SUPPLEMENTARY MATERIAL
METHODSData were compiled from the pre-sorted georoc database ocean island files (http://georoc.mpchmainz.gwdg.de/georoc/). Hot spots such as Aetna or Samoa with geotectonically complex situations and possible direct interference with subduction were avoided; otherwise we used those with sufficiently large data sets. Each file was filtered for analyses with 95-101.5 wt% totals, samples outside this range or listed as altered were excluded. All bulk rock compositions are then normalized to 100 wt% total on a volatile free basis with all Fe as FeO.The histogram of total alkalis for extrusives shows a clear cut-off to <1.4 wt% K 2 O+Na 2 O. Of e.g. 9'000 extrusives of the Hawaiian hotspot only 24 are below this values, these are essentially from three publications (without geographical coherence), suggesting that these rocks were altered or possibly have an analytical problem; this is corroborated by a K 2 O/Na 2 O ranging from 0.01 to 2 for these 24 samples. We have hence excluded all rocks with <1.4 wt% Na 2 O+K 2 O. Similarly, we did not accept trace element concentrations measured by XRF that are below 5-15 ppm (depending on the element).Within each ocean island, rocks were sorted for (i) Those with X Mg >0.77 and/or Ni>650 ppm, this group is mostly olivine cumulative and was excluded from all plots. (ii) Primitive melts were designated when X Mg =0.77-0.65, which corresponds to equilibrium with olivine of Fo 92-88 at 5-15 mol% Fe 3+ (of Fe tot ), and Ni=150-650 ppm, corresponding to roughly 2000-4500 ppm Ni in mantle olivine (note that the K D olivine/melt of Ni strongly depends on temperature, hence this range is chosen rather large). (iii) To enlarge the dataset, rocks with 0.55100 analyses were retained, the dataset sorted for the above criteria contains 124-198 analyses for Pitcairn, Ascension, St. Helena and Tristan da Cunha; 242-331 for GSA Data Repository 2018136