Inversion of propagation speed of internal solitary wave based on moderate resolution imaging spectroradiometer and visible infrared imaging radiometer suite images

Predicting the propagation speed of internal solitary waves (ISWs) is a crucial area in ISW research, and utilizing optical remote sensing images for speed inversion represents a key technique. We establish an ISW speed inversion model based on transfer learning techniques and extreme gradient boosting. Two datasets were created through optical simulation experiments and optical remote sensing observations. A pre-trained model was generated from the extracted 2119 laboratory physics simulation (LPS) samples, and a preliminary inversion analysis was conducted on a total of 1258 ocean remote sensing (ORS) data samples across four hotspot marine areas. The results indicate that linear recalibration based on domain adaptation can effectively correct systematic bias. Before and after calibration, the root mean squared error (RMSE) of the LPS model decreased from 1.08 to 0.22 m/s in four hotspot marine areas. Compared with the joint ORS model, the LPS model can achieve higher accuracy in sea areas with few samples, the RMSE and mean relative error of the LPS model in the test set reached 0.10 m/s and 6.8%, respectively, and the inversion results were in good agreement with satellite observation data. The speed inversion model provides more accurate results under various water depth conditions compared with the nonlinear phase speed calculated by the Korteweg–de Vries equation.