PhD Positions as part of the SNSF Project “From Alps to Arctic: Satellite-based Assessment of Forest Canopy Height across Decades”

Within the SNSF project, the EcoVision Lab will focus on advancing forest parameter estimation, particularly canopy height, at the most detailed level. As a PhD candidate, you will develop novel deep learning and computer vision methods to transform large-scale remote sensing imagery of different satellite missions to maps of canopy height, and further forest parameters and their change over time. Your research will include: Developing deep learning models for satellite image time-series analysis and domain adaption, Developing deep learning models for (guided) super-resolution of historical satellite imagery, Producing calibrated uncertainty estimates for all model outputs, Training models on heterogeneous data sources (e.g., Landsat, Sentinel-2, SPOT, Corona) and exploring multimodal combinations of different data sources. The project offers significant freedom to explore impactful methodological directions in modern AI, including: self-supervised learning, multimodal learning, (guided) super-resolution, uncertainty estimation, time-series regression. We aim for high-impact publications both in machine learning venues (e.g., CVPR, ICCV, ECCV, ICLR, NeurIPS) and leading interdisciplinary journals such as Remote Sensing of Environment, ISPRS Journal, and Nature Sustainability.

The University of Zurich, Switzerland's largest university, offers a range of attractive positions in various subject areas and professional fields. With around 10,000 employees and currently 12 professional apprenticeship streams the University offers an inspiring working environment on cutting-edge research and top-class education. These 2x PhD positions offer: Become part of the EcoVision Lab, a vibrant, exciting, fun place to do research on deep learning for applications to ecology, Close collaborations with leading research groups in machine learning, computer vision, data science, remote sensing, and historical remote sensing image interpretation, A unique opportunity to combine cutting-edge AI research with real-world environmental impact for a yet completely under-explored research topic, Access to diverse, large-scale historical satellite image archives.

• An excellent Master's degree (M.Sc. or equivalent) in Computer Science, Machine Learning, Data Science, or a closely related field (e.g., Electrical Engineering, Applied Mathematics)

• A strong foundation in mathematics and machine learning

• A lot of programming experience, preferably in Python

• Strong prior experience in deep learning and computer vision

• Interest in applying advanced ML methods to ecological and geospatial data

• Fluency in English (written and spoken) is required

• Experience with topics such as self-supervised learning, domain adaption, transfer learning, multimodal learning, uncertainty estimation is a plus - but not strictly required.