Journal of Tropical Oceanography >
Using artificial intelligence for identifying the depth of upper-ocean mixed layer
Copy editor: YIN Bo
Received date: 2018-12-14
Request revised date: 2019-04-10
Online published: 2019-10-09
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)
Copyright
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.
ZHANG Kang , GUO Shuangxi , HUANG Pengqi , QU Ling , LU Yuanzheng , CEN Xianrong , YU Lusha , ZHOU Weidong , ZHOU Shengqi . Using artificial intelligence for identifying the depth of upper-ocean mixed layer[J]. Journal of Tropical Oceanography, 2019 , 38(5) : 32 -41 . DOI: 10.11978/2018137
图1 数据学习过程示意图图中粉色方框表示窗口 Fig. 1 An example of the learning process. The blue line is the temperature profile. The magenta box represents the current window. The green, gray, and red dotted lines represent the fitting curves in the priori, the updating, and the posterior process, respectively |
图5 突变检验示意图a中实线为温度-深度廓线, 三角标注为突变点(混合层下边界); b中实线表示各点发生突变的概率, 虚线表示突变发生的临界概率, 突变发生概率首次超过临界值的点为突变点 Fig. 5 An example for checking the jump. The solid line in (a) denotes the temperature-depth profile, and the triangle is the depth of the jump (the bottom of the mixed layer). The solid line in (b) indicates the probability of mutation at each point, and the dotted line represents the critical probability of the jump occurring. The jump will occur when the probability exceeds the critical value for the first time |
图7 站位23 (28°57'43"N, 166°38'35"E)的温度(a)和密度(b)均匀混合层廓线和对应混合层深度图中三角形表示混合层下边界位置 Fig. 7 Temperature (density) uniform mixed layer and recognition results. The solid line represents the temperature profile (a) and potential density profile (b). The triangle is the lower boundary of the mixed layer |
图8 站位177 (57°22'55"S, 150°8'7"E)的温度(a)和密度(b)存在弱层结的混合层廓线和对应混合层深度图中三角形表示混合层下边界位置 Fig. 8 The temperature (density) profile with gradual pycnocline and recognition results. The solid line represents the temperature profile (a) and potential density profile (b). The triangle represents the lower boundary of the mixed layer |
表1 各方法从温度廓线廓线计算的QI值的结果Tab. 1 QI values calculated from the temperature profile by each method |
方法 | QImean | QI0.25 | QI0.5 | QI0.75 | QIstd |
---|---|---|---|---|---|
阈值法 | 0.795 | 0.708 | 0.899 | 0.962 | 0.239 |
梯度法 | 0.683 | 0.488 | 0.800 | 0.948 | 0.316 |
混合法 | 0.713 | 0.549 | 0.865 | 0.963 | 0.329 |
相对变化法 | 0.900 | 0.870 | 0.945 | 0.982 | 0.129 |
最大角度法 | 0.710 | 0.556 | 0.820 | 0.956 | 0.307 |
最优线性拟合 | 0.782 | 0.725 | 0.927 | 0.985 | 0.314 |
AI法 | 0.936 | 0.922 | 0.969 | 0.990 | 0.104 |
注: QImean为QI值的均值; QI0.25为QI值的1/4分位数; QI0.5为QI值的中值; QI0.75为QI值的3/4分位数; QIstd为QI值的标准差 |
表2 各方法从位势密度廓线中计算的QI值的结果Tab. 2 QI values calculated from the potential density profile by each method |
方法 | QImean | QI0.25 | QI0.5 | QI0.75 | QIstd |
---|---|---|---|---|---|
阈值法 | 0.783 | 0.672 | 0.915 | 0.974 | 0.267 |
梯度法 | 0.727 | 0.596 | 0.808 | 0.926 | 0.251 |
混合法 | 0.734 | 0.594 | 0.872 | 0.962 | 0.306 |
相对变化法 | 0.893 | 0.862 | 0.945 | 0.983 | 0.144 |
最大角度法 | 0.740 | 0.621 | 0.850 | 0.965 | 0.299 |
最优线性拟合 | 0.797 | 0.752 | 0.930 | 0.985 | 0.297 |
AI法 | 0.921 | 0.902 | 0.959 | 0.988 | 0.117 |
注: QImean为QI值的均值; QI0.25为QI值的1/4分位数; QI0.5为QI值的中值; QI0.75为QI值的3/4分位数; QIstd为QI值的标准差 |
[1] |
范聪慧 , 2007. 多因素对海洋上混合层深度影响的数值模拟[D]. 北京: 中国科学院研究生院( 海洋研究所):1-65.
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
[28] |
|
[29] |
|
[30] |
|
[31] |
|
[32] |
|
[33] |
|
[34] |
|
[35] |
|
[36] |
|
[37] |
|
[38] |
|
/
〈 | 〉 |