RATIONALE: The precision obtained in routine isotope analysis of water (δ17 O, δ18 O, δ2 H, 17 O-excess and d-excess values) using cavity ring-down spectroscopy is usually below the instrument specifications provided by the manufacturer. This study aimed at reducing this discrepancy, with particular attention paid to mitigating the memory effect (ME). METHODS: We used a Picarro L2140i analyzer coupled with a high-precision A0211 vaporizer and an A0325 autosampler. The magnitude and duration of the ME were estimated using 24 series of 50 successive injections of samples with contrasting compositions. Four memory correction methods were compared, and the instrument performance was evaluated over a 17-month period of routine analysis, using two different run architectures. RESULTS: The ME remains detectable after the 30th injection, implying that common correction procedures only based on the last preceding sample need to be revised. We developed a new ME correction based on the composition of several successive samples, and designed a run architecture to minimize the magnitude of the ME. The standard deviation obtained from routine measurement of a quality assurance water sample over a seven-month period was 0.015‰ for δ17 O, 0.023‰ for δ18 O, 0.078‰ for δ2 H, 0.006‰ for 17 O-excess and 0.173‰ for d-excess. In addition, we provided the first δ17 O and 17 O-excess values for the GRESP certified reference material. CONCLUSIONS: This study demonstrates the long-term persistence of the ME, which is often overlooked in routine analysis of natural samples. As already evidenced when measuring labelled water, it calls for consideration of the compositions of several previous samples to obtain an appropriate correction, a prerequisite to achieve high-precision data.
RATIONALE: The precision obtained in routine isotope analysis of water (δ17 O, δ18 O, δ2 H, 17 O-excess and d-excess values) using cavity ring-down spectroscopy is usually below the instrument specifications provided by the manufacturer. This study aimed at reducing this discrepancy, with particular attention paid to mitigating the memory effect (ME). METHODS: We used a Picarro L2140i analyzer coupled with a high-precision A0211 vaporizer and an A0325 autosampler. The magnitude and duration of the ME were estimated using 24 series of 50 successive injections of samples with contrasting compositions. Four memory correction methods were compared, and the instrument performance was evaluated over a 17-month period of routine analysis, using two different run architectures. RESULTS: The ME remains detectable after the 30th injection, implying that common correction procedures only based on the last preceding sample need to be revised. We developed a new ME correction based on the composition of several successive samples, and designed a run architecture to minimize the magnitude of the ME. The standard deviation obtained from routine measurement of a quality assurance water sample over a seven-month period was 0.015‰ for δ17 O, 0.023‰ for δ18 O, 0.078‰ for δ2 H, 0.006‰ for 17 O-excess and 0.173‰ for d-excess. In addition, we provided the first δ17 O and 17 O-excess values for the GRESP certified reference material. CONCLUSIONS: This study demonstrates the long-term persistence of the ME, which is often overlooked in routine analysis of natural samples. As already evidenced when measuring labelled water, it calls for consideration of the compositions of several previous samples to obtain an appropriate correction, a prerequisite to achieve high-precision data.