Northern range shift may be due to increased competition induced by protection of species rather than to climate change alone

Abstract Few long‐term, large‐scale studies have been conducted about the factors likely to explain changes in species abundance and distribution in winter. Range shifts are generally attributed to the climate change or land use. This study shows that other factors such as species protection and the ensuing increasing numbers of individuals and competition could be involved. It details the progressive conquest of France, the most important European wintering area for great cormorant, in three decades as its legal protection by the EU Birds Directive. It is based on 13 exhaustive national counts. Cormorants first occupied the farthest areas (Atlantic and Mediterranean lagoons, then larger rivers) from the main‐core European breeding area, with only progressive occupancy of the northeastern part later. This strategy mainly resulted from competition for optimal available feeding areas. Suboptimal areas (smaller wetlands harboring smaller night roosts, colder northeastern French areas) and progressive fragmentation of large night roosts into smaller, better located ones minimized flight costs. The coldest areas were occupied last, once other areas were saturated. Their occupancy was favored locally by the global climate change, but it played a minor role in these strategies. Both factors induced only a small NNE shift of the weighted centroid range of the wintering population (2.6 km/year) which mainly resulted from competition (buffer effect). Only the 2009 cold wave decreased the total number of wintering cormorants at the national scale, once the population had probably reached the carrying capacity of the country, while the previous cold waves had a minor effect. Comparatively, there was a greater SSE range shift of the weighted centroid of the breeding population (4.66 km/year). Range shifts of other recently protected species have been attributed to the sole climate change in the literature, but competition due to the saturation of usual wintering or breeding areas should be considered too.

Therefore, a complementary study at the species level integrating the individual long-term ecological record of the species at a large geographical scale is required. It is more particularly needed in the case of species showing a strong new population dynamics or recovering previous distribution areas as a result of species and/or habitat protection. This is even more relevant since the implementation of the U.S. Migratory Bird Treaty Act Protected Species (1972) and Bird Directive of the European Union (1979).
The continental subspecies of the great cormorant Phalacrocorax carbo sinensis has underwent one of the most greatest bird population booms and dispersals (along with the double-crested cormorant Ph. auritus in the U.S., Wires & Cuthbert, 2006) in less than three decades and is therefore an interesting model for testing these changes (Marion & Le Gentil, 2006). In Europe, the marine subspecies Ph. c. carbo did not globally change its breeding distribution on the marine coasts of Iceland, Norway, the British Isles and northwestern France (with a newly described marine subspecies from Norway to France, Ph. C. norvegicus, Marion & Le Gentil, 2006). In contrast, protection and the increase in food resources due to eutrophication of waters caused the continental P. c. sinensis population to sharply increase. Once located in the refuge of the core area of the Baltic sea and the Netherlands, it extended to a large part of Europe (from around 5,000 breeding pairs in 1970 in northwest Europe to 191,000 in 2012, Van Eerden & Munsterman, 1995;Marion, 1997a, 2014b, Bregnballe et al. 2013, and its wintering population increased concurrently from probably less than 15,000 individuals in 1970 to 600,000 in 2013 in northwest Europe and north Africa. These wintering birds generated new breeding populations within the wintering area (De Juana & Garcia, 2015;Marion, 1997a), probably in relation to progressive saturation of the northern native breeding areas (Marion & Le Gentil, 2006).
France is the main country for wintering and migrating cormorants in Europe. They mainly come from northern Europe (Frederiksen, Korner-Nievergelt, Marion, & Bregnballe, 2018;Marion, 1995), and France is the only European country where their populations have been monitored by exhaustive and regular national winter counts using the same method (this is essential to describe changes in bird populations according to Elmberg et al., 2014). As such, France is particularly adapted for a long-term study (33 years) of the rapid changes in cormorant distribution and numbers. In the daytime, cormorants are dispersed in groups or individually in the feeding areas (web of rivers, lakes, lagoons, seacoasts). By contrast, they are very social at night: they join a few permanent night roosts ( Figure 1) which are generally close to or surrounded by water in sites preferentially undisturbed by humans (safety behavior, Marion, 1995). Similarly to colonies of herons, the radius of the feeding area and thus the mean distance of feeding flights (and associated cost of flight) are thought to be related to the number of birds in the roost, which varies from few birds to more than 2,000 (Marion, 1984(Marion, , 1997a. The aims of this study were (a) to summarize the strategy of the conquest of France by wintering cormorants at different geographical scales, and (b) to test if changes were temperature dependent in the context of the global climate change or mere consequences of the global protection of the species in Europe since the 1970s, because protection induced a strong increase of the population and thereby probably F I G U R E 1 Night roost of great cormorants (P. Marion) increased competition for optimal areas. We tested (a) if the progressive occupancy of the country was geographically oriented; (b) if the optimal areas were occupied first until saturation and before suboptimal areas; (c) which respective roles temperature and water surface areas played in the characterization of optimal wintering areas for the species; (d) whether there was a change in the strategy of cormorant distribution (roost size) to minimize costs of flight and competition.

| ME THODS
Night roosts are generally occupied every year (which facilitates monitoring) except when strong disturbances occur, and most of them gather tens to hundreds of noisy cormorants. Night roosts with less than 10 birds are very rare. To discover the few new night roosts during each national census, counters followed the direction of straight flights of cormorants that flew directly to their night roost, and/or inspected the banks of rivers, lakes or sea coasts. Over 13 winters (named from each January) from 1983 to 2015, national censuses of all the night roosts were coordinated (by Loïc Marion, excepted in 1983 by Eric Pasquet) in the evening or more rarely early in the morning, with a network of counters who knew the local distribution of roosts very well. The roosts were counted simultaneously in mid-January, when migration movements are usually minor. Such counts of night roosts were chosen to avoid underestimation (up to 50%, Marion, 1997c) of daytime counts in feeding areas through the International Waterbird Census and/or in day roosts. However, daytime counts were used in a few rare cases when night roosts were not counted because there were not enough counters or weather was foggy. When neither night nor daytime data were available in a given year, night roosts sizes were estimated by comparing numbers with those of neighboring roosts or neighboring "departments" (administrative French districts) between two successive national censuses. Such estimates only represented 0.5% to 2.87% of the national number of wintering cormorants (Marion, 2012a(Marion, , 2014a(Marion, , 2015a. Data were detailed in national reports for the French Ministry of the Environment (Pasquet, 1983;Marion, 1991Marion, , 1997cMarion, , 1999Marion, , 2001Marion, , 2003Marion, , 2005Marion, , 2007Marion, , 2009Marion, , 2012aMarion, , 2014aMarion, ,b, 2015a. When normality was not present, we used nonparametric tests. The role of temperature on the mean size of the wintering population of cormorants in January was calculated by comparing the average number of wintering cormorants in each department over the 13 censuses with the average temperature in January in each department (Spearman test, climate data from Météo France). The role of temperature in the chronology of the geographical conquest of France by wintering cormorants was tested by classifying the date of appearance of the first night roost in each department into six classes (up to 1983, 1989, 1992, 1997, 1999, and 2001), using ANOVA for independent series and then grouping 1989 and 1992 together. The role of competition for optimal feeding areas in the chronology of occupancy of the departments was tested using their water surface using Carthage database, version 2013. Optimal cormorant feeding areas are large open waters such as coastal lagoons, lakes, large rivers, and estuaries, while the secondary web of rivers only represents suboptimal feeding areas, as regards the fish stock and safety from humans generally present on the banks (Marion, 1983(Marion, , 1995. According to the theory of competition among animals (Krebs & Davis, 1978), suboptimal areas are progressively used only after saturation of optimal feeding areas (Marion, 1997b). So we compared the correlations between the numbers of cormorants in each department and the water surface area, small rivers <50 m wide (suboptimal areas) excluded, over the successive national censuses. A decreasing correlation with time meant an increasing buffer effect of dominated birds rejected toward the suboptimal areas, owing to increasing competition and saturation in optimal areas. The precise distribution of cormorants in the feeding areas in the daytime was not available, so we did not directly correlate the number of cormorants with the types of feeding areas.
We investigated the global relations between cormorant numbers according to years, departments, water surface areas, and annual temperature in January using Generalized Mixed Effect Models (GLMs, R package "lme4"). More precisely, GLMs assuming Poisson's distribution were used to test if cormorant numbers were related to water surface area (fixed effect), temperature (fixed effect), year (fixed effect), and department (random effect). We then calculated the conditional and marginal pseudo-R 2 coefficients of determination of GLMs to represent the variance explained by both fixed and random factors (R package "MuMin"). Analyses of variance (ANOVA) of the GLMs were performed using type 3-ANOVA (R package "car"), and associated p-values were calculated to test the significance of each explanatory variable. These statistical analyses were performed in the free and open-source R platform (R version 3.4.2, R Core Team).

| Population trends
The numbers of wintering cormorants in mid-January strongly increased between 1983 and 1992, and then progressively leveled off afterward, with a temporary decrease during the 2009 cold wave ( Figure 3). The number of night roosts in mid-January was globally correlated to the number of cormorants (Pearson's r = 0.929, p < 0.0013, R 2 86.2%), but did not decrease in 2009.

| Changes in distribution at different scales
In France, wintering cormorants mainly originated from northern European countries (continental subspecies from The Netherlands and Baltic countries, and marine subspecies from the UK, Marion, 1995). Strangely enough, they conquered France preferentially via the more distant wintering areas, that is, the Mediterranean and

| Change of the weighted geographical centroid of the population
The weighted centroid of the wintering population in January

| Were optimal areas occupied first until saturation and before suboptimal areas?
At a large scale (three coastal areas and the inland area), the inland area was mainly occupied after the saturation of the south and west sea-coasts (optimal areas) and then the northern coast ( Figure 4). At a more detailed scale, the strategy of occupancy of the 15 main watersheds between 1983 and 2015 was similar ( Figure 7). There was a rapid conquest of the areas representing the main eastern migration route mainly used by Baltic birds (n°6-11-12), and Brittany (n°2) representing the western migration route used by birds from The NL and the UK. All these populations reached their peak of numbers as early as 1992 and then declined, sometimes sharply (n°2-11-12-15). The 1992 peak was also recorded in areas n°3, 7, and 13, followed by a decrease in 1997 and a subsequent increase. By contrast, the rapid transfer of birds from the declining areas or the additional migrating birds off after the decrease that followed the 1990s peak in numbers (n°2-4-6-11-12), and (b) areas where the numbers of wintering cormorants regularly decreased (n°5-8-9), or conversely (c) those in which the numbers regularly increased (n°7-10-13-14 and above all n°1 = Nord-Picardie, which was first avoided but finally represented the third most occupied area in France). The slope of the mean temperature in January during the studied years was not significantly different from 0 (p < 0.84). The mean number of wintering cormorants between 1983 and 2015 in each department was little correlated to this mean T° (Figure 9, Spearman's r = 0.257, p < 0.013), with a low R 2 (6.6%).
Nevertheless, the departments occupied as early as 1983 were significantly warmer (mean 1983-2015 January temperature 4.87°C) than those occupied later between 1997 and 1999 (3.51°C, ANOVA:

| Global analysis of the respective factors involved in cormorant distribution: geographical area, year, water surface area, and temperature
GLMs results showed that year, water surface area, and temperature explained 43.95% of the variance, and this percentage increased up to 71.93% when taking into account the department' random effect (Model 1, Table 1). GLMs (Models 2 to 7) showed that (a) year, (b) department, (c) water surface area, and (d) temperature (in decreasing order) mainly explained the size of the populations. Models 5-6-7 showed that year, wetland surface area, and temperature explained 37.84%, 13.68%, and 0.44% of the variance, respectively. For each GLM model (from 1 to 7), all tested variables were associated to significant p-value (<0.001) following type-3 ANOVA tests.

15CORSE
was also related to water surface areas (13.68%) and therefore to competition for optimal habitats. Although the role of temperature (0.44%) was significant, it should be mitigated to explain the shift of cormorant distribution across France. This being explained, we finally discuss the role of protecting areas, shooting, and behavior in roost atomization.

| Mechanism of population expansion and role of competition for optimal habitats
In accordance with the Island biogeography theory of MacArthur and Wilson (1963) and subsequent studies about metapopulation models (Levins, 1969;Wiens, 1976), new areas are generally conquered via successive, more and more distant steps from the starting point. There is a hierarchy in the chronology of occupancy from optimal to suboptimal areas, with suboptimal areas being becoming occupied once optimal areas are saturated (Krebs & Davies 1978, Marion, 1984, 1989, 1997bRushing et al., 2015). The Island theory concerns breeding populations, with new individuals (propagules) produced in the successive conquered zones that in turn participate to progressive geographical expansion. The case of wintering cormorants is different because they usually return to their native area located in northern Europe each following spring. Among birds in general the wintering area is usually quite far from the breeding area for climatic reasons. However, we can admit that a similar pattern to the Island theory could occur with the progressive geographical conquest of potential (more distant) wintering areas by cormorants for obvious energy costs of flight because the breeding and wintering populations successively increased in Europe after protection of cormorants by the EC Bird Directive in 1979 (Marion, 1997b;Van Eerden & Munsterman, 1995), as numerous bird species did once legally protected.
Moreover, these wintering cormorants also generated new inland breeding populations, notably in France (up to 118 colonies totaling 7,248 breeding pairs in France in 2015, in addition to the oldest marine population that reached 2,124 breeding pairs distributed across F I G U R E 8 Historical record of the occupancy of each of the 95 French departments from 1983 to 2015. The color used for the first national census in 1983 shows the presence of wintering cormorants at this date; it remains the same (similarly to the other colors of the left column in the legend) for the following censuses as long as cormorant numbers kept increasing. The colors of the right column were used when the numbers decreased by >10% (and remained the same for the following censuses as long as cormorant numbers kept decreasing). The sign "=" was used when numbers leveled off (±10%). A following increase or decrease >10% after such a leveling off is shown by a color change in the corresponding year contributed little (an estimated 14%) to the wintering population in France because part of the birds wintered abroad . These colonies essentially occupied the northern half of France, where the ecological niche of temperature for breeders (a relatively cold climate) favored this situation (Marion, 2014b).
The results of the present study do not follow the Island theory: increase in food resources? One explanation is that the southern and western optimal wintering areas, which had larger wetland surfaces, became saturated, and this induced an increasing shift to suboptimal northern areas where the availability of food resources was lower (smaller water surfaces) and more variable (cold wave, see below).
The cormorants displaced to northern areas were probably young birds (bird dominance hypothesis, Gauthreaux, 1982; see also Rushing et al., 2015 for American redstarts Setophaga ruticilla), supposed to be socially dominated by adults. The recent native adult French cormorants of the Lac de Grand-Lieu did indeed migrate further south than young birds ). Yet, Van Eerden and Munsterman (1995) and Bregnballe et al. (1997) observed the opposite about Dutch and Danish cormorants in the 1980s and 1990s: adult males remained close to the breeding grounds, whereas juvenile females migrated further south to the Mediterranean region.
Such an increase in wintering in northern areas was also recently observed in increasing populations of other birds with an extended wintering range such as the greylag goose (Anser anser) or the great crested grebe (Podiceps cristatus) in The Netherlands (Adriaensen et al., 1993;Loonen & De Vries, 1995;Nilsson, 2006), or the European spoonbill in France (Caupenne & Marion, 2015); for this latter species, it could also be due to saturation of southern wintering areas, and to the climate change that decreased the risk of wintering northernmost for many species (Fiedler, 2003;Issa & Muller, 2015). The survival rate of spoonbills became lower in the usual main southernmost wintering areas than in the northernmost wintering areas (Lok, Overdijk, Tinbergen, & Piersma, 2011).

| Changes in migrating routes and wintering areas due to competition between geographical populations
The decrease in wintering cormorants along the Atlantic coast after 1992 (Figure 7), although in an optimal area for the species, participated to the NNE shift range of the national population and was probably due to competition between the two marine and inland subspecies. Danish cormorants initially migrated mainly through eastern routes in the 1980s and 1990s, but migrated further west afterwards (Frederiksen et al., 2018;Marion, 1995). Maybe there where under the pressure of competition with new northeastern populations migrating for the first time, which could also partly explain the decreasing numbers of cormorants in Tunisia. In western France, Danish cormorants competed with Dutch cormorants, which had previously evicted British cormorants Phalacrocorax carbo carbo from the French sandy Atlantic coasts (Marion, 1983(Marion, , 1994(Marion, , 1995.
Unlike the carbo subspecies, sinensis birds did not use the marine habitat itself (Marion, 1983(Marion, , 1995, so the available feeding area was reduced, and consequently the number of cormorants. Such fluctuating competition between subspecies and probably geographical origins within a same subspecies (Danish and Dutch populations) has not been observed in other bird species. It was also recorded in the breeders of the pioneering and largest French colony of cormorants at the Lac de Grand-Lieu (Loire Atlantique department). The colony was initially created by wintering birds belonging to the continental Ph. c. sinensis subspecies, but rapidly invaded by the marine subspecies Ph. c. carbo (Marion & Le Gentil, 2006). The situation became even more complex with the recent use of inland areas up to 300 km from the sea by the marine subspecies  during wintering (Fonteneau & Marion, 2011;Fonteneau, Paillisson, & Marion, 2009 (Troya & Bernués, 1990). An increase in sinensis birds along the western migrating route was also observed down to Morocco where they competed with African populations of Phalacrocoras carbo maroccanus and Ph. C. lucidus (A. Qninba, pers. com.). Changes in migrating routes have rarely been observed in other birds (Gramet & Dubaille, 1983;Merkel & Merkel, 1983;Feare, 1994;Marion & Marion, 1994;Mewes, 1996;Schmidt, 1998;Fiedler, 2003;Todte et al., 2010;Ławicki, 2014). Europe (Maclean et al., 2008;Pavón-Jordán et al., 2015) and 177 bird species in the USA; 79 bird species even shifted south (Niven & Butcher, 2009) in 1996-1997, 87 in 2002-2003, and 80 in 2008-2009)

| Role of protected areas and shooting
Additionally, the increasing numbers of protected areas (huntingfree reserves) in France over the study period did not explain the increase in the wintering population of cormorant, although it did so for some species (Madsen & Fox, 1997). Protected areas played a minor role in the distribution of winter roosts, and their role was more marked in breeding colonies (Marion, 2015b). A complementary analysis will be needed about fish resources to assess the carrying capacity of the country (Marion, in prep.), to try and explain why the level of 100,000 cormorants has been (momentarily?) overrun since 2013. The shooting of cormorants in France, from 5.55% of the wintering population (in January) in 1996 to 43.51% in 2013, had no effect on the dynamics of the wintering populations from winter to winter at the department scale (Marion, 2012b;Marion, in prep.).

| Role of behavior in roost atomization
The behavior of cormorants (social attraction according to their geographical origin, fidelity to their wintering site, competition between sexes and ages, decreasing fear of humans…cf. Marion, 1994Marion, , 1995Reymond & Zuchuat, 1995;Van Eerden & Munsterman, 1995) also played an important role in their strategies of habitat use. First (until 1989), the mean size of night roosts strongly increased, then the situation reversed (Figure 11), while the increasing rate of the number of roosts became higher than the rate of the number of cormorants (until 2011). This "atomization strategy" (decrease of the biggest roosts, Figure 2) followed the same pattern as the numbers of breeding grey heron colonies in France after the species was first protected in 1975 (Marion, 1997b;Marion & Marion, 1987;Marion, Van Vessem, & Ulenaers, 2000). When wintering cormorants (or breeding grey heron) occupied France again once they were legally protected, safety still outweighed for a time all the other factors for the priority choice of big pre-existing roosts. These roosts played the role of a safety index as compared to creating new but potentially unsafe roosts in vacant areas. When the birds became progressively less afraid of humans and used less safety sites, the biggest roosts were gradually fragmented into many smaller and smaller roosts.
Smaller roosts were better adapted to the distribution of feeding resources, with optimized costs of feeding trips. This strategy allowed for a strong increase in the wintering or breeding populations (Marion, 1997a,b;Marion & Marion, 1987). Paradoxically, the protection of these species caused their biggest roosts or colonies initially used as refuges to decrease.

| CON CLUS ION
The case of wintering cormorants shows that within a few decades protection can logically induce an increase in the population that generates increasing competition for optimal feeding areas. The

ACK N OWLED G M ENTS
We thank all of the 1,500 counters belonging to more than 300 organizations who kindly helped with counting night cormorant roosts during the 13 national censuses in France. We wish to thank Annie Buchwalter for linguistic improvements of the manuscript. The last nine national censuses were supported by the French Ministry of Ecology. The present study was supported by a grant from the SESLG. Climate data were provided by Météo-France.

AUTH O R S CO NTR I B UTI O N
LM conceived and designed the study, collected data, performed analyses except GLMs and wrote the manuscript and made figures; BB performed GLMs analyses.

CO N FLI C T O F I NTE R E S T
None declared.

DATA ACCE SS I B I LIT Y
Species data: available through Dryad (http://datadryad.org/).