We provide a century-scale view of small-mammal responses to global warming, without confounding effects of land-use change, by repeating Grinnell's early-20th century survey across a 3000-meter-elevation gradient that spans Yosemite National Park, California, USA. Using occupancy modeling to control for variation in detectability, we show substantial ( approximately 500 meters on average) upward changes in elevational limits for half of 28 species monitored, consistent with the observed approximately 3 degrees C increase in minimum temperatures. Formerly low-elevation species expanded their ranges and high-elevation species contracted theirs, leading to changed community composition at mid- and high elevations. Elevational replacement among congeners changed because species' responses were idiosyncratic. Though some high-elevation species are threatened, protection of elevation gradients allows other species to respond via migration.

We examine why demographic models should be used cautiously in Population Viability Analysis (PVA) with endangered species. We review the structure, data requirements, and outputs of analytical, deterministic single-population, stochastic single-population, metapopulation, and spatially explicit models. We believe predictions from quantitative models for endangered species are unreliable due to poor quality of demographic data used in most applications, difficulties in estimating variance in demographic rates, and lack of information on dispersal (distances, ages, mortality, movement patterns). Unreliable estimates also arise because stochastic models are difficult to validate, environmental trends and periodic fluctuations are rarely considered, the form of density dependence is frequently unknown but greatly affects model outcomes, and alternative model structures can result in very different predicted effects of management regimes. We suggest that PVA (1) evaluate relative rather than absolute rates of extinction, (2) emphasize short-time periods for making projections, (3) start with simple models and choose an approach that data can support, (4) use models cautiously to diagnose causes of decline and examine potential routes to recovery, (5) evaluate cumulative ending functions and alternative reference points rather than extinction rates, (6) examine all feasible scenarios, and (7) mix genetic and demographic currencies sparingly. Links between recovery options and PVA models should be established by conducting field tests of model assumptions and field validation of secondary model predictions.