Estimating change in species richness from repeated sampling of incidence data

Abstract

Forest tree species richness is an important indicator of sustainability. Forest monitoring allows stake-holders to track species richness over time. To test a hypothesis of whether an observed change in richness is significant or not, the analyst must choose a suitable estimator. Based on experience from comparative studies of potential estimators this study evaluates the performance of one design-based and three model-based estimators of change in species richness. The evaluation is done with Monte Carlo simulations of simple random sampling from four actual populations of forest tree species incidence data collected at two occasions from a fixed set of fixed-area forest inventory plots. The observed change in species richness (design-based) had the lowest root mean squared error, but was often more biased than estimates from the three model-based estimators. The bias issue and poor coverage of 95% confidence intervals dissuade the use of the design-based estimator (observed change). A newly developed urn-type estimator, easily adaptable to processing longitudinal data, was overall best in terms of bias and coverage.