Killer whale (Orcinus orca) population dynamics in response to a period of rapid ecosystem change in the eastern North Atlantic

Abstract This study investigates survival and abundance of killer whales (Orcinus orca) in Norway in 1988–2019 using capture–recapture models of photo‐identification data. We merged two datasets collected in a restricted fjord system in 1988–2008 (Period 1) with a third, collected after their preferred herring prey shifted its wintering grounds to more exposed coastal waters in 2012–2019 (Period 2), and investigated any differences between these two periods. The resulting dataset, spanning 32 years, comprised 3284 captures of 1236 whales, including 148 individuals seen in both periods. The best‐supported models of survival included the effects of sex and time period, and the presence of transients (whales seen only once). Period 2 had a much larger percentage of transients compared to Period 1 (mean = 30% vs. 5%) and the identification of two groups of whales with different residency patterns revealed heterogeneity in recapture probabilities. This caused estimates of survival rates to be biased downward (females: 0.955 ± 0.027 SE, males: 0.864 ± 0.038 SE) compared to Period 1 (females: 0.998 ± 0.002 SE, males: 0.985 ± 0.009 SE). Accounting for this heterogeneity resulted in estimates of apparent survival close to unity for regularly seen whales in Period 2. A robust design model for Period 2 further supported random temporary emigration at an estimated annual probability of 0.148 (± 0.095 SE). This same model estimated a peak in annual abundance in 2015 at 1061 individuals (95% CI 999–1127), compared to a maximum of 731 (95% CI 505–1059) previously estimated in Period 1, and dropped to 513 (95% CI 488–540) in 2018. Our results indicate variations in the proportion of killer whales present of an undefined population (or populations) in a larger geographical region. Killer whales have adjusted their distribution to shifts in key prey resources, indicating potential to adapt to rapidly changing marine ecosystems.


| INTRODUC TI ON
Life history and other population parameters are key elements in status assessments of animal populations. In particular, mortality rate, population size, and geographic range are the main criteria for evaluation of a species' extinction risk (IUCN, 2019), because small populations characterized by restricted geographical ranges are less buffered against losses and face an increased risk of extinction (Purvis et al., 2000).
Time series of abundance estimates can indicate the extent to which a population may be in decline and estimates of survival and birth rates are important components in the evaluation of conservation measures.
Information on abundance is also needed to assess how predators may affect prey populations and how they may respond to fluctuations in prey availability (e.g., Millon et al., 2014;Morissette et al., 2010).
In cetacean research, photo-identification is a noninvasive way to consistently "capture" (first photographic record) and "recapture" (subsequent photographic records) individually recognizable animals over time, using long-lasting natural markings (Hammond, 1990).
This technique offers the possibility to include images from participants other than primary research teams (e.g., citizen science), thus increasing sample size at reduced costs and allowing for data collection in regions for which funding may be limited (Gibson et al., 2020).
Best practices in image manipulation, scoring, and cataloguing are important to allow generation of robust capture history datasets for analysis (Urian et al., 2015). Capture-recapture models fitted to capture histories generated from photo-identification data have been used to obtain estimates of survival rates and abundance for a range of cetacean species (e.g., Arso Civil et al., 2019;Pace et al., 2017;Ramp et al., 2006;Schleimer et al., 2019;Zeh et al., 2002).
Photo-identification was first applied to killer whales (Orcinus orca) in the north-eastern Pacific in 1973 (Bigg, 1982) and has since led to robust estimates of life-history parameters for a number of discrete populations worldwide (Durban et al., 2010;Esteban et al., 2016;Fearnbach et al., 2019;Jordaan et al., 2020;Kuningas et al., 2014;Olesiuk et al., 1990Olesiuk et al., , 2005Pitman et al., 2018;Tixier et al., 2015Tixier et al., , 2017. As time series of photo-identification data have become increasingly available, they have played a central role in identifying population trends and conservation status. For example, a small population size (≤100 individuals), low declining survival rates, and/or low-to-no reproductive output were used as basis for management advice on killer whales at Crozet Poncelet et al., 2010;Tixier et al., 2015Tixier et al., , 2017, Gibraltar (Esteban et al., 2016), Prince William Sound, Alaska (AT1 group, Matkin et al., 2008), and for the Southern resident population in British Columbia, Canada (COSEWIC, 2008). These studies provided an understanding of underlying threats to long-term survival of killer whales and also emphasized the need to account for intrapopulation heterogeneity in behavior when assessing demographic trajectories in this species, otherwise risking false trends being detected (see Esteban et al., 2016;Tixier et al., 2015Tixier et al., , 2017. In Norway, photo-identification studies of killer whales were initiated in the 1980s (Lyrholm, 1988). In this part of the world, killer whales have long been known to mainly feed on Atlantic herring (Clupea harengus) and, more specifically, to follow seasonal movements of the Norwegian Spring Spawning stock (hereafter referred to as the NSS herring; Christensen, 1982Christensen, , 1988Jonsgård & Lyshoel, 1970;Similä et al., 1996). The NSS herring has gone through major changes in abundance and distribution over the past decades, with recruitment of abundant year classes to the spawning stock often resulting in changes in wintering locations (Dragesund et al., 1997;Huse et al., 2010).
Throughout the 1990s, the NSS herring (and killer whales) consistently wintered in the fjord system of Tysfjord-Vestfjord, where they were readily accessible for study, and killer whales were photo-identified annually from 1986 through 2003 (Bisther & Vongraven, 1995;Kuningas et al., 2014;Similä et al., 1996). From these 18 years of data, population size, survival, and reproductive rates were estimated for the first time for killer whales in Norway, which were comparable to other apparently healthy killer whale populations (Kuningas et al., 2014). From 2002 onward, as the inshore winter distribution of NSS herring progressively shifted to a new area further offshore (Holst et al., 2004;Huse et al., 2010), lower numbers of killer whales entered the fjords each year until 2008, after which data collection was interrupted for a few years. After the NSS herring started wintering in coastal fjords of Vesterålen and Troms, annual winter photo-identification surveys were resumed from 2013 (see .
Other major ecological changes occurred in the Norwegian Sea over the past two decades and may have impacted killer whales.
The north-eastern Atlantic mackerel (Scomber scombrus) increased in biomass (from ~2 Mt in 2007 to 9 Mt in 2014) and expanded its geographic range north-and westward (ICES, 2013;Nøttestad, Utne, et al., 2015). The NSS herring declined from ~12 Mt in 2009 to 5 Mt in 2014 and changed feeding and wintering distributions (ICES, 2013(ICES, , 2018. A number of cetacean predators of herring (e.g., pilot whales Globicephala melas and humpback whales Megaptera novaeangliae) seem to have increased in occurrence in the Norwegian Sea (Leonard & Øien, 2020b;Nøttestad, Krafft, et al., 2015), while other abundant baleen whales (e.g., common minke whales Balaenoptera acutorostrata acutorostrata and fin whales Balaenoptera physalus) may have switched from mainly feeding on planktonic prey to pelagic fish such as herring (see Nøttestad, Krafft, et al., 2015;Nøttestad, Sivle, Krafft, Langård, et al., 2014), implying possible variations in resource competition (see . Recent studies in seasons and locations not previously investigated have documented new prey types, that is, Atlantic salmon (Salmo salar; Vester & Hammerschmidt, 2013), Atlantic mackerel (Nøttestad, Sivle, Krafft, Langard, et al., 2014), harbor porpoise (Phocoena phocoena; Cosentino, 2015), lumpfish (Cyclopterus lumpus; Jourdain et al., 2019), and pinnipeds Vongraven & Bisther, 2014) for killer whales in Norway, including for individuals known as herring-eaters (see Jourdain et al., 2019Jourdain et al., , 2020. These new observations could be the result of enhanced research effort but could also reflect behavioral responses to a changing marine ecosystem. Recent toxicological assessments, which analyzed both fish specialists and individuals who consumed various proportions of fish and pinnipeds, showed that killer whales in Norway carried higher pollution levels than previously assumed, with possible impact on survival and population growth . Estimates from line-transect surveys in the Northeast Atlantic are insufficiently precise to explore whether killer whale abundance in this area may have changed in the last 20 years (2002-2007: 18,821 and95% CI: 11,525-30,735;2008: 9563 and 95% CI: 4713-19,403 in Leonard & Øien, 2020b: 15,056 and 95% CI: 8423-26,914 in Leonard & Øien, 2020a. To investigate how killer whales may have responded to this period of rapid ecosystem change in the Norwegian Sea, new estimates of population parameters are needed. In this study, we fitted capture-recapture models to a photoidentification dataset spanning a 32-year period to generate population parameters for killer whales in northern Norwegian waters. The objectives were (1) to estimate survival rates for the period 1988-2019, including investigating any difference between time periods (i.e., 1988-2008 and 2012-2019) and possible underlying factors and (2) to estimate the size of the population at recent herring wintering grounds in 2012-2019 for comparison with estimates published for the period 1986-2003 (Kuningas et al., 2014). The overall aim was to improve understanding of how killer whales may respond to shifting prey populations in rapidly changing Arctic marine ecosystems. in response to the later arrival of herring and killer whales in the fjords.

| Study area and data collection
Hereafter, each field season is referred to by the initial year (e.g., winter 2012-2013 is designated as 2012).
In both periods, surveys were carried out opportunistically or using sighting reports obtained from other vessels in the area, with the aim of maximizing the number of killer whales found. A similar approach was maintained in 2013-2014 when whale-watching rigid inflatable boats were used as research platforms. When a group (defined as individuals in apparent association and acting in a coordinated manner during the observation period) was encountered, left-sided identification photographs were taken following the protocols described by Bigg (1982; Figure 2). Efforts focused on photographing as many individuals as possible in each encountered group (hereafter referred to as an encounter), regardless of individuals' size, behavior, or distinctiveness to minimize heterogeneity of capture probabilities (Hammond, 2010).

| Photograph processing
Processing photographs required the films to be inspected using a stereoscopic microscope until 2000, after which digital images were viewed and enhanced in Adobe Photoshop. The three teams of investigators followed similar photo-identification protocols. For each encounter, individuals were identified from left-sided photographs using nicks, shape, and size of the dorsal fin, alongside scarring and pigmentation patterns of the adjacent gray saddle patch as per Bigg (1982; Figure 2). The best photograph of each individual from each encounter was selected and rated for (1) quality (poor, fair, good, excellent) based on combined criteria of sharpness, TA B L E 1 Total numbers of observation days, killer whale encounters (when known), photographs (total collected regardless of quality), individual killer whales identified (IDs), and of fair-to-excellent quality (see Section 2) identifications including resightings, as contributed by the three research teams (DV/AB: Dag Vongraven and Anna Bisther, TS/SK: Tiu Similä and Sanna Kuningas, NOS: Norwegian Orca Survey) and citizen-science (CS) in 1988-2019 in the study area, and which contributed to building the capture histories of the 1236 killer whales included in this study

| Determining sex and age class
Using clear morphological evidence of physical maturity, adult males were identified based on a distinctively taller dorsal fin (Bigg, 1982;Olesiuk et al., 1990; Figure 2). Other individuals of apparent mature size, seen in close and consistent association with a calf (in echelon position) on at least two encounter days, or showing no development of the dorsal fin in at least 3 years, were categorized as adult females ( Figure 2). Individuals for which sex could not be determined were categorized as "unknowns." These individuals could be either subadult males or females, or adult females.

| Assessing ranging patterns
To assess how the study area compared to ranging capacities of killer whales identified from annual winter surveys, we compiled additional photographic records collected from citizen-science in adjacent coastal and offshore regions for these individuals ( Figure 1).

| Data selection
To be considered marked (re-identifiable) and to be retained for analysis, an individual had to have a minimum of one primary feature, defined as (a) at least three scars on the saddle patch; or (b) at least two nicks in the dorsal fin, or a minimum of two secondary features. Secondary features were defined as (i) one or two scars on the saddle patch, (ii) a single nick in the dorsal fin, and (iii) distinctive pigmentation of the saddle patch ( Figure S1). In addition, only of the saddle patch were visible were retained for analysis ( Figure   S1). Individuals (including calves) lacking permanent markings were excluded from all analyses but were used to estimate the proportion of identifiable individuals in the population (see below).

| Cormack-Jolly-Seber (CJS) models
To estimate annual survival probabilities using CJS models (Lebreton et al., 1992), capture histories were built by pooling sightings recorded during the same annual winter season and by treating each year as a sampling occasion. Prior to running models, we ran  (Pradel et al., 1997), and Test 2.CT tests for equal recapture probability between individuals encountered and not encountered in a given sampling occasion (Pradel, 1993 year. Gap years (2009,2010,2011 with no data available) were included in the full time series by fixing recapture probability to 0 for these years. A set of candidate models was constructed in which apparent survival and recapture probabilities were (using conventional notation): constant over time (.), varied annually (t), or displayed a linear temporal trend (T) (Lebreton et al., 1992). In addition to incorporating a temporal trend, we explored the effect of modeling Period 1 and Period 2 as distinct time periods (period). A sex-effect (s) on both survival and recapture probabilities was also tested. GOF Test 2.CT indicated a behavioral ("trap") response (see Section 3) and justified testing the effect of trap-dependence (td) when modeling recapture probabilities. Trap-dependence was implemented using an individual time-varying covariate comprising dummy variables (0 and 1) depending on whether or not an individual was seen on the previous occasion. Lack of fit in GOF Test 3.SR (see Section 3) justified testing the effect of transience on estimates of apparent survival.
This was achieved by building time-since-marking models with two classes (trans), in which survival probability was estimated for the first annual interval after first capture (first class) and also for all subsequent annual intervals (second class). Additive (+) and interactive (*) models were constructed to test for combinations of effects on φ and p. Overdispersion in the data was evaluated by calculating the variance inflation factor (ĉ, "c-hat") as global GOF test X 2 /degrees of freedom (Lebreton et al., 1992).
The probability of apparent survival is the product of surviving from one sampling occasion to the next and of returning to the study area. We investigated whether differences in residency patterns could influence estimates of apparent survival in Period 2 (2012-2019). Residency groups were identified by categorizing individuals following methods described by Schleimer et al. (2019).
Sighting histories of all individuals (males, females, unknowns) in 2012-2019 were used to calculate individuals' yearly (YSR) and seasonal (SSR) sighting rates with: YSR = number of years in which seen/ F I G U R E 2 Sample of identification photographs showing the persistence of scarring and pigmentation patterns of the saddle patch and nicks in the dorsal fin and thus their reliability for long-term re-identification of individual killer whales in Norway. Distinctively taller dorsal fin for NKW-693, compared to NKW-443 and NKW-619 for which dorsal fin did not develop over the course of the study, further illustrates how sex could be readily determined based on morphological features for most identified individuals total number of years since first identification, and SSR = number of days in which seen/total number of days since first identification in a given season. Individuals first identified in 2018 and 2019 were excluded due to insufficient years with sighting data to reliably evaluate residency patterns. Agglomerative Hierarchical Clustering (AHC) was conducted using the hclust function in R to classify individuals based on similar sighting rates. To allow for direct comparison of the two rates, YSR and SSR were standardized (relative to the median and the median absolute deviation) beforehand using the scale function in R. In the AHC, Euclidean distance was chosen as a measure of dissimilarity and to compute proximity matrices between individuals using Ward's method. This clustering method merges the closest individuals (data points) into clusters based on a proximity matrix. The most appropriate number of residency groups was chosen based on obvious main clusters identified visually in the resulting dendrogram.
CJS models were fitted separately to the residency groups identified by the AHC analysis. For each group, both survival and recapture probabilities were allowed to be constant over time (.), vary annually (t), or display a linear temporal trend (T).

| Robust design models
Robust design (RD) models were fitted to the capture histories of all individuals (males, females, unknowns) in Period 2 to estimate the annual number of killer whales using the study area and to evaluate the extent of temporary emigration from the study area between years (Kendall et al., 1997). Each annual winter season (year) was considered as a primary sampling occasion. Each survey area was covered in every week in all years, so weeks within each winter season were treated as secondary sampling occasions (Table 2). Candidate models were built to incorporate effects that were constant over time (.), varied over time (t), had a linear temporal trend (T), and/or a transience effect (trans) on survival probabilities (φ; GOF Test 3.SR was marginally significant-see Section 3). Capture and recapture probabilities were assumed equal in all models (p = c) and were modeled to vary by primary sampling occasion alone (session) or by both primary and secondary sampling occasion (session*time). The probability of temporary emigration from the study area between years (primary occasions) was modeled using the parameters γ′ (probability of being outside the study area conditional on being outside the study area in the previous year) and γ″ (probability of being outside the study area conditional on being inside the study area in the previous year).
γ″ can thus be interpreted as the annual probability of temporary emigration and 1 -γ′ as the annual probability of re-immigration.

| POPAN models
To obtain an alternative estimate of the size of the "superpopulation," defined as the total number of killer whales that was in the study area at some point in time, the POPAN parameterization of the Jolly-Seber model (Schwarz & Arnason, 1996)  for RD models so neither overall model fit nor overdispersion in the data could be assessed. Therefore, AIC C was used to assess relative model fit.

| Proportion of identifiable individuals
The proportion of identifiable individuals in the population in each year, θ, was estimated by fitting a binomial generalized linear model with logit link function to the number of identified and unidentified individuals in encountered groups, where this could be determined.
Total population size was estimated as: where N is the capture-recapture estimate of the number of identifiable animals, with coefficient of variation (CV) estimated using the delta method as: and 95% confidence intervals calculated assuming a log normal distribution as N total /c to c*N total , where:  Figures 2 and 3). Overall, 691

| Data summary
(56%) individuals were seen in two or more years (Table S3; Figure 4).
While the rate of new identifications leveled off toward the end of Period 1, it increased again after fieldwork was resumed at newly established herring wintering grounds from 2012 (Table 1 females, while 276 were categorized as unknowns (Table S4). Such a disproportionate sex ratio in the ID catalogue may be explained by adult males being more identifiable due to their tendency to bear more markings than adult females and subadults; only 6% of the unmarked individuals in 2015-2019 were adult males. In 1988-2019, 71 (6%) killer whales identified at herring wintering grounds had also been photo-identified in at least one adjacent region, in one or multiple years (Table 3; Figure 1).

| Apparent survival rates 1988-2019
GOF tests 2.CT and 3.SR indicated a lack of fit of the CJS model (Table 4)

| Residency groups and survival rates 2012-2019
Out sighting rates, hereafter referred to as the "Low residency group." A Mann-Whitney Wilcoxon test for non-normally distributed data confirmed a significant difference in both yearly (W = 1850.5, p < .001) and seasonal (W = 601, p < .001) sighting rates between the two residency groups. Figure 7 could be interpreted as showing three clusters, rather than two. We estimated apparent survival rates independently for the three indicated clusters. Results (not shown) indicated that estimated survival was the same for two of the clusters and were thus no more informative that the results for two clusters. No lack of fit of the CJS models was detected from the GOF tests run on each of the two residency groups (  Figure 8a). This is explained by all 159 killer whales in this residency group still being alive at the end of this short second study period (2012-2019). Model-averaged estimates of apparent survival were much lower for the whales assigned to the Low residency group (geometric mean: 0.731 ± 0.075 SE; Figure 8a). In this group, individuals also had consistently lower recapture probabilities than the high residency group, confirming reduced fidelity to the area for these whales (Figure 8b).

| Temporary emigration
Robust design models that included random temporary emigration, either constant or time varying, carried most of the AIC weight (74%), although models featuring Markovian temporary emigration also received some support from the data (26% of the AIC weight,  Table 7). Models with no temporary emigration had no support.
Only models with capture probability varying by both primary and secondary sampling occasions were supported. Model-averaged survival probabilities were similar to those obtained with the CJS ( Figure S5a). Model-averaged capture probabilities varied considerably within and between primary periods, ranging from 0.013 (± 0.006 SE) to 0.392 (± 0.030 SE; Figure S5b). Average annual probability of emigration and re-immigration were γ″ = 0.148 (± 0.095 SE) and 1 -γ′ = 0.760 (± 0.215 SE), respectively.

| Population size
When fitted to data from Period 1, POPAN models that accounted for a temporal trend (T) in recruitment from the super-population into the study area (pent) received >87% of the QAICc weight (Table 8). Models with pent(t) were unable to estimate all parameters and were therefore excluded from consideration. When fitted to data from Period 2, POPAN models that included a temporal trend (T) or a time effect (t) on pent received equal support from the data (48% and 51% of the QAICc weight, respectively), while models with constant pent carried low weight (<2% of the QAICc weight; Table 9).
Annual abundance, estimated from the RD models corrected for the proportion of identifiable individuals, peaked in 2015 at 1061 whales (95% CI: 999-1127) and dropped to 513 whales (95% CI: 488-540) in 2018 ( Figure 9; Table S6). Large standard errors for estimates in 2012-2014 indicated low precision, most likely as a result of a relatively small number of identifications for these years ( Figure   S2). Annual abundance estimates obtained from POPAN models were comparable to those obtained from the RD, but less precise (Table S7). The sex ratio (males/females/unknowns) and the number of individuals assigned to each of the high and low residency groups (Period 2) are also shown. Some individuals may have been seen in multiple zones meaning that "Any Zone" is not the sum of the Z-columns TA B L E 3 Number of photo-identified killer whales that were opportunistically photo-identified in adjacent coastal (zones 1-4) and offshore (zone 5) regions in addition to their winter records in the study area. Zones (Z) match region subdivision from Figure 1 TA B L E 4 Results of the four directional goodness-of-fit tests (GOF), the global combined test of overall CJS model fit, and the variance inflation factor (ĉ, "c-hat") calculated as X 2 /degrees of freedom TA B L E 5 Summary of the best-supported candidate CJS models (≤10 ΔQAICc) for 1988-2019 used for model-averaging the probability of apparent survival (φ) and of recapture (p) accounting for additive (+) or interactive (*) effects of time (t), a linear temporal trend (T), sex (s), periods of time (period), transience (trans), and/or trap-dependence (td) In Period 2, and accounting for an average proportion of identifiable individuals of 0.716 (see Table 10), super-population size was 1894 (95% CI: . Other parameters estimated with POPAN models are given in the supplementary material ( Figure S8).

| Validation of method assumptions
To meet the assumption of correct mark recognition, only good quality photographs of reliably marked individuals were used (i.e., with multiple marks, see Figure S1), thus minimizing the risk of erroneous

TA B L E 6
Most-supported CJS model for the High residency group and best-supported candidate models (≤10 ΔQAICc) for the Low residency group used for model-averaging the probability of apparent survival (φ) and of recapture (p), both allowed to be constant (.), vary by time (t) or display a linear temporal trend (T)

F I G U R E 8
Non-sex-specific probabilities of (a) survival and (b) recapture with 95% CI for the High residency group (black) and the Low residency group (gray), as obtained from the best-supported CJS models listed in Table 6. Note: there were no captures in the High residency group in 2012 identifications. In addition, the photo-identification work was carried out by the same experienced analysts throughout the study period, further minimizing inconsistencies during cataloguing and scoring of images (Urian et al., 2015). Lack of CJS model fit to the data was tested to facilitate model development and obtain robust estimates of apparent survival (Gimenez et al., 2018). In the full dataset, Test 2.CT (Table 4) indicated evidence of trap-dependence in recapture probabilities, which was incorporated in the CJS models (Table 5).
In our study, a behavioral trap response is unlikely because killer whales were photographically and not physically captured. Instead, heterogeneity in recapture probabilities could have been generated by sampling a restricted part of the range of the study population, by individual-or sex-specific differences in behavior, or by nonrandom temporary emigration (Pradel & Sanz-Aguilar, 2012). In most datasets, Test 3.SR indicated a transience effect that could have been caused by the presence of transient individuals (Pradel et al., 1997; Table 4). The strong support for time-since-marking models to account for transience (Tables 5 and 7) and the much lower estimates of apparent survival for transient individuals (see Section 3), confirmed that this was an effective way of dealing with this lack of fit of the CJS model.

| Survival rates
In Period 1, estimates of apparent survival probability for adult killer whales of both sexes exceeded 0.98 and were higher for females than males throughout the study period (Figure 6a) Note: For apparent survival, single, additive (+) or interactive (*) effects were modeled as constant (.), time-specific (t), with a linear temporal trend (T) and accounting for transience (trans). All listed supported models had capture probabilities varying by primary and secondary sampling occasion.
TA B L E 7 Summary of the bestsupported candidate models (≤10 ΔQAICc) obtained when fitting robust design models to the dataset 2012-2019 and used for model-averaging the probability of apparent survival (φ), the probability of being outside the study area conditional on being outside the study area in the previous year (γ′) and the probability of being outside the study area conditional on being inside the study area in the previous year (γ″)  ) and used for model-averaging the probability of apparent survival (φ), of recapture (p) and of recruitment into the study area from the super-population (pent), accounting for single, additive (+) or interactive (*) effects of time (t), a linear temporal trend (T), and transience (trans) or set constant (.) Olesiuk et al., 1990Olesiuk et al., , 2005 and with previous analysis of similar data from northern Norway 1986-2003 (Kuningas et al., 2014). A contributing factor to the sex-specific difference in survival is the extended postreproductive lifespan in females (Foster et al., 2012), which results in a longer mean life expectancy at birth for females than males (46 vs. 31 years in Northern Resident killer whales in British Columbia, Olesiuk et al., 2005). From 2012, after the NSS herring established new wintering grounds (Huse et al., 2010), apparent survival dropped for adult females (geometric mean: from 0.998 ± 0.002 SE to 0.955 ± 0.027 SE), and even more so for adult males (geometric mean: from 0.985 ± 0.009 SE to 0.864 ± 0.038 SE; Figure 6a). In such long-lived species, a decrease in survival of this magnitude is highly unlikely to reflect natural variation in mortality but could be indicative of anthropogenic mortality. For example, survival estimates of killer whales at the Crozet Islands dropped from 0.99 to 0.92 (equivalent to an increase in apparent mortality rate from 1 to 8%) after the illegal Patagonian toothfish longline fisheries started in 1996 (Tixier et al., 2017). In this region, where killer whales depredate longlines as a feeding strategy, illegal vessels were reported to have used lethal means to repel the depredating whales, which led to an increased mortality risk (Poncelet et al., 2010;Tixier et al., 2017). In our study region, there is no evidence of increased mortality to explain the decline in apparent survival rates between the two periods. Thus, it is likely that the detected trend was a result of other features of the data.
Our analysis revealed important differences between the two periods. There was a substantially higher number  TA B L E 9 Summary of the bestsupported candidate models (≤10 ΔQAICc) obtained when fitting POPAN models to Period 2 (2012-2019) and used for model-averaging the probability of apparent survival (φ), of recapture (p) and of recruitment into the study area from the super-population (pent), accounting for single, additive (+) or interactive (*) effects of time (

| Movement patterns and abundance
Robust design (RD) models for Period 2 indicated most support for random temporary emigration (Table 7). When random, and not Markovian, temporary emigration is not expected to bias survival estimates in CJS models, explaining the similarity in estimates of apparent survival from the RD and CJS models (Schaub et al., 2004). confirming an offshore origin for at least some transient individuals.
Notably, the single winter records of these whales were from 2015, the peak year of the number of transients and estimated abundance ( Figures 5 and 9), and during which sampling covered open waters northwest of Andøya ( Figure 1). Therefore, it appears likely that high herring abundance in coastal but open areas in some years, rather than in the inshore fjord system, attracted animals from elsewhere (including offshore) that had previously not been available to be sampled. This explanation is supported by the observation that 60% of the transients seen between 2012 and 2018 were identified at Andøya, even though this area contributed <30% of all captures for this period (EJ, unpublished data). The appearance of large numbers of humpback whales (and fin whales) at the newly established herring wintering grounds, which were not observed at former inshore locations in 1986-2006 , lends further support to this explanation.
As the photo-identification study continues in these dynamic  Figure 9).

F I G U R E 9
Model-averaged estimates of the number of identifiable killer whales (gray) and total abundance (corrected for the proportion of identifiable individuals, black), with 95% CI, that used the study area during the winter months between 2012 and 2019, as estimated from the best-supported robust design models listed in Table 7 Estimated annual killer whale abundance peaked at 731 individuals (95% CI: 505-1059, Kuningas et al., 2014)  As discussed above, it seems likely that this increase in abundance resulted from killer whales responding to the shifting of herring wintering grounds from the strictly inshore fjord system throughout Period 1 to both coastal (Vengsøyfjord and Kvaenangen) and open waters (off Andøya) in Period 2 (Huse et al., 2010) (Figure 1).
A recent period of population growth also cannot be ruled out.
Indeed, a demographic rebound following the end of the culling in 1982 (Øien, 1988) and the recovery of the NSS herring after a nearly total collapse in the late 1960s (Dragesund et al., 1997) may have been expected for killer whales in Norway. However, the combination of limited sampling and dynamic herring wintering grounds preclude the estimation of any meaningful trend in abundance, even in the sampled areas. In addition to a rebound to a precommercial fisheries ecosystem, killer whale population dynamics may be further affected by other ecological changes that are influenced by global warming.
For example, the north-eastern Atlantic mackerel, also a prey of killer whales in the study region (Nøttestad, Sivle, Krafft, Langard, et al., 2014), has greatly increased in biomass in the Norwegian Sea (Nøttestad, Utne, et al., 2015). Total super-population size in 2012-2019 represented roughly 10 to 20% of killer whale abundance estimated from Norwegian shipboard surveys (Leonard & Øien, 2020a(Leonard & Øien, , 2020b. The large-scale migration of the NSS herring, which couples offshore and coastal ecosystems, and the documented wide-ranging capacities of killer whales in Norway (Table 5; Dietz et al., 2020;Similä & Stenersen, 2004;Vogel et al., 2021) suggest that the killer whales studied in northern Norway are part of a larger population of the Norwegian Sea and the wider Northeast Atlantic.

| CON CLUS IONS
Re-identification of individuals over multiple decades confirmed capture-recapture analysis as a suitable tool to monitor long-term changes in the dynamics of killer whales in Norway. Our results show that the NSS herring remains a major ecological driver of killer whale dynamics in this region. We show that killer whales adjust their movement to shifting prey resources, indicating potential to adapt to rapidly changing marine ecosystems, as previously shown in other regions (e.g., Canadian Arctic, Ferguson et al., 2010). By shaping killer whale movement patterns, distributional prey shifts may also influence contact zones between killer whale groups, with We also thank Mònica Arso Civil and Anna Schleimer for advice on analysis. Last, we are very grateful to Graeme Ellis who accepted to review killer whale identification photographs and greatly assisted with data validation.