Effects of Biased and Incomplete Data and Reserve Selection

Efficiently conserving biological diversity in networks of protected areas (reserves) is critical due to the intense pressure from competing land uses that exists in most regions of the world. Computerized reserve selection algorithms that select sites with complementary species compositions create potential reserve networks that contain all biodiversity features requiring protection in the smallest number of sites. These computer programs have greatly increased our ability to meet specified conservation goals and produce reserve network options that are superior to those selected in an ad hoc manner. Unfortunately, the methods are data-intensive and most regions lack accurate distribution maps for the majority of species. Most available species distribution data are biased in some way (i.e. higher sampling intensity closer to roads or within current reserves); however, they are commonly used to select sites for inclusion in reserve networks because they are considered to be the best data available.

Joanna Grand and I along with Mike Cummings (University of Maryland), Taylor Ricketts (World Wildlife Fund), and Tony Rebelo (South African National Biodiversity Institute) are interested in understanding how well we can protect biodiversity when sites are selected based on incomplete and biased data. We are using data on members of the plant family Proteaceae of the Cape Floristic Region of South Africa (purple region on map) as the model system for this project. It is one of the most complete species distribution data sets available at a point locality resolution. In our first analysis, we treated this high-quality database as ‘complete’, and introduced three realistic sampling biases characteristic of biodiversity databases: a detectability sampling bias and two forms of roads sampling bias.

We then used MARXAN, a decision-support system for the complementarity-based design of conservation reserve networks, to compare reserve networks constructed using complete, biased, and randomly-sampled data. To produce a sufficient range of solutions for comparison, we simulated 1000 biased and 1000 random incomplete datasets from the full Proteaceae dataset and conducted 1000 MARXAN runs for each dataset. This study design required a total of 1.2002 x 10⁷ separate MARXAN runs which was possible to complete in only a few weeks by running them asynchronously in parallel on the University of Maryland’s Lattice network.

The results of this analysis demonstrated that all forms of biased sampling performed worse than both the complete dataset and equal-effort random sampling. Biased sampling failed to detect a median of 1-5% of species, and resulted in reserve networks that were 9-17% larger than those designed with complete data. Spatial congruence and the correlation of irreplaceability scores (which measure the importance of each site in the reserve network solution) between reserve networks selected with biased and complete data were low. Thus, reserve networks based on biased data require more area to protect fewer species and identify different locations than those selected with randomly sampled or complete data.