Ecologists have traditionally studied intraspecific variation by sampling species across their geographic ranges. However, whether this classic approach produces samples that accurately represent species’ climatic niches is largely unknown. Alternative, niche-based study designs using species’ climatic niches to inform sampling locations should more efficiently and completely capture the breadth of the niche, but the magnitude of this difference and how it may vary is unclear. Here we use conifers as a model system to explore these issues and reach specific recommendations for future sampling designs. Using an independent dataset of high-quality species’ occurrences, we first show that recent publications examining variation across geographic space do a poor job of capturing the full breadth of species’ niches, such that on average, only 22% of species’ niche space was sampled. This was also true of a large compiled database, the International Tree-Ring Data Bank (ITRDB), which yielded average niche coverage of only 45%. Finally, we simulated common sampling designs (i.e., random points, grids, and transects) in both geographic and niche-based sampling frameworks. Using two sampling metrics, niche coverage and niche undersampling, we measured how completely and evenly these simulated studies characterized the niches of 64 North American conifers. Niche-based sampling better represented species’ niches than geographic sampling, with the magnitude of this difference depending on study design and sample size. Niche-based gridded study designs achieved the most complete sampling at all but the smallest sample sizes, covering ~15-25% more of a species’ niche than similar designs implemented geographically. With fewer than 10 samples, however, all study designs performed poorly, and niche-based transects achieved slightly higher niche coverage. Consequently, when more than a handful of samples are collected, we recommend that studies seeking to characterize variation across a species’ niche consider using a gridded study design implemented in a niche-based sampling framework.

All the data used in this paper were compiled from freely and publicly available sources. Conifer occurrence data was downloaded from the Conifer Database: https://herbaria.plants.ox.ac.uk/bol/conifers ; contact Aljos Farjon ( a.farjon@kew.org ) for use. Climate data was downloaded from the CliMond database: https://www.climond.org/ , and is the same climate dataset as that used in Perret et al. 2019. Metadata for the International Tree-Ring Data Bank (ITRDB) is available from the public data repository: https://www1.ncdc.noaa.gov/pub/data/paleo/treering/

The R code for all novel functions developed as part of this paper's analyses are contained in the attached .R file.