Development of a Coupled Data Assimilation System in the Fully Coupled Model and Its Implications for Seamless Prediction

Abstract

A coupled model considering all atmosphere component is used to extend forecasting, especially for Subseasonal-to-seasonal (S2S) prediction. The coupled model consists of an independent earth system component with a coupler. Thus, independent uncoupled initialization has to accompany imbalance on their interface, and potentially it can reduce forecasting skill in early forecasting time, as known as initialization shock. Coupled data assimilation (CDA) system has been developed to reduce initialization shock to improve forecasting skills. It is expected to attribute it to seamless prediction, which conducts forecasting in a unified earth system model for consistency in all spatial and temporal scales. However, the development of CDA is still a challenging field in a fully coupled model. In this study, a new CDA system is developed on the Global Seasonal Forecasting System Version 5 (GloSea5), a state-of-the-art, fully coupled model. In this system, atmosphere analysis and ocean analysis, including sea ice, are assimilated to coupled model backgrounds using the incremental analysis update (IAU) method. The main objective of this study is (1) intercomparison of coupled reanalysis to existing reanalyses, (2) evaluation of improvement in forecasting skills, and (3) suggests a scientific understanding of the impact of initialization shock so that this study attribute to achieve seamless prediction overcome S2S prediction gap, suggesting implications of CDA to S2S prediction. Developed CDA can provide a reasonable quality of reanalysis dataset compared to existing reanalyses. Further, reanalysis of GloSea5 shows a better representation of probability densities of precipitation intensity on tropics and subseasonal variability of precipitation in zonal wavenumber-frequency space. This improvement of tropical precipitation indicates a potential improvement of reanalysis for atmosphere-ocean coupling phenomena. Initialization with CDA can effectively reduce imbalance shown in uncoupled initialization of GloSea5 with an improvement of forecasting skill. Compared to the Korea Meteorological Administration (KMA) seasonal forecasting system, forecasting skills are significantly improved up to 10 days in the tropics. Moreover, this study suggests evidence for improving S2S prediction, such as Madden-Julian Oscillation (MJO) and El Nino-Southern Oscillation (ENSO) prediction. Moreover, the physical and dynamical mechanism of initialization shock is intensively analyzed. Regardless of initialization, forecasting bias is converged to typical systematic bias. However, the initialization shock occurs in both temperature and moisture fields and affects forecasting skills in early forecasting time. Compared to temperature and moisture initialization shock, initialization shock in moisture has a much longer timescale to modulate convective system, originating from atmosphere stability in initial states. Further, initialization shock in moisture can reduce MJO forecasting skill as measured in Real-time Multivariate MJO Index (RMM) and eastward propagation.