Estimation of Oceanic Particulate Organic Carbon With Machine Learning

Abstract

Abstract. Understanding and quantifying ocean carbon sinks of the planet is of paramount relevance in the current scenario of global change. Particulate organic carbon (POC) is a key biogeochemical parameter that helps us characterize export processes of the ocean. Ocean color observations enable the estimation of bio-optical proxies of POC (i.e. particulate backscattering coefficient, bbp) in the surface layer of the ocean quasi-synoptically. In parallel, the Argo program distributes vertical profiles of the physical properties with a global coverage and a high spatio-temporal resolution. Merging satellite ocean color and Argo data using a neural networkbased method has already shown strong potential to infer the vertical distribution of bio-optical properties at global scale with high space-time resolution. This method is trained and validated using a database of concurrent vertical profiles of temperature, salinity, and bio-optical properties, i.e. bbp, collected by Biogeochemical-Argo (BGC-Argo) floats, matched up with satellite ocean color products. The present study aims at improving this method by 1) using a larger dataset from BGC-Argo network since 2016 for training, 2) using additional inputs such as altimetry data, which provide significant information on mesoscale processes impacting the vertical distribution of bbp, 3) improving the vertical resolution of estimation, and 4) examining the potential of alternative machine learning-based techniques. As a first attempt with the new data, we used some feature-specific preprocessing routines followed by a Multi-Output Random Forest algorithm on two regions with different ocean dynamics: North Atlantic and Subtropical Gyres. The statistics and the bbp profiles obtained from the validation floats show promising results and suggest this direction is worth investigating even further at global scale.

Bridging the gaps between particulate backscattering measurements and modeled particulate organic carbon in the ocean

“…In situ bio-optical measurements are poised to play a key role in monitoring marine POC stocks in layers that cannot be accessed by remote sensing. For example, Sauzède et al (2020) merged BGC-Argo and satellite observations to obtain a dynamic 3D view of particle backscattering. Using a data-driven machine learning approach, they were able to predict the profiles of log 10 b bp700 measured by two BGC-Argo floats in the NASPG and STG biomes (R 2 of ∼ 0.85 and mean absolute percentage deviation of ∼ 12 %) from the sole knowledge of physical properties of the water column and surface ocean color (remote-sensing reflectance).…”

“…Using a data-driven machine learning approach, they were able to predict the profiles of log 10 b bp700 measured by two BGC-Argo floats in the NASPG and STG biomes (R 2 of ∼ 0.85 and mean absolute percentage deviation of ∼ 12 %) from the sole knowledge of physical properties of the water column and surface ocean color (remote-sensing reflectance). Their estimates were recently extended to POC (Sauzède et al, 2021), which can be of great utility for constraining biogeochemical models. Here we took an entirely different approach, based on converting available b bp700 data to POC with a simple empirical algorithm (Fig.…”

The Oceans’ Biological Carbon Pumps: Framework for a Research Observational Community Approach

“…Measurements of TZ variables and processes have been conducted from oceanographic ships for more than a century, and using instrumented moorings since several decades. These measurements cannot be made from satellite remote sensing, although some TZ variables, such as POC (or its proxy backscattering, b bp ), can be estimated from remotely sensed information combined with parameters derived from in situ observations over the water column (EZ + TZ) (Sauzède et al, 2016;Sauzède et al, 2020). These approaches are reviewed in CEOS (2014) and Siegel et al (2016), and one is mentioned in the next section in relation with the biological gravitational pump.…”

“…The wealth of simultaneous TZ and EZ measurements acquired as part of a global coordinated sampling effort could be used to develop models linking TZ processes with those occurring in the EZ. Over the last years, empirical models based on artificial intelligence (i.e., neural networks) have been developed to estimate various biogeochemical variables from physical-chemical water-column drivers (e.g., Bittig et al, 2018) or from surface satellite-based properties (Sauzède et al, 2020). Such artificial intelligence-based approaches could be used to estimate, from globally measurable EZ predictors, metrics of the EZ-TZ carbon flux that are difficult to acquire at the global scale.…”

Bridging the gaps between particulate backscattering measurements and modeled particulate organic carbon in the ocean

“…In situ bio-optical measurements are poised to play a key role in monitoring marine POC stocks in layers that cannot be accessed by remote sensing. For example, Sauzède et al (2020) illustrated the power of merging BGC-Argo and satellite observations to obtain a dynamic 3D view of particle backscattering. Using a data-driven machine learning approach, they were able to predict the profiles of log 10 b bp700 measured by two BGC-Argo floats in the NASPG and STG biomes (R 2 of around 0.85 and MAPD of around 12%) from the sole knowledge of physical properties of the water column and surface ocean color (remote sensing reflectance).…”