Optimal Sampling Designs for Monitoring Plant Frequency

Abstract

Plant frequency is a pragmatic surrogate for plant density in protocols designed for the long-term monitoring of diverse communities. Frequency estimates are based on presence/absence data from plots of fixed size, and plots are usually spatially aggregated into sites (often transects) to reduce field effort. Using a combination of statistical models and computer simulations, we identify sampling designs that maximize statistical power for detecting changes in underlying plant density based on the analysis of plant frequency. The optimal plot size for collecting frequency data decreases both with increasing spatial variation in local density (spatial structure) and with increasing numbers of plots per site. Over realistic ranges for these parameters, plots of optimal size yield mean frequencies that vary from 20% to 80%. However, with the exception of highly overdispersed populations, power is relatively insensitive to plot size; consequently, a plot size that yields a mean frequency of 50% usually provides nearly maximal power. For population monitoring, in which comparisons are made between successive samples from the same population, repeated measures from fixed sites improve statistical power substantially if there is spatial structure among sites, provided that the spatial pattern is at least partially consistent over time. However, there is still a power loss to the extent that the pattern of spatial structure among sites changes over time (a site-by-time interaction). This power loss can be mitigated by increasing the spacing between plots within sites, which has the effect of increasing the within-site structure and reducing the between-site structure. With more than 1 plot per site, there is no statistical advantage to obtaining repeated measures from fixed plots; relocating plots within sites in successive samples may therefore be advisable to minimize disturbance to the community.