Research Scientist, Biosphere Models in London

At Google DeepMind, we value diversity of experience, knowledge, backgrounds and perspectives and harness these qualities to create extraordinary impact. We are committed to equal employment opportunity regardless of sex, race, religion or belief, ethnic or national origin, disability, age, citizenship, marital, domestic or civil partnership status, sexual orientation, gender identity, pregnancy, or related condition (including breastfeeding) or any other basis as protected by applicable law. If you have a disability or additional need that requires accommodation, please do not hesitate to let us know.

At Google DeepMind, we've built a unique culture and work environment where long-term ambitious research can flourish. Our team is part of our Sustainability Program, whose aim is to revolutionize environmental sustainability. We conduct fundamental research which develops novel AI methods, and we translate our research advances into real-world applications and products.

This role focuses on modeling and information retrieval for natural environments, specifically forests and habitats. Our team develops approaches to globally map forest characteristics, understand temporal dynamics, and make predictions about the future. Our goal is to support critical global sustainability efforts, such as the EU Regulation on Deforestation-free Products (EUDR) and the 30x30 conservation targets.

We are looking for a creative thinker with exceptional skills in Geospatial AI and a passion for the natural world. This role may involve building a new system using creative thinking while addressing real-world impact needs. You will bridge the gap between AI/computer vision research and Earth science. Previous technical experience with Geospatial AI for pretraining, complex models design, geospatial data processing, and domain expertise in relevant Earth observation problems are valued.

You will join a team that values rigorous evaluation, open collaboration, and innovation. You will have the opportunity to work with massive datasets, leverage Google’s infrastructure, and see your work translate to verifiable impact for nature protection.

• Design, implement, and train state-of-the-art Geospatial AI models (e.g., multi trucking multi-task) on planetary-scale datasets.
• Develop novel approaches for self-supervised or weakly-supervised pretraining to tackle data scarcity in natural environments.
• Build and maintain scalable data pipelines for ingesting and processing heterogeneous Earth Observation data.
• Lead the technical validation of models against ground truth, contributing to the design of validation campaigns and geospatial annotation strategies.
• Collaborate with domain experts to refine model objectives for downstream application domains.
• Report and present research findings clearly and efficiently, leading to open-source code releases and scientific publications.
• Contribute to team collaborations to meet ambitious research and product goals.
• Engage with application and product needs, to inform research and engineering decisions.

We look for the following skills and experience:

• BSc, MSc or PhD degree in Computer Science, Machine Learning, Remote Sensing, Geoinformatics, or a related technical field, or equivalent practical experience.
• Excellent software engineering skills in Python with a proven ability to build robust and scalable systems.
• Proficiency in deep learning frameworks like JAX, TensorFlow, or PyTorch is essential.
• Experience with either large-scale data processing frameworks (e.g., Apache Beam, Spark) or distributed training infrastructure.
• Demonstrable expertise in Geospatial AI (GeoAI) and Earth Observation (EO) data modalities, specifically working with vision models and satellite imagery (multi/hyper-spectral, SAR, or LiDAR).
• Experience processing and analysing Earth Observation data for natural environments (e.g., LCLU mapping, change detection, vegetation dynamics).
• A proven track record of publications in top-tier conferences and/or journals.

In addition, the following would be an advantage:

• Exceptional expertise in developing and applying multi-modal, multi-task machine learning architectures for remote sensing applications.
• Deep domain experience in natural environments, specifically working on geospatial problems such as land cover/land use (LCLU) mapping, change modeling & detection, and vegetation traits estimation.
• Experience developing foundation models, including techniques for self-supervised pretraining or handling label noise (weak supervision).
• Proficiency with geospatial data processing tools and libraries (e.g., Earth Engine, GDAL/Rasterio, GIS software, GeoPandas) and large-scale data processing frameworks.
• Familiarity with the challenges of data curation, such as geospatial annotation design, validation campaign design, and handling sparse, noisy, or geographically biased ground truth data.
• A strong passion for environmental sustainability and using AI to address climate change and biodiversity loss.