A DRP-4DVar-Based Coupled Data Assimilation Method And Its Application In Decadal Predictions

One of the biggest challenges in decadal predictions is the design of a high-quality and time-saving initialization approach.Currently,the initial conditions(ICs)produced by most initialization methods in the world are generally inconsistent with the coupled model or the used methods are expensive in computing.In this dissertation,we build a coupled data assimilation(CDA)system adopting the DRP-4DVar,a kind of 4DEnVar method proposed by the Chinese scientists,and the FGOALS-g2,a coupled climate system model independently developed in China.This system is one of the earliest 4DEnVar-based CDA systems.It uses historical ensembles to estimate the background error covariance matrix and avoids the adjoint model so as to save large computing cost.The ICs are also in good consistency with the coupled model.It is the first time that the CDA system has the ability to assimilate monthly mean ocean temperature and salinity observations directly(i.e.,the assimilation procedure fits the monthly mean observations in one-month windows along the trajectory of the coupled model).Based on the ICs from the CDA system,we conduct a set of decadal prediction experiments and evaluate the contribution of the CDA system on improvement of the decadal prediction skills.The main works and conclusions are as follows:(1)A DRP-4DVar-based coupled data assimilation system using a simplified off-line localization technique is developed and a 62-year initialization experiment is conducted.Compared with the uninitialized experiment,the climatological mean states of assimilated sea surface temperature,sea surface salinity,sea surface height,sea water velocity,surface air temperature,precipitation and sea level pressure,the seasonal cycle of sea surface temperature,and the climate variabilities such as the ENSO,Pacific Decadal Oscillation(PDO)and Atlantic Multidecadal Oscillation(AMO)are improved.It can provide the ICs that are closer to observations and more consistent with the coupled model for hindcasts and forecasts.(2)Based on the ICs from the aforementioned CDA system,we accomplished the hindcast and forecast experiments initiating from 1961 to 2006 every 5 years.During predictions,the initial shock of surface air temperature anomalies is reduced greatly and is one of the smallest compared with the decadal predictions from the fifth phase of the Coupled Model Intercomparison Project(CMIP5).The decadal hindcast skills of global,ocean and land mean surface air temperature anomalies are better than the twentieth century simulation.(3)A DRP-4DVar-based CDA system using a fast on-line localization technique is developed and a 62-year initialization experiment is carried out.The assimilated sea surface temperature,sea surface salinity,surface air temperature,ENSO,PDO and AMO are closer to observations than those from the previous CDA system using a simplified off-line localization technique.The ICs are more accurate and include more observational variabilities.(4)Based on the ICs from the CDA system developed in Part(3),whose climatology was partly(75%)restored to the model climatology to reduce the initial shock,we accomplished the hindcast and forecast experiments.The decadal hindcast skills of sea surface temperature in the North Pacific,PDO,Aleutian Low,Pacific/North American teleconnection(PNA)and AMO are obviously improved.The forecast shows that,in the next decade,PDO will remain in the warm phase.During the period of 2026-2035,it will probably transition to a cold phase.