Tropical basin interactions reduce spring predictability barrier of ENSO in a deep learning model

The El Niño-Southern Oscillation (ENSO) exhibits a pronounced decline in predictability during boreal spring, referred to as spring predictability barrier (SPB). While tropical basin interactions among the Indian, Atlantic, and Pacific Oceans potentially enhance ENSO predictability, their roles in mitigating SPB within deep learning (DL) frameworks remain underutilized. Here, we introduce GL-Geoformer, a DL model for global tropical ocean-atmosphere prediction. GL-Geoformer captures spatiotemporal evolutions of wind and three-dimensional temperature anomalies across the tropical basins. Our modeling demonstrates that incorporating tropical basin interactions substantially reduces SPB, enabling GL-Geoformer to achieve skillful ENSO predictions up to 16 months in advance when initiated in spring. Pacemaker experiments are performed to quantify individual and synergistic contributing nonlinearities of Indian Ocean Dipole and Atlantic Niño via subsurface heat transport and Walker circulation mechanisms, respectively. This study provides a data-driven framework to represent tropical basin interactions and reduce SPB, thereby deepening understanding of ENSO predictability.