Tracing the Unlabeled Storm: Cross-Variable Transfer in a Lagrangian Atmospheric JEPA Framework

arXiv:2608.22358v1 Announce Type: new Abstract: Deep atmospheric convection governs South Asian monsoon variability, yet attempting to learn its latent world model directly from zero-inflated, heavy-tailed precipitation yields suboptimal predictive representations. Continuous atmospheric proxies, such as outgoing longwave radiation (OLR), express this convective organization far more coherently. We address this mismatch with cross-variable proxy learning: M-JEPA, a multiscale Monsoon Joint-Embedding Predictive Architecture, is pretrained on five continuous proxy fields over Lagrangian patches tracking moving convective systems---without rainfall supervision at any point. The resulting frozen representation is transferred to daily precipitation forecasts through a shared decoder trunk featuring parallel probabilistic and deterministic branches. Because rainfall is strictly unobserved during pretraining, downstream skill directly measures the predictive information captured in the latent rollout. A frozen-backbone probing framework with two controls (an identical architecture trained on rainfall alone, and a randomly initialized backbone) attributes the transfer specifically to proxy pretraining: direct rainfall training exhibits 36% higher CRPS error (7.52 vs. 5.54,mm/day). Against the 51-member operational ECMWF ensemble, the transferred model attains a statistically resolved CRPS advantage (6.81 vs. 6.89,mm/day) and higher Brier skill (+0.05 vs. -0.04) using 15.4M parameters on a single consumer GPU, concentrated at heavy-rain thresholds and fine spatial scales, while the ensemble retains an advantage in neighborhood skill and deterministic references on point metrics. The result provides a competitive monsoon precipitation forecast grounded in intraseasonal dynamics and a diagnostic framework for evaluating transferred atmospheric representations.

Source: arXiv cs.LG | 2026-08-25