This paper addresses the problems of traditional surface current inversion methods, including low model accuracy and insufficient spatial and temporal resolution of the corresponding data, especially the challenges in processing remote sensing image data bearing cloud cover and performing large-scale time series inversion. In this study, several machine learning methods, including Random Forest, LightGBM, and Deep Neural Network, based on Himawari-8 satellite data and AVIS measurements, are employed to realize surface currents' inversion in the South China Sea. The experimental results fully demonstrate that using Himawari-8 full-band feature data for the inversion of sea surface currents even under cloud coverage is feasible. Moreover, the accuracy of the LightGBM model is better than other algorithms, with a correlation coefficient of 0.8856 and a root-mean-square error (RMSE) of 6.15 cm/s. These conclusions clearly show that the LightGBM algorithm is able to overcome the limitations of the traditional methods and realize the inversion of surface currents for the whole region.
The traditional models for inverting sea surface temperature (SST) have relatively low accuracy and are unable to predict the SST in cloud-covered areas of remote sensing data. This study focuses on the South China Sea and its surrounding areas. Based on the 2022 Himawari-8 satellite imagery data and iQuam2 measured data, a 1x1 sampling window was selected to construct the inversion dataset. The support vector regression (SVR), artificial neural network (ANN), and LightGBM algorithms in machine learning were employed for SST inversion under two scenarios: clear sky and entire region, based on the presence or absence of cloud cover. The inversion results during the testing phase indicate: (1) The LightGBM algorithm demonstrates higher inversion accuracy for the entire region compared to the clear sky scenario, indicating its ability to effectively mitigate cloud interference; (2) Under the entire region scenario, LightGBM achieves a correlation coefficient of 0.9407, mean absolute error of 0.2335°C, and mean square error of 0.1803°C, outperforming other algorithms; (3) The inversion accuracy of machine learning algorithms is significantly higher than that of Himawari8 Level 2 products. These conclusions highlight that the LightGBM algorithm can overcome the limitations of traditional methods and achieve high-precision SST inversion for the entire region.
In view of the complexity and low accuracy of traditional sea surface salinity inversion models, this study took the Bay of Bengal sea area (82°~91°E, 9°~15°N) as the research area, and used the L-band sounding SMOS Level2 satellite product data, Argo measured data and characteristic data. Catboost algorithm was used to construct the remote sensing inversion model of sea surface salinity, and grid search algorithm was used to optimize the model parameters, so as to improve the accuracy of SMOS satellite products. The validation results show that the root mean square error (RMSE) of the Catboost model's test set inversion of SSS is 0.2929psu, and the mean absolute error (MAE) is 0.1885psu. Comparison and analysis using measured data show that the Catboost model's inversion of SSS has a high degree of consistency with Argo data. The validation also analyzed the RMSE errors between the model's inversion results and the Argo measured results for each month in 2021, revealing the potential relationship between climate and meteorological variables in the Bay of Bengal and changes in salinity.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.