Site fidelity and determinants of wintering decisions in the Dalmatian pelican (Pelecanus crispus)

Wintering site selection is a major decision in a bird’s annual life cycle since conditions experienced by individuals within a wintering site can influence individual fitness and ultimately population dynamics. If individuals show strong fidelity to specific sites, annual environmental effects can be reinforced. In this study we analyse winter resightings of ringed Dalmatian pelicans, collected during a 35-year-long ringing survey, to identify winter movement patterns of Dalmatian pelicans, assess the level of site fidelity both within and among consecutive wintering events and examine environmental factors that may contribute to an individual’s decision to move towards a specific site during the winter period. Our results showed that the decision of Dalmatian pelicans to move towards a specific site was mainly driven by the site’s size and temperature conditions. Larger wetlands can provide birds with higher resource abundance and offer more opportunities to exploit optimal roosting and feeding habitats, whereas wetlands that do not freeze during the winter are a definite requirement. Wintering movements were more likely to occur at short distances from a bird’s natal colony, whereas it was uncommon for birds born in wetlands located in the Adriatic and Ionian coast to move eastwards during the winter, suggesting a level of spatial substructure. Furthermore, we showed that the species exhibits very strong within-year and among-year wintering site fidelity, with birds moving less during the winter, whereas juvenile birds travel greater distances than the rest of the age classes examined. Our conclusions could guide an effective approach to site-based conservation management at key wintering sites.

Wintering site selection is a major decision in a bird's annual life cycle since conditions experienced by individuals within a wintering site can influence individual fitness and ultimately population dynamics. If individuals show strong fidelity to specific sites, annual environmental effects can be reinforced. In this study we analyse winter resightings of ringed Dalmatian pelicans, collected during a 35-yearlong ringing survey, to identify winter movement patterns of Dalmatian pelicans, assess the level of site fidelity both within and among consecutive wintering events and examine environmental factors that may contribute to an individual's decision to move towards a specific site during the winter period. Our results showed that the decision of Dalmatian pelicans to move towards a specific site was mainly driven by the site's size and temperature conditions. Larger wetlands can provide birds with higher resource abundance and offer more opportunities to exploit optimal roosting and feeding habitats, whereas wetlands that do not freeze during the winter are a definite requirement. Wintering movements were more likely to occur at short distances from a bird's natal colony, whereas it was uncommon for birds born in wetlands located in the Adriatic and Ionian coast to move eastwards during the winter, suggesting a level of spatial substructure. Furthermore, we showed that the species exhibits very strong within-year and among-year wintering site fidelity, with birds moving less during the winter, whereas juvenile birds travel greater distances than the rest of the age classes examined. Our conclusions could guide an effective approach to site-based conservation management at key wintering sites.

INTRODUCTION
Wintering site selection is a major decision in a bird's annual life cycle. Conditions experienced by individuals within a wintering site, such as habitat quality or any energetic costs associated with movement distances can influence life-history traits thus impacting individual fitness and ultimately population dynamics (Marra et al. 1998;Harrison et al. 2011;Sedinger et al. 2011). Resource availability is one of the most important factors shaping wintering movements in waterbirds (Ma et al. 2010) but in many cases, factors that are involved in individual wintering site decisions are probably complex and thus still poorly understood. If individuals show strong fidelity to specific sites, annual environmental effects can be reinforced (Gunnarsson et al. 2005). Winter site fidelity, the faithfulness of individual migrants to their wintering grounds is widespread among waterbirds, with some species exhibiting very high levels (Robertson & Cooke 1999). Such high levels of site fidelity are likely to be associated with advantages related to prior knowledge of high-quality habitats, competition avoidance and even social cues (Newton 2010). However, under a rapidly changing environment, the predictable return to the same sites over the years can expose individuals to increasingly suboptimal sites thus poor habitat conditions (Battin 2004;Matthiopoulos et al. 2005). Therefore, quantifying individual variation in site selection both within and among wintering seasons is key to understanding potentially complex environmental effects on population dynamics and how they can be used to design effective conservation actions.
The Dalmatian pelican (Pelecanus crispus) is a migratory waterbird with a distribution limited to the Palaearctic region, ranging from eastern Europe to Mongolia, where it occurs in three distinct flyways, the Black Sea-Mediterranean Flyway, the Central Asian Flyway, and the East Asian Flyway (Catsadorakis & Portolou 2018;BirdLife International 2021). Birds that belong to the Black Sea-Mediterranean flyway are short-distance migrants that overwinter in the wider area of south-eastern Europe, whereas birds that breed in central and eastern Asia are longdistance migrants overwintering in south Asia and east China, respectively (Crivelli et al. 1991;Catsadorakis & Portolou 2018). The conservation status of the Dalmatian pelican has improved recently, with the species being downlisted in 2017 from "Vulnerable" to "Near Threatened" (BirdLife International 2018). Indeed, in the Black Sea-Mediterranean Flyway, implemented conservation measures have resulted in an increase in breeding numbers in SE Europe and Turkey (Catsadorakis et al. 2015), a fact that is also reflected in the positive trend in the abundance of wintering individuals in the region accompanied by changes in their wintering distribution (Barboutis et al. 2021). Although the species' wintering distribution could be partly shaped by breeding population density and local temperature trends, still a comprehensive examination of the factors affecting the wintering population and distribution of Dalmatian pelicans is lacking.
Over the years, capture-mark-recapture methods have been successfully employed to monitor changes in abundance, distribution and demographic parameters of studied organisms (Lindberg 2012;Schmeller et al. 2012). Specifically, bird ringing has provided undeniable insight into the biology of birds over the years, especially on their movements and migration routes (Bairlein 2001). As a result, bird ringing is widely used in many studies concerning the ecology, behaviour, migration and conservation of bird populations (Peach et al. 1999;Baillie 2001). When data from bird ringing efforts cover long periods of time, they additionally hold great promise for inferring spatiotemporal movement patterns, including changes in relation to several environmental factors (Thorup et al. 2014).
The present study is based on a 35-year ringing survey involving winter resightings of ringed Dalmatian pelicans. The overall aim of this study is to quantify the flexibility of wintering behaviour in relation to different factors. Specifically, we analysed winter movement patterns of Dalmatian pelicans and calculated the level of fidelity both within and among consecutive wintering events. Finally, we examined a variety of environmental factors that may contribute to an individual's decision to move towards a specific site during the winter period.

Data collection and filtering
Birds included in this study originated from six of the major breeding colonies at the western edge of the species distribution ( Fig. 1): Karavasta Lagoon, Amvrakikos Gulf, Mikri Prespa Lake, Kerkini Lake, Srebarna Lake and Gediz Delta. All birds included in the analyses were ringed as chicks in the nest with one or two plastic rings bearing a unique alphanumeric code, from 1984 to 2018. Dalmatian pelican chicks do not exhibit any sexual dimorphism therefore their sex could not be determined. Observations of ringed individuals were extracted from To study wintering behaviour, we only kept records of ringed birds resighted between November and January of each wintering year, therefore recording individual encounter histories. This wintering period for the species was defined according to Barboutis et al. (2021). In short, despite that Dalmatian pelican wintering period was formerly set from December to mid-February (Crivelli 1987), there is evidence of a shift in the species breeding phenology with populations breeding earlier thus subsequently affecting the duration of the wintering period (Doxa et al. 2012). Furthermore, the choice of the November-January wintering period is supported by tracking data (Efrat et al. 2019), whereas November weather conditions could be an important determinant of wintering area decisions (Crivelli et al. 1991). After that first filtering step, we retained 2631 observations of 729 individuals. Then, we kept only wintering sites that had more than one recovery to avoid any rare cases. Finally, in the case that an individual had multiple resightings within a year, only one observation that was closest to the middle of the defined wintering period (i.e. 15 December) was kept for further analyses leading to a filtered dataset with 1169 winter resightings of 718 individuals at 16 different wetlands.

Winter movements and site fidelity
Based on the location and date of ringing and recovery we were able to calculate recovery distance, recovery direction, days elapsed and the age of each individual at each year of recapture. All data analyses were performed in R 4.0.5 (R Core Team 2021) using the packages "birdring" (Korner-Nievergelt & Robinson 2014) and "circular" (Agostinelli & Lund 2011). To evaluate to what extent the breeding sites are also used by the species for wintering, we used a homing rate estimate for each population that is not confounded by survival (Robertson & Cooke 1999). Homing rate is defined as the number of birds wintering in their natal sites divided by the total number of birds resighted anywhere else. The analysis was employed for natal sites that had more than 10 birds recovered anywhere. To assess any differences in site fidelity within a single year and among multiple years we calculated the individual site fidelity index (SFI) developed by Catry et al. (2012): where n i is the number of areas used by individual i, n is the total number of areas surveyed, p i is the number of changes between areas performed by individual i and o i is the total number of observation events of individual i. SFI can take values from zero (no site fidelity) to one (complete site fidelity). To avoid any bias in the estimation of the SFI, only individuals observed at least 3 times during the study period were included, and in order to examine if winter site fidelity changes with age progression, we analysed only individuals where their first wintering area was known. Naturally, we used the full dataset when examining within-year fidelity (prior to removing multiple resightings of any individual per year).

Factors affecting wintering decisions
To determine the most important factors affecting the decision of Dalmatian pelicans to move towards a given wintering area, we used random forests, a machine-learning algorithm based on classification tree ensembles (Breiman 2001). These models do not require assumptions such as linear or nonlinear relationships between predictors and response factors and are not subject to overfitting (Breiman 2001;Cutler et al. 2007). We used each wintering site as the dependent variable and a set of 12 variables as predictors (Table 1): an individual's origin along with latitude and longitude of origin area, the year and month of the wintering event, mean temperature and the North Atlantic Oscillation index (NAO) for November to January, distance travelled and direction from the natal colony to each wintering site, Dalmatian pelican annual number of breeding pairs in Greece, area of each wintering wetland and individual identity to explore for any individual strategies in wintering decisions. Specifically, for each wintering year we calculated the mean temperature recorded across November, December and January based on monthly means of NCEP/NCAR Reanalysis Derived data (Kistler et al. 2001; data provided by the NOAA/OAR/ESRL PSL, Boulder, Colorado, USA; https://psl.noaa.gov/data/gridded/data.ncep.rea nalysis.derived.surface.html). Similarly, the NAO index data were provided by the Climate Analysis Section, NCAR, Boulder, USA (Hurrell et al. 2003). As a proxy of the species population density in the region, we used the number of breeding pairs in Greece the year before each wintering event. Long-term population estimates for Greece were readily available and they make a good proxy since Greece holds the species largest colonies thus the majority of the overall number of breeding pairs in south-eastern Europe (Catsadorakis et al. 2015). Finally, data on the size of each wetland used by wintering Dalmatian pelicans were included, as bigger wetlands are expected to reflect higher habitat diversity thus offering pelicans more chances of finding optimal roosting places and higher resource availability. Area for each wetland was extracted from the Freshwater information system for Europe (https://water.europa.eu/freshwater) and the Hydrolakes dataset (Messager et al. 2016).
Models were fitted using the R package "randomForest" (Liaw & Wiener 2002) using 500 trees and three random variables at each split. Model performance was assessed based on the percentage of misclassifications over all the out-of-bag samples (OOB error rate). Furthermore, we used the R package "rfUtilities" (Evans & Murphy 2015) to perform significance tests on the built models. The importance of the predictors was tested for statistical significance (P < 0.05) using a permutation approach (Altmann et al. 2010) as implemented in the R package "pRF" (Chakravarthy 2016). The number of permutations was set to 1000. The permuted importance values of factors influencing the birds' wintering decisions were rescaled from 0 to 100. Finally, to interpret the effect of the most important variables we used partial dependence plots (Elith et al. 2008). We should note that our dataset was imbalanced regarding wintering recoveries; more than 75% of wintering events took place in Kerkini Lake (Fig. 1), probably due to the implementation of long-term standardised monitoring of birds in the site and therefore our wintering sites were not equally represented in the dataset. To address this issue, we implemented the Synthetic Minority Over-sampling Technique (SMOTE; Chawla et al. 2002) through the SCUT method (Agrawal et al. 2015), which combines SMOTE oversampling of the less represented wintering sites and cluster-based undersampling of the Kerkini Lake events to achieve better classification performance of our random forest model. SCUT was implemented in the "scutr" R package (Ganz 2021).

RESULTS
The majority of birds in this study were recaptured once (489 individuals). A total of 127 individuals were encountered in 2 different winter years, 49 birds were encountered in 3 different winter years and 53 birds were encountered in four or more winters with one individual encountered in 13 different wintering years (Figure. S1 in Supplemental Data). One Dalmatian pelican ringed in Mikri Prespa Lake on 10 June 1999 was recovered in Kerkini Lake on 21 November 2017, aged 19 cy. The longest distance recorded involved an individual born in Mikri Prespa Lake and recovered in November of the same calendar year at Nile Delta, Egypt having travelled 1390 km. Overall wintering recovery statistics of Dalmatian pelican ringing can be found in Table S1 in Supplemental Data.
Most of the Dalmatian pelican winter recoveries took place in Kerkini Lake where birds from five breeding colonies were recorded. Movement patterns and directions of wintering movements analysed per population of origin showed limited communication of birds between wetlands located along the Adriatic and Ionian coast and the rest of them ( Fig. 1(b); Figure. S2 in Supplemental Data). Birds originating from Kerkini Lake showed the highest homing rates, i.e 72.9% of recovered birds born in Kerkini Lake also use the area as a wintering site. Increased homing rates (62.5%) were also observed for birds born in Amvrakikos Gulf. On the other hand, only two birds out of 458 (0.4%) that were born in Mikri Prespa Lake were observed to spend the winter there (Table 2). There was a high individual variation of elapsed distances among and within populations ( Figure. S3 in Supplemental Data); Birds born in Karavasta Lagoon winter at close distances (< 200 km), birds born in Gediz Delta frequently travel more than 400 km to wintering sites, whereas birds born in Mikri Prespa Lake show the highest variability in travelled distances. When examining overall travelling distances in relation to a birds' age, we found that birds cover larger distances during their first wintering year (ANOVA: F = 6.78, P < 0.05; Fig. 2). Of the 222 Dalmatian pelicans that had at least three resightings in different years, most individuals always used the same sites exhibiting a mean among-winter SFI of 0.98 (SD = 0.03; range: 0.82-1). Within each year, 442 birds had multiple resightings (at least three during the given year) and similarly showed very high within-year SFI values (0.99 ± 0.009; range: 0.95-1).
The random forest model used to assess any factors involved in Dalmatian pelican wintering site selection performed well with an OOB error rate of 0.17% over 500 trees. Misclassifications were low and concerned only wintering events that took place in Porto Lagos Lagoon and Kerkini Lake (less than 1% for both sites). The area of the wintering wetland was the most important variable in explaining movement decisions, closely followed by temperature (P < 0.001; Fig. 3). Larger wetlands and higher temperatures seem to positively influence winter use of a site (Fig. 4(a)). Movement direction and distance from the natal colony were the next most important variables.
Wintering movements were more likely to occur at short distances from a bird's natal colony, whereas it was fairly uncommon for birds born in wetlands located in the Adriatic and Ionian coast to move eastwards during the winter (Fig. 4(b)). All other variables included in the model were relatively unimportant (~<< 20% of wetland area importance; Fig. 3).

DISCUSSION
The decision of Dalmatian pelicans to move towards a specific site was mainly driven by the site's area and temperature conditions. Larger wetlands can provide birds with higher resource abundance and offer more opportunities to exploit optimal roosting and feeding habitats, so they have been recognised as important factors governing habitat use and winter movements by waterbirds (Ma et al. 2010;van Eerden et al. 2010;Kleyheeg et al. 2017). Indeed, Dalmatian pelicans as piscivorous waterbirds could be benefitted from moving towards areas with a variety of habitat conditions as, for example, different water levels reflecting differential fish abundance (Ma et al. 2010;Aharon-Rotman et al. 2017). Such favourable local conditions could also be an indirect result of reduced human disturbance in larger wetlands, but further research is needed to evaluate any such relationship. The effect of landscape composition in wintering site use seems to work in synergy with temperature. Climate variability is among the main drivers of population and distributional changes of wintering waterbirds (Lehikoinen et al. 2013;Pavón Jordán et al. 2019) thus suitability of wetlands as wintering sites could be temperature-dependent (Adam et al. 2015;Musilová et al. 2015) especially since harsh weather conditions in winter may directly  Table 1. Variables above the dotted line showed statistically significant importance after 1000 permutations. affect feeding conditions and survival (Newton 1998). The apparent expansion of the species' wintering distribution with wintering individuals progressively appearing around the Black Sea Coast after the 90ʹs, has already been attributed to local warming trends, whereas the species' overall breeding population increase could also play a role (Barboutis et al. 2021). We certainly detected a large effect of temperature in wintering site use in this study, reflecting the species biology and even fluctuations in resources; wintering pelicans require adequate areas of unfrozen water for foraging and thus need to move towards warmer sites where also fish abundance will not be limited by low temperatures (Hurst 2007;Jeppesen et al. 2010). This is further indicated in the extremely low homing rate we found for Mikri Prespa Lake, a wetland that freezes almost every winter rendering it highly unsuitable for wintering waterfowl (Catsadorakis 1997).
Limited food availability during the winter can result in lower energy intake, whereas lower temperatures can lead to increased energy expenditure due to thermoregulatory costs (Swanson 2010). Therefore, birds would be expected to maximise energy intake by wintering at warmer sites with high resource abundance, and our results do partially reflect that. However, our models were able to identify that the probability of wintering at a specific site increased when the site was located proximally to the breeding grounds. Of course, this could reflect the short-distance migratory lifestyle of birds wintering in south-eastern Europe, nevertheless, low winter temperatures provide less suitable flight conditions for soaring birds such as the Dalmatian pelican that could result in increased energy expenditure (Pennycuick 2008) thus posing a trade-off between moving or staying at a particular wintering site. The observed homing rate patterns and extremely high within-year fidelity suggest that Dalmatian pelicans move less during the winter. These results clearly support findings by Efrat et al. (2019) who showed that in winter, long flights between wetlands were rarely undertaken and that energy expenditure was lowest compared to other stages of the annual cycle. Further consideration of the high among year fidelity we calculated in this study, would suggest that the species' wintering behaviour could have evolved according to the "local-knowledge" hypothesis as proposed by Robertson and Cooke (1999). The hypothesis posits that individuals that return to a wintering location could benefit from their previously acquired local knowledge by using it to better exploit food resources and therefore increase their overwinter survival, and has been suggested to explain fidelity patterns in other waterbirds (Guillemain et al. 2009;Chambon et al. 2019;VonBank et al. 2021).
Even though individuals tend to overwinter close to their natal colonies, we did find that juveniles cover slightly larger distances during their first wintering event compared to subadult and adult birds. Young birds often show different wintering movements and site fidelity patterns than older ones (Wilson et al. 1991;Lourenço et al. 2016). One scenario that explains the observed age differential movements could be that young individuals travel farther to avoid any costs related to competition with more dominant adult birds (i.e., dominance hypothesis; Gauthreaux et al. 1982). However, under this scenario, we would also expect high variation in among-year fidelity values, which is not the case in our data. The fact that in the fidelity calculations we only included birds that their first wintering decision was known suggests that the first winter experience may play a role in shaping later life wintering decisions a hypothesis previously suggested in other species (Sanz Aguilar et al. 2012;Chambon et al. 2019). It should be considered that travelled distance patterns could be at some level driven by wetland spatial distribution. Not all wetlands are suitable for Dalmatian pelicans to spend the winter, as they might miss important features as reflected in our variable importance assessments, such as high fish abundance, isolated roosting sites, and not low temperatures for prolonged time intervals. It has actually been suggested that inadequate roosting areas can prevent Dalmatian pelicans from using a site (Crivelli et al. 1996).
Our results provide some evidence of decreased connectivity between western and eastern wetlands during the winter. Birds born in the Adriatic/Ionian coastal wetlands exhibit limited eastward winter movements with only seven out of a total of 77 recovered birds wintering east from the coast. Opposite movements are even rarer with only two out of 628 recoveries occurring in coastal wetlands during the winter. This divide is further supported by our random forest model as the direction of movements from natal to wintering sites was found to play a role in winter site decisions. In fact, there is a lower probability of birds following a 90° direction reflecting a low movement probability "barrier" between western and eastern populations ( Fig. 4(b)). These findings support the current evidence proposing the existence of two Dalmatian pelican meta-populations (Adriatic-Ionian and Eastern), separated by the Pindus mountain range, which are subject to different pressures and since they are demographically independent should be treated as two separate large-scale management units (Saveljić & Rubiniʹ 2008;Catsadorakis et al. 2015). Such a divide could shape the structure and dynamics of Dalmatian pelican populations as it has been suggested that non-breeding distributions can shape population genetic structure (Szczys et al. 2017;Bounas et al. 2018). Actually, any consequences of this geographic divide we observed on Dalmatian pelican population structure could be further reinforced by the high winter site fidelity we found. That is because the formation of pair bonds in the wintering sites is a common behaviour for several waterfowl species (Rohwer & Anderson 1988;Rodway 2007), including Dalmatian pelicans (T. Naziridis, A.J. Crivelli personal observations). The fact that Dalmatian pelicans arrive at their breeding grounds in groups rather than individually and readily begin courtship (Crivelli et al. 1998) could indicate that pair prospecting procedures could happen in the wintering areas. Therefore, the combination of geographic barriers and very high within and among-year fidelity to wintering grounds could lead to different individual fitness patterns and non-random segregation of individuals in specific locations thus shaping population structure and dynamics (Robertson & Cooke 1999;Gunnarsson et al. 2005;Grist et al. 2014).
Overall, our results can be used to provide some insights regarding the species' management and conservation. The reliance of individuals in specific wintering areas can have profound consequences to a species long-term population dynamics, as they are more vulnerable to environmental changes in these sites (Wilson et al. 1991;Newton 2010). The example of Kerkini Lake is characteristic; in the 1960s and the 1970s Kerkini was only a small reservoir hosting very few numbers of Dalmatian pelicans during the winter. After the construction of a new dam in 1982, the lake was enlarged and has eventually become a major wintering site for many bird species including pelicans (Crivelli et al. 1995). This change did not only provide short-term benefits to individuals originating from different breeding populations but positively influenced breeding population dynamics in the long term (Crivelli et al. 1996;Catsadorakis et al. 2015). In fact, Kerkini Lake is an inland delta and is one of the most productive wetlands in Greece (Crivelli et al. 1990;Tatarakis 1995), offering pelicans with accessible fish opportunities due to the existence of extensive shallows, whereas at the same time it rarely freezes during winter. The fact that Dalmatian pelicans seem to repeatedly choose the same areas for wintering, should guide an effective approach to site-based conservation management, through the designation of the species' wintering sites as protected areas and habitat change mitigation at key wintering sites. However, given that Dalmatian pelican wintering movements are short-distance ones and are to some level temperaturemediated along with the high observed among year site fidelity raises some concerns about the amount of flexibility in wintering strategies shown by the species and how it will respond to future climate warming scenarios (Lehikoinen et al. 2013