Basin‐scale pCO2 maps estimated from ARGO float data: A model study

Abstract

A novel method for mapping surface pCO2 on a basin scale using ARGO floats is presented and tested in the framework of an eddy‐resolving biogeochemical model of the North Atlantic. Voluntary observing ship (VOS) and ARGO float coverage of the year 2005 is applied to the model to generate synthetic “observations.” The model‐generated VOS line “observations” of pCO2, SST, and SSS form a training data set for a self‐organizing neural network. The trained neural network is subsequently applied locally to estimate pCO2 from the model‐generated ARGO float SST and SSS data. The local pCO2 estimates at the simulated float positions are extrapolated using objective mapping. The accuracy of the nearly basinwide pCO2 estimates is assessed by comparing against the pCO2 output of the model that serves as synthetic “ground truth.” For an ARGO float coverage of the year 2005, the resulting monthly mean pCO2 maps cover 70% of the considered area (15°N to 65°N) with an RMS error of 15.9 μatm. Compared to remote sensing‐based estimates that suffer from large regional gaps in optical satellite data coverage, the RMS error in reproducing the annual cycle of pCO2 can be reduced by 42% when the more evenly distributed ARGO float‐based data are used.