Multiyear Statistical Prediction of ENSO Enhanced by the Tropical Pacific Observing System

Abstract

The theoretical predictability limit of El Niño–Southern Oscillation has been shown to be on the order of years, but long-lead predictions of El Niño (EN) and La Niña (LN) are still lacking. State-of-the-art forecasting schemes traditionally do not predict beyond the spring barrier. Recent efforts have been dedicated to the improvement of dynamical models, while statistical schemes still need to take full advantage of the availability of ocean subsurface variables, provided regularly for the last few decades as a result of the Tropical Ocean–Global Atmosphere Program (TOGA). Here we use a number of predictor variables, including temperature at different depths and regions of the equatorial ocean, in a flexible statistical dynamic components model to make skillful long-lead retrospective predictions (hindcasts) of the Niño-3.4 index in the period 1970–2016. The model hindcasts the major EN episodes up to 2.5 years in advance, including the recent extreme 2015/16 EN. The analysis demonstrates that events are predicted more accurately after the completion of the observational array in the tropical Pacific in 1994, as a result of the improved data quality and coverage achieved by TOGA. Therefore, there is potential to issue long-lead predictions of this climatic phenomenon at a low computational cost.

The 2018–2019 weak El Niño: Predicting the risk of a dengue outbreak in Machala, Ecuador

“…(2017); Petrova et al . (2020)) to forecast the SST in the Niño3.4 region in 2019. The forecast indicated a continuation of the mild El Niño event from 2018 over the summer, autumn and winter months of 2019, with a slight increase in the warm anomaly at the end of 2019 as compared to the end of 2018.…”

“…The ENSO model used here to predict SST in the Niño3.4 region in 2019 is the dynamics components model described in Petrova et al . (2017, 2020), as well as applied in our previous studies on dengue prediction in El Oro and Machala (Lowe et al ., 2017; Petrova et al ., 2019). The model is composed of unobserved components—trend, time‐varying seasonal cycles, and four‐cycle components corresponding to inter‐annual and decadal variability of SST in the Niño3.4 region, as well as of sets of predictor variables—SST, subsurface temperature and zonal wind stress from different regions in the tropical Pacific Ocean, which are also used at different lead times.…”

“…The model is composed of unobserved components—trend, time‐varying seasonal cycles, and four‐cycle components corresponding to inter‐annual and decadal variability of SST in the Niño3.4 region, as well as of sets of predictor variables—SST, subsurface temperature and zonal wind stress from different regions in the tropical Pacific Ocean, which are also used at different lead times. The model has been successful in predicting retrospectively the ENSO events in the period 1970–2013 at a lead time of more than 1.5 years, and it predicted operationally the events thereafter with the exception of the 2017/2018 La Niña event when the model predicted neutral conditions instead (Petrova et al ., 2017, 2020). Forecasts of the Niño3.4 index in 2019 were run using the observed data up until December 2018.…”

A New Empirical Estimation Scheme for Daily Net Radiation at the Ocean Surface

“…Eight sites from OceanSITES were used. Observations from 66 moored buoy sites have been widely applied in previous studies, such as air-sea interaction studies [13,[39][40][41], ENSO prediction [42], and monsoon variability studies [23].…”

Long-term forecasting of El Niño events via dynamic factor simulations

“…Petrova et al (2017) use a dynamic components model and a number of specially designed subsurface temperature and wind stress predictors (also used in the present methodology) to predict the major EN events in 1996-2015 at lead times of up to 34 months. Furthermore, Petrova et al (2018) show a prediction of the whole ENSO time series between 1972-2016 at a lead time of 2 years with an improved version of the model described in Petrova et al (2017). Finally, Gonzalez and Goddard (2016) use decadal prediction experiments from the Coupled Model Intercomparison Project (CMIP5) to explore the long-term predictability of ENSO with different dynamical schemes.…”