Technical Approach

Background and General WPI information

The Wildlife Picture Index (WPI) was developed jointly between the Wildlife Conservation Society and the Zoological Society of London as an indicator derived from primary camera trap data (O’Brien et al 2010) for ground-dwelling tropical forest medium and large mammals and birds, species that are important economically, aesthetically and ecologically. The WPI was designed to meet the technical requirements of biodiversity monitoring indexes as described by Buckland et al. (2005) and is defined as the geometric mean of the occupancies of the species in the community relative to the first year of sampling (baseline).

The WPI can be aggregated upward from the local site to the global level, and it can be disaggregated to capture trends at regional levels, functional groups of interest, or national level (if adequate national data are available). At all these levels, data can be disaggregated for particular taxonomic and functional groups of species, species with different conservation status, or facing particular threats. For example, the WPI can be calculated for species with different IUCN conservation status categories (EN, CR, LC, etc), CITES levels of threat, and functional groups. The WPI can also be calculated separately for mammals and birds. Currently, the index is calculated from TEAM Network annual camera trap deployments coming from 16 sites across the three main continental forest blocks (Latin America, Africa and South East Asia).

Methods used in HP Earth Insights WPI System

Ahumada et al. (2013) proposed an alternative approach to calculate the WPI using state-space dynamic occupancy models derived from a Bayesian model formulation (Royle & Kéry 2007). Using this approach it is possible to estimate directly the confidence intervals of the WPI, allowing for straightforward inference on changes in the index. This method also allows the inclusion of different factors (climate, landscape configuration, etc) in the occupancy models, enabling interpretation of changes in the WPI as a function of these factors. The WPI Analytics System enables and standardizes the calculation of the WPI, and updates it continuously as new TEAM camera trap data are ingested into the system.

The WPI Analytics System uses HP's Vertica Analytics Platform and R analytical software (http://www.r-project.org/). R integrates tightly with Vertica and this Vertica-R interoperability is used for performing Monte Carlo Markov Chain (MCMC) simulations with JAGS (Plummer 2008) using the R2Jags package (Yu-Sung 2012). The calculation of the WPI starts with a Bayesian model for estimating species occupancy at a given site using species presence-absence data (by processing raw camera trap data) and covariates-climate, forest edge, and human presence data. These models have three state variables (occupancy in the baseline year, extinction probability and colonization probability) and one detection variable (detection probability) each a function of factors mentioned above. Therefore, each model requires a very large number of simulations (250,000 simulations per species) to converge to a stable solution. Once the model converges, we use the final 1000 simulations to calculate the occupancy distributions and aggregate several of these distributions into the WPI. For example, to calculate the WPI for a TEAM Site all of the occupancy distributions for each species (at that site) are aggregated by year. These yearly aggregations result in a WPI distribution over time for that particular TEAM Site. A similar methodology is used to calculate global WPIs (all sites and all species), regional WPIs (e.g., Asia, Africa and Latin America) and for any function group of interest.

How to understand the WPI results

An increase/decrease in the WPI indicates a positive/negative trend in biodiversity (when aggregated across all species) or other component of diversity (when looking at a subgroup of species). Since the WPI is measured relative to the first year of data, values above 1 indicate a proportional increase in diversity and values below 1 indicate a proportional decrease relative to the first year of data. The impacts (effects) help explain how different factors affect the WPI trend as a whole as well as on a yearly basis. Only significant impacts are displayed in the user interface and they denote the sum of the individual impacts from the component occupancy models.

References

  • Ahumada JA, Hurtado J, Lizcano D (2013) Monitoring the Status and Trends of Tropical Forest Terrestrial Vertebrate Communities from Camera Trap Data: A Tool for Conservation. PLoS ONE 8(9): e73707. doi:10.1371/journal.pone.0073707
  • Buckland ST, Magurran AE, Green RE, Fewster RM (2005) Monitoring change in biodiversity through composite indices. Philos Trans R Soc Lond B Biol Sci 360: 243-254.
  • O’Brien TG, Baillie JEM, Krueger L, Cuke M (2010) The Wildlife Picture Index: monitoring top trophic levels. Anim Conserv 13: 335-343.
  • Plummer M (2003). JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DSC 2003). March: 20-22.
  • Royle, J. A., and M. Kéry. 2007. A Bayesian state-space formulation of dynamic occupancy models. Ecology 88(7):1813-1823.
  • Yu-Sung S, Masanao Yajima J (2012) R2jags: A Package for Running jags from R. Package version 0.03-08. CRAN website. Available:
    http://CRAN.R-project.org/package=R2jags Accessed 2013 Aug 7