Reconstructing palaeoclimates from biological proxies: Some often overlooked sources of uncertainty

Abstract Palaeoclimatic reconstructions from biological proxies, remains of organisms, suffer from sources of uncertainty that frequently are unacknowledged or even unrecognised. Primary amongst these is the need to identify, using available biological knowledge and understanding, the variables to which the organism(s) providing the proxy are sensitive, and hence that it is appropriate to reconstruct. These variables generally are not those conventionally recorded at meteorological stations, although they may be derived from these simply or using appropriate models. In the case of smaller organisms living close to the ground, in soil, or in aquatic environments, appropriate variables will relate to the microclimate. Reconstructing inappropriate variables introduces uncertainties and, especially under changed climatic conditions, is likely to give inaccurate results. A second important source of uncertainty arises when variables are reconstructed in isolation. In reality, most organisms respond concurrently to many variables, although two or three of these will usually predominate; these predominant variables also often have interacting effects. Isolating individual variables for reconstruction frequently will result in inaccurate reconstructions, the response to a second variable being aliased as an apparent change in the reconstructed variable. This can be overcome by concurrent reconstruction of the small number of variables that principally determine the distribution and abundance of the organism(s) from which the proxy is derived. Multi-proxy studies may give greater insight, but only when appropriate variables are reconstructed from each proxy. Further reductions in uncertainty, and new insights, are likely in future to be achieved by making concurrent reconstructions from two or more proxies, rather than, as at present, making separate reconstructions that may give incompatible results. The practice of making one-at-a-time reconstructions from individual samples along a core or profile often generates unrealistic sample-to-sample variability. Recent advances in the application of Bayesian modelling frameworks, however, offer concurrent reconstructions of two or more variables made concurrently for all samples in a sediment sequence. Recent improvements in the tools available to derive age-depth models open the possibility that in the near future it will be possible to make concurrent reconstructions from all samples from each of a series of sediment sequences. Knowledge of the physics of the climate system can then provide constraints on the spatial and temporal patterns in reconstructed values, further reducing their uncertainties.