Latitudinal gradients and scaling regions in trait space: Taylor’s power law in Japanese woody plants

Aim: Taylor's power law (TPL) is a variation of the variance:mean ratio and is often used to describe over- or underdispersed ecological distributions. We hypothesize that TPL is also applicable to the distribution of species traits and that the respective power law parameters might determine ecological functioning. Here, we aim to study this hypothesis in detail. Location: East Asian islands, including the Japanese and Ryukyu archipelagos. Time period: 1968–2015. Major taxa studied: Gymnosperm and angiosperm woody plant species. Methods: We used the geographical distribution of 946 Japanese woody plant species at the 10 km × 10 km grid cell level. Based on leaf and wood samples and literature data, we studied 10 important plant traits (maximum plant height, average fruit and seed size, specific leaf area, leaf thickness, wood density, leaf tannin and phenol content, and C/P and C/N ratios) and related trait variability to minimum absolute temperature, land and forest area and to the variability in forest cover and elevation using bi- and multivariate and piecewise regression analysis. Results: The variability in the trait expression was well described by TPL. Average trait expression and the respective variability changed predictably along the latitudinal gradient, resulting in a general tendency towards trait clustering (TPL slopes > 1.0). Piecewise regression detected significant breakpoints in the TPL pattern for most traits. Minimum ambient temperature and latitude were the most important predictors of the variability in the observed TPL slopes. Main conclusions: Taylor's power law appears to be trait specific and cannot be used as a diagnostic ecological character. We propose a new linear model to quantify ecological variability that includes average variable expressions and ecological covariates. We argue that common measures to quantify ecological variability based on the variance:mean ratio might give false impressions about the true degree of variability because they do not account for the variance–mean allometry.