Harvest sustainability assessments need rethinking under climate change: A ringed seal case study from Svalbard, Norway

Abstract

Throughout the Arctic, ice‐affiliated marine mammals constitute local subsistence resources but detrimental effects of declines in their sea ice habitats create a need for harvest sustainability assessments in light of climate change. At the same time, empirical data required for thorough population analysis of these species are often sparse at best, as illustrated by the focal species in this study, ringed seals in Svalbard: the last population survey took place two decades ago (2002–2003), demographic data are limited to age, sex, and reproductive status of a small subset of shot individuals, and harvest reporting is patchy and incomplete. Data sparsity is one of the main reasons why potential biological removal (PBR) became a commonly used tool for assessing sustainability of marine mammal harvests. Herein, we calculated PBR for Svalbard ringed seals using both recommended default parameters and population‐specific parameters obtained from an integrated population model (IPM). PBR estimates were highly uncertain, suggesting the number of sustainably harvestable individuals could lie anywhere between 0 and 91, with a substantial chance of any harvest being unsustainable under current environmental conditions and trends. Subsequent population viability analyses (PVAs) further confirmed that the current harvest was likely unsustainable, even in a scenario in which sea ice conditions would not deteriorate (and therefore lower pup survival) further. However, uncertainty in population projections was high, and forecasts thus not ideal for formulating management advice. Better forecasts will require more frequent population surveys and obtaining more knowledge regarding the links between vital rates and environmental conditions, both of which may be facilitated by the adoption of novel technology (e.g., drone monitoring, genetic studies). The modeling framework created in this study can be readily updated with new data as they become available, and can serve as a tool for adaptive management of this and other marine mammal populations.

INTRODUCTION

During the last four decades the Arctic has warmed at a rate approximately four times faster than the global average and summer sea surface temperatures have increased by about 0.5°C per decade (Meredith et al., 2019; Rantanen et al., 2022; Taylor et al., 2022). Satellite imagery suggests that sea ice losses in the Arctic amount to 7.6 × 10¹² t of ice (Slater et al., 2021). Sea ice thickness, extent, and seasonal duration have all declined precipitously, inducing changes such as increased freshwater inputs in marine systems, increased stratification of the water column, changes in underwater light regimes, general circulation and transport of sediments and nutrients in Arctic marine ecosystems (Walch et al., 2022). Tidewater glaciers, another important marine cryosphere element, are also melting and retracting at unprecedented rates (Geyman et al., 2022). The potential for negative consequences of these impacts of global warming on marine mammals has been predicted since the earliest signs of change in the region (Moore, 2002; Stirling & Derocher, 1993; Tynan & DeMaster, 1997), with particular concern being raised for Arctic endemic species (e.g., ACIA, 2005; Johnston et al., 2005; Moore & Huntington, 2008; Simmonds & Isaac, 2007), all of which are strongly ice-affiliated. Numerous consequences of climate change have already been documented for marine mammals including shifts in species distributions (Hamilton et al., 2015, 2017; Higdon & Ferguson, 2009; Lone et al., 2018; Rode et al., 2015), changes in trophic relationships (e.g., Hamilton et al., 2017; Watt et al., 2016; Yurkowski et al., 2018), and increased disease risk (e.g., Van Wormer et al., 2019). Concomitantly, levels of human activity, including shipping, tourism, commercial fishing, and oil and gas exploration and production, have increased in the Arctic because less sea ice means reduced logistical challenges for these industries (Meredith et al., 2019; Reeves et al., 2014).

The combination of diminishing sea ice and increasing human activity thus creates a complex management situation for ice-affiliated Arctic species, many of which are also important local subsistence resources. The status and trends of polar bear populations are reasonably well known, and harvests are thought to be sustainable. Arctic endemic cetaceans have recently received considerable research attention regarding population sizes, and harvests are overseen by the International Whaling Commission (see Kovacs et al., 2021 for a review). However, the status of seal populations in the Arctic is poorly documented, and harvests are undertaken without knowledge of current trends, despite concerns regarding detrimental climate change impacts on these animals and their habitats (Kovacs et al., 2021; Laidre et al., 2015).

The Arctic ringed seal (Pusa hispida hispida) is a small, northern true seal that is harvested throughout most of its circumpolar Arctic range. Like other Arctic endemic seals (and whales), ringed seals live in tight association with sea ice. They give birth on land-fast ice (and in some few areas drifting pack-ice), molt on the ice in the spring, and rest on it throughout the year (Heide-Jørgensen & Lydersen, 1998). Their diet also contains substantial proportions of sympagic (ice-associated) prey in most regions (Bengtsson et al., 2020). Because of the unique breeding ecology of this species, which involves giving birth in snow caves on top of the sea ice, and their long nursing periods (6 weeks), successful ringed seal reproduction requires sea ice that forms by late winter (so that enough snow is accumulated in time for birthing in March–April) and that stays stable until late in the spring (Lydersen & Kovacs, 1999). This extreme dependence on sea ice makes this species particularly vulnerable (VU) to the dramatic declines in sea ice that have occurred, and continue to occur, throughout the Arctic (Kovacs et al., 2011). This sensitivity to climate warming led the United States to designate the ringed seal as a threatened species under the US Endangered Species act in 2012 (NOAA, 2021) and Norway followed suit in 2015, listing ringed seals as VU on the National Red List (). Despite the obvious conservation concerns, ringed seals are hunted throughout the Arctic without quotas on the numbers taken (Laidre et al., 2015). In many regions harvest statistics are not even recorded, even though harvests are substantial in some areas (e.g., Greenland harvests have been in excess of 50,000 ringed seals, on average per year, in the last two decades (), with no effort expended on regional population assessments).

In the Svalbard Archipelago ringed seal habitat has declined markedly in the last two decades due to increased air and water temperatures (Årthun et al., 2021; Beszczynska-Möller et al., 2012; Isaksen et al., 2016; Lind & Ingvaldsen, 2012). An unpredicted “collapse” in sea ice occurred in 2006, and in most years since there has been little winter/spring sea ice coverage throughout much of the Archipelago, particularly in the west. This has resulted in dramatic reductions in suitable breeding habitat for ringed seals and also marked changes in foraging behavior, territory size, and space use (Hamilton et al., 2015, 2016; Hamilton, Kovacs, & Lydersen, 2019; Hamilton, Vacquié-Garcia, et al., 2019; Lydersen et al., 2017). Consequently, the current population trend of ringed seals in Svalbard is suspected to be negative, but this remains uncertain because the most recent ringed seal population survey was conducted over two decades ago in 2002 (Krafft, Kovacs, Andersen, et al., 2006). Despite the lack of abundance and trend data, local sport hunting is permitted in several areas within the archipelago, the primary area being Isfjorden, the fjord closest to the largest settlement on Svalbard, Longyearbyen. There are seasonal closures of the hunt (to protect seals during the birthing/nursing period and early summer when shot seals would sink), but no quotas are set for the overall harvest. Biodiversity, hunting and wildlife protection legislation in Norway all support sustainable resource use as a principal goal of management, and Norway has set a goal for Svalbard to be one of the world's best managed wilderness areas (Meld. St.22 [2008–2009], Meld. St.32 [2015–2016]). Yet, it is not currently known whether the ringed seal sport hunt in Svalbard is sustainable in light of climate change impacts on the population. The purpose of the present study was to model population dynamics and harvest quotas for this data-poor Arctic seal population in the context of climate warming. To do this, we integrated all available data streams—survey data (Krafft, Kovacs, Andersen, et al., 2006), demographic data collected from harvested seals (Andersen et al., 2020; Krafft, Kovacs, Frie, et al., 2006; Lydersen & Gjertz, 1987), harvest statistics, and time series of sea ice availability, as well as expert knowledge and information from studies from other regions—within a framework combining state-of-the-art integrated population models (IPMs; Schaub & Kéry, 2021) with traditional potential biological removal (PBR) approaches (Nelson et al., 2019; Wade, 1998) and exploratory population viability analyses (PVAs; Morris & Doak, 2002; Reed et al., 2002).

METHODS

Determining levels of sustainable harvests of Arctic seals is challenging because of a general lack of data necessary to assess population dynamics and drivers. We therefore opted to use a framework that maximizes information gain by jointly analyzing multiple types of observational data: long-term demographic data from harvested seals, an abundance survey from 2002/2003, and a time series of sea ice availability from satellite imagery. The core of our framework is an IPM, which mechanistically links population size and (age) structure over time to underlying vital rates (survival, reproduction, etc.) and pulls together information from multiple data types (Schaub & Kéry, 2021). Among the IPM's outputs are estimates of population size and growth rates over time, which can be used for determining the numbers of individuals that can be harvested sustainably via PBR, a method traditionally used for harvest sustainability assessments of marine mammals (Wade, 1998). PBR has recently been applied to the assessment of subsistence harvest of ringed seals in Alaska by Nelson et al. (2019). IPMs also extend naturally into PVAs, which can forecast population dynamics under different harvest and climate change scenarios (Morris & Doak, 2002; Reed et al., 2002).

Below, we first describe the collection and processing of available observational data. The subsequent sections detail the age-structured population model and data likelihoods that make up our ringed seal IPM and the model's implementation in a Bayesian framework, including quantification of among-year variation and sea ice effects and the use of informative priors. Finally, we describe the coupling of the IPM to PBR and PVA for assessing harvest sustainability across a range of scenarios.

Data collection

Seal abundance, demographic, and harvest data

Abundance of ringed seals in Spitsbergen, Svalbard, was last surveyed in June 2002–2003 using aerial digital photographic surveys (see Krafft, Kovacs, Andersen, et al., 2006 for details). The surveys included ice-covered areas in 18 fjords and bays (7 of which occur within the focal area for the current study—Isfjorden). Krafft, Kovacs, Andersen, et al. (2006) estimated the minimum number of seals in each location from manual counts of aerial photographic images, and derived estimates of population size, including uncertainty, by correcting for haul-out probability (estimated from behavior of radio-tagged seals in Carlens et al., 2006).

Relevant data on age structure and reproductive parameters have been collected for ringed seals harvested on Svalbard during three time periods—two scientific hunts in 1981–1982 (Lydersen & Gjertz, 1987) and 2002–2004 (Krafft, Kovacs, Frie, et al., 2006), and from the local sport hunt in Isfjorden, which has been documented since 2012 (the hunt is ongoing; Andersen et al., 2020). During these sampling periods, harvested seals were aged based on cementum layers in the canine teeth, and maturation status of females was determined based on the presence of follicles in the ovaries and/or signs of earlier pregnancies in the uterus (Andersen et al., 2020). Additionally, the number of females that had ovulated in the year of harvest (indicated by the presence of a corpus luteum in an ovary) was recorded for each study period. In the most recent period (2012 onward), pregnancy rates were also assessed based on the presence of a fetus for females shot between August and October (Andersen et al., 2020). While there may be segregation between ringed seals of different sexes and ages during breeding seasons with abundant sea ice (Krafft et al., 2007), the animals are protected during that period and most harvest takes place in the autumn on open water when individuals are distributed more randomly. We therefore do not expect substantial demographic biases in the sample of harvested individuals.

Harvest statistics for ringed seals in Svalbard are incomplete, but records on total numbers taken were available for 2003–2011 from Svalbard Miljøstatistikk (Svalbard Environmental Statistics). In 2006 it became mandatory to report the numbers and species of seals hunted to the Governor of Svalbard, giving rise to a second time series of numbers harvested for 2006–2020. Additionally, we obtained records from individual hunts—including date, location, and number of ringed seals shot—that have been reported via the online platform iNatur () in recent years (2019–2021). There are some discrepancies in the reports from the three sources in some years, suggesting potential biases due to incomplete recording. For the purpose of this study, we assumed that the maximum number reported by any source represented the minimum number of individuals harvested.

Sea ice data

The Norwegian Polar Institute obtains daily sea ice data by processing ice observations available from the Norwegian Ice Service branch of the Norwegian Meteorological Institute (Isaksen et al., 2016). The original data consist of 100 m × 100 m gridded sea ice concentrations (0%–100%), where coastal ice classified as 90%–100% represents land-fast sea ice. An annual coastal sea ice index was obtained by calculating the average percentage of a 5-km coastal mask of the Isfjorden area (Appendix S1: Figure S1) that was covered by land-fast sea ice during the period March–May for each year from 1988 to 2021 (Appendix S1: Figure S2).

IPM construction

Age-class structured population model

Based on the life history of ringed seals on Svalbard (Andersen et al., 2020), we developed a female-based population model with eight age classes: young-of-the-year (YOY); subadults aged 1, 2, 3, 4, and 5 years; and newly and previously mature adults (details below). We limited the model to the female segment of the population, assuming that the sex ratio is balanced and that the number of males does not limit female reproduction. Both assumptions are reasonable considering what is known about the study population, and ringed seals generally (Krafft, Kovacs, Frie, et al., 2006; Lydersen & Gjertz, 1987).

Based on ringed seal breeding biology and the timing of the 2002 population survey, we set the “census” of the population model (= the start of the annual interval) to coincide with the normal time for ovulation—late May (Figure 1A). The model therefore has a postbreeding census, meaning that reproductive individuals will have to survive to the next year before giving birth. Due to limited data, we assumed all vital rates except age-specific maturation rate ($$$ {pM}_{a,t} $$$) and pup survival ($$$ {S}_{pup,t} $$$) to be constant over time (see Table 1).

[IMAGE OMITTED. SEE PDF]

TABLE 1 Overview of vital rate parameters used in the population model (see Figure 1B) including definitions and whether they were estimated from data or assigned informative priors within the integrated population model.

Parameter	Definition	Information origin
$$$ {S}_0 $$$	Young-of-the-year survival from year $$$ t $$$ to $$$ t+1 $$$ $$$ =\exp \left(-\left({mH}_0+{mN}_0\right)\right) $$$a	Informative prior based on literatureb (no annual variation)
$$$ {S}_a $$$	Age $$$ a $$$ subadult survival from year $$$ t $$$ to $$$ t+1 $$$ $$$ =\exp \left(-\left({mH}_{a,t}+{mN}_{a,t}\right)\right) $$$a	Informative prior based on literatureb (no annual variation)
$$$ {S}_M $$$	Mature adult survival from year $$$ t $$$ to $$$ t+1 $$$ $$$ =\exp \left(-\left({mH}_{M,t}+{mN}_{M,t}\right)\right) $$$a	Informative prior based on literatureb (no annual variation)
$$$ {pM}_{a,t} $$$	Age $$$ a $$$ subadult maturation rate in year $$$ t $$$	Estimated from harvested animals, including annual variation
$$$ pO $$$	Ovulation rate of previously mature females in year $$$ t $$$	Estimated from harvested animals, without annual variation
$$$ pP $$$	Pregnancy rate of mature females in year $$$ t $$$	Estimated from harvested animals, without annual variation
$$$ {S}_{pup,t} $$$	Pup survival from birth to weaning in year $$$ t $$$	Informative prior for max. value based on literaturec and explicit linear relationship with annual Mar-May sea ice extent

At the time of census, female seals fall into four categories (Figure 1B). $$$ YOY $$$ are newly weaned pups, born earlier in the year, that may survive from year $$$ t $$$ to $$$ t+1 $$$ with a probability $$$ {S}_0 $$$. Subadults ($$$ Sub{A}_a $$$) are immature females prior to their first ovulation, meaning they have not produced a pup this year, and will not produce a pup next year. The ages $$$ a $$$ of subadult females range from 1 to 5 years. Since some females in this population reach maturity at 3 years of age, subadults aged 2 years or older may not only survive with an age-specific probability $$$ {S}_a $$$ but they also have a probability of maturing (= ovulating for the first time) just prior to next years' census ($$$ {pM}_{a+1,t+1} $$$). Note that the indexing of the maturation rate parameter is $$$ a+1 $$$ for age and $$$ t+1 $$$ for time; this is because by the time the next census time is reached (year $$$ t+1 $$$), the subadults will have become 1 year older ($$$ a+1 $$$). Since immature females in our data were typically no older than 5 years, females in the $$$ Sub{A}_5 $$$ age class are assumed to always reach sexual maturity by the next census. Newly matured adults ($$$ nMatA $$$) can give birth to a pup before next years' census, given that they mate and get pregnant (probability $$$ pP $$$), and survive (probability $$$ {S}_M $$$). Their pups can then be recruited into the population if they are female (probability = 0.5, assuming even sex ratio at birth) and survive from birth to weaning (probability $$$ {S}_{pup,t+ 1} $$$). Finally, the mature adult class ($$$ MatA $$$) consists of females that have been sexually mature for a year or longer. They have the same survival and reproductive rates as the newly matured females ($$$ nMatA $$$), except that their reproductive output is additionally conditional on ovulating in a given year $$$ t $$$ (probability $$$ pO $$$).

The population model's structure is presented as a projection matrix in Figure 1C, and definitions of all parameters can be found in Table 1 and Appendix S1: Table S1. As is common practice for IPMs, the population process model is implemented as a stochastic model, meaning that it accounts for the fact that there is some variation in the individual outcomes of demographic processes (demographic stochasticity, Caswell, 2002). Appendix S1: Table S2 presents definitions for all data nodes linked to the population model described below and Appendix S1: Figure S3 visualizes model structure in a DAG (directed acyclic graph).

Initial population size via abundance survey data likelihood

The population process model introduced above projects the numbers of individuals in each age class in year $$$ t+1 $$$ ($$$ {YOY}_{t+ 1}, Sub{A}_{a,t+ 1}, nMat{A}_{t+ 1}, Mat{A}_{t+ 1} $$$) from the numbers of individuals in each age class in year $$$ t $$$ ($$$ {YOY}_t, Sub{A}_{a,t}, nMat{A}_t, Mat{A}_t $$$). Thus, the first numbers resulting from the projection are for $$$ t=2 $$$, and the initial population size (numbers at $$$ t=1 $$$) need to be set a priori. Within the IPM, we use the population size estimates from Krafft, Kovacs, Andersen, et al. (2006) to initialize the population model under the assumption of a stable age distribution.

By simulating from the population means and SDs estimated by Krafft, Kovacs, Andersen, et al. (2006) for the seven locations representing the Isfjorden area, we obtained a probability distribution for the total number of seals in the Isfjorden area in 2002. We then used that distribution's mean ($$$ mean. estN $$$) and SD ($$$ sd. estN $$$) for setting an informative prior for the total population size (encompassing both sexes and all age classes) in the initial year 2002 ($$$ t=1 $$$): 1 $$$ {estN}_{2002}\sim \mathrm{Normal}\left( mean. estN, sd. estN\right) $$$ The model then uses this estimate to initialize the number of individuals in each age class ($$$ {YOY}_1, Sub{A}_{a, 1}, nMat{A}_1, Mat{A}_1 $$$) by multiplying it by 0.5 (to obtain females only) and with an assumed age-class distribution for 2002. For the $$$ YOY $$$ age class, for example, this looks like: 2 $$$ {YOY}_{2002}\sim \mathrm{Poisson}\left({estN}_{2002}\times 0.5\times {SAD}_1\right) $$$ $$$ {SAD}_1 $$$ is the proportion of the population that is in the $$$ YOY $$$ class at the start of the simulation (2002). The stage proportions ($$$ SAD $$$) were derived from the stable age distribution (= dominant right eigenvector) of the projection matrix in Figure 1C, at $$$ t=1 $$$ (sensu Caswell, 200120).

Maturation, ovulation, and pregnancy rates via reproductive data likelihoods

Data on age and reproductive status from female seals harvested during the three sampling periods (1981–1982, 2002–2004, 2012+) provided information on age- and year-specific maturation rates ($$$ {pM}_{a,t} $$$) and time-average ovulation and pregnancy rates ($$$ pO $$$ and $$$ pP $$$).

Age- and year-specific maturation rates were estimated using a Bernoulli likelihood for individual-level data on reproductive status ($$$ {RepStatus}_i $$$, where index $$$ i $$$ denotes the individual; 1 for mature, 0 for immature): 3 $$$ {RepStatus}_i\sim \mathrm{Bernoulli}\left({pM}_{a_i,{t}_i}\right) $$$ Ovulation and pregnancy rates were both modeled using binomial likelihoods. Ovulation rate describes the number of previously matured females showing signs of ovulation ($$$ NOvl $$$) as a subset of all previously matured females in the samples ($$$ NMat $$$): 4 $$$ N{Ovl}_p\sim \mathrm{Binomial}\left(N{Mat}_p, pO\right) $$$ The subscript $$$ p $$$ stands for sampling period, as we have three datapoints for this likelihood, one from each sampling period. We assumed that ovulation rates did not vary over time since preliminary analyses revealed no substantial differences across time periods (hence there is no time-index on $$$ pO $$$ above).

Pregnancy rate was also set to be constant across time, and the likelihood for the single sample we have from the fall (August–October) harvest data from the most recent sampling period was: 5 $$$ NPrg\sim \mathrm{Binomial}\left( NOvl, pP\right) $$$ where $$$ NPrg $$$ is the number of females with fetuses and $$$ NOvl $$$ is the number of reproductively active females (includes both newly matured females and previously mature females that have ovulated).

Harvest mortality and population structure via age-at-harvest data likelihood

Including likelihoods for data composed of numbers of harvested individuals into IPMs requires that the process model quantifies the latent true number of individuals harvested. If we assume that any female in age class $$$ \mathrm{a} $$$ at the census of year $$$ t $$$ ($$$ {N}_{a,t} $$$) may be harvested before the next census (year $$$ t+1 $$$) with probability $$$ {h}_a $$$, then $$$ {H}_{a,t} $$$, the number of age class $$$ a $$$ females harvested over the interval $$$ t\to t+1 $$$, can be modeled as: 6 $$$ {H}_{a,t}\sim \mathrm{Binomial}\left({N}_{a,t},{h}_a\right) $$$ $$$ {N}_{a,t} $$$ are equal to the classes in the population process model ($$$ {YOY}_t $$$, $$$ Sub{A}_{a,t} $$$, $$$ nMat{A}_t $$$, $$$ Mat{A}_t $$$). The harvest probability $$$ {h}_a $$$ can be expressed as a function of survival probability and harvest and natural mortality hazard rates (Ergon et al., 2018): 7 $$$ {h}_a=\left(1-{S}_a\right)\frac{mH_a}{\left({mH}_a+{mN}_a\right)} $$$ The (relative) abundances of different age classes among harvested individuals ($$$ {H}_{a,t} $$$) can be informed by age-at-harvest data, $$$ AC{aH}_{a,t} $$$ (= number of age class $$$ a $$$ females harvested during the interval June of year t to May of year t + 1), using Poisson or binomial likelihoods, similar to those used for counts of living individuals (Conn et al., 2008; chapter 17 in Schaub & Kéry, 2021). In the present IPM, we opted for a Poisson distribution since we needed to include a scaling factor to account for our age-at-harvest data originating from a larger area (beyond Isfjorden) than our focal study population: 8 $$$ AC{aH}_{a,t}\sim \mathrm{Poisson}\left({H}_{a,t}\times pAC{aH}_t\right) $$$ Here, $$$ {p\mathrm{AC} aH}_t $$$ represents the age-independent ratio of the true harvest in the Isfjorden area and the number of individuals included in the age-at-harvest data (summed over the range of all ages, 1:A) in a given year $$$ t $$$: $$$ pAC{aH}_t=\mathrm{sum}\left( AC{aH}_{1:A,t}\right)/\mathrm{sum}\left({H}_{1:A,t}\right) $$$. Note that the two mature age classes could not be distinguished in the age-at-harvest data and were therefore pooled in the age-at-harvest data likelihood.

Harvest mortality and population size via harvest count data likelihoods

Age-at-harvest data can only inform population structure in this case, and harvest count data are required to quantify relative population size. Since the bulk of our harvest count data ($$$ no.{H}_t $$$) was only available for the whole Svalbard Archipelago and because these data were of relatively poor quality, the corresponding data likelihood had to account for random and systematic errors and be scaled down to represent only harvests from the focal Isfjorden area: 9 $$$ no.{H}_t\sim \mathrm{Poisson}\left(\frac{2\times \mathrm{sum}\left({H}_{1:A,t}\right)\times {r}_t}{prop. Isfj}\right) $$$ Factor 2 in the equation above consolidates the presence of both sexes in the harvest count data. $$$ {r}_t $$$ is a reporting rate representing the probability of a shot seal being reported to the Governor's Environmental Office. This parameter cannot be estimated separately and was therefore set a priori at 0.75 up to and including 2009, and at 0.90 from 2010 onward. The spatial scaling factor $$$ prop. Isfj $$$ is the proportion of the total harvest across all localities on Svalbard that occurred in Isfjorden. Under the assumption that $$$ prop. Isfj $$$ has not changed substantially over the course of the study period, it can be estimated from the subset of harvest reports that were entered into the iNatur database in recent years, which contain information on location: 10 $$$ prop. Isfj=\frac{\mathrm{sum}\left( count. Isfj\right)}{\mathrm{sum}\left( count. all\right)} $$$ with $$$ count.{all}_t $$$ being the total number of harvested individuals reported digitally in the iNatur system for the interval June of year t to May of year t + 1 including location information and $$$ count.{Isfj}_t $$$ being the subset of those with a location in the Isfjorden area.

IPM implementation

Among-year variation in demographic parameters

Demographic rates linked to survival and reproductive output are influenced by environmental conditions and population density and consequently vary among years. Estimating among-year variation in demographic rates requires empirical data collected over several years, and for this study, sufficient relevant data were only available for maturation rates ($$$ {pM}_{a,t} $$$). We used a random-effects model to quantify among-year variation (due to changes in unidentified environmental factors) in maturation rates: 11 $$$ \mathrm{cloglog}\left({pM}_{a,t}\right)=\mathrm{cloglog}\left({\upmu}_a^{pM}\right)+{\upvarepsilon}_t $$$ where $$$ {\upmu}_a^{pM} $$$ is the average maturation rate of age class $$$ a $$$ females and $$$ {\upvarepsilon}_t $$$ are normally distributed random effects. All other demographic rates in the model (except for pup survival, see below) were treated as time-invariant.

Linking pup survival to sea ice conditions

The continuing decline in sea ice is undoubtedly having impacts on ringed seal recruitment, particularly via reducing pup survival (Ferguson et al., 2005; Reimer et al., 2019). At the same time, empirical estimation of relationships between pup survival and sea ice conditions is almost impossible to assess for ringed seals. Other studies facing this problem have used exploratory modeling approaches for forecasting population dynamics under climate change for ringed seals in both the Beaufort Sea (Reimer et al., 2019) and the Baltic Sea (Sundqvist et al., 2012). Their approach consisted of setting a maximum value for pup survival associated with “ideal” ice conditions, and letting it decline linearly toward 0 with deteriorating sea ice conditions. We implemented an equivalent approach in our IPM to mechanistically link annual variation in pup survival to sea ice extent by setting 12 $$$ {S}_{pup,t}={S}_{pup, ideal}\times {p}_{\begin{array}{c} ice,t\ \\ {}\ \end{array}}\kern2.25em {p}_{ice,t}=\left\{\begin{array}{c}\frac{ice_t}{ice_{ideal}},\kern0.75em {ice}_t<{\mathrm{ice}}_{ideal}\\ {}1,\kern4.75em {ice}_t\ge {ice}_{ideal}\hfill \end{array}\right. $$$ Here, $$$ {S}_{pup, ideal} $$$ is pup survival under ideal sea ice conditions, $$$ {ice}_{ideal} $$$ is the threshold sea ice extent that characterizes ideal conditions, and $$$ {\mathrm{ice}}_t $$$ is the sea ice extent in a given year $$$ t $$$. Both $$$ {S}_{pup, ideal} $$$ and $$$ {ice}_{ideal} $$$ were assigned strongly informative priors (see below) and the resulting relationship is visualized in Appendix S1: Figure S2.

Informative priors for survival and mortality parameters

The population model contains several vital rate parameters for which we have no empirical data from Svalbard. For these we employed informative priors: annual survival probabilities ($$$ {S}_{0,t},{S}_{a,t},{S}_{M,t} $$$) and natural mortality hazard rates ($$$ {mN}_{0,t},{mN}_{a,t},{mN}_{M,t} $$$) of all age classes, as well as pup survival under ideal conditions ($$$ {S}_{pup, ideal} $$$) and the associated sea ice threshold ($$$ {ice}_{ideal} $$$).

Strongly informative priors for annual survival probabilities of all age classes were specified via lognormal distributions on the mortality hazard rate scale (loglog link, Ergon et al., 2018). We used log-means that corresponded to the survival probabilities used by Reimer et al. (2019) for ringed seals in Canada: 0.80, 0.82, 0.84, 0.86, and 0.88 for age 1–5 subadults and 0.92 for mature adults. For $$$ YOY $$$ annual survival, we used a mean of 0.75 since Reimer et al.'s (2019) value of 0.65 corresponds to survival from birth to next year's census and hence also includes pup survival ($$$ {S}_{pup}\times {S}_0 $$$). To account for potential differences in survival and cause-specific mortality between different seal populations, we specified our survival priors with an uncertainty of 0.1 log-SD on the hazard rate scale. We tested that Reimer et al.'s (2019) estimates were applicable to ringed seals in Svalbard by running exploratory life table and catch curve analyses using our age-at-harvest data. These analyses are presented in Appendix S2 and show a good match between preliminary estimates of adult survival from our data and Reimer et al.'s proposed value (0.92), suggesting that our priors are appropriate.

We assigned weakly informative priors to natural mortality hazard rates for all age classes using the meta-analytical Hoenig model developed by Porteus et al. (2018, details in SI S2). Note that by setting informative priors for both survival probabilities and natural mortality hazard rates, we also limited the parameter space for harvest mortality hazard rates ($$$ mH $$$) since $$$ mH=-\log (S)- mN $$$ (Ergon et al., 2018). Harvest mortality hazard rates were assumed to differ between first-year and older individuals, but not between subadults and adults.

The informative prior for pup survival under ideal conditions was set using a lognormal distribution for the underlying mortality hazard rate. We chose a mean representing a survival probability of 0.80 (which, multiplied by the average $$$ {S}_0 $$$ of 0.75, is a bit lower than Reimer et al.'s (2019) survival from birth to 1 year old) and a log-SD of 0.2. The sea ice extent threshold for “ideal condition” was given a lognormal prior with a mean corresponding to the 75% quantile of sea ice extent between 1988 and 2005 (i.e., before the dramatic decline in sea ice in 2006) and a log-SD of 0.1 (Appendix S1: Figure S2).

Model implementation with nimble

We implemented and ran the seal IPM (including forecasts) using the software package “nimble” in R (de Valpine et al., 2017; R Core Team, 2022). Assembly and eigenvector calculation for the projection matrix, as well as computation of annual pup survival from sea ice extent, were programed as custom nimbleFunctions to use within the IPM code. We ran five Markov chain Monte Carlo (MCMC) chains of 100,000 iterations, 30,000 of which were discarded as burn-in. The remainder were thinned by a factor of 10, resulting in 5 × 7000 = 35,000 posterior samples.

PBR calculation

We employed traditional PBR analyses to assess potential harvest limits. PBR can serve as an estimate of the maximum number of animals that can be harvested sustainably from a population (Wade, 1998) and is calculated according to: 13 $$$ PBR={N}_{min}\times 0.5{R}_{max}\times {F}_r $$$ where $$$ {N}_{min} $$$ is a minimum estimate of population size, $$$ {R}_{max} $$$ is the maximum rate of population increase ($$$ ={\uplambda}_{max}-1 $$$, with $$$ {\uplambda}_{max} $$$ being population growth rate), and $$$ {F}_r $$$ is a correction factor that can account for both known or unknown population trajectories. $$$ {F}_r $$$ takes on values between 0 and 1, where lower values are assigned for populations that are declining or have highly uncertain population status, and higher values are assigned for populations known to be stable or increasing (NMFS, 2016). Practical applications often use a default value for $$$ {R}_{max} $$$ (0.12 for seals) in combination with a population-specific $$$ {N}_{min} $$$ estimate and an $$$ {F}_r $$$ of either 1 or 0.5 (sometimes both, e.g., Nelson et al., 2019). However, PBR calculations are very sensitive to the choice of parameter values, leading to a substantial risk of bias when using default values for populations whose maximum growth rate is unknown and whose population trajectory is uncertain. We therefore adopted two different approaches to calculate PBR.

Naïve PBR using recommended general values

To obtain a comprehensive picture of the range of possible values for PBR in our focal population, we first adopted an exploratory approach and performed a traditional PBR calculation using combinations of wide ranges of values for $$$ {R}_{max} $$$ and $$$ {F}_r $$$. For $$$ {R}_{max} $$$ we used values from 0 (= population unable to grow even in the absence of harvest) to 0.24 (twice the 0.12 default value). For $$$ {F}_r $$$ we considered the entire range of possible values between 0 and 1. $$$ {N}_{min} $$$ was set to the 20th percentile (following the suggestion of Wade, 1994) of the distribution of estimated population size in 2002 for the seven locations representing the Isfjorden area from Krafft, Kovacs, Andersen, et al.'s (2006) aerial survey (~3019 individuals).

Informed PBR using estimates from IPM

Substantial variation in maximum population growth rates has been found among pinniped populations, even within the same species (Härkönen et al., 2002), and diminishing sea ice has very likely already resulted in decreased population growth for ringed seals (Reimer et al., 2019). Therefore, more representative PBR estimates can be obtained using population-specific estimates of $$$ {R}_{max} $$$ (instead of a default value) and up-to-date $$$ {N}_{min} $$$, both of which can be extracted from the posterior distributions generated by the IPM.

We extracted posterior distributions for several different measures of population growth rate from our IPM. First, we determined the maximum and mean realized growth rates, defined as $$$ N{tot}_{t+ 1}/N{tot}_t $$$ (where $$$ N{tot}_t $$$ = total population size in year $$$ t $$$). Second, we calculated the maximum and mean of hypothetical asymptotic growth rates under an assumption of no harvest by building a projection matrix (Figure 1C) for each year $$$ t $$$ using only the natural component of mortality ($$$ {S}_{noHarvest}={e}^{- mN} $$$) and subsequently extracting its dominant right eigenvector (Caswell, 2001). Finally, $$$ {N}_{min} $$$ was defined as the minimum estimated population size during the main study period (2002–2020).

Using the posterior samples of the four different measures for $$$ {R}_{max} $$$ and $$$ {N}_{min} $$$, we then calculated posterior distributions for PBR assuming F_r = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1. Additionally, we ran the same calculations using the 2002 $$$ {N}_{min} $$$ estimated from the aerial survey that was used in the naïve PBR calculations above.

Population forecasts under different harvest and climate change scenarios

PVAs are forward projections of a population model to explore possible future population dynamics under one or several predefined scenarios (Morris & Doak, 2002; Reed et al., 2002) and can be incorporated directly into IPMs (Schaub & Kéry, 2021). Here, we ran a total of six scenarios consisting of three levels of harvest (unchanged, halved, no harvest) and two types of climate change/sea ice predictions (stable, continuing decline). Different harvests rates were simulated by adjusting the harvest mortality hazard rates ($$$ {mH}_{AgeClass,t} $$$) from 2020 onward by multiplying them with 1 (unchanged), 0.5 (halved), and 0 (no harvest). Sea ice predictions were based on a log-linear trend model fit to the time series of sea ice extent data (Appendix S1: Figure S2A). For the stable scenario, we assumed that future sea ice extent remained as in 2020 on average. For the scenario of continuing decline, we assumed that future average sea ice extent decreased further at the same rate as during the period 1988–2020. We projected population dynamics under all six scenarios for an additional 30 years post-2020 (i.e., until 2050).

Data preparation, IPM fitting, PBR calculations, and PVA projections were run in R version 4.1.2 (R Core Team, 2022). Code for all steps of the workflow is available from Zenodo: .

RESULTS

MCMC converged for all parameters in our IPM within the selected number of iterations (confirmed through visual inspection of trace plots). Posterior summaries for key demographic rates and population parameters are provided in Appendix S1: Table S1. Results based on posterior distributions from the IPM are reported below as median [95% credible interval].

Naive PBR

Figure 2 illustrates that PBR calculations can result in very different numbers depending on the choice of $$$ {R}_{max} $$$ and $$$ {F}_r $$$, suggesting that the number of ringed seals that can be harvested sustainably within the Isfjorden area could lie anywhere between 0 and over 300. When assuming the pinniped default value for population growth rate ($$$ {R}_{max}=0.12 $$$), the number of sustainably harvestable individuals would be 181 when assuming a stable/growing population with little uncertainty about population trajectory (i.e., correction factor $$$ {F}_r=1 $$$), 91 with a more cautious approach using $$$ {F}_r=0.5 $$$, and decreased toward 0 proportionally at lower $$$ {F}_r $$$ values.

[IMAGE OMITTED. SEE PDF]

PBR based on IPM estimates

Posterior distributions for minimum population size and different measures of population growth rates obtained from the IPM are shown in Figure 3. The minimum population size (including males and females) throughout the period 2002–2020 was estimated to be 1942 [1512, 2720] seals. The average asymptotic growth rate without harvest was estimated to be higher (1.00 [0.99, 1.02]) than the average realized growth rate (0.97 [0.96, 0.99]). Maximum asymptotic (1.04 [1.02, 1.05]) and realized growth rates (1.04 [1.01, 1.08]) were more similar.

[IMAGE OMITTED. SEE PDF]

Posterior distributions for PBR calculated from maximum growth rates contained values from 0 to 187 seals that could be harvested sustainably, but 90% of values predicted across the entire range of correction factors ($$$ {F}_r $$$) fell below 49 individuals using realized growth rate and 38 individuals using asymptotic growth rate (Figure 4, first column). With $$$ {F}_r=1 $$$, PBR was estimated to be 39.40 [7.52, 96.9] using maximum realized growth rates and 34.70 [15.00, 67.2] using maximum asymptotic growth rates. With $$$ {F}_r=0.5 $$$, the corresponding estimates were 19.70 [3.76, 48.4] and 17.3 [7.50, 33.6].

[IMAGE OMITTED. SEE PDF]

The posterior distributions of mean growth rates contained estimates of negative population growth rates ($$$ {\uplambda}_{max}<1;{R}_{max}<0 $$$), which resulted in some negative PBR values (Figure 4, second column). Using the mean realized growth rate, all PBR values were negative irrespective of the $$$ {F}_r $$$ value chosen. When using mean asymptotic growth rate, the choice of $$$ {F}_r $$$ was reflected in the degree of uncertainty around an average PBR estimate close to 0.

Results were very similar, albeit with slightly higher PBR estimates, when using the 2002 aerial survey population size estimate instead of the projected minimum population size from the IPM (Appendix S1: Figure S4).

Population forecasts under different harvest and climate change scenarios

Population projections indicated that population size is likely to continue declining over the next three decades under all scenarios (Figures 5 and 6; Appendix S1: Figures S5 and S6). Continuing population declines will also decrease the number of seals that can be harvested sustainably (Figure 5). Reducing hunting pressure by half or completely removing harvest mortality slowed down population decline notably. Without any harvest the population was predicted to decline by 30% [−15, 80] less than when harvest continued unchanged, resulting in a population size of 691[332, 1353] instead of 413 [191, 963] by 2050 for a scenario with stable sea ice condition (Figure 5; Appendix S1: Figure S5). In the more realistic scenarios with continuing deterioration of sea ice conditions, and decreasing pup survival (Figure 5; Appendix S1: Figures S6 and S7), the projected 2050 population sizes were on average ~40% lower than those predicted under stable sea ice conditions (factors −0.40 [−0.80, 0.71], −0.37 [−0.78, 0.786], and −0.38 [−0.76, 0.63] for scenarios of unchanged, halved, and no harvest, respectively, Figure 6).

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

DISCUSSION

PBR (Wade, 1998), a commonly used tool for assessing sustainability of marine mammal harvest (e.g., pursuant to the Marine Mammals Protection Act, NMFS, 2005, 2016), provides conservative estimates of the number of animals that can be harvested sustainably even in data-sparse situations. Calculating PBR naively across a range of possible values of maximum population growth rates and correction factors indicated that the number of ringed seals that can be harvested sustainably in the Isfjorden area of Svalbard could lie anywhere between 0 and 300 (Figure 2). That being said, maximum growth rates for true seals are unlikely to exceed 13% (Härkönen et al., 2002) and 12% is recommended as the default value in PBR for pinnipeds (NMFS, 2016; Wade, 1998). The latter is also what Nelson et al. (2019) used in their recent PBR assessment for ringed seals in Alaska. Following their example, we obtained an annual harvest limit of 91 individuals when also accounting for uncertainty in the population trajectory by setting a correction factor of 0.5, as recommended. This limit is higher than the 15–78 (mean = 48.3) ringed seals reported in the annual harvest statistics in recent years, suggesting that the current harvest would be sustainable if the population could sustain a growth rate of 12% per year in the absence of harvest. There can, however, be substantial variation in growth rates both between and within pinniped species (Härkönen et al., 2002) and diminishing sea ice is likely already constraining the growth potential of ringed seals in Svalbard and elsewhere (Ferguson et al., 2005; Kovacs et al., 2011, 2021; Laidre et al., 2015; Reimer et al., 2019; Sundqvist et al., 2012). This increases uncertainty around the population trajectory and limits potential growth rate of Svalbard ringed seals, which may, in turn, call for further reduction of the correction factor in PBR analyses. This results in an estimate of maximum number harvestable lying somewhere between 0 and 91.

A PBR estimate spanning such a range of values is of limited usefulness for management in practice, and we therefore supplemented the traditional PBR approach with additional information about population size and growth rates from our IPM. Evaluating both realized and theoretical asymptotic growth rates in the absence of harvest, we found that maximum growth rate of the ringed seal population in Svalbard is unlikely to exceed 10% and probably falls below 5% (Figure 3). Simultaneously, the population was also predicted to have declined by about 36% during the two decades since the last survey (Figure 5). The population decline appears to have been initiated in 2006, coinciding with a marked collapse in sea ice (Appendix S1: Figure S2), which was predicted to have led to record-low pup survival (Figure 5). Pups that are born outside a birthing lair, because of ice breakup or premature deterioration of the snow roof, have very poor chances of survival; these small neonates are prey for bears, foxes, and even avian predators (Lydersen & Gjertz, 1986; Lydersen & Smith, 1989). Additionally, exposure to cold, wind, and rain outside the lair also reduces pup survivorship (Smith et al., 1991). The combination of relatively low growth rates and a decreasing population trajectory between 2002 and 2020 (estimated by our IPM) suggest that naively calculating PBR using the default growth rate of 12% and the population estimate from 2002 is almost certainly too optimistic and potentially misleading. Indeed, when we recalculated PBR using predicted growth rates and minimum population size over the period 2002–2020, it became clear that, while uncertainty was high, any harvest exceeding 50 individuals is almost certainly unsustainable (Figure 4).

Average realized population growth rates were estimated to be exclusively below one and even average asymptotic growth rates without harvest had a 43% chance of falling below replacement (Figure 3), strongly suggestive of a population already in decline. This indicates that under current environmental conditions, the ringed seal population in Svalbard is unlikely to grow even in the absence of harvest. For nongrowing populations, the PBR approach breaks down (negative values, Figure 4) and we shifted to a more suitable PVA (Morris & Doak, 2002; Reed et al., 2002). PVA forecasts including harvest predicted further population decline resulting in losses of 58% and 74% from 2020 to 2050 if sea ice conditions remained stable or deteriorated further, respectively (Figure 5). In the absence of harvest, predicted abundance declines by 2050 were lower but still substantial at 28%–55%, depending on the sea ice scenario used. Such rates of decline are very similar to the forecasts by Reimer et al. (2019) for Canadian ringed seal populations over the same time period, assuming medium sensitivity to environmental conditions affecting pup survival (snow depth in their case). Prediction uncertainty was high, but forecasts suggested that a stop to harvest might increase the population's chance of long-term persistence (stable or increasing growth rates) by a factor of 40, to 10.8% with stable sea ice conditions, and by a factor of 30 to 0.3% with further deteriorating climatic conditions.

Taken together, results from both PBR and PVA approaches suggest that any harvest of Svalbard ringed seals may be unsustainable under current conditions. While high uncertainty in estimates and predictions necessitates continued and extended data collection to improve the evidence base for management decisions, a precautionary approach to harvest may be prudent at this stage. A precautionary approach is also appropriate given the fact that this seal population is likely also dealing with additional stressors, such as increasing numbers of harbor seals (Phoca vitulina), which have overlapping dietary preferences with ringed seals in the region (e.g., Bengtsson et al., 2020). Seal hunting on Svalbard is almost exclusively recreational, although some meat makes it into the local restaurants or is used for dog food in the tourist dog-mushing trade. If maintaining opportunities for hunters is to be prioritized, one approach to mediating harvest impacts is setting quotas to avoid the age classes and/or sexes with most impact on population dynamics as is commonly done for ungulates (e.g., Milner et al., 2007, 2011; Peeters et al., 2021; Strand et al., 2012) and recommended for fish (e.g., Birkeland & Dayton, 2005; Gwinn et al., 2015; Nater et al., 2022). For long-lived pinnipeds, such as ringed seals, the most important animals to avoid harvesting would be adult females (e.g., Silva et al., 2021). Unfortunately, it is not easy to separate males from females at large distances, and ringed seals are quite shy, so shooting generally takes place at some hundreds of meters. A more realistic solution for alleviating hunting pressure on ringed seals in Svalbard would be to shift the hunt to other, less vulnerable species. Harbor seals in particular have become more abundant and widespread in Svalbard in recent years (Bengtsson et al., 2021), and small harvests of a magnitude similar to what is practiced on ringed seals today would be unlikely to negatively impact their growing populations (Eldegard et al., 2021).

The potential implementation of new and stricter regulations of ringed seal hunting would also constitute an opportunity for optimizing harvest data collection and hence contribute to reducing uncertainty in population estimates and forecasts. The present study had to make a range of assumptions about reporting and spatiotemporal distribution of hunting activity. The estimation of harvest extent and intensity could be improved greatly by replacing these assumptions with more accurate data on both hunting effort (e.g., time spent hunting, Willebrand et al., 2011) and the time and location of any kills, both of which could be elicited as part of extended mandatory reporting. Collecting more detailed information on hunting effort is also crucial as there is an ongoing transition toward more hunting taking place in Isfjorden as trapper stations further from the town of Longyearbyen become fewer. It would also be useful to require that hunters delive seal jaws for aging, and perhaps even a more comprehensive set of demographic data including sex and, where feasible, reproductive status (Andersen et al., 2020; Krafft, Kovacs, Frie, et al., 2006; Lydersen & Gjertz, 1987) as part of the mandatory reporting. Such individual-based demographic data contributed crucial information in this present study, and collecting such data from a larger proportion of the harvested seals would be valuable in the future (Clutton-Brock & Sheldon, 2010). However, the most important management action would be the design and implementation of additional, nondestructive monitoring.

The last abundance survey of the seal population in Svalbard was carried out more than 20 years ago (Krafft, Kovacs, Andersen, et al., 2006), prior to the dramatic sea ice collapse in the area. There is an urgent need for an up-to-date abundance estimate and—ideally—more regular surveys in the years to come for assessing population status. Resulting data could be used to validate results of model predictions like the ones presented in this study, and also provide baselines for future population forecasts. Manned helicopter surveys combined with manual image analysis (as done in Svalbard previously; Krafft, Kovacs, Andersen, et al., 2006) are costly, logistically challenging, and labor-intensive, but new alternatives combining unmanned aircraft systems, remote sensing, and machine learning are showing great promise (Rodofili et al., 2022; Seymour et al., 2017). Thermal imagery in particular has been shown to be suitable for monitoring abundance of ice-affiliated pinnipeds (Seymour et al., 2017; Young et al., 2019) as long as observation error is properly accounted for (Conn et al., 2014); simulation studies can help with designing optimized surveys (Conn et al., 2016). We therefore suggest exploring options for designing and setting up a new cyclically regular monitoring program for seals in Svalbard, potentially building on automated workflows for ice-seal monitoring developed elsewhere (e.g., SealNet, Gonçalves et al., 2020, 2022). We also note that monitoring approaches based on remote imagery and pattern-recognition algorithms may have the potential to not only just count but also measure individuals. Infantes et al. (2022), for example, developed an algorithm that automatically measures size of harbor seals (P. vitulina) on aerial images, thus allowing distinction of adults versus pups (hence estimation of reproductive output) and a rough assessment of individual body condition. If a similar approach could be implemented for ringed seals in Svalbard, that may be able to at least partially compensate for the lack of demographic data from harvested seals resulting from a reduction or even (temporary) termination of hunting and ultimately allow for efficient, nonlethal population modeling (e.g., sensu Thomas et al., 2019).

Since climate change strongly impacts ice-affiliated seals via pup survival (Ferguson et al., 2005; Reimer et al., 2019; Sundqvist et al., 2012; this study), population forecasts would also benefit greatly from additional knowledge and empirical data on the links between climate, breeding area conditions, and pup survival. In this study, we focused exclusively on relative sea ice extent as a predictor of pup survival, but building and persistence of subnivean birth lairs also depends on snow conditions. Birth lair habitat availability should also be considered in relation to seal abundance as polar bear hunting success and thus pup survival may be linked to both population size and density (e.g., Sundqvist et al., 2012). While proxies for breeding lair habitat availability may be obtained from remote-sensed data (e.g., Lindsay et al., 2021), it may also be worth considering on-site surveys similar to those conducted in the 1980s (e.g., Lydersen et al., 1990; Lydersen & Gjertz, 1986). Snow-lair surveys are central for population estimation for the critically endangered Saimaa ringed seals (Phoca hispida saimensis, Kunnasranta et al., 2021), for example. Notably, monitoring using photographic ID—where individuals are identified (automatically) on photographs taken by people and/or camera traps using natural patterns—constitutes a less invasive approach to obtain data on survival and reproductive rates (e.g., Bolger et al., 2012). Photographic ID is used successfully on a range of pinniped species including harbor seals (Cordes & Thompson, 2014; Mackey et al., 2008), monk seals (Baker et al., 2011), and Saimaa ringed seals (Koivuniemi et al., 2019), and it may be worth exploring opportunities for its use with ringed seals in Svalbard. Similarly, there may be future opportunities for genetic monitoring, for example, to contribute auxiliary information on population size and potentially adult survival through close-kin mark–recapture approaches (Bravington et al., 2016; Marcy-Quay et al., 2020). Not least, genetic monitoring could facilitate studying meta-population structure and the potential role of dispersal in the dynamics of ringed seals in Svalbard. There are suggestions that ringed seals might disperse on a pan-Arctic scale (Martinez-Bakker et al., 2013) and that immigration/emigration might have an influence on the dynamics of ringed seal populations (Chambellant et al., 2012), while data from other areas suggest that some ringed seal populations are regional and not “open” (Hamilton et al., 2022; Kovacs et al., 2021; Rosing-Asvid et al., 2023). Understanding philopatry versus dispersal for ringed seals in Svalbard and including this in modeling frameworks like the one presented in this study would constitute a marked improvement in modeling ringed seal populations and a promising direction for future studies.

CONCLUSION

PBR is a tool commonly used for assessing the sustainability of marine mammal harvests, yet our application of the method to ringed seals in Svalbard showed that the resulting estimates can have high uncertainty and thus be of limited use in practice when data are sparse and population trajectories are unknown or negative. As the latter is likely the case for a range of ice-affiliated seals under climate change, PVAs based on integrated data analysis likely constitute a more suitable framework. Such integrated approaches not only maximize information gain from sparse data but also help identify crucial knowledge gaps. In the case of ringed seals in Svalbard, harvest is likely unsustainable under current conditions, but population forecasts are uncertain. Improving forecasts under climate change will hinge on setting up more frequent population surveys and obtaining more knowledge of the links between vital rates and environmental conditions, both of which may be facilitated by the adoption of novel technology including—but not limited to—drone surveys and genetic monitoring. The modeling framework developed in this study can be updated with new data as they become available, and ultimately serve as a tool for adaptive management of this and potentially other pinniped populations affected by declining sea ice.

AUTHOR CONTRIBUTIONS

Chloé R. Nater: Conceptualization, methodology, software, formal analysis, data curation, writing—original draft, writing—review & editing, visualization. Christian Lydersen: Conceptualization, investigation, data curation, writing—review & editing. Magnus Andersen: Conceptualization, investigation, data curation, writing—review & 719 editing. Kit M. Kovacs: Conceptualization, methodology, investigation, data curation, writing—original draft, writing—review & editing, funding acquisition.

ACKNOWLEDGMENTS

We thank Murray Christian and an anonymous reviewer for constructive comments on an earlier version of the manuscript. We are grateful to Glen Liston for preparing the sea ice data used in the analyses and for helpful discussions of sea ice trends. This study was financed by an Arctic Council grant from the Government of Norway under the auspices of CAFF (Conservation of Flora and Fauna) and the Norwegian Research Council project ARK (project no. 313678).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

Processed data necessary to run the analyses (Nater, 2024a), as well as posterior samples for all presented results, are available from OSF: . Raw data (Kovacs et al., 2024) are available from the Norwegian Polar Data Centre: . R code (v1.2; Nater, 2024b) is available from Zenodo: .

References

ACIA. 2005. Arctic Climate Impact Assessment, Vol. 1042. Cambridge, UK: Cambridge University Press.

Andersen, M., K. M. Kovacs, and C. Lydersen. 2020. “Stable Ringed Seal (Pusa hispida) Demography Despite Significant Habitat Change in Svalbard, Norway.” Polar Research 40: 1–14.

Årthun, M., I. H. Onarheim, J. Dörr, and T. Eldevik. 2021. “The Seasonal and Regional Transition to an Ice‐Free Arctic.” Geophysical Research Letters 48(1): [eLocator: e2020GL090825].

Baker, J. D., A. L. Harting, T. A. Wurth, and T. C. Johanos. 2011. “Dramatic Shifts in Hawaiian Monk Seal Distribution Predicted from Divergent Regional Trends.” Marine Mammal Science 27(1): 78–93.

Bengtsson, O., C. D. Hamilton, C. Lydersen, M. Andersen, and K. M. Kovacs. 2021. “Distribution and Habitat Characteristics of Pinnipeds and Polar Bears in the Svalbard Archipelago, 2005–2018.” Polar Research 40: 773–788.

Bengtsson, O., C. Lydersen, K. M. Kovacs, and U. Lindström. 2020. “Ringed Seal (Pusa hispida) Diet on the West Coast of Spitsbergen, Svalbard, Norway: During a Time of Ecosystem Change.” Polar Biology 43(7): 773–788.

Beszczynska‐Möller, A., E. Fahrbach, U. Schauer, and E. Hansen. 2012. “Variability in Atlantic Water Temperature and Transport at the Entrance to the Arctic Ocean, 1997–2010.” ICES Journal of Marine Science 69(5): 852–863.

Birkeland, C., and P. K. Dayton. 2005. “The Importance in Fishery Management of Leaving the Big Ones.” Trends in Ecology & Evolution 20(7): 356–358.

Bolger, D. T., T. A. Morrison, B. Vance, D. Lee, and H. Farid. 2012. “A Computer‐Assisted System for Photographic Mark–Recapture Analysis.” Methods in Ecology and Evolution 3(5): 813–822.

Bravington, M. V., H. J. Skaug, and E. C. Anderson. 2016. “Close‐Kin Mark‐Recapture.” Statistical Science 31(2): 259–274.

Carlens, H., C. Lydersen, B. A. Krafft, and K. M. Kovacs. 2006. “Spring Haul‐out Behavior of Rigned Seals (Pusa hispida).” Marine Mammal Science 22(2): 379–393.

Caswell, H. 2001. Matrix Population Models: Construction, Analysis, and Interpretation. Sunderland, Massachusetts, USA: Sinauer Associates Incorporated Publishers.

Chambellant, M., I. Stirling, W. A. Gough, and S. H. Ferguson. 2012. “Temporal Variations in Hudson Bay Ringed Seal (Phoca hispida) Life‐History Parameters in Relation to Environment.” Journal of Mammalogy 93(1): 267–281.

Clutton‐Brock, T., and B. C. Sheldon. 2010. “Individuals and Populations: The Role of Long‐Term, Individual‐Based Studies of Animals in Ecology and Evolutionary Biology.” Trends in Ecology & Evolution 25(10): 562–573.

Conn, P. B., D. R. Diefenbach, J. L. Laake, M. A. Ternent, and G. C. White. 2008. “Bayesian Analysis of Wildlife Age‐At‐Harvest Data.” Biometrics 64(4): 1170–1177.

Conn, P. B., E. E. Moreland, E. V. Regehr, E. L. Richmond, M. F. Cameron, and P. L. Boveng. 2016. “Using Simulation to Evaluate Wildlife Survey Designs: Polar Bears and Seals in the Chukchi Sea.” Royal Society Open Science 3(1): [eLocator: 150561].

Conn, P. B., J. M. Ver Hoef, B. T. McClintock, E. E. Moreland, J. M. London, M. F. Cameron, S. P. Dahle, and P. L. Boveng. 2014. “Estimating Multispecies Abundance Using Automated Detection Systems: Ice‐Associated Seals in the Bering Sea.” Methods in Ecology and Evolution 5(12): 1280–1293.

Cordes, L. S., and P. M. Thompson. 2014. “Mark‐Recapture Modeling Accounting for State Uncertainty Provides Concurrent Estimates of Survival and Fecundity in a Protected Harbor Seal Population.” Marine Mammal Science 30(2): 691–705.

de Valpine, P., D. Turek, C. J. Paciorek, C. Anderson‐Bergman, D. T. Lang, and R. Bodik. 2017. “Programming with Models: Writing Statistical Algorithms for General Model Structures with NIMBLE.” Journal of Computational and Graphical Statistics 26(2): 403–413.

Eldegard, K., P. O. Syvertsen, A. Bjørge, K. M. Kovacs, O.‐G. Støen, and J. van der Kooji. 2021. “Pattedyr: Vurdering av Steinkobbe Phoca vitualina for Svalbard (Rødlista for Arter 2021).” Artsdatabanken. http://www.artsdatabanken.no/lister/rodlisteforarter/2021/4135.

Ergon, T., Ø. Borgan, C. R. Nater, and Y. Vindenes. 2018. “The Utility of Mortality Hazard Rates in Population Analyses.” Methods in Ecology and Evolution 9(10): 2046–2056.

Ferguson, S. H., I. Stirling, and P. McLoughlin. 2005. “Climate Change and Ringed Seal (Phoca hispida) Recruitment in Western Hudson Bay.” Marine Mammal Science 21(1): 121–135.

Geyman, E. C., W. J. J. van Pelt, A. C. Maloof, H. F. Aas, and J. Kohler. 2022. “Historical Glacier Change on Svalbard Predicts Doubling of Mass Loss by 2100.” Nature 601(7893): 374–379.

Gonçalves, B. C., B. Spitzbart, and H. J. Lynch. 2020. “SealNet: A Fully‐Automated Pack‐Ice Seal Detection Pipeline for Sub‐Meter Satellite Imagery.” Remote Sensing of Environment 239: [eLocator: 111617].

Gonçalves, B. C., M. Wethington, and H. J. Lynch. 2022. “SealNet 2.0: Human‐Level Fully‐Automated Pack‐Ice Seal Detection in Very‐High‐Resolution Satellite Imagery with CNN Model Ensembles.” Remote Sensing 14(22): [eLocator: 22].

Gwinn, D. C., M. S. Allen, F. D. Johnston, P. Brown, C. R. Todd, and R. Arlinghaus. 2015. “Rethinking Length‐Based Fisheries Regulations: The Value of Protecting Old and Large Fish with Harvest Slots.” Fish and Fisheries 16(2): 259–281.

Hamilton, C. D., K. Kovacs, and C. Lydersen. 2019. “Sympatric Seals Use Different Habitats in an Arctic Glacial Fjord.” Marine Ecology Progress Series 615: 205–220.

Hamilton, C. D., K. M. Kovacs, R. A. Ims, J. Aars, and C. Lydersen. 2017. “An Arctic Predator‐Prey System in Flux: Climate Change Impacts on Coastal Space Use by Polar Bears and Ringed Seals.” Journal of Animal Ecology 86(5): 1054–1064.

Hamilton, C. D., C. Lydersen, J. Aars, M. Acquarone, T. Atwood, A. Baylis, M. Biuw, et al. 2022. “Marine Mammal Hotspots Across the Circumpolar Arctic.” Diversity and Distributions 28(12): 2729–2753. [DOI: https://dx.doi.org/10.1111/ddi.13543]

Schaub, M., and M. Kéry. 2021. Integrated Population Models: Theory and Ecological Applications With R and JAGS. London, UK: Academic Press.

Seymour, A. C., J. Dale, M. Hammill, P. N. Halpin, and D. W. Johnston. 2017. “Automated Detection and Enumeration of Marine Wildlife Using Unmanned Aircraft Systems (UAS) and Thermal Imagery.” Scientific Reports 7(1): [eLocator: 1].

Silva, W. T. A. F., E. Bottagisio, T. Härkönen, A. Galatius, M. T. Olsen, and K. C. Harding. 2021. “Risk for Overexploiting a Seemingly Stable Seal Population: Influence of Multiple Stressors and Hunting.” Ecosphere 12(1): [eLocator: e03343].

Simmonds, M. P., and S. J. Isaac. 2007. “The Impacts of Climate Change on Marine Mammals: Early Signs of Significant Problems.” Oryx 41(1): 19–26.

Slater, T., I. R. Lawrence, I. N. Otosaka, A. Shepherd, N. Gourmelen, L. Jakob, P. Tepes, L. Gilbert, and P. Nienow. 2021. “Review Article: Earth's Ice Imbalance.” The Cryosphere 15(1): 233–246.

Smith, T. G., M. O. Hammill, and G. TaugbØl. 1991. “A Review of the Developmental, Behavioural and Physiological Adaptations of the Ringed Seal, Phoca hispida, to Life in the Arctic Winter.” Arctic 44(2): 124–131.

Stirling, I., and A. E. Derocher. 1993. “Possible Impacts of Climatic Warming on Polar Bears.” Arctic 46(3): 240–245.

Strand, O., E. B. Nilsen, E. J. Solberg, and J. C. D. Linnell. 2012. “Can Management Regulate the Population Size of Wild Reindeer (Rangifer tarandus) through Harvest?” Canadian Journal of Zoology 90(2): 163–171.

Sundqvist, L., T. Harkonen, C. J. Svensson, and K. C. Harding. 2012. “Linking Climate Trends to Population Dynamics in the Baltic Ringed Seal: Impacts of Historical and Future Winter Temperatures.” Ambio 41(8): 865–872.

Taylor, P. C., R. C. Boeke, L. N. Boisvert, N. Feldl, M. Henry, Y. Huang, P. L. Langen, et al. 2022. “Process Drivers, Inter‐Model Spread, and the Path Forward: A Review of Amplified Arctic Warming.” Frontiers in Earth Science 9: [eLocator: 758361].

Thomas, L., D. J. F. Russell, C. D. Duck, C. D. Morris, M. Lonergan, F. Empacher, D. Thompson, and J. Harwood. 2019. “Modelling the Population Size and Dynamics of the British Grey Seal.” Aquatic Conservation: Marine and Freshwater Ecosystems 29(S1): 6–23.

Tynan, C. T., and D. P. DeMaster. 1997. “Observations and Predictions of Arctic Climatic Change: Potential Effects on Marine Mammals.” Arctic 50(4): 308–322.

Van Wormer, E., J. A. K. Mazet, A. Hall, V. A. Gill, P. L. Boveng, J. M. London, T. Gelatt, et al. 2019. “Viral Emergence in Marine Mammals in the North Pacific May be Linked to Arctic Sea Ice Reduction.” Scientific Reports 9(1): [eLocator: 15569].

Wade, P. R. 1994. Managing Populations under the Marine Mammal Protection Act of 1994: A Stragey for Selection Values for Nmin, the Minimum Abundance Estimate, and Fr, the Recovery Factor 26. La Jolla, California, USA: U.S. Department of Commerce, NOAA, National Marine Fisheries Service, Southwest Fisheries Science Center, Administrative Report [LJ‐94‐19].

Wade, P. R. 1998. “Calculating Limits to the Allowable Human‐Caused Mortality of Cetaceans and Pinnipeds.” Marine Mammal Science 14(1): 1–37.

Walch, D. M. R., R. K. Singh, J. E. Søreide, H. Lantuit, and A. Poste. 2022. “Spatio‐Temporal Variability of Suspended Particulate Matter in a High‐Arctic Estuary (Adventfjorden, Svalbard) Using Sentinel‐2 Time‐Series.” Remote Sensing 14(13): [eLocator: 13].

Watt, C. A., J. Orr, and S. H. Ferguson. 2016. “A Shift in Foraging Behaviour of Beluga Whales Delphinapterus leucas from the Threatened Cumberland Sound Population May Reflect a Changing Arctic Food Web.” Endangered Species Research 31: 259–270.

Willebrand, T., M. Hörnell‐Willebrand, and L. Asmyhr. 2011. “Willow Grouse Bag Size Is more Sensitive to Variation in Hunter Effort than to Variation in Willow Grouse Density.” Oikos 120(11): 1667–1673.

Young, B. G., D. J. Yurkowski, J. B. Dunn, and S. H. Ferguson. 2019. “Comparing Infrared Imagery to Traditional Methods for Estimating Ringed Seal Density.” Wildlife Society Bulletin 43(1): 121–130.

Yurkowski, D. J., N. E. Hussey, S. H. Ferguson, and A. T. Fisk. 2018. “A Temporal Shift in Trophic Diversity among a Predator Assemblage in a Warming Arctic.” Royal Society Open Science 5(10): [eLocator: 180259].