基于广义回归神经网络反演悬浮泥沙含量的定量遥感方法

Using General Regression Neural Network to Retrieve Suspend Matter Concentration

利用2003年1月和2004年1月获得的珠江口实测遥感反射率和悬浮泥沙浓度数据,根据MODIS专门的水色通道,建立基于实测遥感反射率的广义回归神经网络模型来反演悬浮泥沙含量,并利用MODIS图像获取悬浮泥沙的浓度图。研究结果表明:广义回归神经网络模型计算方法简单,选取MODIS8～15通道的遥感反射率组合作为输入时,模型的反演精度较理想,平均相对误差为17.01%,相关系数为0.965。

The measured data was obtained from the ocean color experiments in Pearl River Esturay in January, 2003 and January, 2004. General regression neural network model （GRNN） is established to retrieve the suspended matter concentration from remote sensing reflectance. At last, the concentration of suspend matter is derived based on moderate-resolution imaging spectroradiometer data. The result demonstrates that GRNN can get better prediction with a simple algorithm, and the model of using bands of 8-15 is better than other models, the averaged relative error is 17.01% and the correction coefficient is 0.965.