A Wavelet-Neural Network for the Estimation of Chlorophyll-a Concentration in Caspian Sea

Monitoring of vast water bodies, including internal and external waters, is an important issue which is commonly performed by remote sensing as the most economical technology. In this field, the concentration of chlorophyll-a, as a critical water quality index, has attracted most research attentions. In this paper, wavelet neural network are proposed for the estimation of chlorophyll-a concentration in Caspian Sea from multi-date MODIS product MYDOCGA.These networks are evaluated from both aspects of estimation accuracy as well as response stability and are also compared to the classical perceptron neural networks (PNN). In addition, different features are examined as the network input parameters including all the 9 MODIS product MYDOCGA bands, different subsets of these bands and also PCA(Principal Component Analysis) bands in different number. The results, which are obtained and validated based to 55 filed observed samples, proves the effectiveness of WNN (Wavelet Neural Network) in comparison to classical neural networks. The best RMSE=0.07 of these networks reveals that remote sensing can accurately replace field observations to produce thematic maps of water quality parameters provided that appropriate processing techniques are applied.