Contribution of the location and spatial pattern of initial error to uncertainties in El Nio predictions

With the Zebiak-Cane model, the contribution of the location and spatial pattern of initial error in sea surface temperature anomalies (SSTA) to uncertainty in El Nio predictions is investigated using an approach based on conditional nonlinear optimal perturbation (CNOP), which seeks to find the initial error (i.e., the CNOP error) that satisfies a given constraint and that causes the largest prediction error at the prediction time. The computed CNOP error of SSTA has a dipole pattern in the equatorial central and eastern Pacific. The initial error from the equatorial central and eastern Pacific tends to grow more significantly than those from other locations. Because of the contribution of annual mean states the location of the initial error plays an important role in the error evolution; e.g., the shallow annual mean thermocline in the eastern Pacific favors feedback between the thermocline and sea surface temperature. Meanwhile, the specific dipole structure of the initial error is also crucial for optimal error growth. Even with the same magnitude as the CNOP error, random initial error in the equatorial central and eastern Pacific does not evolve significantly over time. Initial errors of SSTA with a similar spatial pattern to the CNOP error (i.e., the dipole pattern of SSTA error) give rise to larger prediction errors than those without similar spatial pattern do. Consequently, the magnitude of the prediction error at the prediction time depends on the combined effects of the location and spatial pattern of the initial error. If additional observation instruments are deployed to observe sea surface temperature with limited coverage, they should preferentially be deployed in the equatorial central and eastern Pacific.