SMAP卫星的RBF神经网络海表盐度遥感反演

Remote Sensing Retrieval of Sea Surface Salinity Based on RBF Neural Network From SMAP Satellite

针对传统海表盐度遥感反演精度不高、影响因素较少等问题,本文基于SMAP(Soil Moisture Active Passive)卫星L2C(Level 2 C)数据、Argo(Array for Real-time Geostrophic Oceanography)数据和其他辅助数据,以太平洋部分海域(160°E~120°W,0°~30°N)为研究区域,综合考虑海面粗糙度以及白冠覆盖率等参量,利用径向基神经网络建立RBF亮温增量模型,并对平静海面亮温进行修正,然后基于Meissner-Wentz介电常数模型得到反演后的盐度值。验证结果表明:模型预测盐度和SMAP卫星盐度相对于Argo实测盐度的均方根误差分别为0.4和0.5,平均绝对误差分别为0.3和0.4。实验证明,利用RBF神经网络建立的亮温增量模型可以提高海表盐度反演的精度,对海表盐度反演具有实用意义。

In order to address the problems of traditional Sea Surface Salinity remote sensing inversion methods,which have poor inversion accuracy and few influencing factors,this paper proposes a new retrieval method using SMAP L2C data,Argo data,and auxiliary data,with a focus on(160°E-120°W,0°-30°N)in the Pacific Ocean.The proposed method considering more influence factors such as sea surface roughness and whitecap coverage,and firstly establishes an RBF bright temperature increment model using RBF neural network and modifies the brightness temperature of the flat sea surface,and then retrieves the SSS based on the Meissner-Wentz dielectric constant model.The validation results showed that the root mean square error of model predicted salinity and SMAP satellite salinity relative to Argo measured salinity were 0.4 and 0.5,respectively,and the mean absolute error were 0.3 and 0.4,respectively.Experiments show that the brightness temperature increment model established by RBF neural network can improve the accuracy of Sea Surface Salinity inversion,and has practical significance for Sea Surface Salinity inversion.