Transfer functions between Biogeochemical-Argo and GO-SHIP to complement and expand observations of nutrient and CO2 dynamics in the open ocean using Argo-O2.
Global repeat hydrography from GO-SHIP and its predecessors provides a unique combination of oceanographic data encompassing a wide variety of parameters. Such cruises fed into the impressive collection of high-quality data accumulated in GLODAPv2. Apart from the spatial distribution, a different view on the data in GLODAPv2 gives information about the interrelation between biogeochemical parameters in different regions and water masses.
These have recently been mapped in CANYON (Sauzede et al., 2017) using a neural network: CANYON predicts the concentration of carbonate system parameters and nutrients from simple inputs of P, T, S, O2, and geolocation, using the GLODAPv2 dataset as training/reference.
Here we use CANYON to transfer that information about biogeochemical interrelations to Argo-O2 observation. Such a transfer has only become possible thanks to recent advances in autonomous O2 measurements and in-situ calibration through optode in-air measurements.
We present examples of (1) validating existing BGC-Argo measurements of, e.g., nitrate or pH, and (2) expanding predictions to floats without other biogeochemical sensors to increase coverage or to parameters that are not yet observable by Argo (e.g., silicate, pCO2).
More recently, we developed CONTENT (Bittig et al., in prep.), a method to better constrain estimates of the carbonate system parameters compared to CANYON. Improvement is most pronounced for pCO2, with a global accuracy estimated for CANYON at 7.6 % (30 µatm at 400 µatm) and up to 3.3 % (13 µatm at 400 µatm) for CONTENT.
CONTENT pCO2 estimates are validated against 6 cruises across the Atlantic, where surface underway measurements of T/S/O2 as well as pCO2 were available. Overall agreement is good, except for regions with significant past productivity where the associated pCO2 drawdown is more persistant than O2 supersaturation (e.g., Patagonian Shelf, Mauritanian upwelling). Here, CONTENT seasonally overestimates in-situ pCO2 by ca. 20 - 30 µatm. On a global scale, CONTENT surface pCO2 estimates are unbiased compared to a SOCAT-based climatology.
Such parameterizations can help to improve ocean CO2 inventory and flux estimates, e.g., in data-sparse regions like the Southern Ocean, albeit accuracy of CONTENT pCO2 may be insufficient in some cases for direct flux calculations. Most promising applications will come from the link of observation systems, e.g., Argo-O2 or Biogeochemical-Argo with CONTENT for the water column structure and the VOS network / SOCAT for accurate surface pCO2.