Skillful subseasonal-to-seasonal ensemble forecasting with a multi-sphere coupled probabilistic model

Accurate subseasonal-to-seasonal (S2S) prediction is critical for disaster mitigation under accelerating climate change, yet remains challenging due to complex multi-sphere interactions and intrinsic atmospheric uncertainty. Here we present TianXing-S2S, a multi-sphere coupled probabilistic deep learning model for global S2S ensemble forecasting. TianXing-S2S first encodes diverse multi-sphere predictors into a compact latent space, then employs a diffusion model to generate daily ensemble forecasts. A novel coupling module based on optimal transport (OT) is incorporated in the denoiser to optimize the interactions between atmospheric and multi-sphere boundary conditions. Across key atmospheric variables, TianXing-S2S outperforms both the European Centre for Medium-Range Weather Forecasts (ECMWF) S2S system and FuXi-S2S in 45-day daily-mean ensemble forecasts at 1.5° × 1.5° resolution. Our model improves the prediction of extreme events including heat waves and anomalous precipitation, with multi-sphere coupling identified as a key contributor. Furthermore, we demonstrate that TianXing-S2S can generate stable rollout forecasts up to 180 days, suggesting its potential as a useful framework for probabilistic S2S prediction under a changing climate.