Stochasticity in Bird Habitat Selection: Year-to-Year Changes in Territory Locations in a Boreal Forest Bird Assemblage

We studied the degree of apparently stochastic variation in the location of the breeding territories of 17 species of birds in a 36-ha study plot in a heterogeneous coniferous forest in southern Finland over six consecutive years; 'apparent stochasticity' was defined as the sum effect of factors that are contingent relative to structural attributes of the habitats as they are measured by human observers. The data were collected by territory mapping, with 50 × 50 m spatial resolution, and the analysis involved three steps: (1) calculation of frequency of use of each 50 × 50 m grid cell in the plot by each species over the six years; (2) construction of logistic regression models (using GLIM) to describe the habitat attributes of cells favoured by each species in the plot; and (3) exploration of stochasticity by analysing how much random models need to be constrained to generate frequency distributions of grid cell occupancy that resemble those observed - for this purpose we used both binomial probabilities and Monte Carlo simulations. For a majority of the species a sufficient constraint was avoidance of such grid cells not used by the birds during the six years at all: the fit between expectations and observations improved in the binomial model, and improved markedly in the simulation model which also included an adjacency constraint and a weak preference constraint on the 'centroid' of the simulated territories. Six of the 17 species showed stronger preference for particular grid cells than predicted by the avoidance models; four of them presumably for nest site and nesting/foraging substrate (Dendrocopos major, Ficedula hypoleuca, Parus major, Certhia familiaris), and two for a particular habitat (Phylloscopus trochilus, Regulus regulus). Four of these species occupied in years with low population size predominantly such grid cells that were occupied in the other years, in accordance with the 'buffer effect' idea. Overall, however, in 11 of the 17 species the territory locations seemed to vary stochastically from year to year once the grid cells not used at all during the study years were excluded. We discuss theoretical and methodological implications of this conclusion for bird habitat relationships.