Detecting ecosystem trends in response to climate and disturbance across continental plot networks: a power analysis

Plant communities are dynamic: cyclical, directional and stochastic changes in species composition make interpretation of responses complex. The multi-dimensionality of ecological change is reduced by ecosystem indicators that quantify functional responses to environmental drivers. However, thresholds of detectable change in indicators may vary by ecosystem or the spatial configuration of monitoring sites within regions. We quantified the power of a continental terrestrial ecosystem plot network (TERN Ausplots, Australia) to detect change in ecological indicators linked to disturbance and climate: community temperature index (CTI), proportional abundance of photosynthetic pathways (C4%), and species abundance distribution (SAD). We simulated trend and noise scenarios and calculated power and effect size across spatial clusters of plots using linear mixed effects models. We then assessed factors that influenced change detection capacity. Power varied substantially across the network. Trends with a minimum magnitude of 0.5-3.0{degrees}C in CTI, 0.01-0.3 in C4% and 0.1-0.5 in the SAD sigma were detectable, depending on cluster identity and noise. Increased plot replication within a spatial cluster increased power and lower variance among baseline replicates increased effect size for CTI and C4% but not SAD. Latitudinal patterns in change detection capacity emerged for CTI and C4%. The effect size of increasing CTI was greater in the tropics due to lower among-plot variance, while power to detect increasing C4% was higher in the temperate south, where C4 is rare. Indicators reveal trends in complex ecological data that can inform decision-making. Increasing spatial replication and minimising spatial variation by stratifying plots to the same ecological community enhance change detection, albeit with different effectiveness by indicator. The results suggest change detection differs by location, irrespective of sampling factors such as number of plots. Strategies and thresholds for change detection may therefore apply unequally across environments and should be considered by region and response parameter.