The accuracy of δ11B measurements of foraminifers

Values of δ11B reported in the literature for Holocene samples of the same foraminiferal species show large variability (6‰) relative to cited precision (< 1‰). This is indicative of significant inter-laboratory biases and raises concerns about the accuracy of foraminiferal proxy pH records. To investigate this problem we have measured the δ11B of modern ocean carbonates using several different analytical procedures. We have used total evaporation negative thermal ionisation mass spectrometry (TE-NTIMS), with various sample preparation and loading procedures and multi-collector inductively coupled plasma mass spectrometry (MC-ICPMS). The δ11B of pure boric acid solutions measured by TE-NTIMS and MC-ICPMS agree well, demonstrating no fundamental biases between the techniques. Yet the δ11B values measured by TE-NTIMS for non-foraminiferal carbonates are about 2‰ lighter than those by MC-ICPMS whereas foraminifers measured by the same standard TE-NTIMS procedure are 2 to 6‰ heavier than those by MC-ICPMS. Significantly, we found that foraminiferal samples kept in acidic solution over several months yielded lower δ11B values (up to 5‰ lower) and better reproducibility when re-measured by TE-NTIMS. We infer that organic material, released on foraminiferal dissolution, causes biases in NTIMS measurements. No residual organics should be present in the MC-ICPMS measurements as matrix is separated from sample before analysis. Moreover we demonstrate by standard addition the absence of any matrix influence in the MC-ICPMS procedure beyond the inherent uncertainties in the standard addition approach (∼ ± 0.35‰ 2sigma). Degradation of the inferred organic residue, either by long term storage in acidic solution or by loading samples in 30% H2O2, reduces but does not totally remove the (variable) bias between TE-NTIMS and MC-ICPMS δ11B measurements. Despite these problems, careful analysis of similar samples by NTIMS may permit data to be obtained with consistent relative differences that still yield valuable proxy records, but there is clearly considerable potential for inaccuracies to result in such an approach from hard to detect changes in matrix.