Abstract. Detecting and explaining differences between palaeoclimates can provide
valuable insights for Earth scientists investigating processes that are
affected by climate change over geologic time. In this study, we describe
and explain spatiotemporal patterns in palaeoclimate change that are
relevant to Earth surface scientists. We apply a combination of multivariate
cluster and discriminant analysis techniques to a set of high-resolution
palaeoclimate simulations. The simulations were conducted with the ECHAM5
climate model and consistent setup. A pre-industrial (PI) climate
simulation serves as the control experiment, which is compared to a suite of
simulations of Late Cenozoic climates, namely a Mid-Holocene (MH, approximately 6.5 ka), Last Glacial Maximum (LGM, approximately 21 ka) and Pliocene (PLIO, approximately 3 Ma)
climate. For each of the study regions (western South America, Europe,
South Asia and southern Alaska), differences in climate are subjected to
geographical clustering to identify dominant modes of climate change and
their spatial extent for each time slice comparison (PI–MH, PI–LGM and
PI–PLIO). The selection of climate variables for the cluster analysis is
made on the basis of their relevance to Earth surface processes and includes
2 m air temperature, 2 m air temperature amplitude, consecutive freezing days,
freeze–thaw days, maximum precipitation, consecutive wet days, consecutive
dry days, zonal wind speed and meridional wind speed. We then apply a
two-class multivariate discriminant analysis to simulation pairs PI–MH,
PI–LGM and PI–PLIO to evaluate and explain the discriminability between
climates within each of the anomaly clusters. Changes in ice cover create
the most distinct and stable patterns of climate change, and create the best
discriminability between climates in western Patagonia. The distinct nature
of European palaeoclimates is statistically explained mostly by changes in
2 m air temperature (MH, LGM, PLIO), consecutive freezing days (LGM) and
consecutive wet days (PLIO). These factors typically contribute
30 %–50 %, 10 %–40 % and 10 %–30 %, respectively, to climate
discriminability. Finally, our results identify regions particularly prone
to changes in precipitation-induced erosion and temperature-dependent
physical weathering.