Visiting Scientist

Planet•San Francisco, CA

2d•Remote

About The Position

We are seeking a highly motivated Visiting Scientist (Postdoctoral Researcher) to join our AI Research (AIR) team for a one-year residency. In this role, you will work directly with Dr. Mirela Tulbure during her sabbatical at Planet to develop our proprietary geospatial foundation models (GFMs). While Planet has historically leveraged external models, we are now focused on building in-house models specifically trained on our unique imagery. As a postdoctoral researcher, you will be the primary technical engine behind creating temporally dense embeddings that capture the dynamic and ephemeral nature of our planet—such as rapid flooding and disaster impacts. You will collaborate with "Planeteers" across data pipelines and analytics to bridge the gap between academic research and operational AI/ML solutions.

Requirements

A recently completed PhD in Geospatial Analytics, Computer Science, Remote Sensing, or a related field.
Demonstrated experience in building AI-based models for environmental change or satellite image analysis.
Hands-on experience with foundation models, contrastive learning, and deep learning frameworks (PyTorch/TensorFlow).
Expert-level Python skills and proficiency with the geospatial scientific stack (e.g., xarray, Dask, Rasterio, GeoPandas).
Experience building automated pipelines for preprocessing and labeling planetary-scale datasets.
Experience working within a research lab environment and a strong desire to apply academic rigor to industry challenges.

Nice To Haves

Prior research in flood-extent mapping, water dynamics, or disaster response.
Direct experience fine-tuning or modifying specific GFM architectures like TerraMind, Prithvi, or Clay.
Proven ability to work with a variety of sensors including PlanetScope, Landsat, and Sentinel-1/2.
A history of developing "human-in-the-loop" workflows or active learning strategies for labeling time-sensitive data.

Responsibilities

Contribute to the design and training of a foundation model specifically optimized for Planet imagery, focusing on the integration of time-series data.
Execute the systematic evaluation of existing GFM architectures (e.g., TerraMind, Prithvi, Clay) against PlanetScope data to identify performance bottlenecks and transferability.
Build and test workflows for detecting short-lived events, such as floods and fires, using high-cadence embeddings.
Develop methods to integrate PlanetScope with Sentinel-1 SAR and other commercial datasets to maintain time-series continuity under cloud cover.
Work closely with Planet’s research scientists to transition experimental prototypes into scalable, operational products.
Co-author findings for publication in top-tier journals and present research at leading conferences like IGARSS or CVPR.