State-Dependent Predictability of ENSO Diversity

Predicting El Niño–Southern Oscillation (ENSO) diversity beyond boreal spring requires more than estimating winter Niño 3.4 amplitude. It requires separating phase sign, Eastern Pacific and Central Pacific spatial expression, and event persistence, because each forecast question can depend differently on Pacific ocean memory and remote tropical information. This thesis tests whether variability in the subtropical South Atlantic (STA) and in the South Atlantic Convergence Zone (SACZ), the clearest low-level expression of the South American Monsoon System (SAMS), provides physically interpretable long-lead information for December–January–February of year +1 (DJF(+1)) ENSO diversity, and whether that information depends on the first-winter Pacific ocean state. The analysis separates pathway discovery from conditional evaluation. It first uses the 1945–2024 ERA5 record to identify candidate STA–SACZ predictive-influence pathways through optimized regional modes, Peter and Clark Momentary Conditional Independence Plus (PCMCI+) with forward-in-time conditioning, physical diagnostics, and out-of-sample prediction tests. It then uses the 1979–2025 satellite-era record, including ORAS5 subsurface fields, to evaluate the retained pathway against Pacific initial-state predictors and established remote precursors with Bayesian model comparison and blocked prediction checks. The pathway analysis identifies a June–July–August of year −1 (JJA(−1)) STA sea surface temperature signal and two March–April–May of year 0 (MAM(0)) SACZ–Amazon Rossby wave source signals. The superposition of the early STA signal and the later SACZ–Amazon signals is consistent with South Pacific circulation adjustment, Walker-circulation modulation, and subsequent growth of Equatorial Pacific sea surface temperature anomalies, and it improves long-lead prediction of the Eastern Pacific and Central Pacific components of ENSO diversity. In retrospective forecasts restricted to the causal parents, correlation skill for the DJF(+1) E-index and C-index improves by approximately 0.15 over canonical precursor sets, with the largest gains before the boreal spring predictability barrier and useful skill extending to leads of about fifteen months; these estimates derive from controlled hindcast tests on a short record and carry no operational claim. The fingerprints are earlier and more coherent for Eastern Pacific events, while Central Pacific events show later and more mixed signals. The conditional evaluation shows that the first-winter Pacific ocean state conditions the forecast value of this remote information. In active years, the contribution of remote predictor families is state dependent and can be shared with established ENSO precursors. For phase emergence from neutral first-winter conditions, the STA–SACZ family provides the clearest remote information about the sign of the developing event. By contrast, La Niña duration is governed more clearly by ocean heat content and broader subsurface conditions. Overall, the thesis reframes South Atlantic and SAMS variability as a conditional component of pantropical ENSO predictability: Pacific recharge–discharge dynamics remain central, while STA–SACZ information becomes most useful under specific initial oceanic conditions.