The NASA GMAO seasonal forecasts are currently produced with the Goddard Earth Observing System (GEOS) Atmosphere-Ocean General Circulation Model and Data Assimilation System Version 3 (GEOS-S2S-3). The new system replaces GEOS-S2S-2 described in Molod et al., 2020 and includes upgrades to many components of the system.

The atmospheric model includes an upgrade from what at the start of development was a GMAO 'forward processing' version of GEOS. The cubed sphere grid configuration is at approximately 1/2 degree resolution with 72 hybrid eta-pressure levels. The important atmospheric model developments are related to the dynamical core (Putman et al., 2011), the moist physics (“two-moment microphysics” of Barahona et al., 2014) and the cryosphere (Cullather et al., 2014). As in the previous GMAO S2S system, the land model is that of Koster et al (2000).
GMAO’s GEOS-S2S-3 (as did GEOS-S2S-3) also includes the Goddard Chemistry Aerosol Radiation and Transport model (GOCART, Colarco et al., 2010) single moment interactive aerosol model that includes prediction of dust, sea salt and several species of carbon and sulfate. The ocean model includes an upgrade in the MOM5 (Griffies 2012) tripolar grid resolution from approximately 1/2 degree resolution in the tropics to an approximately 1/4 degree resolution, and an upgrade in vertical resolution from 40 to 50 vertical levels. As in GEOS-S2S-2, the sea ice model is from the Los Alamos Sea Ice model (CICE4, Hunke and Lipscomb 2010).

The Ocean Data Assimilation System (ODAS), as was the case for GEOS-S2S-2, uses the LETKF (Local Ensemble Transform Kalman Filter, Penny, 2014), and assimilates along-track altimetry,a standard complement of in situ data, and satellite sea surface salinity (SSS) as available (Hackert et al., 2020, 2024). The ODAS also assimilates MERRA-2 SST and sea ice boundary conditions. The “one-way” coupled assimilation includes a “replay” (Orbe et al., 2017 ; Takacs et al., 2018) to the GMAO's GEOS for instrument teams (GEOS_IT) atmospheric fields and an ODAS, all in the context of a coupled model. The atmospheric and oceanic fields for the now seamless forecasts are initialized from the coupled assimilation using the newly developed ensemble perturbation strategy called the Synchronized Multiple Time-lagged (SMT) straategy (Schubert et al., 2019, Borovikov et al., 2026), with an ensemble size that has increased from 10 to 40 members per month.

Seasonal and subseasonal forecasts are submitted to many national and international multimodel comparisons (IRI, APCN, SubC, AR forecast project, for example), and are part of IRI's seasonal outlook for the nation. A large suite of retrospective forecasts have been completed, and are used for forecast calibration and anomaly calculation (the calculation of the model’s baseline climatology and drift, anomalies from which are the basis of the seasonal forecasts) as well as for studies of predictability. Data collections from GMAO’s GEOS-S2S-3 are provided on a regular 0.5°×0.5° longitude-by-latitude grid with 720 points in the longitudinal direction and 361 points in the latitudinal direction.