Posters

Poster Session 3, Wednesday,  October  5, 16:00–18:00

Poster 27

Development of a machine learning approach for the estimation of the marine CO2 partial pressure over the global coastal ocean in the frame of the CO2COAST project

Coastal shelves absorb about 17% of oceanic CO₂ influx. However, large uncertainties in coastal carbon fluxes exist due to the undersampling in both space and time. The main challenge of the CO2COAST project is to evaluate the respective contribution of the estuaries vs. continental shelves to the CO₂ fluxes over the global coastal waters. For that purpose, the surface-ocean CO₂ partial pressure, pCO₂w, has first to be estimated from satellite remote sensing over the global coastal waters at high spatial resolution (i.e. 1 km 2 ). To accomplish this aim, a global coastal database has been built, constituted of in-situ data of pCO₂w, (SOCAT database) presenting 580 10³ satellite match-up (from 1997 to 2020) in coastal waters. This database gathers in-situ pCO₂w, salinity, and temperature, and satellite measurements of remote sensing reflectance, chlorophyll-a, absorption by colored dissolved organic matter, sea surface salinity, and temperature. Such a multidimensional dataset requires machine learning methods to exploit efficiently spatial and temporal embedded structures. For that purpose, we introduce a new clusterwise regression methodology used to classify observations into clusters characterized by their own pCO₂w vs. satellite input parameters relationships. This method allows to identify and characterize regional relationships in a global dataset. The algorithm and its validation based on match-up exercises and temporal series will be presented; We will discuss the dominant driving input parameters according to each delimited region. As a result, the pCO₂w is estimated with an error of 25 uatm, with a correspondence of 93% with in-situ values.

Roy El Hourany, Univ. Littoral Côte d’Opale, Cnrs, Univ. Lille, Laboratoire d’Océanologie et de Géoscience (LOG), Wimereux, France, [email protected], 0000-0002-6454-1645

Hubert Loisel, Univ. Littoral Côte d’Opale, Cnrs, Univ. Lille, Laboratoire d’Océanologie et de Géoscience (LOG), Wimereux, France, [email protected]

Daniel S.F. Jorge, Univ. Littoral Côte d’Opale, Cnrs, Univ. Lille, Laboratoire d’Océanologie et de Géoscience (LOG), Wimereux, France, [email protected]

Sylvie Thiria, Sorbonne Université (Université Paris VI, CNRS, IRD, MNHN), IPSL/LOCEAN, Paris, France, [email protected]

Ndeye Niang-Keita, CNAM, Centre d’études et de recherche en informatique et communications (CEDRIC), Paris, France, [email protected]

Julien Demaria, ACRI-ST, Sophia-Antipolis, France, [email protected]

Antoine Mangin, ACRI-ST, Sophia-Antipolis, France, [email protected]

Marine Bretagnon, ACRI-ST, Sophia-Antipolis, France, [email protected]