Two modelling experiments are described in this article. The first (middle Pliocene control experiment) uses the full set of PRISM2 boundary conditions to investigate the nature of the middle Pliocene palaeoclimate and its relative difference to the present-day climate of the USA. The second experiment (Pliocene sensitivity experiment) utilises global PRISM2 Pliocene SST estimates, sea level and vegetation distributions while setting terrestrial ice cover and orographic parameters to modern conditions. Specifically, this increases ice volume over Greenland and Antarctica by 50 percent and 33 percent respectively, and raises the elevation of the western cordillera of North America by 50 percent. The aim of this approach is to investigate the relative importance of certain climatic boundary conditions in producing the climate changes predicted by the HadAM3 model for the middle Pliocene over the USA.

For both experiments the HadAM3, GCM was integrated for 12 years and climatological means were compiled for the final 10 years only. Time-series analysis of all climate variables for the whole of the 12-year simulation show that disregarding the first two years of simulation is sufficient for the model climatology to reach a dynamic equilibrium. Furthermore, the slowest component to reach equilibrium is soil moisture. An examination of global mean soil moisture showed no significant trend in the upper layers after two years. However, there was a small residual trend in the deepest layer that lasted for approximately five years. Overall, the length of the model run is comparable to previous modelling studies for the middle Pliocene (e.g., Sloan et al. 1996). In the results, statistical tests of the Pliocene control-run model predictions are based on the Student t-test described by Chervin and Schneider (1976). Although this statistical test has many limitations (see Wigley and Santer 1990), it is commonly used and is a simple test. Moreover, the changes in climate between the middle Pliocene and present-day are generally large and so clearly exceed the statistical test.