El Ni?o-Southern Oscillation (ENSO) is the strongest interannual climate mode in the atmosphere-ocean coupling system over the tropical Pacific. Westerly wind bursts (WWBs) are an important precursor of the ENSO events. In the past few decades, several statistical and dynamics-based models have been used to simulate and predict ENSO events. However, these models have some difficulties in simulating WWBs, leading their simulation and prediction performance for ENSO events that are also limited. This study constructs U-Net models that are driven by several atmospheric and oceanic data for zonal wind anomalies (ua) in the tropical Pacific. Furthermore, the spatiotemporal characteristics of WWBs are identified and their relationship with ENSO phenomenon are analyzed. The results indicate that the U-Net model can effectively simulate the spatiotemporal distribution characteristics of WWBs during the testing period (2003-2022). The deviation between the occurrence frequency and accumulated days of WWBs reconstructed by the model and the observed values is less than 4.0 %, and the correlation coefficient between the time series of WWBs occurrence probability (P) simulated by the model and observed values 0.87. Meanwhile, the U-Net model can effectively capture the non-linear relationship between WWBs and ENSO, and there is a significant lead-lag correlation between WWBs and ENSO in both U-Net models and reanalysis data. Besides, the duration, zonal width and the average maximum amplitude of WWBs reach their peak during El Ni?o events, and the peak of probability (P) of WWBs during El Ni?o events reconstructed by the U-Net model is much closer to that of observation. In conclusion, compared with the traditional WWBs parameterization scheme that relies on establishing an approximate linear relationship between WWBs and sea surface temperature fields, the U-Net model has significant advantages in representing WWBs.