About me

Maryam Abdi-Oskouei

GESTAR II Research Scientist, NASA Global Modeling and Assimilation Office (GMAO)

I build satellite data assimilation systems that bring diverse observations into NASA’s GEOS models to improve weather, air quality, and atmospheric composition forecasts. Before joining GMAO in 2025, I spent six years at the Joint Center for Satellite Data Assimilation as a core developer of the JEDI framework. I received my PhD in Civil and Environmental Engineering from the University of Iowa, advised by Prof. Greg Carmichael.

CV (PDF) · Google Scholar · GitHub · LinkedIn · maryam.abdi.oskouei at gmail dot com

Current Work

JEDI-based assimilation for GEOS-CF

I work on transitioning GEOS-CF, NASA’s composition forecasting system, from GSI to JEDI-based data assimilation, and on expanding the range of observations it assimilates. Our JAMES paper (2026) demonstrated JEDI for atmospheric composition applications, using hourly TEMPO NO₂ assimilation as the test case. Ongoing work characterizes TEMPO observation errors and uses JEDI to constrain emissions.

NO2 surface increment

Paper: JAMES (2026)

Swell workflows for GEOS-CF

Swell is GMAO’s workflow system for running JEDI applications. I develop the Swell suites for GEOS-CF: ingesting observations and model backgrounds, computing model-equivalent observations (H(x)), and running assimilation experiments. Data are managed through R2D2. On top of these suites I’m developing a near-real-time monitoring system that compares GEOS-CF forecasts against TEMPO and other observations as they arrive.

I also write training material so others at GMAO can run JEDI through Swell. The HowToJEDI tutorials cover setup, data handling with R2D2, and common failure modes, and I use them in hands-on training sessions.

Repo & Doc: SWELL

LOLA-JEDI: a lat/lon model interface for JEDI

LOLA-JEDI is a lightweight JEDI interface for gridded longitude-latitude model data, built on JEDI’s mist framework. It lets output from any model on a regular lat/lon grid be compared against observations through JEDI’s observation operators, without writing a model-specific JEDI interface. One example application is a CAMS vs. GEOS-CF comparison against TEMPO.

CAMS vs. GEOS-CF in TEMPO observation space

Aerosol assimilation: from PSAS to JEDI 3D-EnVar

I contribute to the transition of GEOS AOD assimilation from the legacy PSAS system to JEDI 3D-EnVar system. This work assimilates VIIRS neural-network-retrieved AOD through an extinction-based observation operator, using a GEOS ensemble-derived background error covariance. The goal is a cycling system that outperforms PSAS at comparable cost. In parallel, I am adapting the Swell workflow to run GEOS aerosol applications.

Selected Publications

Full list on Google Scholar.

Education

Last updated: October 2026