ALBERT

Description: A new feed-forward neural network (FFNN) model is presented to reconstruct surface ocean partial pressure of carbon dioxide (pCO2) over the global ocean. The model consists of two steps: (1) the reconstruction of pCO2 climatology, and (2) the reconstruction of pCO2 anomalies with respect to the climatology. For the first step, a gridded climatology was used as the target, along with sea surface salinity (SSS), sea surface temperature (SST), sea surface height (SSH), chlorophyll a (Chl a), mixed layer depth (MLD), as well as latitude and longitude as predictors. For the second step, data from the Surface Ocean CO2 Atlas (SOCAT) provided the target. The same set of predictors was used during step (2) augmented by their anomalies. During each step, the FFNN model reconstructs the nonlinear relationships between pCO2 and the ocean predictors. It provides monthly surface ocean pCO2 distributions on a 1∘×1∘ grid for the period from 2001 to 2016. Global ocean pCO2 was reconstructed with satisfying accuracy compared with independent observational data from SOCAT. However, errors were larger in regions with poor data coverage (e.g., the Indian Ocean, the Southern Ocean and the subpolar Pacific). The model captured the strong interannual variability of surface ocean pCO2 with reasonable skill over the equatorial Pacific associated with ENSO (the El Niño–Southern Oscillation). Our model was compared to three pCO2 mapping methods that participated in the Surface Ocean pCO2 Mapping intercomparison (SOCOM) initiative. We found a good agreement in seasonal and interannual variability between the models over the global ocean. However, important differences still exist at the regional scale, especially in the Southern Hemisphere and, in particular, in the southern Pacific and the Indian Ocean, as these regions suffer from poor data coverage. Large regional uncertainties in reconstructed surface ocean pCO2 and sea–air CO2 fluxes have a strong influence on global estimates of CO2 fluxes and trends.

Description: A new Feed-Forward Neural Network (FFNN) model is presented to reconstruct surface ocean partial pressure of carbon dioxide (pCO2) over the global ocean. The model consists of two steps: (1) reconstruction of pCO2 climatology and (2) reconstruction of pCO2 anomalies with respect to the climatology. For the first step, a gridded climatology was used as the target, along with sea surface salinity and temperature (SSS and SST), sea surface height (SSH), chlorophyll a (Chl), mixed layer depth (MLD), as well as latitude and longitude as predictors. For the second step, data from the Surface Ocean CO2 Atlas (SOCAT) provided the target. The same set of predictors was used during step 2 augmented by their anomalies. During each step, the FFNN model reconstructs the non-linear relations between pCO2 and the ocean predictors. It provides monthly surface ocean pCO2 distributions on a 1º x 1º grid for the period 2001–2016. Global ocean pCO2 was reconstructed with a satisfying accuracy compared to independent observational data from SOCAT. However, errors are larger in regions with poor data coverage (e.g. Indian Ocean, Southern Ocean, subpolar Pacific). The model captured the strong interannual variability of surface ocean pCO2 with reasonable skills over the Equatorial Pacific associated with ENSO (El Niño Southern Oscillation). Our model was compared to three pCO2 mapping methods that participated in the Surface Ocean pCO2 Mapping intercomparison (SOCOM) initiative. We found a good agreement in seasonal and interannual variabilty between the models over the global ocean. However, important differences still exist at the regional scale, especially in the Southern hemisphere and in particular, the Southern Pacific and the Indian Ocean, as these regions suffer from poor data-coverage. Large regional uncertainties in reconstructed surface ocean pCO2 and sea-air CO2 fluxes have a strong influence on global estimates of CO2 fluxes and trends.

Description: Understanding the relationship between surface marine ecosystems and the export of carbon to depth by sinking organic particles is key to representing the effect of ecosystem dynamics and diversity, and their evolution under multiple stressors, on the carbon cycle and climate in models. Recent observational technologies have greatly increased the amount of data available, both for the abundance of diverse plankton groups and for the concentration and properties of particulate organic carbon in the ocean interior. Here we use synthetic model data to test the potential of using machine learning (ML) to reproduce concentrations of particulate organic carbon within the ocean interior based on surface ecosystem and environmental data. We test two machine learning methods that differ in their approaches to data-fitting, the random forest and XGBoost methods. The synthetic data are sampled from the PlankTOM12 global biogeochemical model using the time and coordinates of existing observations. We test 27 different combinations of possible drivers to reconstruct small (POCS) and large (POCL) particulate organic carbon concentrations. We show that ML can successfully be used to reproduce modelled particulate organic carbon over most of the ocean based on ecosystem and modelled environmental drivers. XGBoost showed better results compared to random forest thanks to its gradient boosting trees' architecture. The inclusion of plankton functional types (PFTs) in driver sets improved the accuracy of the model reconstruction by 58 % on average for POCS and by 22 % for POCL. Results were less robust over the equatorial Pacific and some parts of the high latitudes. For POCS reconstruction, the most important drivers were the depth level, temperature, microzooplankton and PO4, while for POCL it was the depth level, temperature, mixed-layer depth, microzooplankton, phaeocystis, PO4 and chlorophyll a averaged over the mixed-layer depth. These results suggest that it will be possible to identify linkages between surface environmental and ecosystem structure and particulate organic carbon distribution within the ocean interior using real observations and to use this knowledge to improve both our understanding of ecosystem dynamics and of their functional representation within models.