The French convergence institute CLAND (Climate change and land use) and its partner research unit ECOSYS are looking for a motivated candidate for a 12-month postdoctoral position on the mapping of agricultural organic nitrogen sources across France, at the interface between remote sensing, deep learning and agri-environmental sciences.
Scientific background
Organic fertilisation from livestock manure is one of the main sources of diffuse nitrogen in France, with roughly 300 million tonnes of manure produced each year. It is also one of the practices promoted to increase carbon storage in agricultural soils. Any spatially explicit mapping of organic fertilisation pressure first requires an accurate localisation of the sources — livestock buildings and facilities — combined with a quantitative estimate of the manure produced at each site.
The French national livestock identification database (BDNI), updated quarterly and coupled with the Land Parcel Identification System (RPG), already provides the location of farms and their animal numbers, but says nothing about the physical structures housing the animals and storing the manure. Recent work has demonstrated that agricultural facilities can be detected from satellite imagery using deep learning (de Senneville et al., 2025), an approach that can be transferred to livestock buildings. The BDNIÅ~RPG coupling can therefore serve as a georeferenced training set.
The project further aims to add a temporal dimension: the cattle production cycle — alternating between winter housing and summer grazing — strongly changes where and when nitrogen is excreted. The ultimate goal is to identify spatial concentrations (hotspots) and periods of peak production (hotmoments) of reactive nitrogen, and to assess how these match areas of potential NH₃ emission.
This spatially explicit quantification of the available organic residual products (ORP) also opens a second line of enquiry: the carbon storage potential of agricultural soils. The localised manure volumes determine where, when and in what quantities organic inputs can be applied to fields. The resulting maps will therefore make it possible to weigh the expected benefit for soil carbon storage against the associated nitrogen losses, and to inform tradeoffs between these two objectives at the territorial scale.
Tasks
Build a georeferenced training set from the BDNIÅ~RPG coupling, focusing on the cattle sector (the best covered by the BDNI), and use it to train a deep-learning object detector identifying cattle buildings and their associated structures;
Develop and validate a model linking the characteristics of the detected structures (built area, number of buildings, annexes) to the likely herd size expressed in livestock units (LU);
Reconstruct annual herd dynamics per site and approximate the grazing/housing cycle using a satellite proxy;
Produce a spatio-temporal map of organic manure production and identify hotspots and hotmoments;
Disseminate results through publications in peer-reviewed journals and presentations at national and international conferences.
Profile
PhD in remote sensing, geomatics, data science, agronomy, environmental science or a related field (obtained, or to be obtained by the starting date);
Solid practice of geospatial data processing and GIS;
Familiarity with livestock systems, nitrogen biogeochemical cycles or French agricultural databases (BDNI, RPG) is an appreciated asset;
Documented experience in scientific writing and publication in peer-reviewed journals;
Autonomy, a taste for interdisciplinary work and the ambition to pursue a scientific career;
Good communication skills in English; knowledge of French is an asset but is not required.
How to apply
Applications should be sent by email to Raia Silvia Massad (
[email protected]), Bertrand Guenet (
[email protected]) and Marco Carozzi (
[email protected]) before 15 September 2026.
Your application should include a motivation letter, a detailed academic CV (with publication list) and the contact details of two referees.
The selection committee will review applications as soon as the call closes; shortlisted candidates will be invited for an interview.
You will be based at the UMR ECOSYS, on the AgroParisTech campus (22 place de l'Agronomie, 91120 Palaiseau, France).
Reference: de Senneville A. et al. (2025).
Towards Large Scale Geostatistical Methane
Monitoring with Part-based Object Detection. arXiv:2507.18513.