Codes for Deep learning ENSO prediction dictated by nonlinear atmosphere convection

This repository contains Python 3.11 code and NetCDF files for generating the key findings and figures in the article by Cai et al. *ERA5_Nino3_olr_SSTA.nc contains the time series of SSTA and OLR anomalies in the Nino 3 region from ERA5. *model_nonlinear_convection_para.nc contains the nonlinear convective coefficients for each CMIP6 model used. *model_prediction_Nino34_ensemble_mean.nc contains the Nino3.4 index forecast results for each CMIP6 model, generated using our AI system. *model_true_Nino34.nc contains the Nino3.4 index in each CMIP6 model during 1970-2014. *obs_prediction.nc contains the observed and hindcast (from our AI system) Nino3.4 indices. *nonlinear_convection.py This script contains the code for calculating the nonlinear convective coefficient β, as well as the code for plotting the nonlinear convection in observations. *predictability_barrier.py This script contains the verification of ENSO forecast results against observations. It can generate the complete Extended Figure 1. It also includes validation of the key finding: that nonlinear convection helps weaken the Spring Predictability Barrier (Fig1E).