Journal cover Journal topic
Ocean Science An interactive open-access journal of the European Geosciences Union
Journal topic

Journal metrics

Journal metrics

  • IF value: 2.289 IF 2.289
  • IF 5-year value: 2.756 IF 5-year 2.756
  • CiteScore value: 2.76 CiteScore 2.76
  • SNIP value: 1.050 SNIP 1.050
  • SJR value: 1.554 SJR 1.554
  • IPP value: 2.65 IPP 2.65
  • h5-index value: 30 h5-index 30
  • Scimago H index value: 41 Scimago H index 41
Volume 13, issue 2 | Copyright
Ocean Sci., 13, 303-313, 2017
© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

Technical note 19 Apr 2017

Technical note | 19 Apr 2017

Technical note: Evaluation of three machine learning models for surface ocean CO2 mapping

Jiye Zeng1, Tsuneo Matsunaga1, Nobuko Saigusa1, Tomoko Shirai1, Shin-ichiro Nakaoka1, and Zheng-Hong Tan2 Jiye Zeng et al.
  • 1Centre for Global Environmental Research, National Institute for Environmental Studies, Tsukuba, Ibaraki, Japan
  • 2Institute of Tropical Agriculture and Forestry, Hainan University, Haikou, Hainan, China

Abstract. 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.

Publications Copernicus
Short summary
Three machine learning models were investigated for the reconstruction of global surface ocean CO2 concentration. They include self-organizing maps (SOMs), feedforward neural networks (FNNs), and support vector machines (SVMs). Our results show that the SVM performs the best, the FNN the second, and the SOM the worst. While the SOM does not have over-fitting problems, it is sensitive to data scaling and its discrete interpolation may not be good for some applications.
Three machine learning models were investigated for the reconstruction of global surface ocean...