Reconstructing Global Chlorophyll-a Variations Using a Non-linear Statistical Approach

 Monitoring the spatio-temporal variations of surface chlorophyll-a concentration (Chl, a proxy of phytoplankton biomass) greatly benefited from the availability of continuous and global ocean color satellite measurements from 1997 onward. These two decades of satellite observations are however still too short to provide a comprehensive description of Chl variations at decadal to multi-decadal timescales. This paper investigates the ability of a machine learning approach (a non-linear statistical approach based on Support Vector Regression, hereafter SVR) to reconstruct global spatio-temporal Chl variations from selected surface oceanic and atmospheric physical parameters. With a limited training period (13 years), we first demonstrate that Chl variability from a 32-years global physical-biogeochemical simulation can generally be skillfully reproduced with a SVR using the model surface variables as input parameters. We then apply the SVR to reconstruct satellite Chl observations using the physical predictors from the above numerical model and show that the Chl reconstructed by this SVR more accurately reproduces some aspects of observed Chl variability and trends compared to the model simulation. This SVR is able to reproduce the main modes of interannual Chl variations depicted by satellite observations in most regions, including El Niño signature in the tropical Pacific and Indian Oceans. In stark contrast with the trends simulated by the biogeochemical model, it also accurately captures spatial patterns of Chl trends estimated by satellite data, with a Chl increase in most extratropical regions and a Chl decrease in the center of the subtropical gyres, although the amplitude of these trends are underestimated by half. Results from our SVR reconstruction over the entire period (1979–2010) also suggest that the Interdecadal Pacific Oscillation drives a significant part of decadal Chl variations in both the tropical Pacific and Indian Oceans. Overall, this study demonstrates that non-linear statistical reconstructions can be complementary tools to in situ  and satellite observations as well as conventional physical-biogeochemical numerical simulations to reconstruct and investigate Chl decadal variability.

Disciplines

Biological oceanography, Physical oceanography, Cross-discipline

Keywords

machine learning, phytoplankton variability, satellite ocean color, decadal variability, global scale, Support Vector Regression

Data

FileSizeFormatProcessingAccess
Satellite surface Chl from the Climate Change Ocean Color Initiative
77 MoNetCDFProcessed data
Surface Chl from the PISCES biogeochemical model
189 MoNetCDFProcessed data
Reconstructed surface Chl from a SVR trained on satellite OC-CCI Chl
189 MoNetCDFProcessed data
Reconstructed surface Chl from a SVR trained on PISCES model surface Chl
189 MoNetCDFProcessed data
How to cite
Martinez Elodie, Gorgues Thomas, Lengaigne Matthieu, Fontana Clément, Sauzède Raphaelle, Menkes Christophe, Uitz Julia, Di Lorenzo Emanuele, Fablet Ronan (2020). Reconstructing Global Chlorophyll-a Variations Using a Non-linear Statistical Approach. SEANOE. https://doi.org/10.17882/75639
In addition to properly cite this dataset, it would be appreciated that the following work(s) be cited too, when using this dataset in a publication :
Martinez Elodie, Gorgues Thomas, Lengaigne Matthieu, Fontana Clement, Sauzède Raphaëlle, Menkes Christophe, Uitz Julia, Di Lorenzo Emanuele, Fablet Ronan (2020). Reconstructing Global Chlorophyll-a Variations Using a Non-linear Statistical Approach. Frontiers In Marine Science. 7 (464). (20p.). https://doi.org/10.3389/fmars.2020.00464, https://archimer.ifremer.fr/doc/00641/75314/

Copy this text