Multi-model climate projections for biodiversity risk assessments

Species distribution models, linked to climate projections, are widely used in extinction-risk assessment and conservation planning. However, the degree of confidence that we can place on future climate-change projections depends on global climate-model performance and involves uncertainties that need to be assessed rigorously via climate-model evaluation. Performance assessments are important because the choice of climate model influences projections of species' range movement and extinction risk. A consensus view from the climate modeling community is that no single climate model is superior in its ability to forecast key climatic features. Despite this, the advantages of using multi-model ensemble-averaged climate forecasts to account for climate-model uncertainties have not been recognized by ecologists. Here we propose a method to use a range of skill and convergence metrics to rank commonly used atmosphere–ocean general circulation models (AOGCMs) according to their skill in reproducing 20-year observed patterns of regional and global climates of interest, and to assess their consistency with other AOGCMs. By eliminating poorly performing models and averaging the remainder with equal weights, we show how downscaled annual multi-climate-model ensemble-averaged forecasts, which have a strong regional focus, can be generated. We demonstrate that: (1) model ranking (match of simulated to observed conditions) differs according to the skill metric used, as well as the climate variable and season considered; (2) although the multi-model averaged result tends to outperform single models at a global scale, at the continental scale at least some models can perform better than the multi-model average; and (3) forecasts for the Australian region, which are often based on a single AOGCM (CSIRO-3.0), show spatial patterns of change that differ noticeably from ensemble-average projections based on a subset of better-performing AOGCMs. Our suggested approach—novel in the ecology discipline—provides a straightforward, consistent, and defensible method for conservation practitioners and natural-resource managers to generate estimates of future climate change at a spatial resolution suitable for biodiversity impact studies.