JMA routinely operates Ensemble Prediction Systems (EPSs) to support from medium-range to seasonal prediction. In JMA, the Global Ensemble Prediction System (Global EPS; GEPS) is operated to support for issuing typhoon information as well as one-week, two-week and one-month forecasts, while the Seasonal Ensemble Prediction System (Seasonal EPS) is operated to support issuing three-month and warm/cold season forecasts as well as El Niño monitoring and outlook.
The numerical prediction model used in the Global EPS is an atmospheric general circulation model (AGCM) that is a low-resolution version of the Global Spectral Model (JMA-GSM) used for global deterministic numerical weather prediction. The numerical prediction model used in the Seasonal EPS is a coupled ocean-atmosphere general circulation model (CGCM), which consists of the AGCM (based on a low-resolution version of the GSM) and the ocean general circulation model (OGCM).
The specifications of the Global EPS for one-month prediction and the Seasonal EPS are shown in the following table.
| Global EPS for one-month prediction | Seasonal EPS (JMA/MRI-CPS3) | |
| Last upgrade | March 2025 | February 2022 |
| Model | Atmospheric general circulation model (AGCM) with 2 Tiered SST approach | Coupled ocean-atmosphere general circulation model (JMA/MRI-CGCM3) |
| Atmospheric general circulation model (AGCM) |
JMA-GSM Horizontal resolution: TQ479 (approx. 27 km)(up to 18 days), TQ319 (approx. 40 km)(after 18 days) Vertical levels: 128 levels up to 0.01 hPa |
JMA-GSM with wide-ranging updates for CPS3 Horizontal resolution: TL319 (approx. 55 km) Vertical levels: 100 levels up to 0.01 hPa |
| Oceanic general circulation model (OGCM) |
N/A | MRI.COM Horizontal resolution: 0.25º x 0.25º Vertical levels: 60 levels |
| Integration range | 34 days | up to 7 months |
| Initial conditions | Atmosphere: JMA Global Analysis Ocean: N/A Land: JMA land surface analysis |
Atmosphere: JMA Global Analysis Ocean: MOVE/MRI.COM-G3 (Low-res. 4DVAR + High-res. downscaling) Sea ice: MOVE/MRI.COM-G3 (3DVAR) Land: Offline model runs (forced by JRA-3Q and JMA Global Analysis) |
| Boundary conditions | Sea Surface Temperature (SST): Prescribed SSTs using persisted anomaly with daily climatological SST up to 6-day forecast. After that, SST is relaxed to the ensemble-mean SST operationally predicted by JMA's coupled Seasonal EPS Sea ice: Prescribed sea ice concentration using persisted anomaly with daily climatological sea ice concentration |
SST: One-tier method Sea ice: Interactive sea ice model |
| Ensemble Perturbation method | Combination of Singular vectors (SVs) and Local Ensemble Transform Kalman Filter (LETKF), Stochastic Perturbation of Physics Tendency (SPPT) and Stochastic Humidity Profile for Convective parametrization (SHPC), SST perturbations and Lagged Average Forecast (LAF) |
Combination of Breeding of Growing Modes (BGM), Stochastic Perturbation of Physics Tendency (SPPT) and Lagged Average Forecast (LAF) |
| Ensemble size | 50 | 51 |
| Freq. of model product creation | Once a week (Model operation) |
Once a month (around 20th of every month) (Model operation) |
| Hindcast period | Jan 1991 - Dec 2020 | Jan 1991 - Dec 2020 |
| Hindcast ensemble size | 13 / initial date Initial dates: 15th and the end of calendar month |
10 / month |