Artificial Intelligence powered forecast of oceanic mesoscale phenomena: A typhoon cold wake case occurring in Northwest Pacific Ocean

Abstract

The phenomenon of typhoon cold wake has always been a key research object in the field of ocean–atmosphere interaction. Different from the physical equation-based numerical model which has long been dependent on, inspired by largely available forecasting and reanalyzed data and the advance of artificial intelligence technology, we build RC_ViT deep learning model for forecasting the typhoon cold wake. The total RMSE of grid points at each moment was between 0.5 and 0.9 °C, indicating that the model was accurate to forecast the sea surface temperature after the pass of typhoon. During the testing period of 11 years (2010–2020), using the deep learning method, our artificial intelligence model accurately and efficiently forecasts the sea surface temperature field. The cold wake of typhoon Megi (2010) was simulated by a Regional Ocean Model System (ROMS) and the artificial intelligence model separately. The results show that before the typhoon passing, the SST forecasted by the artificial intelligence model is more accurate, and after the typhoon passes, the ROMS forecast result has a smaller RMSE value. The reason may be that the intensity of typhoon Megi was too strong, and the artificial intelligence model training samples were few, resulting in artificial intelligence model forecast accuracy lower than traditional numerical models after the Megi passing. This study demonstrates the strong potential of the data-driven artificial intelligence model as an alternative to traditional numerical models for forecasting oceanic phenomena.