Climate change is impacting ecological systems. The predicted impacts, however, are highly sensitive to the choice of bioclimatic variables. As a result, an essential step in using climate projections is to identify those variables properly aligned with the study's objectives. Here, we propose a fundamental method for variable selection. We seek to identify variables that capture biophysical properties of a system of interest. This is done by identifying variables where the range of values occupied is relatively narrow compared to the full range of values over the considered domain. The absolute difference between the two distributions is a measure of its ecological relevance. This enables us to select an optimal subset of variables for the system. We assess this approach by applying it to the climatic niches of 287 European tree species. More specifically, we construct climatic niches based on variables selected with our method and compare them with niches derived from two alternatives: a simple variable set based only on temperature and precipitation, and ab unrestricted set including a large number of climatic variables. The choice of variable sets has a strong influence on the inferred climatic niches, with simple or unrestricted sets producing systematically broader and less informative niches. In contrast, the proposed selection framework yields more consistent and ecologically meaningful representations of species–climate relationships. Furthermore, optimal variable sets for individual species are close to the one optimised for the 287 considered tree species.