Advancing Hurricane Forecasting: Model Utilizing Transformers Trained on MERRA2 and ERA5 Data

Transformers, in general, have been explored in the domain of language and vision models to understand long sequences and can be utilized in different Earth Science applications. Foundation models in the domain of vision and language models utilizing transformers have shown great promise for generalization over different downstream applications. Fourier neural operator-based token mixer transformers with a backbone of ViT have been used for predicting wind and precipitation on the ERA5 dataset. Our goal is an attempt to forecast a hurricane and its spatial distribution of its structure using the MERRA2 data.

The initial study was performed using Fourcastnet architecture, which was trained over the MERRA2 data set with 38 variables. We trained on data from 2000 to 2015 and for prediction, we used the initial condition for the model based on 406 different initial conditions for 41 named hurricanes formed from 2017 to 2022. The tracks and intensities of multiple cat 4-5 hurricanes were forecasted, which produces a track error range of 50-70 km and 70-100 km for the 6-hour and 18-hour predictions, respectively. Though there was a relatively large error in estimating the hurricane intensity, this error was 5-7 hPa and 10-15 hPa for 6-hour and 18-hour predictions. The tracks and intensities are validated against the observed HURDAT track and Weather Research and Forecasting (WRF) model (at 50 km spatial resolution), and have the similar track and intensity error.

With respect to the baseline model which has been trained on the ERA5 dataset, track and intensity for the hurricane Florence (2018) and Micahel (2018) has been improved in the MERRA2 trained model. Track error in ERA5 and MERRA2 trained models is 327 km and 85 km respectively for the hurricane Florence (2018). Fourcastnet showed promise in forecasting-based tasks with a lead time of up to 5 days. The MERRA2 trained model shows the correlation of more than 0.9 until the 5 day forecast for wind speed, SLP, T2M, Z500 etc. Overall, FourcastNet model trained on MERRA2 data can be utilized to forecast the hurricane track and intensity.