing temperature, American Black Duck has shown signs of delayed migration and wintering in more northern areas in several studies (Brook et al., 2009; Notaro et al., 2016). Inter-annual variations in anthropogenic food availability and extreme weather events (heavy snowfall) that ultimately drive this species southward may be responsible for the species' irregular fall migration patterns. As a species of critical importance to harvesters, the American Black Duck has often been used as a key species for conservation, management, and monitoring of waterfowl (Bordage et al., 2017; Conroy et al., 2002; Zimmerman et al., 2012). The  American Black Duck's irregularity in fall migration phenology, which differs from the majority of species in the study, suggest that relying on a single species to guide conservation and management for a suite of species can be highly problematic.

Comparatively the other two species that have shown no significant changes in the fall migration phenology, the Blue-winged Teal and the Long-tailed Duck, are both long-distance migrants (Bellrose, 1980). Long-distance migrants appear to use photoperiod as a migration cue and as such may be less able to adapt to a changing climate (Bellrose, 1980; Notaro et al., 2016). Other studies linking a changing climate to changes in fall migration phenology have found similar findings for a lack of influence on long-distance migrants that rely on photoperiod as a migratory cue (Notaro et al., 2016; Thurber et al., 2020).

A final exception from the general shifts of delays in fall migration among waterfowl was the significant advancement of abundance peaks for Common Goldeneye in our study region. The Common Goldeneye is a short-distance migrant that is often one of the first to arrive north in the spring and the last to move southward in the fall, and also performs a moult migration in the summer following breeding (Eadie et al., 2020). While abundance peaks for the species in the 1970s to 1990s show a clear, unimodal late season peak, this shifts to a bimodal pattern in the late 1990s to the end of the study period (Figure S15). We hypothesize that Common Goldeneye are more frequently migrating for moult and/or overwintering in our study region in the later part of the study period, leading to an earlier shift in abundances over time in our models.

Using a hierarchical, nonlinear modelling framework allowed us to explore inter- and intra-annual modelled abundance distributions of species during fall migration. Some species showed uniform inter-annual abundance distributions (e.g., Hooded Merganser, Ring-necked Duck), where the shape of the distribution barely changed between years except for a shift to a later peak of migration over time. For example, the Snow Goose also shows a uniform and very distinct migration peak from about 1979–1995, after which the shape of the distribution changes demonstrating not only a delay in migration peak but a longer passage period. Comparatively Mallards, known opportunistic feeders that can switch to anthropogenic food sources when wetlands are restricted and delay migration (Bellrose, 1980; Notaro et al., 2016), had high inter-annual variation of abundance distribution despite an overall shift in peak migration. Patterns of Mallard abundance also suggested that in certain years there were multiple migration pulses, perhaps capturing differing movements of various populations or changes in migration connectivity. The Common Eider had a sizable shift in migration peak and showed a very interesting and stepwise change in abundance distribution. For the first 10–15 years, it was most abundant in the St. Lawrence Lowlands at the end of the summer/early fall. Over time, this shifted to a clearer migratory signal with a unimodal peak that grew stronger over time. Common Eiders are short-distance migrants and supplanting, a phenomenon where northern birds replace those that have bred locally and migrated south, is common throughout eastern North America (Goudie et al., 2020). As southern populations of this species are declining (Noel et al., 2021; Savard et al., 2015), these phenology changes over time might be influenced by a declining number of more ‘local’ breeders that pass through the St Lawrence region at the beginning of the migration period. At the same time, the occurrence of more northern breeders moving into the study region to overwinter increases with milder winters (Robertson et al., 2021).

Using a large community-science dataset such as ÉPOQ, which in our study covered 43 years of data, including >189,000 lists of >54 million individuals, allows for the exploration of long-term trends and shifts of animal movements, such as migration phenology. However, certain signals change in migration phenology were stronger in some species than others, despite model or variable significance under frequentist statistics. For example, in some species the trends of changes in peak abundance during migration were steady over time (e.g., Snow Goose, Common Goldeneye, Common Eider), while for others the trend was more changeable and would benefit from additional years of data despite being significant in our analysis (American Wigeon). As other community-science datasets grow (e.g., eBird and iNaturalist), we can continue and adapt these models to best inform conservation and management actions.

Ongoing climate change leading to delayed fall migration of waterfowl and northward shift of their winter distribution will have drastic ecological and economic outcomes in North America. Foraging pressure on remaining northern and central wetland habitats will be increased, highlighting the critical need to protect and restore wetlands in these areas (Notaro et al., 2016). Specifically to our study area, the Lac St. Pierre Biosphere Reserve designated by UNESCO and internationally recognized wetland of importance by the RAMSAR convention, has undergone significant annual variation in its water levels related to climate change and faces increased pressure of habitat loss and degradation due to agricultural activity (SLAP, 2015). Waterfowl hunting, which occurs in the fall and varies geographically, is a multi-billion-dollar industry in North America (Henderson, 2005). Species-specific changes in migration timing and demography may lead to changes in hunting pressures and waterfowl availability to hunters. Our study suggests there are already significant changes in the fall migration phenology of common waterfowl species with high ecological and economic importance (e.g., Mallard, Snow Goose, Ring-necked Duck), which may simply increase in the context of a changing climate. Due to data limitation in our dataset our analysis ended in 2012, but these trends are clearly continuing as climate continues changes. These changes need to be quantified and forecasted to allow for adaptive management. Our results also highlight the variation in the species-specific response to climate changes and the inherent difficulties that managers will face to develop global management and conservation strategies. If not addressed, climate change could induce mismatches in management that may lead to hunting windows to be biased towards certain species or populations or even age and sex classes, for which there is huge variation across the annual cycles of migratory birds, including the timing of migration (Newton, 2011). In addition, winter surveys or harvest management tools such as banding programs, part collections surveys, or hunter surveys that are sometimes used to inform population models could also provide biased information over time (Devers et al., 2021; Robertson et al., 2021). Yet, there may be an unseen benefit to the fact that a changing climate is responsible for less ducks for hunting in the southern United States. Herein lies an opportunity for dialogue with a group that generally engages in less climate change mitigation actions (Ghasemi & Kyle, 2021), but has strong potential for conservation action, as the tremendous success in wetland conservation and waterfowl management has shown (Anderson & Padding, 2015).

AUTHOR CONTRIBUTIONS

B.F., A.R.C. and C.R. conceived the ideas and designed methodology, B.F., A.R.C. and A.C.M. analysed the data, B.F. led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

ACKNOWLEDGEMENTS

We thank QuébecOiseaux for sharing the Étude des populations d'oiseaux du Québec (ÉPOQ) database, Jean-Sébastien Guénette for his insight on the database, and most notably the thousands of community scientists that contributed to ÉPOQ. We thank Christine Lepage and Josée Lefevre for providing insightful comments on the work. Earlier drafts were greatly improved thanks to the detailed comments and insights of two anonymous reviewers. Funding was provided by Environment and Climate Change Canada.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Open Research

DATA AVAILABILITY STATEMENT

Data available from the Dryad Digital Repository https://doi.org/10.5061/dryad.wwpzgmsrd (Frei et al., 2024).