Reconsidering climatic predictors for high‐resolution niche models of alpine plants

The increasingly acknowledged and consequently also better understood microclimatic variability in terrestrial ecosystems has motivated a call for finer spatial resolution in species distribution modelling, especially in the case of sedentary low‐stature organisms such as plants. In contrast, less attention has so far been paid to the way climate should be represented in these models. In fact, most modelling applications rely only on a handful of so‐called bioclimatic variables (i.e. essential climatic variables designed for ecological applications), which is at odds with the hypothesised variation in the sensitivity of individual plant species to different facets of climate. We argue that the recent shift towards microclimate modelling provides a window of opportunity for re‐evaluating the predominant reliance on the small set of bioclimatic variables. We used a unique dataset of 895 1‐m 2 plots with vascular plant species observations and in situ soil temperature measurements across a high‐mountain landscape spanning 1700 elevational metres. From the hourly temperature measurements, we calculated bioclimatic variables as well as 188 additional ‘grid variables' arising from an aggregation of various summary statistics over different parts of the year. We then used those ‘grid variables', their subsets, and the bioclimatic variables to fit species distribution models for 101 plant species with combinations of one, two or three predictors. We found that bioclimatic variables consistently delivered less accurate models than many of the grid variables. Models based on subsets of grid variables only slightly decreased in accuracy and remained superior over bioclimatic variables even after the set was narrowed from the initial 188 to only six variables on the basis of a cluster analysis. These results highlight that modelling species distributions with only a few climatic variables is a viable strategy. However, the most suitable variables may often be different from those that are commonly used nowadays.