DeepONet based online climate bias correction

DeepONet

Consider an operator \(\mathcal{G}\), that maps from the input function \(v\) to the output function \(u\), i.e., \(\mathcal{G}: v \rightarrow u\). DeepONet tries to learn the operator \(\mathcal{G}\) by approximating the basis function for expressing the output functional space. The physics of the problem is enforced using the labelled input-output dataset pairs for the conventional architecture of the proposed operators.

• DeepONet for E3SM
  • Problem Setup for E3SM
  • DeepONet Setup
  • Data
  • Code Setup
  • Results
  • Online Integration of DeepONet with E3SM
  • Transfer Learning
  • Results for pre-trained DeepONet in different Zones
  • Loss comparision strategy-I
  • Loss comparision strategy-II
  • Encoder Decoder parameter analysis
  • DeepONet parameter analysis
  • Conclusion
• DeepONet for QG Model
  • Problem setup for beta model
  • Results for beta model