A machine learning approach on ARGO floats for the correction of degraded salinity.

Argo floats are autonomous floats designed to measure temperature and salinity of the world oceans. Ever since its inception, the Argo program is complementing the ocean observations which are otherwise taken by ship-based CTD's and XBT/XCTDs. Once deployed these floats goes to as deep as 2000 m and while coming up measure temperature and salinity of the underlying ocean automatically. These floats act as a substitute to the ship-based data sets and currently as many as ~3800 are active in the global oceans. The data from these platforms have grown enormously and the state of the art techniques like big data and data analytics are needed to effectively handle them. These instruments being autonomous in nature measure and transmit data seamlessly irrespective of the weather, season, and region. However, the salinity sensors on these floats are sensitive to bio-fouling and can cause degradation to the data. As these are one time deployed and data is continuously obtained they are not available for calibration unlike the instruments on the ship. In this work machine learning approach is used as alternate method for delayed mode quality control (DMQC) to check the degradation of the sensors and correct the same so that the data can be used in scientific analysis.