pling explanatory models of carnivore extinction risk from comparative analyses with human population projections to identify species whose conservation status was likely to worsen.

Here, we enlarge this approach in a preliminary analysis of two drivers and all mammals. Across ecoregions, the proportion of species with risk status higher than LC was modeled (as a binomial denominator) as a function of two drivers and two summaries of biological vulnerability by using generalized additive models (Wood, 2006) to avoid forcing any particular form on the relationship. A smooth relationship was fitted to link risk level to mean human population density (Center for International Earth Science Information Network and Centro Internacional de Agricultura Tropical, 2005) and the proportion of land converted to urban or cropland (European Commission, Joint Research Centre, 2003; Center for International Earth Science Information Network et al., 2004). A second smooth relationship was fitted to control for two biotic variables [proportion of species weighing >3 kg, the size at which ecology and life history begin to influence risk strongly (Cardillo et al., 2005), and the proportion of species in the lowest quartile of global range size (K.E.J., J.B., A.P., C.D.L.O., Susanne A. Fritz, Christina Connolly, Amber Teacher, J.L.G., R.G., Elizabeth Boakes, Michael Habib, Janna Rist, Chris Carbone, Christopher A. Plaster, O.R.P.B.-E., Janine K. Foster, Elisabeth A. Rigby, Michael J. Cutts, Samantha A. Price, Wes Sechrest, Justin O’Dell, Kamran Safi, M.C., and G.M.M., unpublished data)]. Fig. 14.4 shows the marginal effect of the drivers on extinction risk. The two drivers are strongly correlated across ecoregions and interact strongly. As expected, risk is low when both drivers are at the very lowest levels. However, risk rises rapidly as either driver increases. Medium levels of land conversion and low density are associated with high levels of risk, but risk falls as land conversion rises further. This suggests that land conversion is an extinction filter (Balmford, 1996), removing one set of species sufficiently thoroughly that highly converted regions can again have low levels of risk, with only the more bulletproof taxa remaining. Scenario analysis will obviously need to count projected extinctions as well as declines and may need to consider historical as well as present driver patterns. As human density reaches high levels, risk levels become uniformly higher.

A more refined model, perhaps incorporating other drivers, could be combined with projected future driver intensity to predict where a high proportion of species will decline. Such an approach uses the spatial heterogeneity in present driver intensity as a surrogate time series, with high-intensity ecoregions suggesting what will happen elsewhere as conditions deteriorate. However, incorporating climate change into this modeling approach presents major challenges. Because it has not been a major driver of present risk patterns, we have not yet seen how intensity