On the use of automated cameras to estimate species richness for large‐ and medium‐sized rainforest mammals

The question of ‘how many species occur,’ whether at a study site, in a park or in a country, is of relevance to the management and conservation of biodiversity. Biodiversity is often equated with species diversity and species richness at a site (Schluter & Ricklefs, 1993; Lande, 1996). Species richness frequently is used as a state variable in evaluating the impact of management interventions or anthropogenic disturbance on biodiversity (Yoccoz, Nichols & Boulinier, 2001). As Tobler et al. (2008) correctly observe, knowledge of the presence and distribution of species is critical to planning and evaluating strategies for conservation of biodiversity. Restricting the scope of their study to inventories of terrestrial forest mammals larger than 1 kg, they discuss the use of automated camera traps to detect forest mammal species, and give an example of how to use an inventory to answer questions about species richness. They conclude with a discussion of issues related to designing species surveys. The authors used a standard survey design appropriate for estimating jaguar Panthera onca and ocelot Felis pardalis abundance to obtain records of occurrence for all species during two surveys. Using two cameras at each point allowed them to estimate the reliability of the cameras in detecting a species when it passed through the sampling point and to relate the per cent of detections to body mass of the species. They compared eight species richness estimators with a species list for their site, considered to be the ‘true’ number of species. Finally, they used the frequency of detection to estimate the sampling effort necessary to obtain photographic evidence of rare species at a site. I enjoyed the article and agree with much of it including the idea that a survey designed for one purpose may produce data that are useful for another. I do, however, have several concerns and observations about this study and its inferences. The treatment of the species list as the ‘true’ species richness warrants caution. Although species lists are informative about the potential richness of a site, local species lists or pooled species lists represent the accumulated knowledge of a site rather than a snapshot estimate of current species richness. Local or historical species lists are composed of samples from a variety of sources and are subject to all of the limitations of multi-season count data due to differences in detectability among methods and species. Historical presence does not equate to presence today and it is incorrect to conclude that a species richness estimate at a point in time is biased because it is less than the potential species richness. Tobler et al. (2008) found that jackknife estimators of species richness performed better than others they examined. Colwell & Coddington (1994) reviewed asymptotic and non-asymptotic methods of calculating species richness based on species accumulation curves. They point out that the basic problem with extrapolation is that different models give different asymptotic species richness estimates for the same data. Because the appropriate curve depends on the rate of species accumulation, which in turn depends on the distribution of species abundance, no single model is expected to work across a range of accumulation patterns. This is problematic because the answer depends on the method and the choice of methods is subjective. Other reasons exist to prefer jackknife estimators (Burnham & Overton, 1978, 1979). First, they allow a measure of the precision of the estimate so that we can evaluate whether differences between the estimate and some reference value (perhaps a species list, or last year's estimate) are significant. Second, they can be expanded to include spatial and temporal models of species richness that are especially useful for estimating extinction, colonization and turnover at a site or between sites (Boulinear et al., 1998; Nichols et al., 1998a, b; Hines et al., 1999). Cam et al. (2000) applied jackknife estimators with regional species lists to estimate relative species richness – the ratio between the estimated number of species at a single site (α-diversity) to the number of species in the regional species pool (γ-diversity). When species richness is used as a state variable in a monitoring program, these statistics are indispensable. Estimators based on asymptotic curve fitting or species abundance distributions such as the log-normal distribution (Preston, 1948) or the log-series distribution (Williams, 1964) do not include estimates of precision nor are they easily compared across space or time (Colwell & Coddington, 1994). A recent development in species richness estimation takes advantage of the fact that species richness may be estimated from detection/non-detection data using occupancy models to estimate species richness at a site or at a collection of sites (MacKenzie et al., 2006). Occupancy models are especially useful for estimation involving site-level species richness when a historical or a regional species of potential species occurring at the site or in the region is available. When the regional species pool is known, each species may serve as a ‘site’. Replicated sampling is carried out and a detection history is constructed for each species in the regional pool. Some species from the regional pool species are never detected in the study area. The proportional occupancy of the species is then interpreted as the probability that a member of the regional pool is present in the sample, which is the same as the relative species richness described by Cam et al. (2000). Occupancy modeling allows tracking of changes in species richness over time, allows estimation of detection probability, extinction and colonization and allows the direct incorporation of covariates that might affect species detectability, extinction or colonization (MacKenzie et al., 2006). An occupancy-based analysis of species richness would allow the incorporation of body size (Tobler et al., 2008) as a covariate in the construction of detection probabilities. Given the opportunities to use jackknife- and occupancy-based estimators, the discussion of survey design considerations needs to be placed in perspective. While trap spacing may not be an important consideration when using a jackknife estimator of species richness, it is a fundamental consideration if the survey is multi-purpose, as the Tobler et al. (2008) survey was, or if an occupancy-based estimator is used. A good rule of thumb for trying to detect rare species is to sample extensively as well as intensively. Detecting a species with an expected occurrence of 1/1000 trap days requires 3000 trap days of effort (Tobler et al., 2008): 50 trap sites for 60 days, or 100 trap sites for 30 days. Placing sampling point too close together essentially duplicates effort and wastes resources when sampling for rare or elusive species. Trap placement considerations are also important. Surveys designed for single species abundance estimation often sample at points that maximize the likelihood of encountering the target species. These locations may not be the best or even optimal for a species richness survey, and the species richness estimates may be biased due simply to species avoiding the travel paths of the target species. The use of a systematic trap placement design or a design stratified by habitat may be desirable when species richness estimation is the primary goal of the survey. I agree with Tobler et al. (2008) that camera traps are a powerful new tool in the quest for better methods to monitor components of biodiversity that are hard to observe, such as forest mammals. Because camera traps can be left in the field for extended periods and operate 24 h day−1, they are superior to sampling methods that require human observation, in areas where opportunities for traditional sampling are limited, and in areas where nocturnal animals make up a significant component of the terrestrial community. The next challenge for biologists will be to design surveys that maximize the potential of camera trapping for biodiversity monitoring.