基于PSO-BPNN算法的CFOSAT SWIM有效波高修正方法  

Calibration of CFOSAT Off-Nadir SWIM SWH Product Based on PSO-BPNN Model

作　　者：张锐[1] 张杰[1,2] 万勇 ZHANG Rui;ZHANG Jie;WAN Yong(College of Ocean and Space Informatics,China University of Petroleum(East China),Qingdao 266580,China;First Institute of Oceanography,Ministry of Natural Resources,Qingdao 266061,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580 [2]自然资源部第一海洋研究所,山东青岛266061

出　　处：《海洋技术学报》2024年第6期1-11,共11页Journal of Ocean Technology

基　　金：国家自然科学基金资助项目(61931025)。

摘　　要：高精度的有效波高(Significant Wave Height,SWH)观测对于海洋领域的研究至关重要。中法海洋卫星(China-France Oceanography Satellite,CFOSAT)搭载的波谱仪(Surface Waves Investigation and Monitoring,SWIM)可提供全球范围内的波浪数据,但非星下点有效波高数据经验证后与真实值相比存在偏差,因此有必要对其进行修正。本文通过利用CFOSAT搭载的散射计(wind Sactterometer,SCAT)传感器提供的风场数据,建立基于粒子群优化算法(Particle Swarm Optimization,PSO)的反向传播神经网络(Back Propagation Neural Network,BPNN),发展了对SWIM 6°波束有效波高数据修正的模型,并使用8°和10°波束有效波高验证了模型的适用性。结果表明:修正后的SWIM中6°波束有效波高均方根误差为0.2329 m,相关系数R为0.9858,模型对8°和10°有效波高修正也达到较高精度。该模型在中等海况下修正效果最好,但在低海况和高海况下分别存在高估和低估现象,通过增加低海况和高海况的训练数据可减小该部分误差。最后对比不同机器学习模型,证明了PSO优化算法可以提高模型精度。本文提出的PSO-BPNN网络模型有效地提高了有效波高参数的精度,可为参数修正研究提供一种实验思路。High-precision significant wave height(SWH)observation is crucial for Marine research.The Surface Waves Investigation and Monitoring(SWIM)aboard the China-France Oceanic Satellite(CFOSAT)provides the global wave data.However,the SWH data obtained from SWIM off-nadir measurements exhibit a relatively high bias when compared with true value,thereby demonstrating the need for calibration.In this paper,the Back Propagation Neural Network(BP)based on Particle Swarm Optimization(PSO)is established,and the wind field data provided by the wind Sactterometer(SCAT)sensor carried by the CFOSAT is used to correct the SWH by the SWIM 6°offnadir.The applicability of the model to correct the SWH of 8°and 10°was verified.The proposed model has been compared with various machine learning models,and validated across different sea conditions.The results indicate that the root mean square error of the corrected SWH is 0.2329 m,with a correlation coefficient(R)of 0.9858,and that the SWH at 8°and 10°can also be corrected with high precision.The model has the highest accuracy under medium sea state,but there are overestimation and underestimation in low and high state.This bias can be reduced by increasing the training data of low and high sea state.Finally,comparing different machine leaning models,it is proved that the PSO optimization algorithm can improve the accuracy of the model.The PSO-BPNN model proposed improves the accuracy of the SWH parameters,and provides an experimental idea for the study of parameter correction.

关 键 词：波谱仪 散射计 时空匹配 有效波高修正 粒子群优化算法 反向传播神经网络 

分 类 号：P717[天文地球—海洋科学]