反向传播神经网络的太平洋海域温跃层反演*

温跃层是反映海洋温度场的重要指标，针对太平洋中东部海域复杂多变的水文情况以及传统温跃层分析模式的局限性，本文基于BOA_Argo历史网格，通过BP神经网络，建立温度剖面的经验正交系数与海面遥感数据、少量深度处海水温度之间的非线性映射关系，实现海洋垂向温度剖面的实时反演，最后利用垂向梯度法获得海洋温跃层的相关参数。实验结果表明，相比于传统方法，该方法反演得到的跃层深度与测量值更加吻合，其中上层深度平均反演误差从10.3m下降到5.7m，下层深度平均反演误差从16.8m下降到8.8m。

Inversion of Thermocline in Pacific Ocean Based on back propagation neural network

The thermocline is an important parameter to the ocean temperature field. The traditional thermocline analysis method is limited due to the complex and changeable hydrological conditions in the central and eastern Pacific Ocean. In this paper, based on the historical gridded BOA_Argo data, the BP neural network is used to establish the nonlinear mapping relationship between the empirical orthogonal coefficient of the temperature profile and the remote sensing data of the sea surface and the sea water temperatures at a few depths, so as to realize the real-time inversion of the vertical temperature profile. Finally, the vertical gradient method is used to obtain the relevant parameters of the ocean thermocline. The experimental results show that, compared with the traditional method, the inversion of the thermocline depth obtained by this method is more consistent with the measured values. The average inversion error of the upper layer depth decreases from 10.3 m to 5.7 m, and the average inversion error of the lower layer depth decreases from 16.8m to 8.8m.