Marine surface latent heat flux prediction model based on multi-scale attention mechanism and data assimilation

The invention provides an ocean surface latent heat flux prediction model based on a multi-scale attention mechanism and data assimilation, and relates to the field of oceanography and climate dynamics. The method is mainly divided into three steps: feature extraction based on a U-Net architecture, feature enhancement of a multi-scale attention mechanism and ensemble Kalman filter (EnKF) data assimilation. In the first step, the model adopts an encoder-decoder structure, and spatial features are systematically extracted, converted and reconstructed from ocean surface latent heat flux data. In the second step, a multi-scale attention mechanism is introduced, so that the recognition capability of the model is remarkably improved. In the third step, real-time observation data is assimilated through EnKF, continuous adjustment is carried out based on new data input, and prediction robustness and accuracy are ensured. Experimental results show that the model has excellent performance in the aspect of predicting th