A Study Of Ensemble Forecating For Indian Ocean Dipole Using The Climatically Relevant Singular Vector Method

An Indian Ocean Dipole(IOD)ensemble forecast system is preliminarily established based on the global climate prediction system of the National Marine Environment Forecasting Center using the climatically relevant singular vector method by perturbing the sea surface temperature.Results show that the ensemble forecast system has certain ability of deterministic and probabilistic skills.And then the system is compared with the random perturbed system.Finally,this paper discussed the influence of perturbing different ocean layers and of different amplitudes of perturbations on the forecasts.This paper provides meaningful references for the operational IOD ensemble forecast.The main conclusions of this paper are as follows:(1)CSV1 is sensitive to different start months and perturbations of different layers.Generally speaking,when the lead time is one month,the perturbation patterns are similar to each other.They all show a three poles pattern,but not exactly the same.In addition,compared with the CSV1 calculated at other start month,the CSV1 calculated from the perturbed SST and the perturbed subsurface temperature are more similar,but also not identical.In the operational forecast,the corresponding CSV needs to be calculated for different start months and different perturbation variables.(2)Results of the deterministic forecast of EM_SST show that forecast skills have obvious seasonal variation characteristics,and the prediction is greatly affected by the IOD winter forecast barrier.The correlation skill for the start month of March is the lowest while the September is the highest.The forecast skills of the East and the West poles are higher than the skills of IOD.(3)Results of probabilistic forecasts for EM_SST show that the dispersion of ensembles in EM_SST is low,which may lead to overconfidence of the system,and may cause false alarms in operational forecasts.Through the test of BS score,hit rate,false alarm rate and ROCA of probabilistic forecasts of three events of DMI index,it shows that there is a certain probabilistic prediction skill in EM_SST.(4)Comparing EM_SST with Control Run,the results show that the climatically relevant singular vector method is better than the random disturbance method in the deterministic prediction.Among them,the improvement of correlation coefficient forecast skill is better,and the maximum improvement can reach about 0.1.The improvement of RMSE is relatively smaller.The improved period is mainly in the months with low predictability.In terms of probabilistic prediction,the spread of EM_SST and Control Run is basically the same.Examinations using other probabilistic forecasts methods also show that EM_SST has no significant improvement on Control Run.(5)The comparison of EM_SST and EM_Sub shows that EM_SST is better at two months and three months lead,but EM_Sub is better after 3 months forecasts.Compared with EM_SST and EM_Sub,it is found that the spread of ensembles is sensitive to the sea temperature perturbation of different layers.(6)The sensitivity of forecast skills to different random perturbation amplitudes is investigated.The results show that the forecast skills are more sensitive to the magnitude of perturbation,whether for correlation coefficient,root mean square error or dispersion.