SEARCH

SEARCH BY CITATION

Keywords:

  • Area;
  • climate correlate;
  • Europe;
  • grid cells;
  • GAMs;
  • grain;
  • model accuracy;
  • species richness;
  • variable selection;
  • 3D area

ABSTRACT

Aim  Species richness–area theory predicts that more species should be found if one samples a larger area. To avoid biases from comparing species richness in areas of very different sizes, area is often controlled by counting the numbers of co-occupying species in near-equal area grid cells. The assumption is that variation in grid cell size accrued from working in a three-dimensional world is negligible. Here we provide a first test of this idea. We measure the surface area of c. 50 × 50 km and c. 220 × 220 km grid cells across western Europe. We then ask how variation in the area of grid cells affects: (1) the selection of climate variables entering a species richness model; and (2) the accuracy of models in predicting species richness in unsampled grid cells.

Location  Western Europe.

Methods  Models are developed for European plant, breeding bird, mammal and herptile species richness using seven climate variables. Generalized additive models are used to relate species richness, climate and area.

Results  We found that variation in the grid cell area was large (50 × 50 km: 8–3311 km2; 220 × 220: 193–55,100 km2), but this did not affect the selection of variables in the models. Similarly, the predictive accuracy was affected only marginally by exclusion of area within models developed at the c. 50 × 50 km grid cells, although predictive accuracy suffered greater reductions when area was not included as a covariate in models developed for c. 220 × 220 km grid cells.

Main conclusions  Our results support the assumption that variation in near-equal area cells may be of second-order importance for models explaining or predicting species richness in relation to climate, although there is a possibility that drops in accuracy might increase with grid cell size. The results are, however, contingent on this particular data set, grain and extent of the analyses, and more empirical work is required.