Spatial variability and power to detect regional-scale trends

An important issue in regional-scale trend detection is the degree of concordance in trend between sites in relation to an average regional trend. If there is much or more variability between sites than within sites over time, inter-site variance can overwhelm the effects of other sources of variation in a system, resulting in low power in trend detection across that scale, despite precise methods of measurement and long time series of data. We used published data from repeated regional-scale surveys of Adelie penguin breeding population size in east Antarctica to assess the degree of spatial concordance in population trends within the scale of notional small-scale management units, and given these estimates of trend concordance, then used the power analysis program MONITOR to predict power for various multiple-site monitoring scenarios. Under the conservative, hypothetical scenarios we examined, monitoring at only 2 or 3 sites provided insufficient power to detect a trend. At the other extreme, monitoring at all sites in a region (a census) may be an unnecessary use of resources because the gains in power over a design using a sample of sites were very marginal. Reasonable power was achieved by monitoring around 6 sites in a region under conservative criteria of a 2-tailed test and alpha = 0.10. However, fewer sites were required for less conservative criteria; using a 1-tailed test instead of a 2-tailed test meant that a few less sites were able to achieve the same power for a fixed duration for detection, and increasing the significance or Type I error level from 0.10 to 0.20 improved power such that a few less years or sites were required to detect a trend. Monitoring every 3 years instead of annually reduced power only very marginally.