The terrestrial biosphere is a key player in slowing the accumulation of carbon dioxide in the atmosphere. While quantification of carbon fluxes at global land scale is important for mitigation policy related to climate and carbon, measurements are only available at sites scarcely distributed in the world. This leads to using various methods to upscale site measurements to the whole terrestrial biosphere. This article reports a product obtained by using a Random Forest to upscale terrestrial net ecosystem exchange, gross primary production, and ecosystem respiration from FLUXNET 2015. Our product covers land from −60°S to 80°N with a spatial resolution of 0.1° × 0.1° every 10 days during the period 1999–2019. It was compared with four existing products. A distinguishable feature of our method is using three derived variables of leaf area index to represent plant functional type (PFT) so that measurements from different PFTs can be mixed better by the model. This product can be valuable for the carbon-cycle community to validate terrestrial biosphere models and cross check datasets.

The global ocean is a major sink of anthropogenic carbon dioxide (CO₂) emitted into the atmosphere. Machine learning has been actively used in the past decades to estimate the oceanic sink, but it is still a challenge to obtain an accurate estimate due to scarcely available CO₂ measurements. One of the methods to deal with data scarcity was normalizing multiple years’ CO₂ values to a reference year to increase the spatial coverage. The practice assumed a constant CO₂ trend for the normalization. Here, we used three machine learning models to extract variable ocean CO₂ trends on a decadal scale and proposed a method to use the extracted ocean CO₂ trends to correct the decadal atmospheric CO₂ trends for data normalization. The method minimizes assumptions of using the extracted ocean CO₂ trends directly. Comparisons of our CO₂ flux estimate with machine learning products included in Global Carbon Budget 2021 indicates that using the variable trends improved the bias resulted from using a constant trend and that the trends are a critical factor for machine learning methods. Our dataset includes monthly distributions of surface ocean CO₂ concentration and air-sea flux in 1980-2020 with a spatial resolution of 1×1 degree.

A Support Vector Machine (SVM) for regression is a popular machine learning model that aims to solve nonlinear function approximation problems wherein explicit model equations are difficult to formulate. The performance of an SVM depends largely on the selection of its parameters. Choosing between an SVM that solves an optimization problem with inequality constrains and one that solves the least square of errors (LS-SVM) adds to the complexity. Various methods have been proposed for tuning parameters, but no article puts the SVM and LS-SVM side by side to discuss the issue using a large dataset from the real world, which could be problematic for existing parameter tuning methods. We investigated both the SVM and LS-SVM with an artificial dataset and a dataset of more than 200,000 points used for the reconstruction of the global surface ocean CO₂ concentration. The results reveal that: (1) the two models are most sensitive to the parameter of the kernel function, which lies in a narrow range for scaled input data; (2) the optimal values of other parameters do not change much for different datasets; and (3) the LS-SVM performs better than the SVM in general. The LS-SVM is recommended, as it has less parameters to be tuned and yields a smaller bias. Nevertheless, the SVM has advantages of consuming less computer resources and taking less time to train. The results suggest initial parameter guesses for using the models.

Reconstructing surface ocean CO₂ from scarce measurements plays an important role in estimating oceanic CO₂ uptake. There are varying degrees of differences among the 14 models included in the Surface Ocean CO₂ Mapping (SOCOM) inter-comparison initiative, in which five models used neural networks. This investigation evaluates two neural networks used in SOCOM, self-organizing maps and feedforward neural networks, and introduces a machine learning model called a support vector machine for ocean CO₂ mapping. The technique note provides a practical guide to selecting the models.

This dataset includes the monthly distributions of CO₂ fugacity in the world surface oceans reconstructed using a feed‐forward neural network model and the CO₂ measurements of the Surface Ocean CO₂ Atlas version 2.0. It has a spatial resolution of 1 × 1° and spans a period of 22 years, from January 1990 to December 2011. The dataset also includes necessary parameters for the reconstruction and an estimate of the CO₂ fluxes between the ocean and the atmosphere. The aim of this work is to provide a dataset for estimating the oceans' contribution to the global carbon budget.