Abstract 093

Abstract ID: 093

How Can Quasi-Periodic Signals Privilege S2S Operational Forecast? From a Perspective of Deep Learning

Lead Author: Masashi Sumitomo
JMA, Japan

Keywords: JMA/MRI-CPS3, CPS3, Coupled model, Ensemble Prediction System, S2S data provider

Abstract: In the first quarter of 2023, the Japan Meteorological Agency (JMA) has changed its product for S2S Project from the ensemble prediction system using an atmospheric model (GEPS: Global Ensemble Prediction System) to that using the atmosphere/ocean/land/sea-ice coupled model (CPS: Coupled Prediction System).
The latest version of CPS is JMA/MRI-CPS3 (hereafter “CPS3”) which has been used for the operational seasonal forecasting in JMA since February 2022. CPS3 consists of an atmosphere model of approximately 55 km and ocean model of 0.25 degrees. Although the atmospheric resolution of CPS3 is lower than that of GEPS, CPS3 has some advantages for sub-seasonal and seasonal forecasting, because CPS3 uses own physical schemes optimized for sub-seasonal and seasonal forecasting, such as convection, cloud, land surface and sea-ice. In addition, CPS3 introduces 3D-VAR sea-ice and 4D-VAR ocean data assimilation for initialization. CPS3 also has increased the frequency of operation. CPS3 produces five ensemble members every day, while GEPS produces fifty ensemble members once a week. Therefore, S2S researchers can choose more optimized ensemble structure by the lagged average forecasting method as they like.
We found that the CPS3 outperforms the GEPS in terms of the general S2S forecast skill. Atmosphere-ocean coupled phenomena such as Madden Julian Oscillation are important to improve sub-seasonal prediction skills. The verification of operational CPS3 forecast indicated better prediction skills of the tropical atmospheric variability than GEPS such as 200-hPa velocity potential especially for two-week or longer lead time. For the forecast skill of 500-hPa geopotential height in the extratropics, CPS3 presented a slightly lower forecast skill than GEPS for two-week forecasting and comparable for three to four week forecasting. These results promise the advantage of a coupled model for the S2S forecast.
We hope that the new JMA product based on CPS is more useful for S2S researchers.

Yutaro Kubo (Japan Meteorological Agency)
Takashi Yamada (Japan Meteorological Agency)