UC Berkeley / Caltech / Stockholm University

Position Title/Short Description

Full Text of JOE Listing:

The Aerial History Project is a multi-institution collaboration that aims to extract long-run data on human development and environmental change from an archive of historical aerial photographs. The archive surveys more than 60 developing countries repeatedly from the 1940s through the 1990s, providing a window back in time to observe large-scale, structural transformations in economic development, population growth, political institutions, and natural resources over the 20th century. Our team is developing a pipeline to digitize the photographs, assemble them into basemaps resembling modern satellite imagery, and extract structured information such as maps of land cover, infrastructure, and population. The project is co-led by Solomon Hsiang at UC Berkeley, Hannah Druckenmiller at Caltech, and Andreas Madestam and Anna Tompsett at Stockholm University.

We seek to hire one or more postdoctoral scholars to help develop state-of-the-art approaches for accurately orthorectifying the aerial photographs and/or extracting structured information from them at scale. Candidates should have research interests and demonstrated AI/ML technical skills in computer vision, satellite/aerial imagery, remote sensing, change detection, or photogrammetry/orthorectification. While the projects will ultimately be co-developed by the postdocs and PIs, possible directions include:

1. Apply existing or develop new models to extract new information from the imagery, such as estimates of human migration, maps of land degradation and abandonment, trade infrastructure, or building height.

2. Apply existing or develop new methods that allow for change detection in repeated aerial surveys that are irregularly spaced over time and imperfectly georeferenced.

3. Develop techniques for better transferring models trained on modern satellite imagery to historical aerial photography when historical ground-truth labels are sparse or unavailable.

4. Improve orthorectification of 2D imagery using camera calibration and stereo pairs

5. Develop end-to-end AI/ML pipelines (including data ingestion, data pre-processing, modeling, evaluation, and deployment) on cloud services such as GCP or AWS.

The appointment is for 2-3 years and could be at UC Berkeley, Caltech, or Stockholm University (location is to be mutually agreed upon by the candidate and PIs). Applications will be reviewed starting on December 1, 2023 and will remain open until the position is filled.

Application requirements:
- Cover letter
- CV
- Sample research paper
- Sample code (e.g. share a GitHub repo)
- Letters of recommendation (2)