The observational scale sensitivity of pCO2 reconstruction uncertainties and biases in the Southern Ocean: going from the observations we have to the observations we need

The Southern Ocean (SO) is a vital buffer against climate change, absorbing about 75% of excess heat and 50% of the oceanic uptake and storage (42±5 PgC) of anthropogenic CO2 emissions to the atmosphere since the pre-industrial age (Frölicher et al., 2015; Landschutzer et al., 2015). However, machine learning (ML) reconstructions of surface CO2 (pCO2) and CO2 fluxes have plateaued in accuracy, a limit we call the "wall" (Ref). This study investigates one of the causes of these limitations and advances some new insights related to the observing scale sensitivity of the reconstructions. Current limitations are attributed to sparse data and a summer sampling bias (Gregor et al., 2019). While increased observations with autonomous platforms (Floats, WaveGliders, Saildrones) are underway, their effectiveness remains untested. This study utilizes an Observing System Simulation Experiment (OSSE) approach using a mesoscale-resolving Peri-Antarctic (BIOPERIANT12) forced ocean model (NEMO-PISCES 3.4) to examine the sensitivity of  ML reconstructions to observing scale choices. We hypothesized that addressing critical spatial and temporal sampling scales would overcome the "wall." Analyses identified two critical missing scales: the seasonal cycle of meridional gradients and intra-seasonal variability (Djeutchouang et al., 2022). Results demonstrate that resolving the seasonal cycle of meridional gradients is the primary requirement for improved accuracy, followed by resolving intra-seasonal variability. This research lays the groundwork for optimizing future CO2 observation strategies in the Southern Ocean, using a more cost-effective combination of ship and autonomous platforms to have a decisive impact in reducing the uncertainties and biases of the SO carbon budget and a better understanding of this crucial region's role in mitigating climate change.