Prithvi-EO Foundation Model for Forest Structure, Biomass, and Fire-Relevant Metrics

About the role

The successful candidate will contribute to one or more core research thrusts. One thrust focuses on developing a self-supervised masked autoencoder for GEDI LiDAR data to learn transferable structural embeddings, creating a foundational model for forest structure representation learning. A second thrust centers on multi-modal fusion, designing and training shallow fusion networks that combine the GEDI structural embeddings with spectral–temporal features from an HLS/Landsat foundation model to produce robust, scalable predictors. A third thrust involves metric-specific modeling, either via multi-headed architectures or per-metric models, to estimate a suite of forest structure and fuel-relevant variables. Ground-truth supervision and evaluation will draw on sources such as NSF NEON, Forest Observation System data, and FIA inventory datasets.

Description

We are recruiting a Postdoctoral Fellow to support a NASA-aligned research effort developing continual, improved, high-resolution maps of forest structure and derived metrics relevant to fire and carbon-cycle modeling. The project will build a unified modeling framework that uses GEDI LiDAR and Landsat/HLS data to train deep learning models capable of predicting forest structure variables such as above-ground biomass, with a design that can readily incorporate new inputs (for example SAR) and new outputs (for example fuel categorization). The work explicitly leverages NASA's Foundation Model, Prithvi-EO, to make the system extensible.

Field of Science

Earth Science

Advisors

• huikyo.lee@jpl.nasa.gov
• (626) 864-0557
• Olga.Kalashnikova@jpl.nasa.gov
• (818) 393-0469

Eligibility

• U.S. Citizens;
• U.S. Lawful Permanent Residents (LPR);
• Foreign Nationals eligible for an Exchange Visitor J-1 visa status;
• Applicants for LPR, asylees, or refugees in the U.S. at the time of application with 1) a valid EAD card and 2) I-485 or I-589 forms in pending status

Questions about this opportunity?

Please email npp@orau.org

Qualifications

• Strong experience in deep learning;
• Comfort working with NASA's satellite observations;
• The ability to design experiments and evaluate models rigorously;
• Experience with remote sensing modalities such as LiDAR and optical imagery;
• Familiarity with GEDI, Landsat/HLS, biomass estimation, forest fuels, or foundation model/self-supervised methods.

Skills

• Experience with natural-language interaction with geospatial forest products;
• Ability to generate text annotations and question-answer style supervision derived from predicted metrics and domain heuristics;
• Training the projection layer so the system can describe and answer questions about pixel-level conditions using the fusion outputs.

Benefits

• Opportunity to lead peer-reviewed publications aligned with the project’s research components.

Pay

• N/A

Schedule

• N/A