A DRP‐4DVar‐Based Coupled Data Assimilation System With a Simplified Off‐Line Localization Technique for Decadal Predictions. (21st April 2020)

• Record Type: Journal Article
• Authors: He, Yujun; Wang, Bin; Liu, Li; Huang, Wenyu; Xu, Shiming; Liu, Juanjuan; Wang, Yong; Li, Lijuan; Huang, Xiaomeng; Peng, Yiran; Lin, Yanluan; Yu, Yongqiang
• Is Part Of: Journal of advances in modeling earth systems. Volume 12:Number 4(2020)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2020-04-21
• Subjects: Geological modeling -- Periodicals; Climatology -- Periodicals; Geochemical modeling -- Periodicals; 551.5011
• DOI: 10.1029/2019MS001768
• Languages: English
• ISSNs: 1942-2466

Abstract

Abstract: A new weakly coupled data assimilation (CDA) system based on the dimension‐reduced projection four‐dimensional variational data assimilation (DRP‐4DVar) with a simplified off‐line localization technique and a fully coupled model, i.e., the Grid‐point Version 2 of Flexible Global Ocean‐Atmosphere‐Land System Model (FGOALS‐g2), was developed for the initialization of decadal predictions. A 1‐month assimilation window was adopted for the CDA system, in which monthly mean temperature and salinity analyses were assimilated along the trajectory of the coupled model during the initialization for the period of 1945–2006. The system is efficient because the 62‐year initialization only takes about 2.375 times of the time cost of the uninitialized run for the same period. Compared with the uninitialized simulation and the initialization without localization, ocean temperature and salinity, sea surface elevation, surface air temperature, and precipitation are in better agreement with the verification data. Furthermore, climate variabilities in the Pacific and Atlantic regions such as El Niño‐Southern Oscillation, Pacific Decadal Oscillation (PDO), and Atlantic Multidecadal Oscillation (AMO) are more realistically captured. Starting from the initial conditions (ICs) generated by the initialization, 10‐member ensemble decadal prediction experiments were conducted each year from 1961 to 1996. The results demonstrate that higher decadal prediction skills of surface air temperature anomalies averaged over the globe, ocean, land, and the North Pacific subpolar gyre are achieved than those obtained from persistence, the uninitialized simulation, and the prediction initialized from the ICs without localization. Besides, PDO and AMO indices exhibit significant correlation skills in most lead times. Plain Language Summary: The decadal prediction is important for medium‐ and long‐term planning in various fields, e.g., infrastructure, agriculture, fishery, electric power, and tourism. It is usually made by a climate model which includes the ocean, atmosphere, land, and sea ice components. The decadal prediction is initialized from a close‐to‐observed initial state which consistently combines observations and model simulations using a kind of methods called "data assimilation (DA)." However, DA methods for initializing decadal predictions currently around the world are still at an exploring stage, which leads to the inconsistency between the model and the initial state, possibly degrading the prediction skill. This study focuses on proposing a novel DA method for initializing decadal predictions, which is time‐saving and produces initial states consistent with the model. This method is able to well reproduce the multiyear mean ocean temperature and salinity, sea surface elevation, surface air temperature, and precipitation accurately. It also accurately represents the notable climate oscillations in the Pacific and Atlantic. The decadal prediction initialized by the new method shows high skills in presenting surface air temperature and the climate oscillations in the Pacific and Atlantic. Key Points: An efficient DRP‐4DVar‐based coupled data assimilation system with a simplified off‐line localization technique is developed Climatological states and climate variabilities are well captured by the new system Higher prediction skills of surface air temperature, PDO, and AMO are achieved than those of the predictions without localization