Research Engineer
Research Engineer to Develop collaborative software to produce forest maps at scale
Contexte
The development of satellite imagery and LiDAR, combined with recent progress in AI, are disrupting the way forests are being monitored. While forests are in particular essential for carbon sequestration and biodiversity, they are profoundly affected by climate change. Accurate and granular forest maps are essential for forest managers and public institutions in order to adapt management practices and policies.
The European research project AI4Forest and the national project One Forest Vision have assembled an international team of established researchers in machine learning, remote sensing, forest ecology (Paris Laboratoire des Sciences du Climat et de l’Environnement and Ecole Normale Supérieure, INRAE, IRD, Munster University, Technical University of Munich, Berlin University) in order to produce accurate and periodically updated maps at global scale for forest structure (height), biomass carbon stocks, and activity data related to forest loss and gains (disturbances, including degradation, fires, clearcut) using cutting edge artificial intelligence models driven by satellite and field observations. The Paris team is collaborating with public institutions such as Office National des Forêts (ONF), Institut Géographique National (IGN).
The Paris team is looking for an experienced and motivated Research Engineer who will actively contribute to the production and validation of forest maps and the associated academic publications, with a focus on code scalability, reusability, and reproducibility.
Description
Missions
The missions will cover the full stack of AI product development, including data integration, model design, fitting, inference.
Simplify and automate access and processing of satellite imagery and LiDAR by consolidating existing community tools for a team of ten researchers and national and international collaborators in research projects (Sentinel 1&2, Gedi, LiDAR HD, Spot).
Implement tools for pre-processing and inclusion of input data sources from multiple spaceborne and airborne platforms as input or validation of AI models (Alsar, Nisar, Icesat 2, orthophotos, airborne LiDAR).
Lead the development of the common pipeline to fit AI models on space/aerial data, with a focus on scalability and reusability inside the group. Minimal requirements include:
- Usability on a high performance computing environment (SLURM on national HPC facility Jean-Zay) Testing environment.
- Versioning with a minimalist continuous integration (code does not need to be production ready, but must be readily usable, while keeping a fast iterations pace)
- Experiments’ tracking and reproducibility.
- Assist researchers in using common tools in order to produce maps and analysis.
- Promote and diffuse the use of coding good practices inside the group.
- Maintain core tools of the group.
- Creation on new maps, evaluation and diffusion on institutional platforms (Data Terra and Theia).
- Design and implementation of new methods with researchers (data fusion, models…).
Location
Laboratoire des Sciences du Climat et de l’Environnement (Saclay, in the Orme des Merisiers green area). Remote friendly.
LSCE is a world-class research laboratory established and a collaboration between CEA, CNRS and the University of Versailles Saint-Quentin (UVSQ). The LSCE hosts approximately 300 researchers, engineers and administrative staff including many PhD and master’s students. This project will provide the employee with the opportunity to work directly on advanced methods with researchers from the LSCE and other institutions.
Contract duration
24 months, with an extension possible.
Starting date
The position is available from March 1st 2024 and will remain open until filled. The expected start of the position is May-June 2024.
Salary
Competitive salary, full social and health benefits, commensurate with work experience.
Supervision
Main supervisor: Philippe Ciais, Research director at LSCE
Co-supervisors: Ibrahim Fayad (LSCE), Fajwel Fogel (ENS), Martin Schwartz (LSCE)
How to apply
Applicants should submit a complete application package by email to the contacts below. The application package should include (1) a curriculum vitae including e.g. important recent publications / projects, (2) statement of motivation (3) answers to the selection criteria above (4) names, addresses, phone numbers, and email addresses of at least two references.
Contacts
Philippe Ciais • et Fajwel Fogel •
References
Example of height maps before/after fires at Teste-de-Buch from Schwartz, Martin, et al. FORMS: Forest Multiple Source height, wood volume, and biomass maps in France at 10 to 30 m resolution based on Sentinel-1, Sentinel-2, and Global Ecosystem Dynamics Investigation (GEDI) data with a deep learning approach. Earth System Science Data 15.11 (2023): 4927-4945.
Compétences requises
- You have a masters degree in Remote Sensing / Machine Learning / Computer Science or any other relevant field.
- PhD is a plus but not a requirement. You will contribute to scientific publications, but your main focus will be to deliver tools and products (e.g., maps).
- You have experience in building ML pipelines for production use (in Python) and are fond of MLOps.
- You have experience in training and testing large deep learning models, preferably on vision tasks.
- You have experience in remote sensing or are motivated to develop technical skills in that field.
- Working with tens of terabytes of data is not a problem for you.
- You like problem solving, you are autonomous, but want to work in a collaborative environment.
- You are able to lead and structure implementation projects.
- You are curious, enjoy learning and a research environment.
- You are able to communicate efficiently with the team on their needs and the projects you’re involved in.
- You want to work on problems that can benefit the environment.
