Biased niches – Species response curves and niche attributes from Huisman-Olff-Fresco models change with differing species prevalence and frequency

The study aimed to examine the effects of different numbers of presences and frequencies (proportions) of occurrences of species in a plot data set of forest vegetation on the species response curves and their niche attributes, based on Huisman-Olff-Fresco models (HOF). We modeled responses of 72 to 105 herbaceous forest species along a pH gradient under 14 different random sampling scenarios by varying the number of presences and absences used for model fitting. Mean niche attributes were calculated from 100 repetitive runs for each scenario and species. Re-prediction success of HOF models among the repetitive runs was highest when the total number of plots was high and the frequency of occurrences was low. With low plot numbers and high frequencies, less complicated model types (no response or monotonically increasing/decreasing responses) predominate. Measures of species niche boundaries (limits & borders) and niche width were strongly influenced by changes in sampling characteristics. With an increasing number of presences and an increasing frequency, limits and borders shifted to more extreme values, leading to wider niches. In contrast, species optima showed almost no change between the scenarios. Thus, the detected ecological response of a species often depends on the size of the data set and the relation between presences and absences of a species. In general, high data quantities are required for reliable response curve modeling with HOF models, which prevents the assessment of the responses of many rare species. To avoid undesired bias by differing sampling characteristics when comparing niches between different species or between data sets, the data basis used for model fitting should be adjusted according to the niche attribute in question, for example by keeping the frequency of the species constant.