The various types of grassland (pastures and meadows) offer undeniable benefits to tackle the
challenges of agro-ecological transition: reduced inputs, enhanced vegetation, maintenance of
biodiversity, etc. However, they face a number of challenges (climate change, contact with wildlife,
etc.). The project “Artificial Intelligence for the Holistic Efficiency of Rangelands in terms of Breeding
Animals and Grassland Exploitation” aims to develop new algorithms for better grassland
management in terms of ecological factors and herd efficiency.
Our approach is holistic, taking into account aerial data (satellite, drone, etc.), data from animal
sensors (location, weighing, etc.) and field data (plants, invertebrates, etc.). A first objective is to
provide algorithms, mainly based on learning, to complete and merge these various sources of
data. The aim is to increase the potential, and therefore the use, of new technologies in this
agricultural practice. Through existing data, and new data produced during the project, we will
obtain, in addition to these numerical methods, new datasets rich in diverse information.
A second objective is to provide stakeholders (such as breeders, shepherds...) with concrete
decision-support tools. To this end, strong interaction with stakeholders is planned throughout the
project, in order to provide better grazing management tools. Based on the multi-annotated data
formed during the project, we will develop various algorithms and operational methods. These
include, in particular, information-enriched maps, sensor-based early warning systems as well as
best-practice guidelines for landscape and resource management..
Several scientific challenges are tackled by the project, mainly concerning numerical aspects
(mathematical and computational), particularly in the fields of unsupervised learning and
optimization. We are aiming for hybrid approaches that take into account differences in expert
knowledge of the mechanisms a
