Reducing uncertainty in projections of extinction risk from climate change

Aim Concern over the implications of climate change for biodiversity has led to the use of species-climate 'envelope' models to forecast risks of species extinctions under climate change scenarios. Recent studies have demonstrated significant variability in model projections and there remains a need to test the accuracy of models and to reduce uncertainties. Testing of models has been limited by a lack of data against which projections of future ranges can be tested. Here we provide a first test of the predictive accuracy of such models using observed species' range shifts and climate change in two periods of the recent past. Location Britain. Methods Observed range shifts for 116 breeding bird species in Britain between 1967 and 1972 (t(1)) and 1987-91 (t(2)) are used. We project range shifts between t(1) and t(2) for each species based on observed climate using 16 alternative models (4 methods x 2 data parameterizations x 2 rules to transform probabilities of occurrence into presence and absence records). Results Modelling results were extremely variable, with projected range shifts varying both in magnitude and in direction from observed changes and from each other. However, using approaches that explore the central tendency (consensus) of model projections, we were able to improve agreement between projected and observed shifts significantly. Conclusions Our results provide the first empirical evidence of the value of species-climate 'envelope' models under climate change and demonstrate reduction in uncertainty and improvement in accuracy through selection of the most consensual projections.