Study Of The Vertical Distribution Of Chlorophyll Concentration In The Ocean Based On Artificial Neural Network

A method of retrieval vertical profile of chlorophyll concentration from sea surface parameters based on artificial neural network technique is proposed in this thesis. To study the biomass, primary production in the water column and carbon fixation, the estimation of the vertical distribution of chlorophyll concentration is very important.Remote satellite sensors provide the chlorophyll concentration over one light penetration depth. The vertical profiles of chlorophyll can not be detected. To estimate the vertical profiles of chlorophyll from sea surface parameters, one three-layer artificial neural network model is developed. A shifted Gaussian function proposed by Marlon R. Lewis (1983) is used to represent the chlorophyll vertical profile with four parameters Bo, h,Zm . ANN model is trained by back-propagation algorithm. Six input variables of ANN model include day, latitude, longitude, sea surface temperature, mixed layer depth and sea surface chlorophyll concentration. The outputs are four parameters shown in Gaussian model, In order to accelerate the back-propagation learning process, the inputs is preprocessed by using principal components analysis. After comparing with the regression model, the ANN model shows good ability in generalization. Finally, the possibility of the ANN model application in the field of satellite ocean remote sensing is discussed through the sensitivity analysis and error transfer analysis of the ANN model.The vertical distribution of chlorophyll concentration can be well reconstructed by the ANN model proposed by this thesis, and the more analysis and verification of the ANN model results are required specially the satellite data as the model inputs due to the limitation of sea surface parameters retrieved from satellite data.