Estimation Of Ocean Mixed Layer Depth In Indian Ocean Based On Artificial Intelligence Algorithm

The total area of the ocean accounts for 70.8%of the earth’s surface area,and ocean resources are an important material basis for human survival and sustainable development.Only by fully knowing and understanding the ocean can human beings reasonably control,efficiently exploit and scientifically develop the ocean.However,at present,there are still deficiencies in human’s understanding of the ocean.Among many ocean phenomena,mixed layer is an important ocean phenomenon which plays a significant role in the study of ocean dynamics and global climate change.However,the methods of estimating mixed layer depth(MLD)still have limitations due to the sparse resolution of the observed data.In recent years,with the development of remote sensing technology and the Argo project,multisource ocean data have become more and more abundant,the era of big data has been ushered in by ocean science.Based on multisource ocean data,the estimation of ocean internal parameters by artificial intelligence has become a hot issue in oceanography and statistical science.At present,the research results in this field are still relatively insufficient,the existing estimation models have poor accuracy.Based on this,the integration of multiple artificial intelligence method and using multisource sea surface data for the estimation of MLD is of great theoretical significance and practical application.In this paper,based on a variety of artificial intelligence algorithms,multisource sea surface parameters are used to estimate MLD in different areas of the Indian Ocean.On the basis of preprocessing multisource sea surface data,a hybrid estimation model,namely BO-ET model,which combines Bayesian optimization algorithm(BO)and extreme random tree(ET)algorithm,and a hybrid estimation model,namely K-ANN model,which combines K-means clustering algorithm and artificial neural network(ANN)are established.They are used to estimate MLD in the typical areas of the Indian Ocean and the whole Indian Ocean respectively.The estimation accuracy is verified and analyzed by calculating the root mean square error(RMSE)and determination coefficient(R~2)of the model.The influence of sea surface parameters on MLD in different seasons is analyzed by Pearson correlation coefficient.The main research contents and results are as follows:(1)Study on MLD estimation in typical areas of Indian Ocean based on BO-ET modelThis paper focus the typical areas of the Indian Ocean where the physical ocean motion is complex and changeable.Based on the satellite remote sensing data and Argo observational data from January 2012 to December 2018,five sea surface parameters:sea surface temperature(SST),sea surface salinity(SSS),sea surface height(SSH),horizontal component(USSW)and vertical component(VSSW)of sea surface wind(SSW)are selected as input variables to establish a BO-ET hybrid estimation model.After preprocessing the data,the gridded monthly average data from 2012 to 2018 is selected as the training set of the model,and the gridded monthly average data of 2019is selected as the test set of the model.In order to verify the validity of the selected sea surface parameters,experiments with five different parameter combinations are designed in this paper.The experimental results show that the five selected parameters have a positive effect on improving the accuracy of the model.In order to verify the positive effect of BO algorithm on the ET model,this paper compares the estimation accuracy of ET model before and after applying BO algorithm.The results show that BO algorithm can significantly improve the estimation accuracy of the ET model and reduce the RMSE of three typical areas from 2.50 m to 2.14 m.In addition,the performance of this model is significantly better than that of decision tree(DT)and random forest(RF).The model can accurately estimate MLD distribution characteristics and seasonal variation in the local area of the Indian Ocean.(2)Study on MLD estimation in Indian Ocean based on K-ANN modelTaking the whole Indian Ocean as the study area,this paper selects the satellite remote sensing data and Argo observational data from January 2012 to December 2018as the training set.Combining K-means clustering algorithm and ANN model,a new K-ANN hybrid estimation model is established to retrieve the annual average value of MLD in 2019.The model takes five sea surface parameters SST,SSS,SSH,USSW and VSSW as inputs to retrieve MLD in the whole Indian Ocean.The estimation results show that the average RMSE and R~2 of the model in the whole Indian Ocean are 3.79 m and 0.67 respectively,which indicates that the hybrid estimation model can reveal the characteristics of MLD distribution in the Indian Ocean.On this basis,five comparative experiments with different input combinations of sea surface parameters are designed to quantitatively analyze the impact of different sea surface parameters on the K-ANN model.The results show that all sea surface parameters have a positive effect on the model,and the estimation effect of K-ANN model with five input parameters(SST,SSS,SSH,USSW and VSSW)is the best.The estimation accuracy of the model decreases with the decrease of parameters.Compared with the traditional multiple linear regression model(MLR)and Hybrid-Coordinate Ocean Model(HYCOM)model,the estimation effect of this model is better and the generalization ability is stronger.