Extended Data Fig. 3: The statistical approach was implemented via one-stage generalized linear mixed models (GLMMs) in which the response variable was species composition dissimilarity among years.
From: Warming and cooling catalyse widespread temporal turnover in biodiversity
a) The simplest model included the relationship between dissimilarity and temporal distance among observations so that, for example, dissimilarity could increase with time. The slope of this relationship is the turnover rate. Random intercepts and slopes helped account for variation among studies and time series (not shown). b) We tested the hypothesis that faster rates of temperature change (Tchange) were associated with faster accumulation of dissimilarity through time (compare red vs. blue line). This hypothesis was statistically tested as an interaction (Tchange × Years). c) We additionally tested the hypothesis that the influence of temperature change on the turnover rate depended on average baseline temperatures. For example, the slope of dissimilarity over time could be steeper in areas with hotter average temperatures and fast rates of temperature change than in areas with colder average temperatures and fast rates of temperature change (compare dashed red vs. solid red line). Statistically, this was tested as a three-way interaction (Tchange × Tave × Years). d) Turnover rates as a function of temperature change rates, showing an increase in turnover rate with increasing rates of temperature change (i.e., the same relationship as panel b but summarized as rates). The slope of this relationship was termed sensitivity (Δturnover rate/Δtemperature change rate). e) Turnover rates as a function of temperature change rates and average baseline temperatures, showing a faster increase in turnover rate with temperature change at hotter average baseline temperatures (i.e., summarizing the same relationship as panel c). f) Sensitivity as a function of average temperatures, showing an increase in sensitivity at hotter average temperatures (i.e., summarizing the same relationship as in panels c and e). The x-axis could also be other environmental covariates, such as microclimates or non-climate human impacts (as in Fig. 3).
