%0 Journal Article
%9 ACL : Articles dans des revues avec comité de lecture répertoriées par l'AERES
%A Moussa, H.
%A Benallal, M. A.
%A Goyet, C.
%A Lefèvre, Nathalie
%T Satellite-derived CO2 fugacity in surface seawater of the tropical Atlantic Ocean using a feedforward neural network
%D 2016
%L fdi:010066132
%G ENG
%J International Journal of Remote Sensing
%@ 0143-1161
%K ATLANTIQUE ; ZONE TROPICALE
%M ISI:000368723800006
%N 3
%P 580-598
%R 10.1080/01431161.2015.1131872
%U https://www.documentation.ird.fr/hor/fdi:010066132
%> https://www.documentation.ird.fr/intranet/publi/2016/02/010066132.pdf
%V 37
%W Horizon (IRD)
%X A feedforward neural network is used to quantify the fugacity of CO2 in surface seawater (f (CO2sw)) of the tropical Atlantic Ocean, exclusively from satellite data: sea-surface temperature, sea-surface salinity, and chlorophyll-a (chl-a), at a 4 km x 4 km spatial resolution, for the period of spring (March and April). The model was constructed using 7188 in situ data provided by the 'Surface Ocean CO2 ATlas' (SOCAT) products, and the 'EC-funded project CARBOOCEAN IP program' products, available for the years 2001, 2002, 2004, 2006, 2007, and 2009. The model was tested using remote sensing data of the Moderate Resolution Imaging Spectroradiometer Aqua. This approach was validated over the area extending from 8 degrees N-61 degrees W to 23 degrees N-20 degrees W. A comparison with multiple linear regression model was established. The neural network has provided better results (root mean square error (RMSE) of 8.7 atm (0.881 Pa)) than linear regression (RMSE of 9.6 atm (0.973 Pa)) for f (CO2sw) interpolation using remote sensing data. Since the required input data are available, this approach could be applied to the whole tropical Atlantic Ocean and for the remaining seasons (summer, fall, and winter).
%$ 032 ; 126