ABSTRACT This study conducts ensemble forecasting experiments for El Niño–Southern Oscillation (ENSO) events spanning 1982–2015, comparing two perturbation generation methods: the coupled condition nonlinear optimal perturbation (C‐CNOP) and singular vector (SV). We specifically focus on the sea temperature component of the C‐CNOP, referred to as CP‐T. The results demonstrate that the CP‐T ensemble mean forecast outperforms the SV ensemble mean forecast in capturing both the temporal evolution of Niño3.4 sea surface temperature anomalies (SSTAs) and spatial patterns of SSTAs across the tropical Pacific. This is especially pronounced during El Niño events with strong nonlinearity and at longer lead times, effectively extending the lead times for skillful forecasts. Furthermore, it is revealed that the CP‐T ensemble‐mean perturbations, which incorporate nonlinear effects, can better capture the nonlinear development of analysis errors and appropriately adjust the feedback between sea temperature and wind field, resulting in higher forecast skill than the SV ensemble mean forecast. Therefore, the C‐CNOP method is a valuable approach that not only appropriately considers the effect of initial coupling uncertainties but also incorporates the effect of nonlinearity, significantly improving ENSO forecasting skill.
Hu et al. (2026) studied this question.