Abstract's details
The quasi-operational 4D-Var ocean data assimilation/prediction system for the western North Pacific at JMA
CoAuthors
Event: 2017 Ocean Surface Topography Science Team Meeting
Session: Application development for Operations
Presentation type: Type Poster
Contribution: PDF file
Abstract:
Japan Meteorological Agency (JMA) has a plan to introduce a new coastal ocean assimilation/prediction system (MOVE/MRI.COM-JPN (referred here after as MOVE-JPN)) in 2020 after an update of super-computer systems of JMA. In MOVE-JPN, developed by Meteorological Research Institute (MRI/JMA), a high-resolution (2 km) prediction model covers whole Japan coast and a 4D-Var assimilation system (MOVE-4DVAR) covers the North Pacific with an eddy-resolving (10 km) model. As a prototype of MOVE-JPN, MRI/JMA also developed a coastal prediction system for a limited area (MOVE-Seto) to be able to calculate with fewer computer resources. It consists of a MOVE-4DVAR covering the western North Pacific and a 2 km model covering western part of Japan around the Seto Inland Sea. This system, MOVE-Seto, has been quasi-operational since June 2016 at JMA. Its output is used as reference for the development of the next operational system, MOVE-JPN.
Our presentation focuses on MOVE-4DVAR that consists in both MOVE-JPN and MOVE-Seto. Usui et al. (2017) has already detailed MOVE-4DVAR and Kuragano et al. (2016) presented the western North Pacific reanalysis for 30 years (FORA-WNP30), a dataset generated by MOVE-4DVAR, in OSTST 2016. We present some modifications made to suit operational requirements. In the quasi-operational mode, 10-days assimilation and subsequent 11-days prediction are executed in a daily basis. This leads to usage of the latest observation data in the analysis. The near-real time observational data such as satellite sea level anomalies (SLA), in-situ temperature and salinity profiles, and analyzed SST data (prompt analysis of MGDSST (Merged satellite and in situ Global Daily Sea Surface Temperature)) are assimilated. The satellite SLA observations are the along-track data of Jason-3, SARAL/AltiKa, Cryosat-2, which are produced and distributed by the Copernicus Marine and Environment Monitoring Service (CMEMS). In-situ observation of temperature and salinity are obtained via GTS, e-mail and facsimile. Another main difference is an external forcing. The quasi-operation system adopts the latest NWP model (GSM: 20km, 3 hourly data), otherwise reanalysis (FOR A-WNP30) adopts the Japanese 55-year Reanalysis (JRA-55: 55km, 6 hourly data).
The validation results against in-situ data show that Root Mean Square Error of 100m depth temperatures for MOVE-4DVAR is significantly reduced in the region with large temperature variability such as the Japan Sea, Oyashio, Kuroshio, and Kuroshio Extension region compared to MOVE-3DVAR (the current operational 3D-var ocean assimilation system). However, positive bias is generally seen in the both systems (MOVE-4DVAR and MOVE-3DVAR) and the bias of MOVE-4DVAR is somewhat larger south of 25°N.
We speculate that the positive biases of both systems are attributed to the method for SLA assimilation. The low frequency variations of observed SLA are mainly caused by 1) steric sea level change (caused by the temperature and salinity change of water column), 2) ocean water mass variation by net surface water flux and 3) ocean water mass variation by wind stress. The cost function includes a term of difference between steric height anomalies estimated from temperatures and salinity in the model and SLA observation, therefore 2) and 3) components of observed SLA possibly contribute to analysis errors or biases of subsurface temperatures and salinity. We will show a simple method for treating 2) and 3) components and preliminary result of the corrected assimilation.
Our presentation focuses on MOVE-4DVAR that consists in both MOVE-JPN and MOVE-Seto. Usui et al. (2017) has already detailed MOVE-4DVAR and Kuragano et al. (2016) presented the western North Pacific reanalysis for 30 years (FORA-WNP30), a dataset generated by MOVE-4DVAR, in OSTST 2016. We present some modifications made to suit operational requirements. In the quasi-operational mode, 10-days assimilation and subsequent 11-days prediction are executed in a daily basis. This leads to usage of the latest observation data in the analysis. The near-real time observational data such as satellite sea level anomalies (SLA), in-situ temperature and salinity profiles, and analyzed SST data (prompt analysis of MGDSST (Merged satellite and in situ Global Daily Sea Surface Temperature)) are assimilated. The satellite SLA observations are the along-track data of Jason-3, SARAL/AltiKa, Cryosat-2, which are produced and distributed by the Copernicus Marine and Environment Monitoring Service (CMEMS). In-situ observation of temperature and salinity are obtained via GTS, e-mail and facsimile. Another main difference is an external forcing. The quasi-operation system adopts the latest NWP model (GSM: 20km, 3 hourly data), otherwise reanalysis (FOR A-WNP30) adopts the Japanese 55-year Reanalysis (JRA-55: 55km, 6 hourly data).
The validation results against in-situ data show that Root Mean Square Error of 100m depth temperatures for MOVE-4DVAR is significantly reduced in the region with large temperature variability such as the Japan Sea, Oyashio, Kuroshio, and Kuroshio Extension region compared to MOVE-3DVAR (the current operational 3D-var ocean assimilation system). However, positive bias is generally seen in the both systems (MOVE-4DVAR and MOVE-3DVAR) and the bias of MOVE-4DVAR is somewhat larger south of 25°N.
We speculate that the positive biases of both systems are attributed to the method for SLA assimilation. The low frequency variations of observed SLA are mainly caused by 1) steric sea level change (caused by the temperature and salinity change of water column), 2) ocean water mass variation by net surface water flux and 3) ocean water mass variation by wind stress. The cost function includes a term of difference between steric height anomalies estimated from temperatures and salinity in the model and SLA observation, therefore 2) and 3) components of observed SLA possibly contribute to analysis errors or biases of subsurface temperatures and salinity. We will show a simple method for treating 2) and 3) components and preliminary result of the corrected assimilation.