Applications in ecological statistics: plausibility in linking species and climate data Statistics Seminar: Mark J Brewer (Biomathematics and Statistics Scotland)
Commonly in ecological statistics we model species’ data in space or time as a function of climate (or weather) variables. Ecologists have recently favoured Generalised Additive Models (GAMs) or piecewise linear forms (in MaxEnt) for their flexibility, however, resulting estimates of species response to climate are often implausible.
Rather than try to force structure on GAMs, we have preferred to define flexible model forms which encode ecological knowledge and in effect, increase our chances of producing a plausible (or “believable”/”realistic”) species response function. We illustrate these ideas with two case studies: relating bird abundance indices to historical weather data; and fitting climate envelopes to distribution data on vascular plants.