Error correction of model analog forecasts using a linear inverse model for improving statistical ENSO prediction skill

 Accurately predicting the El Niño-Southern Oscillation (ENSO) remains a central challenge in seasonal climate forecasting. Statistical approaches, such as the Linear Inverse Model (LIM) and the Model Analog (MA), have been widely applied to predict sea surface temperature and sea surface height anomalies in the tropical Indo-Pacific, but each approach has intrinsic limitations. Although their weighted combination, MA-LIM, improves forecast skill over LIM and MA individually, it is still affected by residual biases from both methods. To address this limitation, this study introduces a new statistical prediction framework, NEW MA-LIM, which more optimally unifies MA and LIM by explicitly modeling the temporal evolution of MA forecast errors within the LIM operator and applying dynamic corrections at each forecast lead. Hindcast experiments for the period 1961–2023, using observational datasets and 15 CMIP6 preindustrial control simulations, show that NEW MA-LIM consistently outperforms LIM, MA, and MA-LIM across 1–12 month leads. In particular, it substantially alleviates the spring predictability barrier. A key finding is that MA forecast errors exhibit significant linear predictability, and their spatiotemporal patterns can be effectively reproduced by LIM. This enables more reliable ENSO prediction within a low-dimensional dynamical framework.