Sensitivity of functional diversity metrics to sampling intensity

Abstract

Summary Functional diversity (FD) metrics are increasingly used in ecological research, particularly in the studies of community assembly and ecosystem functioning. However, studies using the FD metrics vary greatly in the intensity by which ecological communities were sampled and it is largely unknown how sensitive these metrics are to low sampling intensity (undersampling). In this study, we used a combination of simulations with theoretically assembled communities and three comprehensive, independent, empirical datasets on plant, ground beetle, and bird communities to investigate the sensitivity of nine commonly used FD metrics to undersampling. Simulations with both theoretical communities and empirical data showed that in a wide range of contexts, the measurement of various FD metrics requires a much higher sampling effort to reach an ‘adequate’ precision (defined as an r2 of at least 0·7 between different subsets of the same population), than that required for commonly used taxonomic diversity metrics (e.g. species richness), although the ‘accuracy’ (their deviation from the diversity value of a completely sampled community) of their measurements is not more sensitive to undersampling than species richness. We also found that some FD metrics (e.g. Functional Dispersion) are consistently less sensitive to undersampling than others (e.g. nearest neighbour distance‐metrics). Problems of undersampling were generally most severe in datasets with high overall species richness and low overall abundances. We found that the precision of many FD metrics is highly sensitive to undersampling, and more so than commonly used taxonomic diversity metrics. Therefore, to ensure reproducible results in functional biodiversity research, we recommend that thorough sampling designs are used to sample communities and that datasets originally collected for studying taxonomic diversity should only be used for FD when it can be shown that undersampling is not a major issue. In cases where undersampling is suspected or logistically unavoidable, the FD metrics that are relatively insensitive to its effects (e.g. Functional Dispersion) should be prioritized.