Fig. 4: Mg/Ca and Sr/Ca cluster analysis and ln–ln PCP test.

Elemental ratios for a slow-growing stalagmites and b fast-growing stalagmites. c Comparison of ln(Mg/Ca) and ln(Sr/Ca) with regressions and bedrock_sp alongside the theoretical PCP slope of Sinclair et al.³⁴. The Sulawesi average PCP slope is calculated using the average of each stalagmite regression. Slopes shown for individual records were calculated from resampled data. Regressions on original (unevenly spaced) data made no significant difference in slopes (see Supplementary Fig. S3 and Supplementary Table S1). Data pairs in a section of AC11-06, with offset Sr/Ca (grey asterisks), show limited PCP and are not included in the average Sulawesi PCP regression. k-means cluster analysis of Mg/Ca and Sr/Ca for all Sulawesi stalagmites: Cluster 1 (blue ellipse, 1.5σ) grouping near average bedrock_sp (yellow square with uncertainty range) is interpreted as minimal PCP while Cluster 2 (orange ellipse, 1.5σ) has evolved away from bedrock_sp towards higher values, indicating enhanced PCP. Large uncertainties on bedrock_sp correspond with the uncertainties in the Sinclair PCP slope equation used to determine the Sulawesi partition coefficient, Kd_Sr (Supplementary Methods). K-means cluster analysis (k-means clustering), ellipse determination (ELLIPSATE) and linear regression (Linear Regression plot with Confidence Intervals) were performed and plotted in Matlab (https://au.mathworks.com/matlabcentral/fileexchange/).