Predicting Species Occurrence with Habitat Network Models: Why Topological Placement Matters

Authors and Affiliations: 

Damian Ortiz-Rodríguez (1,2), Maarten J. van Strien (2), Adrienne Grêt-Regamey (2), Antoine Guisan (3), Rolf Holderegger (4)

(1) Biodiversity and Conservation Biology Research Unit, Swiss Federal Research Institute WSL, Birmensdorf (Switzerland)
(2) Planning of Landscape and Urban Systems PLUS, Institute for Spatial and Landscape Development, ETH Zürich (Switzerland)
(3) Spatial Ecology Group, Department of Ecology and Evolution, University of Lausanne (Switzerland)

Corresponding author: 
Damian Ortiz-Rodríguez

In fragmented, human-dominated landscapes, both suitability and connectivity of animal habitats can be significantly affected by anthropogenic developments. In such spatial situation, it becomes important for conservation and landscape planning to have access to models that can predict the occurrences of species under different development scenarios.
Although species distribution models successfully predict habitat suitability, the occurrence of species depends not only on the suitability of a habitat patch itself, but also on factors such as its connectedness to surrounding patches and its specific topological placement within the entire habitat network. All these factors can be addressed by a predictive network modelling approach, yet very few studies have explored the potential of using habitat networks to predict species occurrence.
We present the first results of a network-based model developed to predict the occurrence status of a species in all the patches of its habitat network (i.e., the network's occupancy-state configuration). For this first model, our focus was the European tree frog (Hyla arborea L.) and its distribution in the densely populated Swiss Plateau, the study area of the CHECNET project, of which this study is an integral part.
The nodes of the network are patches obtained from ensemble habitat suitability models, themselves based on readily available species observation data from data collecting institutions. The edges are defined by cost surfaces limited by dispersal distance. The complete predictive network model takes into account node attributes as suitability variables, edge attributes as connectivity variables, and whole network properties as topological variables. The predictions were contrasted against a more traditional non-network based site occupancy model. We found that topological properties enhance considerably the predictive power of the model.
We discuss the implications of the different variables considered, especially the generalization potential of topological variables, and address the ways this model could be used with other species, in other areas and in a broader social-ecological context.

Oral or poster: 
Oral presentation
Abstract order: