Figure 2 salmon=south). The number at each location corresponds with sequence numbers in ESM 2. The sample scores are plotted with the oldest (bottom) at the far left and the youngest (top) at the far right. The individual sample scores are weighted averages of the response (pollen taxa) variable scores (also applies to ESM 8) Fig. 2 Map of the 47 analysed Eemian pollen sequences and indi- vidual sample scores (standard deviation units) of selected detrended canonical correspondence analysis (DCCA) axis 1 plots for 18 sequences (the remaining 29 sequences are in ESM 8). Colours of locations indicate assigned region (blue=north; green=central; within-sample standard deviation is unity along the ordina- tion axes which are here constrained by sample depth or order. The change in weighted average (WA) sample scores (CaseR sensu ter Braak and Smilauer 2012) reflects com- positional change or turnover in standard deviation (SD) units. PCs are more “neutral” than DCCA in that they make fewer assumptions of the data than DCCA does. In the PC approach, a PC is fitted to the entire Eemian dataset of 2,840 samples. Sample locations along the final PC are determined and scaled to 0-1. Maximum difference of sample scores within a sequence is a relative turnover measure (Simpson and Birks 2012). For each sequence, total compositional turnover is estimated and within each sequence we explore patterns of turnover. Emphasis here is placed on the DCCA results for both total turnover and changes within a sequence because they are expressed in ecologically interpretable SD units of taxon turnover (Figs. 2, 3; ESM 4). The PC results for total turnover are summarised in Fig. 3b. Compositional turnover is estimated using DCCA con- strained by depth (or order) plus depth” (Birks 2007; ter Braak and Smilauer 2012) and principal curves (PCs) (Simp- son and Birks 2012). DCCA directly scales variables’ (in our case pollen taxa) ordination scores such that their average