热带海洋学报 ›› 2019, Vol. 38 ›› Issue (5): 32-41.doi: 10.11978/2018137

• 海洋水文学 • 上一篇    下一篇

一种海洋混合层深度的智能识别方法研究

张康1,2,郭双喜1,3,黄鹏起1,2,屈玲1,3,鲁远征1,3,岑显荣1,3,于璐莎1,2,周伟东1,3,周生启1,3()   

  1. 1. 热带海洋环境国家重点实验室(中国科学院南海海洋研究所), 广东 广州 510301
    2. 中国科学院大学, 北京 100049
    3. 中国科学院南海生态环境工程创新研究院, 广东 广州 510301
  • 收稿日期:2018-12-14 修回日期:2019-04-10 出版日期:2019-09-20 发布日期:2019-10-09
  • 通讯作者: 周生启
  • 作者简介:张康(1991—), 男, 安徽省亳州市人, 硕士研究生, 从事物理海洋学研究。E-mail: zhangkang16@mails.ucas.ac.cn
  • 基金资助:
    国家自然科学基金项目(91752108);国家自然科学基金项目(41476167);国家自然科学基金项目(41706029);国家自然科学基金项目(41606010);广东省自然科学基金(2016A030311042);广东省自然科学基金(2016A030310114);广州市科技计划重点项目(201804020056);中科院战略性先导专项资助项目(XDA11030302);中国科学院南海生态环境工程创新研究院课题(ISEE2018PY05)

Using artificial intelligence for identifying the depth of upper-ocean mixed layer

Kang ZHANG1,2,Shuangxi GUO1,3,Pengqi HUANG1,2,Ling QU1,3,Yuanzheng LU1,3,Xianrong CEN1,3,Lusha YU1,2,Weidong ZHOU1,3,Shengqi ZHOU1,3()   

  1. 1. State Key Laboratory of Tropical Oceanography (South China Sea Institute of Oceanology, Chinese Academy of Sciences), Guangzhou 510301, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. Institution of South China Sea Ecology and Environmental Engineering, Chinese Academy of Sciences, Guangzhou 510301, China
  • Received:2018-12-14 Revised:2019-04-10 Online:2019-09-20 Published:2019-10-09
  • Contact: Shengqi ZHOU
  • Supported by:
    National Natural Science Foundation of China(91752108);National Natural Science Foundation of China(41476167);National Natural Science Foundation of China(41706029);National Natural Science Foundation of China(41606010);Natural Science Foundation of Guangdong Province(2016A030311042);Natural Science Foundation of Guangdong Province(2016A030310114);Guangzhou Science and Technology Program Key Project(201804020056);Strategic Priority Research Program of Chinese Academy of Sciences(XDA11030302);Institution of South China Sea Ecology and Environmental Engineering, Chinese Academy of Sciences(ISEE2018PY05)

摘要:

文章提出了一种识别混合层深度的人工智能方法。该方法在温度(密度)与压强(或深度)间建立线性模型, 并且将其系数和方差做成一组表征廓线特征的统计量。初始时为模型设定一个主观的先验分布, 在一个自海表向下移动的窗口内通过贝叶斯链式法则和最小描述长度原理学习新数据, 得到系数均值的最大后验概率估计。用F-检验识别系数发生突变的位置, 以此确定混合层的存在性及其深度。通过2017年2月太平洋海域的地转海洋学实时观测阵(Array for Real-time Geostrophic Oceanography, ARGO)数据进行测试, 并且以质量因子(Quality Index, QI)值作为判断识别混合层深度结果准确性的依据, 发现该方法相比于梯度法、阈值法、混合法、相对变化法、最大角度法和最优线性插值法在识别结果上具备更大的QI值。表明该方法能够准确识别混合层深度。

关键词: 海洋混合层, 人工智能方法, 贝叶斯链式法则, 最小描述长度原理

Abstract:

An artificial intelligence (AI) method for identifying upper-ocean mixed layer depth (MLD) is proposed. In this method, a linear model, whose coefficient and variance are made into a set of statistics to characterize the profile, is established between temperature (density) and pressure (or depth). A subjective priori distribution is set in an initial window. The maximum posterior probability estimate of the mean coefficient value is obtained when the window is moved down from the sea surface, by learning the new data through the Bayesian chain rule and the minimum description length principle. The existence and depth of the mixed layer are determined when the jump of the coefficient is found by using the F-distribution. Using the Argo buoy data measured in the Pacific Ocean in February 2017, and taking the value of the quality index (QI) to estimate the accuracy of the MLD results, we find that this AI method is superior to the gradient method, the threshold method, the Hybrid method, and the relative-variant method.

Key words: upper-ocean mixed layer, artificial intelligence method, Bayesian chain rule, minimum description length principle

中图分类号: 

  • P731.26