Smart-Biodiv
Smart-Biodiv project : Smart AI technologies for Biodiversity Research
Project overview
Marine environments undergo rapid changes and the monitoring of their ecosystem status becomes critical. Such a monitoring requires gathering data, processing them and extracting indicators summarizing the status of the environment. However, the data in environmental sciences are often sparse and imbalanced, which constitute challenges for AI algorithms.
This leads to the two directions followed in the SMART-BIODIV proposal:
- Harnessing the power of machine learning algorithms to complete and process sparse and imbalanced data that we often encounter in environmental sciences ;
- Designing indicators to qualify the ecological status of the considered environments. We will also exploit the large image databases collected by the partners on marine plankton and make them available to the challenge participants.
Main objectives
- Designing AI methods optimized for biodiversity research:
We will focus here on the development of robust supervised learning methods applied to:
- Quantitative data (occurrences, physico-chemical parameters, etc.) used for regression or classification tasks ;
- Spatio-temporal data (time-series, geolocalized samples) used to model dynamics, interpolate and/or extrapolate parameters over space and time based on forcing parameters ;
- Image data (e.g. of individual plankton) to detect and identify taxa or traits and support occurrence monitoring. In these three AI variants, the challenges of explainability, limited amount of annotations, and incremental learning will be addressed. Additionally, an important focus will be put on the use of noisy/uncertain labels and heavily unbalanced datasets, which are typical of environmental science datasets in our experience.
- Designing models and predictive indicators for biodiversity research:
Our combined expertise with marine ecosystems and bioindication, in particular in freshwater environments, will be brought to bear on the definition of new indicators adapted to the marine ecosystem, using not only taxonomic occurrences but also trait-based approaches. Although there are many ways to recover functional traits, this aspect will specifically benefit from the image processing methods resulting from the previous point.
- Designing hybrid AI models:
Biodiversity research combines a lot of sub-symbolic challenges, which will be mostly addressed when considering of the first point of these objectives, with symbolic priors expressing the networks of trophic relations, competitions, co-occurrences, … between taxa exposed to environmental pressures. By joining forces between biodiversity research and AI research, we intend to use domain knowledge to express, model and formally represent these networks of relationships, then use data-driven methods to identify the parameters of these models, and finally use this combined knowledge as predictive assembly models.
The consortium
IRL
The IRL2958 GT-CNRS is an International Joint Laboratory founded in 2006 by the National Center for Scientific Research (CNRS) and the Georgia Institute of Technology (Georgia Tech) based in Atlanta, USA. This laboratory is the first and only IRL entity located in France. It is based on the campus of Georgia Tech-Lorraine, the European campus of Georgia Tech in Metz. Activities at the Atlanta campus were begun in 2010. The IRL2958 also involves associated laboratories, among them : the Laboratory for the Optical Properties of Materials (LMOPS), jointly operated by CentraleSupélec and the University of Lorraine, the Laboratory of Microstructures and Mechanics of Materials (LEM3), jointly operated by Arts & Metiers Paris Tech and the University of Lorraine, and the Laboratory for MEMS, Photonics, Metrology and Mechanics (FEMTO-ST), associated to the University of Franche-Comté.
The DreamLab team, specialised in robotics for natural environments and computational perception applied, in particular, to environment sciences, will use its expertise in AI (Artificial Intelligence) technologies to develop, adapt and evaluate AI methods for the specific challenges of biodiversity monitoring. Here, the challenge will be to extract useful information from the large and heterogeneous datasets covering various ecosystem components. The team will be leader of WP2 (AI for biodiversity) and WP5 (Project management, challenge-wide activities & Outreach). The team will also be involved in all WPs.
LORIA
Loria, the Lorraine Laboratory for Research in Computer Science and its Applications, is a joint research unit (UMR 7503), shared by several institutions: the CNRS, the University of Lorraine and Inria.
Loria’s mission is fundamental and applied research in computer science, and has been since its creation in 1997. Loria is a member of the Charles Hermite Federation, which brings together the three main research laboratories in mathematics and ICST (information and communication sciences and technologies) in Lorraine. The laboratory is part of the AM2I (Automatic, Mathematics, Computer Science and their Interactions) scientific cluster of the University of Lorraine. The scientific work is carried out by 29 teams organized into 5 departments, 15 of which are shared with Inria, representing a total of over 430 people. Seven transversal axes structure the laboratory around major themes such as computer security, health, E-education, cyber-physical systems, automatic language processing and artificial intelligence, energy and the factory of the future. Loria is one of the largest laboratories in the Lorraine region.
Loria and CentraleSupélec bring their expertise in computer science and more particularly in artificial intelligence applied to the particular theme of biodiversity. The team involved will lead the WP3 on hybrid AI and will also participate in the other WPs. The challenge of WP3 will be to incorporate business knowledge in the form of a relationship graph into predictive models of biodiversity.
LIEC
The Interdisciplinary Laboratory for Continental Environments (LIEC) is a joint research unit (UMR 7360) between the CNRS and the University of Lorraine. It gathers nearly 140 members spread over 3 geographical sites in Metz and Nancy (Eastern France).
The LIEC conducts research in environmental sciences in order to understand the functioning of continental environments strongly impacted by human activity, aiming at their rehabilitation. Within the EcoSe team (stress ecology), LIEC researchers involved in Smart-Biodiv (M Laviale, P Usseglio-Polatera) will bring their expertise in bioindication of aquatic ecosystems, internationally recognized through the development of operational tools, trait-based, already successfully implemented for various biological compartments of freshwater ecosystems (bacteria, diatoms, macroinvertebrates, fish).
In Smart-Biodiv, LIEC is leading WP4 (Ecosystem models and bioindicators). In connection with WPs 2 (AI for biodiversity) and 3 (hybrid AI), the objectives are to produce synthetic indicators but also functional models to predict the ecological state of marine environments.
LOV
the UMR 7093, Villefranche sur Mer Oceanography Laboratory (LOV) is a joint research unit of Sorbonne University (SU) and the National Center for Scientific Research (CNRS), which conducts research at the interfaces between physical and biological oceanography.
Using sophisticated sensors designed by the laboratory teams and combining data acquired with satellite observations, the LOV monitors the major biological and biogeochemical cycles of the world ocean. It is also specialized in monitoring plankton ecosystems, in the bay of Villefranche as in many oceans around the world. The LOV is involved in numerous offshore and coastal oceanographic campaigns and its expertise in the management and modeling of marine data is recognized.
The LOV is responsible for WP1, the LOV researchers involved in Smart-Biodiv (F. Lombard, J.O. Irisson) from the Complex team, will bring their expertise on the analysis of large sets of observations coupling environment and planktonic ecosystem to the using quantitative imaging techniques. The proposed datasets cover both the Mediterranean ecosystems via the Villefranche point B time series (sampled since the 1960s) and the island ecosystems of the Pacific Ocean via the data provided during the Tara Pacific expedition.
LOCEAN
The UMR 7159, Laboratory of Oceanography and Climate: Experiments and Numerical Approaches (LOCEAN) is a joint research unit in partnership with Sorbonne University (SU), the National Center for Scientific Research (CNRS), the Research Institute for Development (IRD) and the National Museum of Natural History (MNHN). It brings together around 185 people, including 121 permanent staff. The LOCEAN conducts studies on the physical and biogeochemical processes of the ocean and their role in the climate in interaction with marine ecosystems. Its teams, widely recognized internationally, address a wide range of time and space scales for a better understanding of the dynamics and variations of the ocean within the climate system as well as its present, past and future evolution. Its teams also contribute to the development of analysis, modeling and observation methods, as well as to the systematic observation of the ocean, in situ or from space.
Within the SmartBiodiv project, the PROTEO team is involved in WP1 – Data integration (WP leader: SD Ayata). Within this WP, and in collaboration with the LOV, the LOCEAN will contribute to the co-supervision of an assistant engineer based at the LOV. LOCEAN will also contribute to WP4 on marine ecosystem models and the development of bioindicators in collaboration with the LIEC and the LORIA. Finally, the LOCEAN will contribute to WP5 via the co-organization of summer school, workshop and hackathon.
