Japan Geoscience Union Meeting 2023

Presentation information

[J] Online Poster

A (Atmospheric and Hydrospheric Sciences ) » A-CG Complex & General

[A-CG44] Future global ocean observation system: complementarity of autonomous and shipboard observations

Fri. May 26, 2023 10:45 AM - 12:15 PM Online Poster Zoom Room (9) (Online Poster)

convener:Shigeki Hosoda(JAMSTEC), Shota Katsura(Atmosphere and Ocean Research Institute, The University of Tokyo), Yosuke Fujii(Meteorological Research Institute, Japan Meteorological Agency), Shuhei Masuda(Japan Agency for Marine-Earth Science and Technology)

On-site poster schedule(2023/5/25 17:15-18:45)

10:45 AM - 12:15 PM

[ACG44-P05] Estimation of QC flag for Argo data using machine learning method

*Shinya Kouketsu1, Nozomi Sugiura1, Shigeki Hosoda1, Kanako Sato1 (1.JAMSTEC Japan Agency for Marine-Earth Science and Technology)

Keywords:Global marine observation networkobservation network, Autonomous observations, Quality control

The ocean observation network using autonomous profilers, which is coordinated, maintained, quality-controlled, and data released under the Argo project, is a major foundation for the modern ocean observation network. The number of profiles released under this project amounts to 12,000 profiles per month. These profiles are released through immediate quality control for use of quick view, and then released as a more reliable data set through delayed quality control that withstands detailed scientific analysis. The status of the profile data to date indicates that more than 20% of the quality control flags attached during immediate quality control are replaced during delayed quality control (depending on the data center), suggesting that delayed quality control has a certain impact on data flow as well. In addition, the importance of delay quality control has recently become apparent again due to the problem of quick temporal drift of salinity sensors on floats and impact analysis using data assimilation. On the other hand, Argo data is also used for quick view (ex. operational data centers). As quick quality control is also considered important, it would be useful to generate data sets more quickly and with better quality. Based on this background, JAMSTEC is developing a machine learning-based quality control flagging method with the aim of enabling to produce data sets that bridge the gap between immediate and delayed quality control in a relatively immediate manner by learning from existing delayed quality control results. We report on the progress of the project.