Simulation Of Temperature By Decadal Prediction Experiments With BCC_CSM1.1

Decadal prediction experiment, as a new method of climate prediction, has attracted muchattention in recent years. The difference between this experiment and tranditional experiment ofclimate model is that the former considered the influence of observation ocean data through theinitialization. Decadal prediction on10-30year time scale is one of the exprement contents of the5th phase of the Coupled Model Inter-comparison Project (CMIP5). However, there are fewstudies related to the results of decadal prediction experiment in our country. Using the output ofdecadal prediction experiment with BCC_CSM1.1participating in CMIP5, this study evaluatedthe model’s prediction capability in regional and global surface temperatures on decadaltime-scale. The model’s dependences on the initial observed states of ocean is explored throughthe comparison between decadal experiment and historical experiment in20th century.Simulation ability of tropical sea surface temperature (SST) and ENSO is further evaluated in thebias correction of the model output of decadal prediction experiments. The main conclusions areas following:The decadal prediction experiment can obviously reduce the global warming trend simulatedby BCC_CSM1.1that under the condition of oceanic initialization, which is closer to theobservation compared to the historical experiment that without oceanic initialization. This featureis much more remarkable in the area between50°S and50°N where the observation data areabundant. Since the ‘nudging’ method is used to initialize the model with the SODA temperaturedata before the prediction date, the simulation of ocean and land mean surface temperature isclosed to the observation in the first year of prediction. After a period of2-7years, the surfacetemperature is restored gradually to the model simulation state from the observation state.1. On the simulation of regional scale surface temperature, the decadal predictionexperiment has a better ability than the tranditional method to a certain degtree. From the view of10-year mean climate anomalies, the high correlations with the CRU observations are mainly onthe tropical west pacific and the tropical Atlantic in the North Hemisphere, and the middle-andhigh-latitude Indian Ocean in the South Hemisphere. Further, analysis shows that the variation of10-year mean SST in decadal prediction experinment is closely correlated with the surface heatflux. In the tropical and subtropical region, the net long wave radiation and sensible heating fluxhas larger influence on the10-year mean SST variation than the net short wave radiation and thelatent heating flux. But in oceans at higher latitude, the variation of decadal mean SST is mostlydetermined by the latent heating flux. 2. Bias correction of the decadal-predictive SST can improve the predictive skill. Thecorrelation analysis of360monthly series between decadal-predictive SST and SODA shows thatthe high-predictvive area depends on the initialization time. In the decadal experiments startsbefore1985, there is a high predictive skill on the west Pacific. The decadal experimentsinitialized in1985,1990and1995have higher predictive skill in the middle and east tropicalPacific. After the bias correction of SST in the decadal experiments, the predictive skill isimproved in modest area in the ocean, especially in the middle and east Pacific.3. BCC_CSM1.1model is poor in the simulation of ENSO. After the bias correction of theSST in the decadal experiments, the simulated spatial mode of ENSO is improved. Waveletanalysis of Nino3.4index reflecting that the model can only capture the obviously period featuresin the observation. But the amplitude is weaker in the simulation and the2-year period is toostrong. Bias correction can promote the4-year period cycle of ENSO closely to the observation.After the bias correction of SST, the correclation between Nino3.4index and SST is closer to theobservation. The time series of1961-2007monthly Nino3.4index composite from10cases ofdecadal hindcast experiments has high correlation (0.44) with the observations. The El Ninoevents possibly occur in2012-2013,2017-2018, and2023.