珠江口海域叶绿素a质量浓度SAR反演模型

Inversion models of chlorophyll a mass concentration in Pearl River Estuary using SAR image

以珠江口海域的Radarsatt-2全极化SAR数据和海域表层水面叶绿素a质量浓度实测数据为基础,利用微波散射原理及Cloude Pottier理论对SAR图像进行分解,得到平均散射角a、散射熵H及VH、VV、HH、HV等6个参数;采用BP人工神经网络模型建立上述6个参数与叶绿素a质量浓度的数学关系模型,并结合实测数据对叶绿素a质量浓度进行分类。结果表明:当隐含层节点数为9,输入层和隐含层传递函数分别为tansig和logsig,学习速率和动量系数均为0.2时的网络结构对叶绿素质量浓度反演取得了较好的效果,叶绿素a质量浓度实测值与预测值之间的决定系数(R^2)为0.826。将模型应用于不同时期的2幅图像进行验证,效果良好,与实际情况基本相符。

Base on the Radarsat-2 full polarimetric SAR data and the observed chlorophyll a data from the sea surface of the Pearl River Estuary, and using microwave scattering theory and the Cloude-Pottier theory, the SAR images were decomposed, which resulted in the average scattering angle a and scattering entropies H and VH, VV, HH and HV. A mathematical relationship model between the six parameters and the chlorophyll a mass concentration was established using the BP artificial neural network model, and combined with the measured data the chlorophyll a mass concentration was classified. The results shows that the network structure has a good result of the inversion to the chlorophyll a mass concentration when the hidden layer nodes is 9, the transmit function of the input layer and the hidden layer is tansig and logsig respectively and the learning rate and the momentum coefficient are both 0.2, namely, the determination coefficient between the measured data and the predicted data of the chlorophyll a mass concentration （Rz） is 0. 826. When the model was applied to two images of different period for verification, it worked well, and the results were in good accordance with the actual situation.