The use of camera traps to sample populations and communities of birds and mammals shows great promise as a tool for monitoring terrestrial species that often avoid easy observation (Kéry, 2010). Use of camera traps in ecology is growing exponentially (Rowcliffe & Carbone, 2008), including an expanding roster of inventories, surveys and monitoring programs. The commentaries on O'Brien et al., (2010a) in this issue (Dobson & Nowak, 2010; Nichols, 2010) show that camera trap photographs not only provide data on cryptic species but can also be used to sway public opinion and policymakers in a way that is different from, and clearly complementary to, the most compelling graph or table. Dobson, Nowak and Nichols point out several aspects of monitoring and camera trapping that I want to highlight for researchers interested in biodiversity monitoring and conservation, and the policy makers who rely on their data and information.
Use of primary data
A key feature of the wildlife picture index (WPI) is reliance on primary data collected for use as a biodiversity monitoring index within a specified analytical framework. The 2010 indicators developed for the Convention on Biological Diversity promoted indicators based on data already collected for monitoring, research and other purposes (Balmford et al., 2005; Dobson, 2005). This data mining approach has produced some useful indicators, but, as Collen et al. (2008) point out, the surveillance approach under represents the majority of tropical mammalian and avian biodiversity. Nichols (2010) discusses some of the problems associated with unfocused monitoring, also called omnibus surveillance monitoring (Nichols & Williams, 2006) and advocates that monitoring be goal directed and useful for management. Primary data are essential to targeted monitoring and informed adaptive management. The 2010 targets were universally not met and as we move to developing the 2020 targets and indicators use of primary data will be essential.
Consideration of detection probability
Both Dobson & Nowak (2010) and Nichols (2010) refer to the need to incorporate detection probabilities (P) in analyses of species diversity and richness using camera trap data. Failure to control for species-specific variation in detection probabilities forces one to assume either that detection is perfect (P=1) or that detection is constant among species (P1=P2=···=Pk) in order to make inferences and comparisons in space and time. Neither assumption is likely to be true. Detection probabilities can vary unpredictably among species inducing bias of unknown magnitude and direction in estimates of species diversity. Despite the development of unbiased estimators of species richness based on capture–recapture models (Nichols & Conroy, 1996; Boulinier et al., 1998) and mixture models (Dorazio & Royle, 2005; MacKenzie et al., 2006), camera trap practitioners have been relatively slow to apply these models to community level surveys although there are exceptions (Kéry & Royle, 2008; Tobler et al., 2008a,b; O'Brien, Kinnaird & Wibisono, 2010b).
Camera trap data are ideal for applying these models to spatial and temporal analysis of species richness of terrestrial birds and mammals. Species identification using camera trap photographs is relatively straightforward for birds and mammals, trapping designs easily allow for replicated spatial and temporal point sampling, and estimation methods can control for variability in detection among species due to species differences and due to covariates, such as body size and habitat preference (Kéry, 2010; O'Brien et al., 2010b).
Spatial scale and sampling intensity
Dobson & Nowak (2010) lament the paucity of baseline data and express concern that the scale and intensity of later sampling may distort underlying trends. The WPI is designed for application at a landscape scale and is both appropriate to the most wide-ranging members of the community of interest, and to control for differences in detection. In principle, trends in indices that incorporate detection probabilities should not be subject to distortion due to sampling intensity. The ability to leave camera traps in place for weeks means that species detection probabilities are often quite high (O'Brien, 2010a,b) and proper analyses result in unbiased occupancy and species richness estimates, eliminating distortion. The issue of area sampled and allocation of trap locations is more problematic for comparison of a WPI survey to baseline data collected for other purposes. Many early camera trap studies were designed for population estimation of large, spotted or striped cats (e.g. Karanth & Nichols, 1998; Silver et al., 2004). These studies normally were conducted at the scale of hundreds of square kilometers and may be appropriate baselines for a WPI analysis. Only one study has used a large cat camera trap survey to estimate species richness (Tobler et al., 2008a), but I am optimistic that these datasets can serve as baselines for the WPI. The issue to be resolved is whether there is bias in species detection arising from trap placements that maximize detection of a target species at a cost to detection of other nontarget species. WPI estimates may be biased if targeted camera placement excludes a portion of the community of interest. If species are systematically missed by targeted sampling, trend monitoring of the subset community may still give us interesting insights into community dynamics and macro-level impacts on the community. The UK Common Bird Census is an example of a monitoring scheme characterized by a subset of potential species, non-random site selection and changing spatial coverage over time, but still yielded useful population trends when calibrated to the UK Breeding Bird Survey (Buckland et al., 2005).
Alternative formulations and other issues
Nichols (2010) makes a number of useful comments about alternative analytical approaches for the WPI. In addition to the recommendation to base the WPI on an odds ratio of occupancy estimates, one might also base the analysis on the ratio of point abundance estimates (Royle & Nichols, 2003; Royle, 2004; O'Brien, 2010b). The recent work of Dorazio & Royle (2005), Royle & Dorazio (2008), Kéry & Royle (2008) and others provides a hierarchical modeling framework for species richness estimation, improving the precision of estimates of species occurrence for rare species compared with species-specific estimates, proposed by O'Brien et al. (2010a). Nichols correctly points out that O'Brien et al. (2010a) proposed an implicit dynamics model (MacKenzie et al., 2006) whereas an explicit dynamics model would allow incorporation of a wider range of covariates and heterogeneity in modeling occupancy. These are all ideas that are worth exploring in future development of the WPI as they would improve the quality of the initial estimates that enter into the WPI and thus increase the applicability of the WPI as an indicator in the decade to come.
As Rowcliffe & Carbone (2008) point out, there is still a great need for better access to the large pool of existing camera trap data to test the robustness of the WPI. In many studies of single species, non-target species are frequently considered the equivalent of fisheries ‘by-catch’ and the data are filed away and not available for analyses. Of the hundreds of camera trap studies published and unpublished, few have made the data publicly available. The development of open access databases by the Tropical Ecosystem Assessment and Monitoring Program, Smithsonian Institution and others, promises to improve access to the vast amounts of camera trap data and metadata (present and future).