| Summary: | The Southern Ocean plays a pre-eminent role in the global carbon-climate system. Model studies show that since the start of the preindustrial era, the region has absorbed about 75% of excess heat and 50% of the oceanic uptake and storage (42±5 PgC) of anthropogenic CO2 emissions. However, due to the spatial and seasonal sparseness of the Southern Ocean CO2 observations (biased toward summer), this role is poorly understood. The seasonal sampling biases have hampered observation-based reconstructions of partial pressure of CO2 at the surface ocean (pCO2) using machine learning (ML) and contributed to the convergence of the root mean squared errors (RMSEs) of ML methods to a common limit known in the literature as the “wall”. The hypothesis here is that addressing the critical missing sampling scale will get the community reconstructions of pCO2 “over the wall”. In this study, I explore the sensitivity of pCO2 reconstructions to these observational scale gaps. Using a scale-sensitive sampling strategy means adopting a sampling strategy which addresses these observational limitations including intra-seasonal as well as seasonal sampling aliases in high eddy kinetic energy and mesoscale-intensive regions. In increasing CO2 sampling efforts in the Southern Ocean using autonomous sampling platforms such as floats, Wave Gliders and Saildrones, the community has tried to answer this problem, but the effectiveness of these efforts has not yet been tested. This study aims to do this evaluation and advance our understanding of the sampling scale sensitivities of surface ocean pCO2 reconstructions from machine-learning techniques and contribute – through a scale-sensitive sampling strategy of observing platforms in the Southern Ocean – to breaking through the proverbial “wall”. This aim was achieved through a series of observing system simulation experiments (OSSEs) applied to a forced mesoscale-resolving (±10km) ocean NEMO-PISCES physics-biogeochemistry model with daily output. In addition to underway ships, the sampling ...
|
|---|