Classification, similarity and other methods for presence-only data
Janet Franklin · Cambridge University Press eBooks · 2010
Introduction Many of modeling methods described in Chapters 6 and 7 require observations of species presence and absence, preferably a lot of them (in order to characterize complex response functions), well distributed in space and along environmental gradients. These data are required to estimate model parameters or to derive decision rules for supervised classification. If presence and absence data are available, the modeling approaches that are designed for binary response variables, discussed in the previous chapters, generally give more accurate predictions than models based on presence-only data (e.g., Brotons et al ., 2004), but not always (Hirzel et al ., 2001). But what if only observations of species presence (but not absence) are available, or what if there are no georeferenced observations at all, but simply some expert knowledge on species habitat requirements? The methods described in this chapter can be, and have been, applied to SDM in these situations. It is actually a very common predicament, to have species presence data, or no species location data at all, and the reasons for that are reviewed here. If a species has been recorded as being present in a location, we can be fairly certain that it occurs there (except for taxonomic misidentifications). We then make the assumption the occurrence of an organism indicating habitat use , occupancy or suitability for the purpose of modeling.