Inversion Of Ocean Subsurface Temperature Based On Machine Learning

The thermal structure of the ocean subsurface is of great importance to ocean circulation and global climate change,yet improving the accuracy of its inversion remains a great challenge.In recent years,with the development of remote sensing technology and the Argo project,satellite remote sensing data and in situ observation data have become more and more abundant,and the era of big data has been ushered in by marine science,and the use of artificial intelligence methods to invert the thermal structure of the ocean subsurface based on multi-source ocean data has become a hot issue in oceanography.At present,the research results in this field are still relatively insufficient,and the existing inversion models have single types and the inversion accuracy needs to be improved.Based on this,the integration of multiple machine learning methods for the inversion of ocean subsurface temperature is still of great theoretical significance and practical application.In this paper,a new ocean subsurface temperature(OST)estimation model combining the Gaussian mixture model(GMM)and light gradient boosting machine(LightGBM)algorithm was proposed.The model used multisource sea surface parameters including sea surface temperature(SST),sea surface salinity(SSS),sea surface height(SSH),northward and eastward components of sea surface wind(USSW and VSSW)to retrieve the subsurface thermal structure of the Indian Ocean.In addition,the CatBoost model was used to invert the global ocean subsurface temperature by adding longitude and latitude information(LON,LAT)as input variables to the five sea surface parameters.Based on the constructed subsurface temperature inversion model,the inversion of the subsurface temperature was realized and the accuracy of the model inversion results was analyzed and verified using root mean square error(RMSE)and coefficient of determination(R~2).The main research content and results are as follows:(1)Study on the inversion of subsurface temperature in the Indian Ocean based on GMM clustering and LightGBM algorithmThe inversion model integrating GMM clustering and LightGBM algorithm was developed and subsurface temperature inversion was carried out using the Indian Ocean as the study area,multi-source remote sensing parameters and Argo measurement data from January 2005 to December 2018.The inversion results show that the model proposed in this paper can accurately reflect the distribution characteristics and seasonal variation of the OST in the Indian Ocean.On this basis,three comparative experiments with different input combinations of sea surface parameters were designed to quantitatively analyze the influence of different input variables on the LightGBM model.The experimental results show that all sea surface parameters have a positive effect on the model.Moreover,the LightGBM model with five input parameters(SST,SSS,SSH,USSW,VSSW)has the best estimation effect,followed by the LightGBM model with three input parameters(SST,SSS,SSH)and two input parameters(SST,SSH).The LightGBM model has better simulation capabilities than another machine learning method which is the e Xtreme gradient boosting(XGBoost).(2)Study on the inversion of global ocean subsurface temperature based on CatBoost algorithmThe inversion model based on the CatBoost algorithm was developed by selecting the multi-source remote sensing parameter data and Argo measured data from 2010 to December 2018,using the global ocean as the study area.The inversion results show that the performance of the model varies at different depth levels,and the RMSE of the model first gradually increases and then gradually decreases,with a vertically averaged RMSE value of 0.2719°C and R~2 value of 0.9947.Compared with another machine learning method which is the gradient boosting decision tree(GBDT),the CatBoost model has better inversion accuracy and can accurately invert the OST distribution characteristics and seasonal variation patterns of the global ocean.