Ontogenetic niche shifts as a driver of seasonal migration

Ontogenetic niche shifts have helped to understand population dynamics. Here we show that ontogenetic niche shifts also offer an explanation, complementary to traditional concepts, as to why certain species show seasonal migration. We describe how demographic processes (survival, reproduction and migration) and associated ecological requirements of species may change with ontogenetic stage (juvenile, adult) and across the migratory range (breeding, non-breeding). We apply this concept to widely different species (dark-bellied brent geese (Branta b. bernicla), humpback whales (Megaptera novaeangliae) and migratory Pacific salmon (Oncorhynchus gorbuscha) to check the generality of this hypothesis. Consistent with the idea that ontogenetic niche shifts are an important driver of seasonal migration, we find that growth and survival of juvenile life stages profit most from ecological conditions that are specific to breeding areas. We suggest that matrix population modelling techniques are promising to detect the importance of the ontogenetic niche shifts in maintaining migratory strategies. As a proof of concept, we applied a first analysis to resident, partial migratory and fully migratory populations of barnacle geese (Branta leucopsis). We argue that recognition of the costs and benefits of migration, and how these vary with life stages, is important to understand and conserve migration under global environmental change. Electronic supplementary material The online version of this article (10.1007/s00442-020-04682-0) contains supplementary material, which is available to authorized users.


Defining the different time steps and life stages
In order to be able to compare the different populations of the barnacle goose, we divided the year into the different time steps recognized in the model (Fig. S1). Time step 1 (the beginning of the breeding period, in which chicks are produced) ranged from the 15 th of May to the 15 th of June. Time step 2 (the remaining breeding period until migration to the wintering range) ranged from the 15 th of June to the 15 th of August. Time step 3 (migration towards the wintering range) was estimated from 15 th of August to the 31 st of October. Time step 4 (wintering) ranged from the 31 st of October to the 15 th of March. Finally, time step 5 (migration towards the breeding grounds) ranged from 15 th of March to the 15 th of May. Note that this division of periods in solely for comparison between the different strategies. The absolute time ranges are likely to differ between the different strategies. The long-distance migrants have to migrate further than the short-distance migrants, making migration periods for this population longer. This is not taken into account here.
The different life stages were distinguished according to the model. The young produced during time step 1 were the number of eggs laid by an individual female. The period in time step 2 ranged from eggs to the point of migration (including a post-fledging period). One month of this period was considered survival until fledging. The other month was considered to be post-fledging survival. Subadults were considered as the chicks which were still alive after migration. Even though barnacle geese do not breed until they are 2 of 3 years old, we did not consider the subadults which were left at the wintering grounds. Firstly, since those subadults migrate with the adults to the breeding range and secondly because data on this group of birds is not widely available.

Data used and calculation of vital rates
Parameters in the model were established using vital rates from different sources of literature. For the Netherlands, data originated from van der Jeugd (2013). These data were collected over a period from 2004 to 2012. For this population yearly survival was available (0.88). However, there was no separate data for the survival of different life stages or periods of the year. We did distinguish between a hunting and non-hunting period. Since 2005-2006 this population is hunted during the summer months (time steps 1, 2, 3 and 5). During the winter months hunting is prohibited. Since we have data available from years in which the geese were not hunted (2004)(2005)(2006) and from the years in which the geese were hunted (2006 onwards), we have two different annual survival rates: with hunting the annual survival is 0.8346. Without hunting the annual survival is much higher (0.9543). We used the annual survival with hunting to calculate survival during the time steps 1, 2, 3, and 5, as follows: 0.8346 12 � , where x is the number of months associated with the different time steps (for time step 1: x=1, for time step 2: x=2, for time step 3: x=2.5, for time step 5: x=2). This gave the following survival rates for the different periods: 1: 0.985, 2: 0.970, 3: 0.963, 5: 0.970). We used annual survival without hunting to calculate survival during time step 4 (the winter period, which lasts 4.5 months), by 0.9543 4.5 12 � , which gives a survival during time step 4 of 0.983. Since there was no data on subadult survival during the months after fletching, we also used annual survival of the whole population to calculate subadult survival in time step 3 (0.963), as well as to calculate survival during the last phase of the breeding period, when fledging already took place. The number of eggs (in time step 1) was obtained from one wellstudied population (  . Again, no data was available on survival per area. Therefore, we used annual adult survival from the first winter onwards (0.861). We calculated the survival per time step in the same way we did for the Baltic population, which led to the following survival rates for the adults in the different time steps: 1: 0.988, 2: 0.975, 3: 0.969, 4: 0.945, 5: 0.975.

Calculations population dynamics
All the vital rates calculated above were put into matrices (Fig. 5). These matrices were used to determine the population dynamics (Fig. S1). At the start of the calculations, we assigned a number of individuals to R0 (100 in this case, but the exact value is of no influence on the outcome of the model). With this number and the matrix of time step 1 (Fig. 5), we could calculate R1 and J1. S1 remained 0, since we did not consider subadults remaining at the wintering grounds. After that, we calculated J2 and R2, by multiplying those with the matrix of time step 2. We continued like this until time step 5. The results of time step 5 were then inserted in the first time step, for a calculation of the second year and so on. We calculated λ, which is a value indicating the growth (when >1) or decline (when <1) of a population from one year to the next. This was calculated by dividing the total number of individuals (the sum over all life stages) in one time step in one year with the total number of individuals in that time step the previous year. All different populations had a λ which was larger than 1, indicating population growth. For the resident population, λ was 1.139, for the Baltic population 1.157 and for the Russian population 1.034.
Sensitivity (= change in λ by a change in the parameter) and elasticity (= relative change in λ by a relative change in a parameter) of λ were calculated for a 1% change in the matrix elements.

Fig. S1
Example life cycle bird: brent goose. The main interactions in the ecosystem contexts are indicated. The arrow indicating the fecundity is both blue and green, since an important carry-over effect has been established in brent geese, in which the ecosystem context of the wintering grounds strongly impacts the reproductive output on the breeding grounds. Interactions in the ecosystem context are indicated with arrows.

Fig. S2
Example life cycle fish: Pacific salmon (Oncorhynchus spp.). Important interactions are indicated. Reproductive adult salmon die after spawing in the freshwater spawning streams. This lack of survival is compensated for by high fecundity. However, survival of reproductive adults salmon, as well as migration back to the oceans are lacking from the diagram.

Table S1
Sensitivity (= change in λ by a change in the parameter) and elasticity (= relative change in λ by a relative change in a parameter) of λ for a 1% change in the matrix elements. See Fig. 2