ENSO Intensity Prediction Based On Asymmetric Tropical Air-sea Precursor Signal

El Ni?o-Southern Oscillation（ENSO）,which is the predominant mode of year-to-year climate variability in the tropical Pacific,can exert profound impacts on the local ocean-air conditions and farther climate variability over the globe via atmospheric teleconnection.Considering the distinct climate impacts for the ENSO events with different amplitude,it is of great importance to predict the amplitude of ENSO accurately for better regional disaster prevention and mitigation.Current climate models have relatively high skills in predicting ENSO phase once leaping over the spring predictability barrier.However,it is still a big challenge to realistically forecast the ENSO amplitude for both statistical and dynamical models currently.In this study,we focus on the accurate prediction of ENSO amplitude at its mature phase on the grounds of the physical understanding in ENSO dynamics as well as the great efforts in ENSO prediction.The main conclusions are summarized as follows:（1）The North America Multimodel Ensemble（NMME）has great room for improvement in ENSO amplitude forecast skill which is significantly correlated with the mean state bias,although it has relatively high skills in predicting ENSO phase at the lead time of six months.Based on the NMME hindcast data,we assess the forecast skill of the ENSO-related sea surface temperature（SST）anomaly and investigate its relationship with the mean state bias.In general,the ensemble mean succeeded in predicting most ENSO years（10of 13 El Ni?o and 10 of 14 La Ni?a）six months in advance during the period of 1982-2020,suggesting the relatively high skills of NMME in predicting the ENSO phase.However,the root-mean-square error（RMSE）between the predicted and observed ENSO amplitude,defined as Ni?o3.4 index averaged from November to the following January(Ni?o3.4_NDJ),is near 0.6℃.Accordingly,the ENSO amplitude might be largely overestimated or underestimated in the 6-month-lead prediction.The ENSO amplitude forecast error is significantly correlated with the mean state bias which includes an excessively westward-shifted cold tongue,too strong easterly wind in the western-central Pacific,and the double ITCZ problem.The forecast error is significantly reduced with the weaker easterly wind bias and double ITCZ problem.However,the cold bias in the cold tongue region does not undermine the ENSO amplitude forecast skill,suggesting a complicated physical linkage between them.（2）Both the preconditioning high-frequency zonal wind and the oceanic recharged/discharged states in association with ENSO amplitudes exhibit highly asymmetric spatiotemporal features.El Ni?o amplitude is closely associated with the recharged state in the western equatorial Pacific（130°-160°E）during February and the accumulated westerly wind events（WWEs）over the eastern equatorial Pacific（140°-170°W）from the previous December to May.In contrast,the amplitude of La Ni?a events is sensitive to the discharged state extending from the equatorial western to central Pacific（130°E-130°W）during February and the accumulated easterly wind events（EWEs）over the equatorial western Pacific（120°-150°W）from the previous November to April.The asymmetric features of the recharge-discharge process for El Ni?o and La Ni?a are associated with the ENSO evolution asymmetry.Almost all first-year La Ni?a is in cyclic transitions,preceded by El Ni?o in the previous winter.In contrast,El Ni?o is generally episodic,which is preceded by neutral state on average.The accumulated WWEs in the region east of the strongest WWEs can be identified as a good indicator of the eastward extension of the WWEs which could lead to the farther eastward extended warm pool and favor stronger El Ni?o.Likely for La Ni?a,the accumulated EWEs in the region west of the strongest EWEs can well measure the westward extension of the EWEs,suggesting that the farther westward extension of EWEs tend to favor a stronger La Ni?a via the farther westward extension of the warm pool.（3）Based on the asymmetric oceanic and atmospheric conditions,we established an effective ENSO amplitude forecast model which exhibits relatively high stability and feasibility.Considering these asymmetric preconditions,we here develop the amplitude forecast model for the El Ni?o and La Ni?a events separately.For the El Ni?o model,the expected winter Ni?o3.4 SST anomaly is computed using the linear method based on two parameters:the 130°-160°E oceanic heat content during February and the accumulated 140°-170°W WWEs from the previous December to May.In the La Ni?a model,two parameters used are the 130°E-130°W oceanic heat content in February and the accumulated 120°-150°E EWEs from the previous November to April.In the hindcast model of El Ni?o amplitude for the full period,the preceding oceanic recharged state and accumulated WWEs at the lead time of six months can explain about 90%of the total variance in the Ni?o3.4_NDJ SST anomaly and the RMSE between the prediction and observation is 0.22℃.As for La Ni?a,about 94%of the total variance can be explained by the preceding discharged state and accumulated EWEs,and the corresponding RMSE is about 0.14℃.Due to the limited sample size,a 1-yr cross validation and different independent sample tests are applied to test the overfitting of our model.When excluding any one El Ni?o/La Ni?a year from the training periods,the explained variance is always greater than 85%and the RMSE is always below 0.25℃.For different independent ENSO years,the correlation coefficient between the predicted and observed amplitude is always greater than 0.95,and the RMSE is always below 0.35℃.Our model can provide a much more accurate prediction of ENSO amplitude for the inconsecutive ENSO events which are correctly predicted in phase,showing a better performance respective to the NMME’s prediction.（4）In both the established statistical model and NMME,the atmospheric and oceanic variabilities outside the tropical Pacific during boreal summer and autumn of ENSO developing year have very limited effects on the forecast of ENSO amplitude in the following winter.In contrast to the moderate events,super El Ni?o（La Ni?a）events are always accompanied with an Atlantic Ni?a（Atlantic Ni?o）in summer,a positive（negative）Indian Ocean dipole（IOD）in autumn and increased（decreased）tropical cyclone（TC）activity over the western North Pacific.However,the difference between super and moderate ENSO events has already been apparent in the oceanic and atmospheric preconditions in the tropical Pacific before June.Based on the observed and reanalysis datasets,CMIP6 historical simulations as well as NMME hindcast data,this work proved that the atmospheric and oceanic variabilities outside the tropical Pacific,including Atlantic Ni?a（Atlantic Ni?o）,IOD,and TCs,plays a minor role on the ENSO amplitude forecast at the six-month lead in both statistical and dynamical models,which further emphasize the decisive role of tropical atmospheric and oceanic conditions.