Enhanced Variational U-Net for Weather Forecasting

This work describes our third-place solution in both the core and transfer learning challenges of Weather4cast – IEEE BigData Cup. The solution builds on our success in Weather4cast - Stage 1 [1], and uses the same Variational U-Net architecture. Building on the lessons learned from Weather4cast - Stage 1, we enhanced the model’s performance through the use of data augmentations and model blending. Furthermore, we explored transfer learning between the two competitions. The code for this solution is available at https://github.com/qiq208/w4c-2021-IEEE