Minding the protection gap: estimates of species' range sizes and holes in the Protected Area network

Abstract

In a paper in this issue (Beresford et al., 2011), we set out to quantify the extent to which site-based conservation initiatives overlapped with the ranges of globally threatened bird species in Africa, using a GIS approach. We considered Protected Areas and Important Bird Areas (IBAs), both of which cover c. 7% of the land surface of Africa. The former, as noted by both Brooks & Matiku (2011) and Rodrigues (2011), carry significant governmental weight despite not necessarily being identified for their conservation value, while the latter (which are of conservation value to other taxa in addition to birds – Pain et al., 2005) have no legal standing, but are identified using objective criteria that include their value for globally threatened species (Fishpool & Evans, 2001). We find that Protected Areas perform poorly at protecting the ranges of Africa’s most threatened birds, covering just 13.9% on average, and even where a Protected Area overlaps part of an IBA, the part excluded from the Protected Area has higher value for these species than the protected part. As Rodrigues (2011) notes, the limitations of GIS studies that utilize Extent of Occurrence (EOO) maps are well documented. We therefore attempted to reduce the commission errors inherent in the use of EOO maps by analysing extent of suitable habitat (ESH: the area of potentially suitable vegetation types within the altitudinal preferences of the species (Rondinini, Stuart & Boitani, 2005; Buchanan et al., 2008)). While it is desirable to move beyond interpolation approaches based on EOO or ESH and utilize only presence/absence data derived from recent complete field surveys, these are not available for the majority of species. Even in the most well-watched countries, such as the United Kingdom, the most reliable information on the distribution of birds comes from systematic atlas surveys in which ranges are interpolated from sampled field counts which themselves record only a tiny proportion of the individuals present. At least for the foreseeable future therefore, we will have to continue to assess geographical distributions using interpolative approaches, whilst acknowledging the limitations of these methods. ESH estimates were considerably smaller than EOOs for most species, with a concomitant reduction in commission errors and increase in omission errors. Reducing commission errors reduces the likelihood that a species is assumed to be conserved elsewhere (Rodrigues & Gaston, 2001), and hence in our study, the danger of overestimating the coverage of species’ ranges by Protected Areas or IBAs. Omission errors, on the other hand, reduce the estimated overlap between sites and species’ ranges. Considering the information we presented, Rodrigues (2011) suggests that the ESH approach was not well able to distinguish between occupied and unoccupied areas of the EOO, because both approaches have similar ratios of true presences to total predicted presences. We agree that by focusing on other aspects of the analysis and by presenting only a broad overview of the accuracy of the ESH approach compared with EOOs (table 1; Beresford et al., 2011), we did not show the full advantage of the ESH method. A species-level analysis indicates that although the ESH comprises, on average, 28% of the EOO – hence reducing the chance of the species being assigned to areas where it did not occur – this did not make any difference to the accuracy of the estimated distribution for 74 species [47%, assessed by comparing the total numbers of errors (commission plus omission) using ESH vs. EOO]. Importantly, however, it did improve accuracy for 58 species (37%), decreasing commission errors by, on average, 11.8 IBAs per species. This was at the expense of a reduction in the accuracy of maps for 25 species, due to an increase in omission errors, although these averaged just 1.8 sites per species (Fig. 1). While considerable inaccuracies remain, especially in commission errors (table 1; Beresford et al., 2011), we suggest that the increase in accuracy for over a third of