New metrics for distinguishing the skill of long-range ENSO forecasting models

Long-range seasonal ENSO (El Niño Southern Oscillation) forecasts are provided by operational dynamical and statistical models and the skills of these models are a matter of contention. In this work, new skill metrics are proposed for determining whether or not the model skills are significantly different. Using an ENSO idealized data set, it is shown that the newly developed metrics RIP (Rotated Index Positive) and RIN (Rotated Index Negative) were capable of distinguishing between under-prediction and over-prediction whereas other popular metrics ACC (Anomaly Coefficient Correlation) and RMSE (Root Mean Squared Error) metrics failed. These metrics were also applied to perform skill assessment on the ENSO operational data set. RIP, RIN, ACC and RMSE metrics successfully differentiate model skills based on the ENSO phase. Dynamical models need more improvement in reducing their false alarms. The two models do not significantly differ in predicting the La Niña phase. It is recommended that both RIP and RIN should be used to complement ACC and RMSE in model skill assessment.