Weakly Supervised EO ML Scientist for Climate Impacts

Research Institutes of Sweden; Göteborgs kommun; On-site; SEK 420,000 - 560,000; Full time

Benefits offered by this job

Global research environment

Industry collaboration

PhD stipend

Job summary

Research Institutes of Sweden (RISE) invites applications for a postdoctoral researcher to advance an Earth observation ML framework that detects, localizes, and quantifies climate-extreme impacts from satellite sequences. The project emphasizes weak supervision, data efficiency, and robust uncertainty estimates to guide decision-makers.

The role focuses on architectural modules atop frozen geospatial foundation models, addressing distribution shifts while leveraging multi-sensor data to

Qualifications

• PhD in Computer Science, ML, Data Science, Applied Mathematics, Physics or highly quantitative equivalent.
• Strong theoretical and hands-on coding mastery of deep learning frameworks (PyTorch or TensorFlow), with experience in CV or spatio-temporal modeling.
• Excellent English communication (written and spoken) required.

Responsibilities

• Develop data-efficient ML models for geospatial change detection using multi-sensor sequences (Sentinel-1/2, Landsat/HLS, VIIRS/MODIS).
• Incorporate uncertainty quantification and distribution-shift handling (e.g., SNGP) into model design.
• Collaborate with cross-disciplinary teams to translate research into actionable climate impact quantification.

Skills

Deep learning

PyTorch

TensorFlow

Remote sensing

Geospatial ML

English communication

Education

PhD in Computer Science / ML / Data Science

Tools

Python

PyTorch

TensorFlow

Geospatial libraries

Job description

Research Institutes of Sweden (RISE) invites applications for a postdoctoral researcher to advance an Earth observation ML framework that detects, localizes, and quantifies climate-extreme impacts from satellite sequences. The project emphasizes weak supervision, data efficiency, and robust uncertainty estimates to guide decision-makers.

The role focuses on architectural modules atop frozen geospatial foundation models, addressing distribution shifts while leveraging multi-sensor data to