hide
Free keywords:
Glacial-interglacial cycles; Mid-Brunhes event; Mid-Brunhes transition; Machine learning; Echo state network
Abstract:
The Mid-Brunhes Transition (MBT) refers to the change in the amplitude of glacial-interglacial cycles around 430 ka BP, with more pronounced, warmer interglacials after ca. 430 ka BP. Despite the advances in the understanding of glacial cycles, the cause and mechanism of the MBT are still not entirely clear. In this study we examine (i) whether the MBT is caused by a change in the intrinsic dynamics of glacial cycles and (ii) how important the systematic changes of the orbital elements across the MBT are for the occurrence of the MBT. In order to address these questions, we take a pure machine-learning approach. We develop an artificial neural network model which provides a skilful 21-ka ahead prediction of glacial-interglacial changes in the LR04 benthic δ18O stack record as well as a sea level reconstruction obtained by an inverse model. This allows us to predict the interglacial levels from glacial conditions. Although the neural network model is trained over a pre-MBT period of 900–450 ka BP, it exhibits the intensification of interglacials after 450 ka BP. This suggests that the dynamical characteristics generating the stronger post-MBT interglacials is inherent already before the MBT. When the neural network model is forced by a hypothetical insolation for which the amplitude of the obliquity cycles is kept at pre-MBT level, the MBT-like phenomenon does not appear in our simulations. In line with earlier suggestions, our results thus give quantitative evidence that the MBT is caused by amplitude changes of the obliquity forcing. For comparison, our results suggest that the change in the mean eccentricity level across the MBT has a smaller impact on the appearance of the MBT.
