A method to the madness: Plant trait sampling strategies matter in community ecology
At its most basic, niche theory argues that many of the phenotypic differences we see between species are important, influencing interactions with different species, tolerance of environmental conditions, and exploitation of a range of resources. Although species have historically been the currency of much of community ecology, the importance of intraspecific variation for ecological processes is increasingly recognized, as part of a larger conceptual trait-based shift in ecology. Key ecological questions addressing community assembly, competitive dynamics, and the relationship between biodiversity and ecosystem function require an understanding of species’ niche overlap, and traits are ever more used as a proxy. Despite this recent focus on traits, outstanding questions include: how many and which traits to measure, and at what spatial scale trait variation should be examined. We need an understanding of how sampling decisions affect our ability to capture relevant phenotypic variation, reliably judge species’ similarities and therefore compare findings across studies. Here, we used field-sampled trait data of monkeyflower (Mimulus) congeners in California to investigate how sampling scale, trait types, and trait dimensionality impact estimations of phenotypic similarity. Using this data, we constructed a set of recommendations for future sampling.
Trait dimensionality, trait choice, and sampling scale all influenced estimates of species’ similarity. Generally, the ability of Linear Discriminant Analysis to correctly assign individuals to species (“LDA success”) increased as we included more traits; however, certain species were poorly discriminated regardless of trait dimension, depending on whether floral or vegetative traits were being used. Unintuitively, vegetative traits on average outperformed floral or combined trait sets in discriminating these herbaceous, moisture-loving species. For a given number of traits, LDA success varied up to 25 per cent based on the particular trait combinations used, though ‘logical’ trait combinations thought a-priori to capture more distinct phenotypic axes fared no better than others on average. Lastly, the amount of variation in LDA success seen for a given population using different numbers of traits approximates the variation observed between populations of the same species. That is, population choice has a substantial influence on our perceptions of species’ similarity. Taken together, these results emphasize the importance of sampling to capture intraspecific variation and suggest that, although the ‘best’ combinations of traits at lower trait dimensionality can recover similar success in discriminating species, these combinations may not be predictable, and a ‘more is better’ trait sampling approach may indeed be the best way forward.