ENSO Forecast Skill in a Changing Climate (Invited Presentation)

Diagnosing El Niño-Southern Oscillation (ENSO) predictability within operational forecast models has been challenging — due to the computational expense of generating suites of retrospective ensemble forecasts, and the need for accurate, balanced initialization, typically via three-dimensional global data assimilation. In this study, we leverage an alternative approach that sidesteps these limitations, by using an inexpensive yet powerful model-analog technique to examine multi-year ENSO predictability since the late 1800s. First, we demonstrate that 20th-century ENSO forecast skill using model-analogs is comparable to that of twice-yearly hindcasts produced by a state-of-the-art European operational forecasting system. Over the period examined, ENSO exhibits its highest amplitude and forecast skill towards the end of the 19th century, and again in recent decades. To better understand the extent to which these multi-decadal variations in ENSO forecast skill can be attributed to climate change versus internal variability, we conduct a perfect model-analog experiment, where the initial model-analogs are selected directly from historical large ensemble simulations generated by various coupled GCMs, and the subsequent model-analog trajectories (forecasts) are verified against those models' own actual evolutions. Although the ENSO variations differ somewhat among the individual ensemble members, the resulting ensemble forecast skill is found to mirror the evolution of the ENSO variance over the historical period. Moreover, the forced response of ENSO appears to have increased in the past, as well as ENSO forecast skill.