Abstract
The Southern Ocean is a complex system yet is sparsely sampled in both space and time. These factors raise questions about the confidence in present sampling strategies and associated machine learning (ML) reconstructions. Previous studies have not yielded a clear understanding of the origin of uncertainties and biases for the reconstructions of the partial pressure of carbon dioxide (pCO2) at the surface ocean (pCO2ocean). Here, we examine these questions by investigating the sensitivity of pCO2ocean reconstruction uncertainties and biases to a series of semi-idealized observing system simulation experiments (OSSEs) that simulate spatio-temporal sampling scales of surface ocean pCO2 in ways that are comparable to ocean CO2 observing platforms (Ship, Waveglider, Carbon-float, Saildrone). These experiments sampled a high spatial resolution (±10 km) coupled physical and biogeochemical model (NEMO-PISCES) within a sub-domain representative of the Sub-Antarctic and Polar Frontal Zones in the Southern Ocean. The reconstructions were done using a two-member ensemble approach that consisted of two machine learning (ML) methods, (1) the feed-forward neural network and (2) the gradient boosting machines. With the baseline observations being from the simulated ships mimicking observations from the Surface Ocean CO2 Atlas (SOCAT), we applied to each of the scale-sampling simulation scenarios the two-member ensemble method ML2, to reconstruct the full sub-domain pCO2ocean and assess the reconstruction skill through a statistical comparison of reconstructed pCO2ocean and model domain mean. The analysis shows that uncertainties and biases for pCO2ocean reconstructions are very sensitive to both the spatial and temporal scales of pCO2 sampling in the model domain. The four key findings from our investigation are the following: (1) improving ML-based pCO2 reconstructions in the Southern Ocean requires simultaneous high resolution observations of the meridional and the seasonal cycle (< 3 days) of pCO2ocean; (2) Saildrones stand out as the optimal platforms to simultaneously address these requirements; (3) Wavegliders with hourly/daily resolution in pseudo-mooring mode improve on Carbon-floats (10-day period), which suggests that sampling aliases from the low temporal frequency have a greater negative impact on their uncertainties, biases and reconstruction means; and (4) the present summer seasonal sampling biases in SOCAT data in the Southern Ocean may be behind a significant winter bias in the reconstructed seasonal cycle of pCO2ocean.
Highlights
Since the early 2000’s, the Southern Ocean (SO) carbon sink has undergone a reinvigoration characterized by a substantial strengthening as reported by Landschützer et al (2015), following a decade of weakening trends (Le Quéré et al, 35 2007; Canadell et al, 2021)
SHIP(smr+wtr) performed better (e.g., root mean square error (RMSE) = 6.8 μatm) than the SHIP(smr) + FLOAT(PFZ) and SHIP(smr) + WG(SAZ) experiments that produced RMSEs of 8.0 μatm and 6.88 μatm, respectively. These results suggest that while resolving the local seasonal cycle of the surface ocean pCO! with the WG and the FLOATs had a decisive impact on the 600 RMSEs and mean biases (MBEs), an additional scale is being resolved by the SHIP experiment in winter, which is not addressed by the sampling scales of the two autonomous sampling platforms WG (1-day period) and FLOAT (10-day period)
790 5 Conclusions In this study we propose that in order to advance the uncertainties and biases from machine learning reconstructions in the Southern Ocean “beyond the wall”, the seasonal cycle of the meridional gradient of pCO! needs to be resolved through a combination of high frequency observations spanning the meridional axis of the Southern Ocean
Summary
Since the early 2000’s, the Southern Ocean (SO) carbon sink has undergone a reinvigoration characterized by a substantial strengthening as reported by Landschützer et al (2015), following a decade (the 1990s) of weakening trends (Le Quéré et al, 35 2007; Canadell et al, 2021). Many empirical approaches such as statistical interpolations and regression methods (Rödenbeck et al, 2014; Iida et al, 2015; Jones et al, 2015) gained attention as alternative methods to ocean biogeochemical models (Lenton et al, 2013) until recently where Machine Learning (ML) approaches have been used increasingly as an alternative (Landschützer et al, 2013, 2014, 45 2016; Gregor et al, 2017, 2019; Denvil-Sommer et al, 2019) These novel mapping methods all seek to fill the spatial and temporal sampling gaps from existing ship-based surface ocean CO2 observations by extrapolating the CO2 partial pressure (pCO!) at the surface ocean (pCO!"#$%&) using prognostic proxy variables (such as satellite-observed and re-analysis-based sea surface temperature, sea surface salinity, mixed layer depth, chlorophyll-a, etc.). The feasibility of these extrapolations is justified through the non-linear relationships between surface ocean pCO! and the above-mentioned prognostic variables that 50 may drive changes in the surface ocean pCO! (Takahashi et al, 1993)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
