Benchmark-based assessments of best-practice automatic quality control for ocean salinity in-situ observations

Abstract. Ocean salinity is an Essential Ocean Variable that underpins our understanding of Earth’s climate system and the global hydrological cycle. Reliable and accurate salinity observations are therefore critical, yet historical profile archives contain heterogeneous data quality, and the community lacks benchmark-based recommendations for salinity automatic quality control (AutoQC). Under the International Quality-controlled Ocean Database (IQuOD) framework, we developed a benchmark-based approach to compare suites of salinity AutoQC tests from seven international oceanographic institutions. The suites comprised 30 individual checks and were evaluated based on several manually (expert) quality-controlled benchmark datasets. We assessed ~6,400 QC configurations in receiver operating characteristic (ROC) space and identified three best-practice schemes for distinct user priorities. These comprised a high-detection scheme with a high true-positive rate (HTPR), a conservative scheme with a low false-positive rate (LFPR), and a balanced scheme with a cut-off between HTPR and LFPR. We further examined their spatial and vertical performance and investigated factors underlying their contrasting behaviour, with local climatological range check and the constant-value checks accounted for most between-scheme differences. This result suggests that regional salinity structure, observational sampling density, and threshold design may strongly affect the trade-off between detecting erroneous observations and retaining valid data. Our results provide a transparent framework for salinity AutoQC best-practice recommendations while highlighting where an additional regional and depth-dependent validation remains necessary. The resulting schemes could support generating internally consistent QC flags for the first release of the IQuOD salinity database and, more broadly, could allow varying data centers and producers to select best-practice QC procedures for ocean reconstruction, modeling and operational oceanography, as well as allow the general data users to choose salinity QCed datasets flagged using the HTPR, LFPR, or Compromise schemes based on their research objectives and tolerance for potentially data errors.