Towards optimal observational array for dealing with challenges of El Niño-Southern Oscillation predictions due to diversities of El Niño

This paper investigates the optimal observational array for improving the initialization of El Niño-Southern Oscillation predictions by exploring the sensitive areas for target observations of two types of El Niño predictions. The sensitive areas are identified by calculating the optimally growing errors (OGEs) of the Zebiak–Cane model, as corrected by the optimal forcing vector that is determined by assimilating the observed sea surface temperature anomalies (SSTAs). It is found that although the OGEs have similar structures for different start months of predictions, the regions covered by much large errors for the SSTA component tend to locate at different zonal positions and depends on the start months. Furthermore, these regions are also in difference between two types of El Niño events. The regions covered by large errors of OGEs represent the sensitive areas for target observations. Considering the dependence of the sensitive areas on related El Niño types and the start months of predictions, the present study propose a quantitative frequency method to determine the sensitive areas for target observations associated with two types of El Niño predictions, which is expected to be applicable for both types of El Niño predictions with different start months. As a result, the sensitive areas that describe the array of target observations are presented with a reversal triangle-like shape locating in the eastern Pacific, specifically the area of 120°W–85°W, 0°S–11°S, and an extension to the west along the equator and then gathering at the 180° longitude and the western boundary. “Hindcast” experiments demonstrated that such observational array is very useful in distinguishing two types of El Niño and superior to the TAO/TRITON array. It is therefore suggested that the observational array provided in the present study is towards the optimal one and the original TAO/TRITON array should be further optimized when applied to predictions of the diversities of El Niño events.