Can subsets of species indicate overall patterns in biodiversity?

Abstract

AbstractResource limitations often allow only a subset of species to be counted. But using subsets may bias inferences on spatial or temporal trends in biodiversity. Using data from a video survey on reefs in the Gulf of Mexico for which all fish species observed were counted (243 species), we investigated how the use of reduced species lists (RSLs) can impact perceived patterns in biodiversity. We estimated four common biodiversity metrics (species richness, and Margalef's, Shannon's, and Simpson's indices) at each of 2115 sampling locations, using the total species list and RSLs. For all diversity metrics, correlations between estimates using the total species list and RSLs increased with the number of species in the list. Using a bootstrap approach, we randomly generated hypothetical lists equal in length to each empirical RSL to evaluate their performance; empirical RSLs tended to perform similar to random lists of equivalent length when estimating species richness or Margalef's index, and tended to outperform most hypothetical RSLs when estimating Shannon's and Simpson's indices. To understand how to create better performing RSLs, we extended the bootstrap analysis to select RSLs of all possible lengths, using four different selection methods related to species commonness; the functional relationships between correlation and number of species in an RSL were similar among metrics but were very different among selection methods. With each hypothetical RSL, we tested common biodiversity hypotheses such as relationships with depth and latitude and compared the outcomes with the best estimate of true relationships identified using the total list. Longer lists comprised of the most common species more often identified the true relationship, but results showed complex patterns. Many short lists of the most common species yielded results opposite the true relationships, and many lists of intermediate length failed to identify any relationship while the total list showed a significant trend. Overall, these analyses show that sampling methods used for biodiversity studies should be as unselective as possible, and datasets based on more selective methods should be interpreted carefully and should not be expected to reflect true patterns in biodiversity.