Sensitivity of paleoclimate simulation results to season definitions

According to the Milankovitch theory, slow variations of the Earth's orbital parameters change the amplitude of the seasonal cycle of insolation and are considered to be the main forcing mechanism of glacial-interglacial cycles. Because of the precession and changes in eccentricity the length of seasons also varies. No absolute phasing is then possible between the insolation curves of two different periods. Various solutions to compare different periods have been given either for astronomical computations [e.g., Berger and Loutre, 1991; Laskar et al., 1993] or for model simulations [e.g., Kutzbach and Otto-Bliesner, 1982; Mitchell et al., 1988], but the sensitivity of model results to the different possible solutions has never been quantified. Our results, based on simulations of the last interglacial climate, 126 kyr B.P., where changes in the length of the seasons are large, clearly show that phase leads or lags between the various solutions used introduce biases in the analysis of insolation and climate change of the same order of magnitude as the Milankovitch forcing. Our main conclusions are that (1) when comparing various model simulations, the date of the vernal equinox (i.e., the phasing of the seasonal cycle of insolation) as well as the definition of seasons must be the same for all models in order to avoid artificial differences; (2) seasons based on astronomical positions are preferred to seasons defined with the same lengths as today, since they better account for the phasing of insolation curves. However, insolation is not the only forcing in most atmospheric general circulation model simulations. We also discuss the impact of the calendar hidden behind the definition of the seasonal cycle of the other boundary conditions, such as sea ice or sea surface temperatures.