%0 Journal Article
%9 ACL : Articles dans des revues avec comité de lecture répertoriées par l'AERES
%A Gueye, M.B.
%A Niang, A.
%A Arnault, Sabine
%A Thiria, S.
%A Crépon, M.
%T Neural approach to inverting complex system : application to ocean salinity profile estimation from surface parameters
%D 2014
%L fdi:010067232
%G ENG
%J Computers and Geosciences
%@ 0098-3004
%K ATLANTIQUE ; OCEAN INDIEN
%M ISI:000343631600018
%P 201-209
%R 10.1016/j.cageo.2014.07.012
%U https://www.documentation.ird.fr/hor/fdi:010067232
%> https://www.documentation.ird.fr/intranet/publi/depot/2016-07-12/010067232.pdf
%V 72
%W Horizon (IRD)
%X A neural network model is proposed for reconstructing ocean salinity profiles from sea surface parameters only. The method is applied to the tropical Atlantic. Prior data mining on a complete dataset shows that latitude and sea surface salinity are the most relevant surface parameters in the prediction of salinity profiles. A classification using a self-organizing map learned on a large multivariate dataset is able to retrieve the most probable vertical salinity profiles from the surface parameters only. Both in situ and modelled oceanic data are used to evaluate the results. The reconstruction misses some salinity features in areas with high time-space variability in which the limited available dataset was unable to provide the complete variability ranges during the learning process. However, apart from these restricted areas, the salinity profiles are reproduced with correlations greater than 0.95 for most of the profiles of the test set.
%$ 032 ; 020 ; 122