Data for manuscript "The 3D-Geofomer for ENSO studies: a transformer-based model formulation with an integrated gradient method for enhancing its explainability"

Version1 is missing the 22nd compressed file, resulting in incomplete data. This version has been corrected. This public library contains the training code of the model, the parameters of the trained model, and the hindcast results from Jan 1983 to Dec 2023 for 20 lead months. You can directly use the hindcasts for ENSO predictability analysis and skill comparisons, or use the trained models for sensitivity experiments.