ENSO Predictability Study Based On The Conditional Nonlinear Optimal Perturbation Approach And IOCAS ICM

The El Ni?o,an ocean-atmosphere coupled phenomenon characterized by an abnormal warming in the central and eastern tropical Pacific,has been much focused on because of its effects on natural disasters around the world.In recent decades,continued in-depth studies of El Ni?o events have deepened our understanding of its dynamics,modelings and predictions.However,significant uncertainties still exist in real-time ENSO predictions.Generally,the prediction uncertainties are mainly attributed to errors in the initial conditions and numerical models.In this study,based on an intermediate coupled model(ICM),the conditional nonlinear optimal perturbation(CNOP)approach was employed to study the optimal initial condition and model parameter errors and also their combinded effects on largest error growth in the El Ni?o predictions.Then,we investgated the extent to which ENSO predictions can be improved by removing CNOP-related initial condition errors in the ICM.Lastly,some suggestions with respect to model improvement are presented.The main contents and conlusions are summaried as follows:(1)The simulative and predictive skills of the tropical Pacific SST were investigated with the ICM.The ICM can successfully depict a dominant four-year oscillation period of ENSO cycle and phase locking.The high prediction skill region is located in the central and eastern equatorial Pacific.Similar to other ENSO models,the “spring predictability barrier”(SPB)is also strong in the ICM.(2)The optimal initial condition errors(as represented by CNOP-I)in sea surface temperature anomalies(SSTAs)and sea level anomalies(SLAs)were obtained with seasonal variation.The CNOP-induced perturbations tend to evolve into the La Ni?a mode beacused it casuses the Bjerknes-like positive feedback and thermocline feedback.The CNOP-I was found to induce the SPB phenomenon.Based on the characteristic distributions of the CNOP-I,it implies that the upper layer in the central equatorial Pacific and subsurface in the eastern Pacific are the most sensitive areas for El Ni?o prediction in the ICM.Given the season-dependence of the CNOP-I,targeted observing strategies are suggested to be implemented seasonally.(3)The extent to which ENSO predictions can be improved by removing CNOPrelated initial condition errors in the ICM was investigated.Observing system simulation experiments(OSSEs)indicate that additional obsevations in the central equatorial Pacific are more effective for improvement of prediction skills than in other areas: The forecast errors can be reduced by 25%.It is worth noting that removing initial errors in certain areas may worsen the prediction due to the imbalance of initial fields.Further,CNOP-I-related targeted observation strategies are employed.It was found that seasonal varying observational network can effectively limt the prediction error growth: On the premise of the central Pacific observations,an additional observation in the easern Pacific during April to October can futher improve the forecast skills by more than 62%.Particullarly,CNOP-I-related targeted obsevation can weaken the SPB phenomenon and vice versa.(4)The roles playe by model parameter errors [relative ocean-atmosphere coupling coefficient()and thermocline effect(0))],initial condition errors and their optimal combination in ENSO prediction uncertainties were investigated.It reveals that the prediction errors induced by CNOP-P,which are found to be depedent on El Ni?o itself,show great uncertainties.However,despite all that,CNOP-type parameter errors have localized distributions: the component errors are mainly located in the central equatorial Pacific,whereas 0)component errors are mainly located in the eastern Pacific cold tongue region.CNOP-P can strengthen the Bjerknes feedback and subsurface thermal forcing to surface,so that the prediction results are deviated from the reference ENSO events.Furthermore,the optimal combinations of parameter and initial condition errors(C-CNOPs)were calculated.Seasonal C-CNOPs-induced error evolutions are similar to those of CNOP-I but have larger amplitude for their intensified Bjerknes feedback and thermocline feedback.Addtionally,more significant SPB phenomena are induced by C-CNOPs than by CNOP-P,CNOP-I or even CNOPP+CNOP-I(the simple combination of the CNOP-I and CNOP-P).It indicates that the coexisting initial and model errors are more likely to lead to a significant SPB phenomenon,thus contaminating the prediction results.(5)Revealing a new way to improve the ENSO prediction skill on model perspective.The parameter errors derived frome C-CNOPs or CNOP-P dominate a few areas: errors are concentrated on the central equatorial Pacific,and 0)errors are mainly in the eastern Pacific cold tongue region.That is to say,the El Ni?o simulations and predictions are significantly sensitive to the ocean-atmosphere coupling in the central Pacific and thermocline effect in the eastern Pacific.Therefore,except for providing accurate observations,improving these two dynamic representions in the central and eastern Pacific,especially their parameterization in numerical model,can more effectively enhance the El Ni?o prediction skills.