Modern Methods of Estimating Biodiversity from Presence-Absence Surveys

Abstract

Communities of species are often sampled using so-called “presence-absence” surveys, wherein the apparent presence or absence of each species is recorded. Whereas counts of individuals can be used to estimate species abundances, apparent presence-absence data are often easier to obtain in surveys of multiple species. Presence-absence surveys also may be more accurate than abundance surveys, particularly in communities that contain highly mobile species. A problem with presence-absence data is that observations are usually contaminated by zeros that stem from errors in detection of a species. That is, true zeros, which are associatedwith the absence of a species, cannot be distinguished from false zeros, which occur when species are present in the vicinity of sampling but not detected. Therefore, it is more accurate to describe apparent presence-absence data as detections and non-detections, but this terminology is seldom used in ecology. Estimates of biodiversity and other community-level attributes can be dramatically affected by errors in detection of each species, particularly since the magnitude of these detection errors generally varies among species (Boulinier et al. 1998). For example, bias in estimates of biodiversity arising from errors in detection is especially pronounced in communities that contain a preponderance of rare or difficult-to-detect species. To eliminate this source of bias, probabilities of species occurrence and detection must be estimated simultaneously using a statistical model of the presence-absence data. Such models require presence-absence surveys to be replicated at some – but not necessarily all – of the locations selected for sampling. Replicate surveys can be obtained using a variety of sampling protocols, including repeated visits to each sample location by a single observer, independent surveys by different observers, or even spatial replicates obtained by placing clusters of quadrats or transects within a sample location. Information in the replicated surveys is crucial because it allows species occurrences to be estimated without bias by using a model-based specification of the observation process, which accounts for the errors in detection that are manifest as false zeros. Several statistical models have been developed for the analysis of replicated, presence-absence data. Each of these models includes parameters for a community’s incidence matrix (Colwell 13