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

Based on the bio-optical dataset acquired in the Bohai Sea in June, 2005, neural network (NN) model for remote sensing inversion of chlorophyll-a concentration is developed. The three-layer backpropagation NN model employs hyperbolic tangent sigmoid transfer function and 7 Neurons in hidden layer, taking seawater remote sensing reflectance of 443nm, 490nm, 555nm and 670nm as model input. Model precision is evaluated by in situ data, with the Pearson correlation coefficient R2=0.8829, Mean Relative Error (MRE) =22.32 % and Root Mean Square (RMS) =0.5385, which is satisfactory. NN model is proved to have a better performance than that of statistical inversion model (R2=0.7400, MRE =27.92%, RMS =0.9986). NN Model sensitivity to noise is tested to show its stability. Considering of the complexity of bio-optical properties of the Bohai Sea, the neural network model developed is suggested to be tested and validated using much more in situ data