Spatial variation in population size is affected by many factors, which makes it hard to evaluate the appropriateness of empirical models of population sizes or range dynamics. To complicate matters further in dynamic and spatially structured populations, such as metapopulations, spatial interactions via dispersal as well as local extinctions and colonizations confound the effects of environmental factors. Additionally, while a wealth of ”coarse” environmental data are available for most terrestrial ecosystems it is difficult to know how adequate such data are for explaining abundance compared to situations where more detailed habitat and demographic data are also available. The acquisition of such detailed data is costly and might not be available in reasonable time, but can at the same time be crucial for useful modelling outcomes.
To quantify the performance of models incorporating different environmental and demographic data, we analyzed sixteen years of population surveys of the Glanville fritillary butterfly (Melitaea cinxia) in the Åland Islands archipelago in Finland. We concentrated on the relative role of landscape heterogeneity and habitat quality in population dynamics of the butterfly as contrasted to the spatial configuration of habitat — issues of some contention between landscape and metapopulation ecology. For landscape effects we used habitat land use as well as distance-based and multi-scale measures of surrounding land use and topography. Habitat quality incorporated different measures of host-plant abundance and vegetation quality in the dry meadows that host the larval stages of the butterfly. To quantify the role of different sources of variation we grouped explanatory variables into (meta)population, habitat quality and landscape structure -related variables. We then assessed their relative importance by comparing spatiotemporal random effects models incorporating permutations of these variable groupings as covariates. The models were implemented using INLA. To mimic situations where only less detailed demographic data are available, we compared the results of the modelling with abundance data to presence–absence data.
Models incorporating both measures of habitat quality and spatial configuration of the habitat performed better than combinations with landscape structure, though the model incorporating all classes of covariates still performed best. These results further confirm that for species with short generation times and life stages involving specialized habitat, some data on habitat and demography is essential. From the viewpoint of metapopulation ecology, these results show that large scale occupancy patterns and abundance are predicted well by metapopulation theory even in a highly heterogeneous insular environment. Nonetheless, the inclusion of landscape structure in spatial population models improves the predictive capability of simpler models and influences estimates of metapopulation persistence.