It offers a theoretical and methodological review of the literature of the past 15-20 years in a field of growing interest within ecology, biogeography, and conservation biology: predictive habitat distribution modelling has undoubtedly become a hot topic (again). This comparably comprehensive synthesis was probably lacking, and was published just at the right time. How, and how well, can we simulate the spatial distribution of organisms with a range of statistical methods? What are the theoretical assumptions behind this general approach? What are the advantages and limitations of the various methods? These are some of the points that are discussed in our paper.

Although it does not develop any new discovery or methodology, the paper attempts to critically review the whole model-building process. Such a synthesis was not yet available, and this might have contributed to its success and usefulness. Besides, it summarized the theoretical framework this topic is based on, and it discusses gaps in the area of predictive distribution modelling.

Predictive distribution modelling aims at simulating the geographic distribution of organisms with the aid of computers, a set of explanatory variables, and statistical models. Once a statistical model has been formulated, and if the explanatory variables are available in the form of maps, we can predict the distribution and/or the abundance of species or habitats in space. At the same time, such models allow us to test the relevance of individual variables to explain the geographical distribution of species and habitats. Moreover, it allows us to detect areas of high or low sensitivity with respect to change, or areas of high or low suitability/risk of survival or extinction. This field in ecological research was arising mostly in the late 80s and early 90s. Since this approach does not really include dynamical processes, many scientists turned to more dynamical approaches, and this method seemed to fade away a bit. However, recently it became a hot topic again. Several factors might explain this recent rise: 1) new statistical models were developed and became more easily available and user-friendly; 2) geographical information systems (GIS) became a widely used management tool; and 3) applied ecology itself is presently on the rise. Besides, it became more and more obvious that process models (i.e., more dynamic models) cannot easily solve many urgent management questions within a reasonable time. Management and planning, on the other hand, often require simple and flexible models and spatial predictions. Also, the increasing availability of data stored in large biological databases has facilitated the re-analysis and testing of biogeographical hypotheses and has boosted the predictive distribution modelling. Controversially, such large data sets usually have inherent sampling biases that often require the use of alternative or specially adapted approaches. This was well discussed and developed in the recent, specialized literature. With the development of new statistical methods, it became more and more difficult for applied managers and non-expert users to decide upon which technique to use. Few overviews of the techniques were available at the time we published our review. Exceptions were the important paper of Janet Franklin (1995) and the pioneering work of Mike P. Austin and co-workers, published over the last 20 years. Our review tried to fill this gap, and to summarize a conceptual and theoretical framework for the whole process of model development and model evaluation in this field. One could ask, why then are such predictive models (so) important? Their development tightly followed the development of geographical information systems in ecology. Most ecological questions clearly have a spatial component. Hence, drawing and—more importantly—explaining the potential distribution of a species constitutes clearly a first step in the sequence of resolving more complex fundamental or applied research, such as testing biogeographical hypotheses, assessing ecological risks (e.g. invasive species, impact of climate change, locating populations of rare species), or planning suitable locations for new nature reserves.

We both used such approaches extensively during our Ph.D. projects, with A. Guisan focusing more on individual species and N.E. Zimmermann more on habitat types in the European Alps at that time. Interestingly, both studies were initiated to perform risk assessment of the potential ecological implications of climate change upon the distribution and abundance of plant species and habitat types. No synthesis existed at the time we started, and we had to develop or adjust most tools and methods ourselves. Modelling in a mountainous landscape presented additional difficulties (rugged terrain, mosaic-like vegetation, small-scale variation) that required a careful evaluation of methods and approaches. As we both kept working in this field, we maintained a tight and fruitful collaboration. The idea of writing a review emerged as an almost logical consequence of the many discussions, the long collaboration, and the idea to make available to others what we thought we had developed in this field in the recent past.

Dr. Antoine Guisan
Assistant Professor
Institute of Ecology
University of Lausanne
Lausanne, Switzerland