Environmental drivers of productivity explain population patterns of an Arctic‐nesting bird across a half‐century

Abstract

Joint estimation of demographic rates and population size has become an essential tool in ecology because it enables evaluating mechanisms for population change and testing hypotheses about drivers of demography in a single modeling framework. This approach provides a comprehensive perspective on population dynamics and how animal populations will respond to global pressures in future years. However, long‐term data for such analyses are often limited in quantity and quality. We developed an integrated population model combining data on demography and population size from nine different sources to understand the population ecology of the lesser snow goose (Anser caerulescens caerulescens) in the Pacific Flyway in North America from 1970 to 2022. We divided the flyway population into Wrangel Island and Western Arctic subpopulations and assessed demographic mechanisms for population change and environmental and anthropogenic drivers that influenced demography. During 1970–2022, the estimated spring population of snow geese in the Pacific Flyway increased from ~300,000 to ~2,300,000. Short‐term changes in population growth rate were primarily driven by changes in productivity in the Western Arctic and productivity and immigration in Wrangel Island. Changes in hunting and natural mortality had less influence on short‐term but likely contributed to the pronounced long‐term population growth. Early snowmelt positively influenced per capita productivity in both regions, and warm, rainy weather during the non‐breeding season was associated with high per capita productivity in the Western Arctic. In the Western Arctic, per capita productivity was negatively associated with population size, and adult natural mortality was positively associated with population size, indicating density‐dependent regulation in this subpopulation. In Wrangel Island, warm weather in early fall decreased juvenile natural mortality. Our results demonstrate that per capita productivity and immigration, rather than adult survival, were the primary mechanisms of short‐term population change in this long‐lived species. Our results also indicate that environmental conditions and density‐dependent effects can impact population dynamics more than harvest, even for a long‐lived, commonly harvested species. We demonstrate that a warming climate can have multiple effects on demography, emphasizing the importance of assessing a variety of spatial and temporal factors when predicting how populations might respond to large‐scale environmental changes. This emphasizes the importance of conservation plans that consider these environmental drivers, although this may complicate direct management of such populations.

INTRODUCTION

Estimating the demographic mechanisms of population change is fundamental in ecology because it is the first step in understanding why and how animal populations change over time and across regions (e.g., Newton, ¹⁹⁹⁸). Demography (i.e., survival and productivity) is influenced not only by the life‐history characteristics of a species but also by the environmental conditions individuals encounter throughout their annual cycle. To gain a more comprehensive understanding about population dynamics and predict how populations might react to future environmental changes, it is essential to use the best available data and analytical methods to quantify which, and to what extent, environmental conditions affect demography (Ellner & Fieberg, 2003; Selwood et al., ²⁰¹⁵). In addition to environmental factors, both direct (such as harvest) and indirect (such as climate change) anthropogenic impacts can affect demography (e.g., Brashares et al., ²⁰⁰⁴; Ceballos et al., ²⁰²¹). Furthermore, demographic outcomes of natural and anthropogenic processes are often interrelated (Gibson et al., 2023; Riecke et al., ^{2022a, 2022b}). Therefore, jointly estimating demographic rates and their contributions to population change, as well as their environmental and anthropogenic drivers, provides comprehensive understanding of population ecology and informs conservation and management efforts (Koons et al., 2017).

Long‐term monitoring programs provide the foundational data needed for assessing population changes and their environmental correlates (Lindenmayer et al., 2012). However, for many species, the data needed to estimate demographic rates and trends, including survival, productivity, and population size, are often limited in quantity and quality, and data gaps limit rigorous assessments of population status and trends (Conde et al., 2019). Long‐term monitoring programs are critical for assessing species status and trends due to their long‐term, multi‐species, and broad‐scale focus (Rosenberg et al., 2019). Programs such as breeding bird surveys or bird banding programs have provided decades of data about demography and population sizes of North American waterfowl and led to a wealth of ecological knowledge (Cooch et al., 2014). However, because of logistical and financial constraints, many long‐term research programs contain gaps in data collection or protocol changes. Few monitoring programs can be sustained indefinitely, limiting the scope and scale of inferences possible from such programs.

Disentangling the impacts of harvest, environmental effects, and density‐dependent regulation on the demography of harvested populations is complicated. When focusing only on relationships between harvest and, for example, annual survival rates, the influence of harvest on demography can be confounded with environmental changes or density‐dependent processes (Riecke et al., ^2022b). To rigorously quantify the effect of harvest on demographic rates and population change, it is important to use as much available data as possible to inform estimates of population size and demographic rates. Data integration, which combines multiple datasets into a single model, offers a powerful approach for leveraging all available data to improve parameter estimation and increase the scope of inference from individual datasets (Zipkin et al., 2021). Integrated population models (IPMs) combine demographic data with a time series of population‐ or site‐level count data to mechanistically link population processes to demographic rates (Schaub & Kéry, 2022). IPMs can be used to combine datasets with different spatial or temporal coverage, allowing estimation of parameters for which little to no data exist. Another benefit is improved precision on parameter estimates, which is most beneficial when using complex models with large numbers of parameters but limited data (Rhodes et al., 2011).

Lesser snow geese (Anser caerulescens caerulescens, hereafter snow geese) in the Pacific Flyway of western North America are a prime example of a harvested population where uncertainties remain about population dynamics and their environmental and anthropogenic drivers despite long‐term monitoring and research efforts. The Pacific Flyway population comprises two subpopulations: one breeding on Wrangel Island in Russia and the other across the North Slope of Alaska and Western Canadian Arctic. The size of the Pacific Flyway snow goose population has substantially increased in recent decades (Baranyuk et al., 2018; Burgess et al., ²⁰¹⁷; Kerbes et al., ²⁰¹⁴), but the mechanism(s) for this change are unknown. Interestingly, the increase in snow goose numbers in the Pacific Flyway coincided with a decrease in the adjacent hyperabundant midcontinent snow goose population (USFWS, ²⁰²³). Researchers have documented substantial immigration into some breeding colonies in the Pacific Flyway (Burgess et al., 2017; Johnson, ¹⁹⁹⁵), but movement between the midcontinent population and the Pacific Flyway is thought to be minimal (Alisauskas et al., 2022). While much effort has been devoted to understanding harvest and population dynamics of midcontinent lesser snow geese, greater snow geese (Anser caerulescens atlantica), and Ross's geese (Anser rossii), the impact of harvest mortality on Pacific Flyway lesser snow goose population dynamics remains largely unknown (Alisauskas et al., 2011; Lefebvre et al., ²⁰¹⁷; LeTourneux et al., ²⁰²⁴ and references therein). Moreover, while relationships between environmental conditions and demography are well‐studied for midcontinent snow geese (e.g., Baldwin et al., ²⁰²²; Ross et al., ^{2017, 2018} and references therein), they are mostly unknown for Pacific Flyway snow geese (but see Ruthrauff et al., ²⁰²¹; Samelius et al., ²⁰⁰⁸).

There are 50+ years of data collected on the population dynamics of Pacific Flyway snow geese, although the long‐term, cooperative, multi‐national program to monitor breeding snow geese on Wrangel Island, Russia, was ended in 2022. Reductions in monitoring and banding programs on the North Slope of Alaska and in the Western Canadian Arctic, where the remaining Pacific Flyway snow geese breed, would further erode long‐term data. Most demographic data collected about these birds has been on breeding areas in the Arctic, causing concern in the conservation community about the confidence in management decisions if various monitoring programs are diminished or ceased. These events magnify the importance of understanding the causes and consequences of the recent rapid increase in Pacific Flyway snow geese to guide future data collection schemes and conservation plans.

In this study, we developed an IPM to jointly analyze nine datasets representing population size, survival, and productivity, together with data on changes in environmental conditions and anthropogenic pressures, to understand population dynamics of Pacific Flyway lesser snow geese from 1970 to 2022. We investigated mechanisms of population change and to what extent environmental factors and hunting affect demography. We also estimated the impact of harvest regulations and hunter availability on changes in harvest mortality. We hypothesized that per capita productivity was influenced by both direct effects of local weather during the breeding season and carry‐over effects from weather conditions during the non‐breeding season (Baldwin et al., 2022; Cunningham et al., ²⁰²¹; Ross et al., ²⁰¹⁸). We also hypothesized that adult and juvenile natural mortality was affected by density‐dependent effects and local weather conditions before the fall migration (Menu et al., 2005). Last, we hypothesized that harvest mortality of snow geese was influenced by harvest regulations and hunter availability (Riecke et al., ^2022a).

We anticipate that this work will provide important insights into the ecological and anthropogenic factors affecting dynamics of our study population, as well as those affecting other arctic‐nesting geese and species. Our study also serves as a useful example for practitioners similarly facing challenging conservation and management decisions for species where population dynamics, and response to management actions, are poorly understood.

METHODS

Study population

Snow geese in the Pacific Flyway breed in the Western Canadian Arctic, the North Slope of Alaska, and Wrangel Island in Russia. Two of the primary breeding colonies are on Wrangel Island, three in the North Slope of Alaska, and three in the Western Canadian Arctic (Figure 1). Geese breeding in the latter two regions comprise the Western Arctic snow goose subpopulation and are typically considered together for conservation plans because they share staging and wintering sites (Pacific Flyway Council, 2013). We follow this delineation and categorize Pacific Flyway snow geese into Wrangel Island and Western Arctic subpopulations. The birds breeding in Wrangel Island mainly winter around the Fraser River and Skagit River deltas of British Columbia and Washington and the Central Valley of California (Baranyuk et al., 2018; Williams et al., ²⁰⁰⁸; Figure 1). The main wintering site for the Western Arctic birds is the Central Valley of California, but some of these birds winter in Oregon, some interior US states, and Mexico (Williams et al., 2008; Figure 1).

FIGURE 1.

Breeding colonies (open circles), band recovery locations (dots), and the key areas (rectangles) used by snow geese breeding in the Western Arctic and Wrangel Island. Recoveries are from birds banded in 1970–2021 across the study area. Red and blue dots denote recoveries from birds banded in the Western Arctic and Wrangel Island, respectively. Dots are transparent to highlight the regions with most recoveries. Rectangles highlight the approximate locations of the largest breeding colonies (on Wrangel and Banks islands) and the main wintering (Central Valley of California, Fraser, and Skagit river deltas) and stopover (Alberta and Western Saskatchewan) sites. Arrows denote main migratory directions. The dashed arrow denotes a minor migration corridor between Wrangel Island and the stopover site in Alberta‐Saskatchewan. Small number of recoveries south of 25° N are left outside the map.

Population size data

We used four different monitoring datasets to estimate the size of the Pacific Flyway snow goose population (see Appendix S2: Table S1 for the temporal coverage of datasets). For Wrangel Island birds, we used ground‐based surveys of the total spring population size, which were available almost annually since 1960 (see Baranyuk & Kraege, ²⁰¹⁷; Kerbes et al., ¹⁹⁹⁹ for methods). For Western Arctic birds, we incorporated three different measures of population size: aerial counts of nesting birds at primary breeding colonies (Western Arctic Photo Survey, available sparsely between 1976 and 2013; Kerbes et al., 2014), index of winter population size (Midwinter Survey Index, available almost annually since 1979; Olson, 2022), and Lincoln estimates (Lincoln, 1930) derived from harvest and band‐recovery data (calculated for most years between 1988 and 2019). The Western Arctic Photo Survey data include the size of nesting colonies where most of the Western Arctic subpopulation breeds.

Midwinter survey

The number of snow geese and Ross's geese (collectively referred to as light geese) wintering in Oregon and the Central Valley of California has been indexed during an aerial Midwinter Survey since 1979 (Pacific Flyway Council, 2006). Both species have predominantly white plumage and cannot be separately identified during the aerial surveys. Counts include Ross's geese, snow geese from the Western Arctic and Wrangel Island, as well as low numbers of snow geese from the midcontinent population (Olson & Sanders, 2017). The proportion of Ross's geese is typically estimated every third year from a ground‐based survey, and we estimated the annual proportion of Ross's geese by fitting a second‐order polynomial function to the proportion data. To derive the annual winter estimate of total snow geese, we subtracted the estimated annual proportion of Ross's geese from the total light goose estimate and assumed that the number of midcontinent snow geese in the counts was negligible (Alisauskas et al., 2022). To derive the annual winter estimate of just Western Arctic snow geese, we subtracted the estimated annual number of Wrangel Island snow geese wintering in Oregon and Washington. To do this, we subtracted the number of snow geese in the winter counts in the Fraser‐Skagit River Delta from the Wrangel Island spring counts and considered the remaining Wrangel Island birds to be wintering in California and Oregon.

Lincoln estimates

We calculated Lincoln estimates for adult Western Arctic snow geese during 1988–2019 using a bias‐corrected estimator and estimated variances using the delta method (see Appendix S1: Section S1.1.3; Alisauskas et al., ^{2009, 2014} for equations and details). We used harvest data from Olson (2022), partitioned to consider only Western Arctic snow geese following Dooley (2017). We accounted for overestimation bias in harvest estimates using a correction factor by Padding and Royle (2012). To decrease potential bias in Lincoln estimates caused by a small number of banded birds and variable harvest rate of juveniles, we calculated them only for the adult population in the Western Arctic and only in years when at least 400 adult snow geese were banded in the region (n = 21 years).

Productivity data

To estimate per capita productivity of Wrangel Island snow geese, we used several data sources: proportion of juveniles in midwinter surveys around the Fraser‐Skagit River deltas (see Olson & Sanders, ²⁰¹⁷), proportion of juveniles in spring counts on breeding areas (see Baranyuk & Kraege, ²⁰¹⁷), and the ratio of juveniles and adults at harvest (hereafter age‐ratio at harvest). See Appendix S2: Table S1 for the temporal coverage of different productivity datasets. We used harvest data from the Parts Collection Survey (PCS; Raftovich et al., ²⁰²³) and from a local harvest survey (HS) in the Fraser Valley (retrieved from the Canadian Wildlife Service). For Western Arctic snow geese, we estimated per capita productivity using the age‐ratio at harvest (PCS) because it was the only long‐term dataset available. To estimate the age‐ratio at harvest separately for Wrangel Island and Western Arctic snow geese, we considered age‐ratios among birds harvested in Washington and British Columbia (where most Wrangel Island birds winter) to represent productivity of the Wrangel Island subpopulation and those harvested in California and Oregon (where most Western Arctic birds winter) to represent productivity of the Western Arctic subpopulation (Dooley, 2017).

Band‐recovery data

Biologists captured snow geese by herding them into portable corral traps during summer while adults were flightless due to wing molt and juveniles were too young to fly (Baranyuk & Kraege, 2017). Birds were captured for banding at brood rearing sites and molting sites of non‐breeding birds at all main colonies within the study area but were not continuous at any of the colonies (see Appendix S2: Table S2 for the annual banding totals). Birds were aged based on plumage, sexed using cloacal examination, and fitted with a uniquely engraved metal leg band (Cooke et al., 1995). We restricted the analysis to birds caught between June and September. After filtering, our banding data in the analysis included 76,228 banded adults and 35,859 juveniles. We also obtained records of all banded lesser snow geese that were shot and reported to the US Geological Survey Bird Banding Lab between 1970 and 2022 (n = 17,739; https://www.pwrc.usgs.gov/BBL). Preliminary analysis indicated that differences in survival between males and females were relatively small, so we combined sex classes in our analyses to increase the sample size.

Explanatory variables

Following our hypothesis, we tested the effect of (1) timing of snow melt in spring, (2) extreme weather conditions around hatch, and (3) weather conditions during winter on per capita productivity. Early snow melt is known to increase per capita productivity of Arctic‐nesting geese by enhancing breeding propensity and success (e.g., Boom et al., ²⁰²³; Reed et al., ²⁰⁰⁴; Ross et al., ²⁰¹⁷; Samelius et al., ²⁰⁰⁸), so we predicted that timing of snow melt in spring (SNOW _r,t , defined annually as the earliest date with 50% snow cover) would negatively affect productivity. Harsh weather conditions around hatch can increase mortality of small goslings (Ross et al., 2017). We predicted that cold and wet weather conditions around hatch (EXT _r,t , defined as the number of days with either freezing minimum temperature or minimum temperature below +2°C and daily precipitation ≥10 mm during the 21 days following the estimated annual hatch date in each colony) would increase mortality of goslings and decrease productivity (see Appendix S3 for the estimation of hatch dates). Favorable weather conditions during the non‐breeding period are thought to affect the body condition of breeding geese through decreased energy expenditure or increased feeding opportunities, influencing productivity (Baldwin et al., 2022; Cunningham et al., ²⁰²¹; Ross et al., ²⁰¹⁷). Therefore, we predicted that warm weather between December and March (ONI _t , the values of Oceanic Niño Index) would increase productivity. In Arctic‐nesting geese, warm temperature at the breeding grounds during early fall can decrease juvenile mortality by extending the period for fueling before fall migration (Menu et al., 2005). We tested the effect of mean temperature in September (TEMP _r,t ) at the breeding sites on juvenile natural (non‐harvest) mortality and predicted that it negatively influenced this parameter.

To assess these predictions, we used the ERA5‐Land reanalysis dataset (Muñoz‐Sabater, 2019) for Wrangel Island and Banks Island snow goose colonies to derive time series of the environmental covariates described above. We used data only from Banks Island to represent environmental conditions for Western Arctic birds as more than 90% of the subpopulation breeds there. We retrieved values of Oceanic Niño Index (ONI) from the National Oceanic and Atmospheric Administration (NOAA; Barnston et al., ¹⁹⁹⁷). The ONI is an indicator for monitoring phases of the El Niño‐Southern Oscillation (ENSO) climate pattern. It has high values during El Niño, when temperature and precipitation are high at the main wintering sites of Western Arctic snow geese in California. Conversely, the index has low values during La Niña, when weather conditions are colder and drier in California, but wetter in the wintering sites of Wrangel Island snow geese in Washington and British Columbia (e.g., Barnston et al., ¹⁹⁹⁷). We used ONI values from December to March, covering the wintering and the beginning of spring migration (when most birds are close to their wintering sites).

Regulation of hunting aims to maintain harvest mortality of the target species within desired levels. In addition, harvest mortality can be influenced by the number of available hunters (Riecke et al., ^2022a). We predicted that snow goose harvest mortality is negatively influenced by the number of available hunters and more liberal hunting regulations. To assess these predictions, we utilized data from the Mail Questionnaire Survey (MQS) during 1970–1999 (retrieved from Trost & Drut, ²⁰⁰³) and the Harvest Information Program (HIP) during 1999–2021 (retrieved from Olson, ²⁰²²), which describe the number of active adult duck hunters in the US Pacific Flyway (which constitutes the vast majority of hunters within the entire Pacific Flyway) and serve as a proxy for goose hunter availability. Due to methodological differences between these surveys, we aligned the MQS and HIP data using three overlapping years (1999–2001; see Appendix S4: Figure S1). We assessed annual hunting regulations by calculating the product of the total length of the waterfowl hunting season (in days) and the bag limit (allowable harvested geese per hunter per day) in each province or state and then summed these values to derive an index for the entire Pacific Flyway (Appendix S4: Figure S2).

We also examined density‐dependent effects on natural mortality and per capita productivity by modeling them as a function of population size. We did not include any correlated predictors (r > 0.4) in the models (see Appendix S4: Figures S3 and S4 for correlations between predictors). We standardized all predictors to the mean of zero and SD of one (z‐standardization).

Modeling approach

We combined all data into an IPM, which comprised of three sub‐models: population process, productivity, and survival. These sub‐models were connected in a joint likelihood in the IPM (Figure 2). We describe each sub‐model here but included some details in Appendix S1.

FIGURE 2.

Graphical representation of integrated population models (IPM) for Wrangel Island and Western Arctic snow geese. The dotted squares denote the individual sub‐models (i.e., likelihoods), and the gray squares denote the joint likelihood. The green squares denote datasets: mark‐recovery (m), count data (C _x ), age‐ratios at counts (AR _x ), and age‐ratios at harvest (PCS, HS). Orange circles denote the parameters: kill rate (h ^[κ]), natural mortality probability (h ^[η]), annual survival (S), population size (N), per capita productivity (f), and immigration rate (ω). Variables in the black squares are covariates used to explain variation in natural and harvest mortality probability and per capita productivity: hunter availability (N _HUNTER), harvest regulation (REG), log‐population size [log(N)], timing of snow melt (SNOW), extreme weather conditions around hatch (EXT), and ONI‐index (ONI). The subscripts linc, win, spr, waps denote Lincoln estimates, winter counts, spring counts, and Western Arctic Photo Survey, respectively.

Population process sub‐model

We used a state‐space population process to describe the development of spring (pre‐birth) population size in time. The population process is described separately for the two subpopulations (r) within the study area: Wrangel Island (WRI) and Western Arctic (WA). The spring population is comprised of juveniles ($N_{\mathrm{JUV},r,t}$, birds hatched in previous summer in the study area), adults ($N_{\mathrm{AD},r,t}$, birds survived from the previous year), and immigrants ($N_{I,r,t}$, birds which originate outside the given study region and are assumed to be adults). The population process can be expressed as follows:

$$N_{r,t+1}=\left[\begin{array}{c}{N}_{\mathrm{JUV},t+1}\\ N_{\mathrm{AD},t+1}\end{array}\right]=\left[\begin{array}{cc}{\mathrm{\gamma }}_{r,t}{S}_{\mathrm{JUV},r,t}& {\mathrm{\gamma }}_{r,t}{S}_{\mathrm{JUV},r,t}\\ S_{\mathrm{AD},r,t}& S_{\mathrm{AD},r,t}\end{array}\right]\left[\begin{array}{c}{N}_{\mathrm{JUV},r,t}\\ N_{\mathrm{AD},r,t}\end{array}\right]+\left[\begin{array}{c}0\\ N_{I,r,t+1}\end{array}\right]$$ (1)

where ${\mathrm{\gamma }}_{r,t}$ is the region‐specific per capita productivity (see productivity sub‐model) and S _a,r,t is the annual survival probability for each age‐class (a). To inform population size estimates with spring counts, winter counts, Western Arctic Photo Surveys, and Lincoln estimates, we assumed that natural mortality was uniformly distributed across the year and harvest mortality was uniformly distributed across the hunting season (September–January before 2009, September–March from 2009 onward). We then multiplied spring population size estimates $\left(N_{r,t}\right)$ with the natural and/or harvest survival over the period between the spring and each survey to derive a population size estimate at the time of each survey. We used zero‐truncated normal distributions as observation models for all population size datasets. We modeled the number of immigrants entering each subpopulation annually as temporal random effects using log‐normal distribution. For the details of population size likelihoods, see Appendix S1: Sections S1.1.1–S1.1.4. For the details of immigration model, see Appendix S1: Section S1.1.5. We scaled population sizes by dividing them by 100,000 to improve optimization of algorithms used in model fitting.

Survival sub‐model

We formed the likelihood for the band‐recovery data using the Brownie model (Brownie, 1978, see Appendix S1: Section S1.2.1 for details of the Brownie model). We parameterized the band‐recovery model with age‐ (a), time‐ (t), and region‐specific (r) survival $\left(S_{a,r,t}\right)$ and band‐recovery $\left(f_{a,r,t}\right)$ rates. We derived kill rates (κ, hunting mortality probability) using band‐recovery rates, crippling rate (c, probability that a bird is killed by a hunter but not retrieved) [prior: c ~ Beta(20,80); USFWS, ²⁰¹⁹], and band reporting rates (rr _t ) such that

$$f_{a,r,t}={\mathrm{\kappa }}_{a,r,t}\left(1-c\right){\mathrm{rr}}_t.$$ (2)

We used informative priors for reporting rates (rr _t , i.e., the probability that a hunter reports a banded birds that is shot and retrieved) based on reporting rates estimated for mallards, which were available for years 1970–2010 (Arnold et al., 2020) and 2017–2022 (USFWS, ²⁰²¹). For years 2011–2016, we derived the prior for rr _t by fitting a Gaussian process model (see Piironen et al., ²⁰²², and references therein) to the previously published reporting rates and used model estimates as data in years with missing information to form year‐specific prior distributions for rr _t (see Appendix S4: Figure S5).

We parameterized the survival sub‐model with age‐, time‐, and region‐specific harvest $\left(h_{a,r,t}^{\left[\mathrm{\kappa }\right]}\right)$ and natural $\left(h_{a,r,t}^{\left[\mathrm{\eta }\right]}\right)$ mortality hazard rates (Ergon et al., 2018), and we derived associated annual harvest (κ) and natural (η) mortality rate, as well as annual survival rate (S) as a function of the cause‐specific mortality hazard rates. This allowed us to estimate harvest and natural mortality rates without specifying a functional relationship between the two (i.e., without specifying whether the hunting mortality is additive or not, see Ergon et al., ²⁰¹⁸):

$${\mathrm{\kappa }}_{a,r,t}=\left(1-S_{a,r,t}\right)\frac{h_{a,r,t}^{\left[\mathrm{\kappa }\right]}}{h_{a,r,t}^{\left[\mathrm{\kappa }\right]}+h_{a,r,t}^{\left[\mathrm{\eta }\right]}}$$ (3)

$${\mathrm{\eta }}_{a,r,t}=\left(1-S_{a,r,t}\right)\frac{h_{a,r,t}^{\left[\mathrm{\eta }\right]}}{h_{a,r,t}^{\left[\mathrm{\kappa }\right]}+h_{a,r,t}^{\left[\mathrm{\eta }\right]}}$$ (4)

$$S_{a,r,t}=e^{-\left(h_{a,r,t}^{\left[\mathrm{\kappa }\right]}+h_{a,r,t}^{\left[\mathrm{\eta }\right]}\right)}.$$ (5)

We modeled $h_{a,r,t}^{\left[\mathrm{\kappa }\right]}$ as a function of the number of active hunters $\left(N_{\text{HUNTER}}\right)$ and hunting regulations (REG), $h_{\mathrm{AD},r,t}^{\left[\mathrm{\eta }\right]}$ as a function of region‐specific population sizes, and $h_{\mathrm{JUV},r,t}^{\left[\mathrm{\eta }\right]}$ as a function of region‐specific population sizes and fall temperature as follows:

$$\log \left(h_{a,r,t}^{\left[\mathrm{\kappa }\right]}\right)={\mathrm{\beta }}_{a,r}^{\left[\mathrm{\kappa }\right]}+{\mathrm{\beta }}_{a,r}^{\left[\text{HUNTER}\right]}{N}_{\text{HUNTER}}+{\mathrm{\beta }}_{a,r}^{\left[\mathrm{REG}\right]}\mathrm{REG}+{\mathrm{\epsilon }}_{a,r,t}^{\left[\mathrm{\kappa }\right]}$$ (6)

$$\log \left(h_{\mathrm{AD},r,t}^{\left[\mathrm{\eta }\right]}\right)={\mathrm{\beta }}_{\mathrm{AD},r}^{\left[\mathrm{\eta }\right]}+{\mathrm{\beta }}_{\mathrm{AD},r}^{\left[\text{DENS}\right]}\log \left(N_{r,t}\right)+{\mathrm{\epsilon }}_{\mathrm{AD},t}^{\left[\mathrm{\eta }\right]}$$ (7)

$$\log \left(h_{\mathrm{JUV},r,t}^{\left[\mathrm{\eta }\right]}\right)={\mathrm{\beta }}_{\mathrm{JUV},r}^{\left[\mathrm{\eta }\right]}+{\mathrm{\beta }}_{\mathrm{JUV},r}^{\left[\text{DENS}\right]}\log \left(N_{r,t}\right)+{\mathrm{\beta }}_r^{\left[\text{TEMP}\right]}{\text{TEMP}}_{r,t}+{\mathrm{\epsilon }}_{\mathrm{JUV},t}^{\left[\mathrm{\eta }\right]}$$ (8)

where are the age‐ and region‐specific intercepts, are the regression slopes, and are the residual errors (see Appendix S1: Section S1.2.2 for modeling of residual errors).

Productivity sub‐model

Our productivity datasets represent the proportion of juveniles among all birds, so we estimated per capita productivity ($\mathrm{\gamma }$) in each region (r, WRI or WA) and year (t) as the number of juvenile females produced per adult female in the spring population. We modeled age‐ratios at midwinter surveys $\left({\mathrm{AR}}_{\mathrm{WIN}}\right)$ along the Fraser and Skagit River deltas (wintering site of the Wrangel Island subpopulation) as follows:

$$n_{\mathrm{JUV},t}^{\left[\mathrm{win}\right]}~\text{Binomial}\left({\mathrm{\theta }}_{\mathrm{JUV},t}^{\left[\mathrm{win}\right]},\left(n_{\mathrm{JUV},t}^{\left[\mathrm{win}\right]}+n_{\mathrm{AD},t}^{\left[\mathrm{win}\right]}\right)\right)$$ (9)

where $n_{a,t}^{\left[\mathrm{win}\right]}$ is the number of juvenile and adult snow geese observed in the counts and ${\mathrm{\theta }}_{\mathrm{JUV},t}^{\left[\mathrm{win}\right]}$ is the estimated proportion of juveniles in the population in February (see calculation of ${\mathrm{\theta }}_{\mathrm{JUV},t}^{\left[\mathrm{win}\right]}$ in Appendix S1: Section S1.3.1). Age‐ratios from spring counts on Wrangel Island $\left({\mathrm{AR}}_{\mathrm{SPR}}\right)$ are based on aging 4000–5000 birds in the field (Baranyuk & Kraege, 2017), and hence, we used a sample size of 4000 total birds for the likelihood. Some juvenile snow geese have already molted most of their body feathers by the time they turn 1‐year old and they can be difficult to reliably distinguish from adults. In addition, although 1‐year‐old Wrangel Island snow geese are believed to visit breeding sites in the spring (Baranyuk & Kraege, 2017), migration from wintering sites directly to the molting sites (observed in greater snow geese; Reed et al., ²⁰⁰³) could bias the proportion of juveniles in spring at the breeding sites low. To allow flexibility in the spring age ratios, we used a beta‐binomial observation model for spring age‐ratio data:

$$n_{\mathrm{JUV},t}^{\left[\mathrm{spr}\right]}~\text{Binomial}\left({\mathrm{\mu }}_{\mathrm{JUV},t}^{\left[\mathrm{spr}\right]},4000\right),$$ (10)

$${\mathrm{\mu }}_{\mathrm{JUV},t}^{\left[\mathrm{spr}\right]}~\text{Beta}\left(\left(\frac{N_{\mathrm{JUV},\mathrm{WRI},t}}{N_{\mathrm{WRI},t}}\right){\mathrm{\tau }}^{\left[\mathrm{spr}\right]},\left(1-\frac{N_{\mathrm{JUV},\mathrm{WRI},t}}{N_{\mathrm{WRI},t}}\right){\mathrm{\tau }}^{\left[\mathrm{spr}\right]}\right)$$ (11)

where $n_{\mathrm{JUV},t}^{\left[\mathrm{spr}\right]}$ is the observed number of second‐year birds within the sample of 4000 birds and ${\mathrm{\tau }}^{\left[\mathrm{spr}\right]}$ is the precision parameter [prior: ${\mathrm{\tau }}^{\left[\mathrm{spr}\right]}~\text{Half Cauchy}\left(0,\mathrm{SD}=2.5\right)$].

Following the delineations in 2.3, we considered age‐ratios among snow geese harvested in different parts of the non‐breeding range to separately represent per capita productivity in Wrangel Island and Western Arctic. We accounted for differences in vulnerability to harvest between juveniles and adults and modeled age‐ratios in harvest as follows:

$$n_{\mathrm{JUV},r,t}^{\left[\text{tails}\right]}~\text{Binomial}\left({\mathrm{\theta }}_{r,t}^{\left[\text{HARVEST}\right]},\left(n_{\mathrm{JUV},r,t}^{\left[\text{tails}\right]}+n_{\mathrm{AD},r,t}^{\left[\text{tails}\right]}\right)\right)$$ (12)

where $n_{\mathrm{JUV},r,t}^{\left[\text{tails}\right]}$ and $n_{\mathrm{JUV},r,t}^{\left[\text{tails}\right]}$ are the number of juveniles and adults in harvest data, and ${\mathrm{\theta }}_{r,t}^{\left[\text{HARVEST}\right]}$ is the estimated proportion of juveniles in the harvest at the onset of hunting season in September in each region (for the calculation of ${\mathrm{\theta }}_{r,t}^{\left[\text{HARVEST}\right]}$, see Appendix S1: Section S1.3.2).

To assess predictions about the drivers of per capita productivity, we modeled region‐specific productivity $\left({\mathrm{\gamma }}_{r,t}\right)$ as a function of annual timing of snow melt $\left({\text{SNOW}}_{r,t}\right)$, number of days of extreme weather conditions around hatch $\left({\mathrm{EXT}}_{r,t}\right)$, non‐breeding weather $\left({\mathrm{ONI}}_{r,t}\right)$, and population size such that

$$\log \left({\mathrm{\gamma }}_{r,t}\right)={\mathrm{\alpha }}_r+{\mathrm{\beta }}_r^{\left[\text{SNOW}\right]}{\text{SNOW}}_{r,t}+{\mathrm{\beta }}_r^{\left[\mathrm{EXT}\right]}{\mathrm{EXT}}_{r,t}+{\mathrm{\beta }}_r^{\left[\mathrm{ONI}\right]}{\mathrm{ONI}}_t+{\mathrm{\beta }}_r^{\left[\text{DENS}\right]}\log \left(N_{r,t}\right)+{\mathrm{\epsilon }}_{r,t},$$ (13)

where ${\mathrm{\alpha }}_r$ and are region‐specific intercepts and slopes, respectively [priors for all $\text{Normal}\left(0,\mathrm{\sigma }=1\right)$], and ${\mathrm{\epsilon }}_r$ are residual errors [priors: ${\mathrm{\epsilon }}_r~\text{Normal}\left(0,{\mathrm{\sigma }}_r\right),{\mathrm{\sigma }}_r~\text{Half Cauchy}\left(0,\mathrm{SD}=2.5\right)$].

Model implementation

We fit the IPM in a Bayesian framework using NIMBLE (de Valpine et al., ²⁰¹⁷), gplite (Piironen, 2021), and related packages in R version 4.4.1 (R Core Team, ²⁰²⁴). We sampled three MCMC chains of 550,000 iterations, removed the first 50,000 iterations from each chain as burn‐in, and thinned the chains by retaining every 50th sample. We assessed convergence of the MCMC sampling using R‐hat statistics and Gelman–Rubin diagnostics (Brooks & Gelman, 1998). R‐hat values for all parameters were <1.1, and visual inspection of chains indicated convergence and good mixing of chains. We assessed the support for covariate relationships based on whether 95% of the posterior distribution for each regression coefficient was above or below zero and report the mean 95% credible interval (CrI).

Post hoc analysis

To retrospectively assess contributions of various demographic rates to population change, we used a transient life‐table response experiment (tLTRE, Koons et al., ^{2016, 2017}) to decompose the changes in population growth rate between successive years to the realized temporal variation in demographic rates, immigration, and age structure (juveniles/adults) of the population. We calculated the difference in the realized population growth rate $\left({\mathrm{\lambda }}_t\right)$ between years t and t + 1 as a function of demographic rates, immigration, and population structure as follows:

$$\mathrm{\Delta }\left({\mathrm{\lambda }}_t\right)\approx \sum _i\left({\mathrm{\theta }}_{i,t+1}-{\mathrm{\theta }}_{i,t}\right){\left.\frac{\mathrm{\delta }{\mathrm{\lambda }}_t}{\mathrm{\delta }{\mathrm{\theta }}_{i,t}}\right|}_{\overline{{\mathrm{\theta }}_i}},$$ (14)

where θ is the vector with the annual demographic rates and ${\left.\frac{\mathrm{\delta }{\mathrm{\lambda }}_t}{\mathrm{\delta }{\mathrm{\theta }}_{i,t}}\right|}_{\overline{{\mathrm{\theta }}_i}}$ are the sensitivities of the population growth rates toward different demographic rates, calculated at the means between successive years.

RESULTS

Population estimates and contributions of demographic rates to population change

The population size (total number of adults and juveniles in the spring before breeding, rounded to a thousand individuals) on Wrangel Island varied around 100,000 birds without a trend before 2010, after which it rapidly increased to an estimated 750,000 snow geese in 2022 (95% CrI: 735,000, 765,000; Figure 3A). The Western Arctic population increased throughout the study period from 146,000 birds (95% CrI: 122,000, 171,000) in 1970 to 1,638,000 birds (95% CrI: 1,179,000, 2,039,000; Figure 3A) in 2022. Altogether, the estimated number of snow geese in the Pacific Flyway in 2022 was 2,389,000 birds (95% CrI: 2,029,000, 2,789,000). Mean estimates for adult survival increased from ca. 0.70–0.80 in the 1970s to 0.85–0.90 in the 2010s in both subpopulations (Figure 4). Year‐to‐year variation in juvenile survival was large, but the estimates increased until the 1990s, after which they slightly decreased on Wrangel Island and stabilized in the Western Arctic (Figure 4). Increases in survival were simultaneous with decreases in kill rates, whereas natural mortality rates did not show clear trends during the study period (Figure 4). Annual variation in per capita productivity was high in both regions, with a decreasing trend in the Western Arctic and no discernible trend on Wrangel Island (Figure 5A,B). Immigration to Wrangel Island was relatively low, except for the high peak in 2019 and 2020 when the population increased rapidly (Figures 3C and 5C). Immigration estimates for the Western Arctic subpopulation were low and relatively stable throughout the study period (Figures 3D and 5D). However, the immigration estimates should be interpreted cautiously (see Immigration ).

FIGURE 3.

Spring population size and population growth rates of snow geese in the Wrangel Island and Western Arctic subpopulations. Subplots (A) and (B) show the posterior mean (black line) and the 95% credible interval (CrI, shaded gray), along with region‐specific population survey data (dots). Subplots (C) and (D) show the estimated mean population sizes, split into three categories: survived adults (birds that survived in the region from the previous year; purple bar), second‐year birds (birds hatched in the region the previous summer and survived over winter; turquoise bar), and immigrants (birds older than second‐year, hatched outside the region and moved to the region after the previous spring; yellow bar). In subplot (B), summer counts (blue dots) represent combined estimates from the Western Arctic Photo Survey (Kerbes et al., 2014) and the Arctic Coastal Plain Survey (Amundson et al., 2019). Lincoln estimates for Western Arctic birds represent the number of adult birds during summer after breeding (at the time of banding). Winter counts represent the numbers of all Western Arctic (adults and juveniles) birds in winter, whereas summer counts represent the numbers of these birds during breeding. Subplots (E) and (F) show the realized population growth rates (mean and 95% CrI) for Wrangel Island and Western Arctic subpopulations, respectively.

FIGURE 4.

Annual survival, harvest, and natural mortality probabilities of Wrangel Island (left column) and the Western Arctic (right column) snow geese during 1970–2022. The lines and shaded areas denote means and 95% credible intervals of the integrated population model (IPM) estimates, respectively.

FIGURE 5.

Annual per capita productivity and number of immigrants of Wrangel Island and Western Arctic snow geese during 1970–2022. In subplots (A) and (B), the dots and whiskers denote the mean and 95% credible interval of the integrated population model (IPM) estimates, respectively, and the dashed lines illustrate the temporal trend in per capita productivity. In subplots (C) and (D), the line and the shaded areas denote the mean and 95% credible interval of the IPM estimates, respectively.

Population growth rate of both subpopulations varied substantially (between 0.25 and 1.5) during the study period (Figure 3E,F). When change in the realized population growth rate was large between consecutive years $\left({\mathrm{\Delta \lambda }}_t>\left|0.15\right|\right)$, per capita productivity was the most important contributor in 17 out of 21 cases (81%) in both Wrangel Island and Western Arctic. Changes in immigration contributed substantially to the rapid population growth in recent years in the Wrangel Island but were less important in the Western Arctic (Figure 6C,D). Influence of adult and juvenile mortality to the short‐term variation in population growth rate was less important in both subpopulations, but the overall increases in juvenile and adult survival during the study likely contributed to the longer term, positive trends in population growth. Population structure had negligible impact on population dynamics in both subpopulations.

FIGURE 6.

Overall contributions of demographic rates to changes in realized population growth rate between successive years, derived using transient life‐table response experiment (tLTRE). Subplots (A) and (B) show the difference in realized population growth rate between successive years. Subplots (C) and (D) show the overall contributions of changes in demographic rates to the changes in the population growth rate. The annual sum of these contributions equals approximately the difference in the realized population growth rates.

Effects of environmental covariates on per capita productivity and juvenile mortality

The per capita productivity of Wrangel Island snow geese was negatively associated with timing of snow melt (SNOW_RU,t ; −0.63; 95% CrI: −1.1, −0.16; Figure 7), but not with extreme weather conditions during early brood rearing (EXT_RU,t ; −0.44; 95% CrI: −0.90, 0.03) or the ONI index (ONI _t ; −0.15; 95% CrI: −0.61, 0.32). For Western Arctic birds, per capita productivity was negatively associated with timing of snow melt (SNOW_WA,t ; mean −0.27; 95% CrI: −0.45, −0.09; Figure 7), positively associated with extreme weather conditions during early brood rearing (EXT_WA,t ; 0.19; 95% CrI: 0.02, 0.38), and positively associated with the ONI index (ONI _t ; 0.22; 95% CrI: 0.05, 0.40; Figure 7). Warm early fall temperature at the breeding sites decreased juvenile natural mortality of Wrangel Island birds (TEMP_WRI,t ; −0.37; 95% CrI: −0.66, −0.12) but not for the Western Arctic birds (TEMP_WA,t ; 0.17; 95% CrI: −0.15, 0.53).

FIGURE 7.

The relationship between per capita productivity of snow geese and the timing of snow melt (SNOW _r,t ) and weather conditions during winter and early spring (ONI _t ) in the two study regions. Dots denote the mean integrated population model (IPM) estimate of per capita productivity, and error bars denote the associated 95% credible intervals. The lines and shaded areas denote the mean 95% of the fitted log‐linear model, respectively. Note that vertical axes are truncated to better illustrate trends, and hence, all error bars are not fully visible. ONI, Oceanic Niño Index.

Density dependence of productivity and natural mortality

We found evidence of density‐dependent regulation of per capita productivity and adult natural mortality in the Western Arctic but not in Wrangel Island. Per capita productivity was negatively associated with the population size in the Western Arctic (−0.52; 95% CrI: −0.80, −0.27, Appendix S4: Figure S6). Adult natural mortality increased with increased population size in the Western Arctic (0.67; 95% CrI: 0.21, 1.31) but not in Wrangel Island (0.36; 95% CrI: −0.13, 1.04; Figure 7). There was no association between the natural mortality hazard rate and the population size for juveniles from Wrangel Island (0.28; 95% CrI: −0.25, 0.86), but there was a negative association in the Western Arctic (−0.76; 95% CrI: −1.37, −0.21).

Effect of hunter availability and harvest regulations on harvest mortality

Harvest mortality hazard rates were positively associated with hunter availability for both age classes within both subpopulations (Figure 8). Regression slopes were positive for all groups and greater for Western Arctic than Wrangel Island, with little difference between age classes within subpopulations. For Wrangel Island snow geese, the slopes were 0.50 (95% CrI: 0.34, 0.66) for juveniles and 0.53 (95% CrI: 0.39, 0.66) for adults. In the Western Arctic, slopes were 0.60 (95% CrI: 0.45, 0.76) for juveniles and 0.58 (95% CrI: 0.47, 0.70) for adults. Harvest regulations did not affect juvenile harvest mortality hazard rates (Wrangel Island 0.02; 95% CrI: −0.11, 0.14; Western Arctic 0.003; 95% CrI: −0.15, 0.10), but they weakly affected adult harvest mortality hazard rates (Wrangel Island −0.17; 95% CrI: −0.27, −0.07; Western Arctic −0.10; 95% CrI: −0.20, −0.002).

FIGURE 8.

The relationship between hunter availability and harvest mortality hazard rates (h _κ, on top row) and log‐population size and adult natural mortality hazards rates (h _η, on bottom row). Dots and whiskers denote the mean and 95% credible interval for the estimated cause‐specific mortality hazard rates, and the line and shaded area denote the mean and 95% credible interval of the model fit, respectively. In all subplots, turquoise denotes juveniles and purple adults.

DISCUSSION

We combined several data sources into one modeling framework using an integrated population model to comprehensively estimate snow goose demographic rates, their contribution to population change, and their environmental and anthropogenic drivers in the Pacific Flyway of Western North America. Changes in annual and short‐term population growth rates for both Wrangel Island and Western Arctic snow geese were primarily driven by changes in per capita productivity, and immigration also contributed substantially in some years to the rapid increase in the Wrangel Island subpopulation in recent decades. Per capita productivity was positively influenced by early snowmelt in both regions, and warm, rainy weather during the non‐breeding season (i.e., high ONI‐index values) was associated with high per capita productivity in the Western Arctic. Despite this, per capita productivity showed an overall decreasing trend in the Western Arctic likely due to density‐dependent regulation. Similarly, adult natural mortality was regulated by density dependency in the Western Arctic but not in Wrangel Island. Kill rates notably decreased during the study period, mostly driven by a decline in the number of hunters in the Pacific Flyway. These findings provide important perspective about the interplay between environmental variation and harvest dynamics contributing to the population dynamics of a long‐lived, migratory species.

Harvest mortality on long‐lived species

Snow goose kill rates in the Pacific Flyway were relatively high before the 1990s but have decreased and stabilized to lower levels during the past 20 years (Figure 4). We note that our kill rate estimates depend on the prior information used to inform band reporting and crippling rates (see Survival sub‐model ) and are vulnerable to biases in these parameters. Hence, these estimates should be interpreted with some caution, especially prior to the 1990s after which band reporting rates increased due to improvements in reporting methodology and were better estimated based on more reward band studies (Arnold et al., 2020). In addition, banding efforts for juveniles altered between the breeding colonies among Western Arctic (more juveniles were banded in the colonies located in Alaska than in the colonies located in Western Canadian Arctic). This might have created inconsistencies in juvenile survival and mortality estimates and their covariate relationships. Given the decreasing number of hunters (Appendix S4: Figure S1), the strong relationship between hunter numbers and harvest mortality (Figure 8), and the low impact of harvest regulation on harvest mortality (see Effect of hunter availability and harvest regulations on harvest mortality ) in recent years, it is unlikely that kill rates will notably increase under current harvest regimes and when the population is at or above its current size. Harvest mortality plays a minor role currently in the population dynamics of Pacific Flyway snow geese, but its role was more important in earlier years when hunters were more numerous, the snow goose population was smaller, and kill rates were substantially higher than currently (Figures 4 and 6). We note that the impact of harvest regulations on harvest mortality can be complex and might be imperfectly described by the variable used here (i.e., season length × daily bag limit). For example, harvest later in the season might have a greater effect on population development (LeTourneux et al., 2024). Previous experiences on the influence of introduction of later spring hunting seasons on population growth of other abundant goose populations in North America are divergent (e.g., Alisauskas et al., ²⁰¹¹; Lefebvre et al., ²⁰¹⁷). A prospective analysis regarding the potential effect of later spring hunting would likely benefit managers regarding future population management and harvest regulations.

The small contributions of harvest and natural mortality to short‐term variations in population change (in comparison with productivity) are interesting from a life‐history perspective; snow geese are a long‐lived species, and the population dynamics of such species are typically most sensitive to changes in adult survival (e.g., Sæther & Bakke, ²⁰⁰⁰). However, temporal variation in productivity is typically higher, which manifests as a stronger influence on population dynamics (e.g., Manlik et al., ²⁰¹⁶). Our results demonstrate the latter and add to recent studies showing the importance of short‐term changes in population growth rate also in long‐lived species. However, long‐term trends in survival should not be underrated. Adult survival in both subpopulations steadily increased over the study period, which inevitably contributed to the long‐term population growth. This was likely the case especially in the Western Arctic, where the size of the population increased during the study period despite a decreasing trend in productivity. From a conservation viewpoint, influencing an increase or decrease in productivity (to affect short‐term population changes), especially for broadly dispersed, Arctic‐nesting birds, can be challenging. Altering mortality through harvest management is a more direct, tangible action than altering reproductive output which, as in our case, appears to be mainly affected by large‐scale weather patterns and density‐dependent effects. Hence, future research on the extent to which conservation actions can meaningfully impact productivity through scenarios of changing environmental variation and across the full annual cycle would be valuable.

Immigration

Immigration appeared to be an important mechanism of recent, rapid increase in Wrangel Island snow geese (Figure 3), which aligns with some previous studies reporting high immigration to new and rapidly growing snow goose colonies on the North Slope of Alaska (Burgess et al., 2017; Johnson, ¹⁹⁹⁵; Pearce et al., ²⁰²²). However, we cautiously interpret our immigration estimates because they were modeled as latent parameters, which can absorb inconsistencies in datasets in IPMs and lead to overestimation of its contribution to population change (Paquet et al., 2021). This especially applies to the Western Arctic subpopulation, where data on population size was imprecise in many years. However, the rapid population increase in Wrangel Island cannot be explained without substantial immigration in some years. Immigration estimates were highest around 2020, when adult survival and productivity were also high. Adequate data on both demography and population size were available at the time, and we believe that the high immigration to Wrangel Island at the time was not caused by biases in other datasets or in the model.

The most likely source of immigrants is the midcontinent snow goose population, although movement between the Pacific and midcontinent flyways is thought to be small (Alisauskas et al., 2022). Generally, immigration is more important for the dynamics of small study populations, and its influence decreases with larger spatial scales (Millon et al., 2019). Previous studies with geese have documented that immigration and emigration can be important mechanisms to drive changes in the size of local breeding colonies (Alisauskas et al., 2022; Weegman et al., ²⁰²²) or separate wintering flocks (e.g., Weegman et al., ²⁰¹⁶). Our results demonstrate that immigration can, at least in some years, substantially contribute to population change of subpopulations (birds with breeding and wintering sites separate from other subpopulations), which are used as management units in waterfowl conservation and management in North America and Europe. In other species that have lower site fidelity than geese, immigration and emigration are known to have a greater contribution to population change (Millon et al., 2019). Immigration can also compensate for losses caused by mortality, which can complicate detecting the influence of harvest mortality on populations (Hörnell‐Willebrand et al., 2014). Hence, future monitoring schemes could be planned to facilitate more spatially explicit estimation of demographic rates and population size to aid conservation planning and the ability to monitor metapopulation dynamics of interest.

Environmental effects and density dependence

The environmental factors affecting per capita productivity (timing of snow melt, precipitation, and temperature during winter) and juvenile survival (early fall temperature at the breeding sites) are affected by climate change and will likely continue to benefit the productivity of snow geese in the Pacific Flyway. Warmer springs are advancing snow melt in Arctic breeding areas, enhancing breeding propensity and success (e.g., Boom et al., ²⁰²³), although these climatic changes may increase the phenological mismatch between peaks in hatching date and forage quality (Ross et al., 2017). Mild and moist winters will likely increase in frequency due to climate change and are likely to decrease energy expenditure during winter, providing more feeding opportunities and even shortening migration distances, all of which can increase breeding performance. In addition, increasing fall temperatures due to climate change will allow birds to depart later from the breeding sites, which provides more feeding opportunities for juveniles and decreases their mortality during their first fall migration. By benefiting from climate change in many cases, geese differ from most other Arctic‐nesting birds, whose productivity and survival are negatively impacted by a warming climate due to reductions in food availability (caused by phenological mismatches), northwards spread of pathogens and parasites, and increases in predation (Kubelka et al., 2022). Although the warming climate already favors goose reproduction through these conditions, per capita productivity has shown a declining trend in the Western Arctic, suggesting that density‐dependent regulation, potentially together with unknown drivers, is counteracting the positive effects of a changing climate. This phenomenon has been observed in other Arctic‐nesting goose populations (e.g., Baldwin et al., ²⁰²²; Layton‐Matthews et al., ²⁰¹⁹), indicating that density‐dependent processes are decoupling the well‐established relationship between spring phenology and productivity in Arctic‐nesting geese. Furthermore, density dependence can also act among multiple species, and sympatric species with relatively similar life histories can share drivers of demography (e.g., Weegman et al., ²⁰²⁴). In the future, our approach could be extended by incorporating several sympatric species into one modeling framework, which could provide an understanding of the interactions between species and their habitats.

Future directions

We had multiple datasets describing productivity, survival, and population size across the breeding colonies of the Pacific Flyway snow geese for our analysis. However, we anticipate less data will be collected on these birds in future years. Planning cost‐efficient data‐collection schemes for the future would likely benefit from an assessment of the relative contributions of different datasets toward the estimation of population size, demographic rates, and mechanisms of population change. Practitioners across the world are repeatedly faced by the same challenges (e.g., Johnson et al., ²⁰²²), and such analysis would likely benefit data collection for various taxa. In addition, as per capita productivity was the primary demographic mechanism causing variation in population growth rates, scenarios that consider the linkages between productivity and population growth rate will be useful for equipping federal, provincial, and state agencies responsible for the management of this species with the most robust information to make informed decisions.

CONCLUSIONS

Our study demonstrated how multiple, diverse, partially sparse, and long‐term demographic datasets can be integrated into a single hierarchical model to understand the importance of several demographic rates on population change, as well as their environmental and anthropogenic drivers. We found that even for a long‐lived, commonly harvested game species, environmental conditions and density‐dependent effects on productivity played a more important role than harvest toward annual variations in population growth rate. Increases in survival likely contributed to long‐term population change. Our study magnifies the importance of conservation plans that consider these environmental drivers, although this complicates direct management of such populations through harvest regulation. Additionally, our results showed that a warming climate can have both direct and cross‐seasonal effects on demography, highlighting the need to holistically consider environmental conditions when predicting how populations might respond to large‐scale environmental changes. Practitioners who are equipped with multiple datasets that were asynchronously collected in space and time could adapt our framework for their ecological questions to robustly assess population dynamics and customize conservation plans.

AUTHOR CONTRIBUTIONS

Mitch D. Weegman, Jeffrey M. Knetter, Kyle A. Spragens, Joshua L. Dooley, and Antti Piironen conceived the ideas and designed the methodology. Vijay Patil, Eric T. Reed, Kyle A. Spragens, Megan Ross, and Todd A. Sanders collected the data. Antti Piironen analyzed the data. Mitch D. Weegman and Daniel Gibson supervised. Mitch D. Weegman secured funding for the work; Antti Piironen led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Appendix S2.

Appendix S3.

Appendix S4.

Piironen, Antti , Knetter Jeffrey M., Spragens Kyle A., Dooley Joshua L., Patil Vijay, Reed Eric T., Ross Megan, et al. 2025. “Environmental Drivers of Productivity Explain Population Patterns of an Arctic‐Nesting Bird across a Half‐Century.” Ecological Applications 35(5): e70067. 10.1002/eap.70067

DATA AVAILABILITY STATEMENT

Data and code (Piironen, 2025) are available on Figshare at https://doi.org/10.6084/m9.figshare.28077731.v3.