NEURAL NETWORK MODEL FOR CHLOROPHYLL-A CONCENTRATION RETRIEVAL IN THE BOHAI SEA

摘要

Based on the bio-optical dataset acquired in the BohairnSea in June, 2005, neural network (NN) model forrnremote sensing inversion of chlorophyll-a concentrationrnis developed. The three-layer backpropagation NNrnmodel employs hyperbolic tangent sigmoid transferrnfunction and 7 Neurons in hidden layer, taking seawaterrnremote sensing reflectance of 443nm, 490nm, 555nmrnand 670nm as model input. Model precision is evaluatedrnby in situ data, with the Pearson correlation coefficientrnR~2=0.8829, Mean Relative Error (MRE) =22.32% andrnRoot Mean Square (RMS) =0.5385, which is satisfactory.rnNN model is proved to have a better performance thanrnthat of statistical inversion model (R~2=0.7400, MRErn=27.92%, RMS =0.9986). NN Model sensitivity to noisernis tested to show its stability.rnConsidering of the complexity of bio-optical propertiesrnof the Bohai Sea, the neural network model developed isrnsuggested to be tested and validated using much more inrnsitu data.