Tropical cyclone intensity forecasting using model knowledge guided deep learning model

Abstract

This paper developed a deep learning (DL) model for forecasting tropical cyclone (TC) intensity in the Northwest Pacifi c. A dataset containing 20,533 synchronized and collocated samples was assembled, which included ERA5 reanalysis data as well as satellite infrared imagery, covering the period from 1979 to 2021. The u-, v- and w-components of wind, sea surface temperature, infrared satellite imagery, and historical TC information were selected as the model inputs. Then, a tropical-cyclone-intensity-forecast-fusion (TCIF-fusion) model was developed, in which two special branches were designed to learn multi-factor information to forecast 24- hour TC intensity. Finally, heat maps capturing the model's insights are generated and applied to the original input data, creating an enhanced input set that results in more accurate forecasting. Employing this refi ned input, the heat maps (model knowledge, MK) were used to guide TCIF-fusion model modeling, and the model- knowledge-guided TCIF-fusion model achieved a 24-hour forecast error of 3.56 m/s for Northwest Paci fic TCs spanning 2020-2021. The results show that the performance of our method is signi ficantly better than the official subjective prediction and advanced deep learning methods in forecasting TC intensity by 4% to 22%. Moreover, in contrast to approaches by operational agencies, the model-knowledge-guided approach can forecast TC intensity landfall better.