Biodiversité, changements climatiques et environnement

Mots-clefs :

• Biodiversité, changements climatiques, télédétection, drone, imagerie, vision par ordinateur
• (Ajout 22/07) système climatique, intelligence artificielle, apprentissage profond, apprentissage machine, sciences géomatiques, télédétection, programmation, analytique, sciences des données, big data, faune et flore, modélisation, inférence, statistiques, cartographie, système d’information géographique, sécurité civile, militaire, météorologie, assurances, R&D, éducation, construction, gouvernement, archéologie, télécommunications, LIDAR, radar, satellite, photos aériennes, photogrammétrie, orthoimage, haute résolution, drone, station, capteur, base de données, données vectorielles, raster, requêtes spatiales, aide à la décision, temporalité, zones à risque, hydrologie, réseaux, ingénierie, indicateurs, aide à la décision
• (Ajout 22/07) Vision par ordinateur, Apprentissage automatique
• (Ajout 22/07) sustainability, nature-based solutions, evolution, environmental DNA, genomics.

Organisations pertinentes :

• Observatoire Aérien Canadien de la Biodiversité (CABO), Centre sur la biodiversité de l’Université de Montréal, Centre de la science sur la biodiversité du Québec (CSBQ), GEO BON, NEON, MELCC, MFFP, Hydro-Québec, HoromaAI, MAPAQ, Microsoft (AI for Earth, Planetary Computer), Google Earth Engine, NEON, TERN, GBIF, Canadensys.
• (ajout 22/07) John Deere, Bayer, Monsanto, Sollio, UPA, FADQ, CNRC PARI, AAFC, CRAAQ, OBVIA, IVADO, MILA, CRSNG, Mitacs, U. Laval, U. Sherbrooke, McGill U., U. Montréal, ENAP, Zone Agtech, Géomont, LBM Agtech, horoma AI, NorthStar Ciel & Terre, HiLDEN, NASA, Agence Spatiale du Canada.
• (ajout 22/07) les acteurs de ces domaines sont impliqués : sécurité civile, santé publique, militaire, météorologie, assurances, R&D, éducation, construction, gouvernement, ingénierie, réseaux, archéologie, les banques, les assurances, télécommunications
• (ajout 22/07) There are several groups operating at different levels that are relevant to this theme and projects in Quebec: Group on Earth Observations Biodiversity Observation Network (GEO BON) is an international network and flagship of GEO. The Quebec Centre for Biodiversity Science (QCBS, a FRQNT strategic cluster). Both organizations have their secretariats in Montreal. As co-director of these institutions I am working closely with Microsoft and their AI for Earth program and have received funding on AI applications to biodiversity assessments and monitoring.

Personnes pertinentes suggérées durant la consultation : 

Les noms suivants ont été proposés par la communauté et les personnes mentionnées ci-dessous ont accepté d’afficher publiquement leur nom. Notez cependant que tous les noms des professeur.e.s (qu’ils soient affichés publiquement ou non sur notre site web) seront transmis au comité conseil pour l’étape d’identification et de sélection des thèmes stratégiques. Notez également que les personnes identifiées durant l’étape de consultation n’ont pas la garantie de recevoir une partie du financement. Cette étape sert avant tout à présenter un panorama du domaine, incluant les personnes pertinentes et non à monter des équipes pour les programmes-cadres.

• Etienne Laliberté
• Mickaël Germain
• Timothée Poisot
• Anne Bruneau
• Guillaume-Alexandre Bilodeau
• Teja Kattenborn
• Wassim Bouachir

botanist: AI to monitor plant biodiversity from drone imagery

bot   an autonomous program on the internet […] that can interact with systems or users  |   botanist   a scientist who studies plants

Our planet is home to nearly 400,000 plant species that collectively power the biosphere, on which our lives ultimately depend. Human activities, notably via climate change, are rapidly and fundamentally altering this precious biodiversity. Actionable data about plant biodiversity ーwhich plant species grow where, now and in a warmer futureー are in very high demand across a large number of sectors, because the species composition of vegetation controls the functioning of ecosystems and the many services they provide us with. However, acquiring plant biodiversity data is severely constrained by the fact that it requires sending highly trained botanists in the field to visually identify and survey plant species. This human-powered process is extremely slow, laborious, costly, and limited to very small localities (roughly 10ー100 m² day^-1 botanist^-1), whereas land management decisions often involve surfaces millions of times larger. As a result, we are left completely in the dark about how climate change and other major environmental changes are impacting plant biodiversity across the overwhelming majority of the Earth’s surface, and land managers are essentially playing blind.

Advances in drone technology could boost our ability to monitor plant biodiversity by a factor 10⁴ー10⁶, thus providing actionable data at spatial scales relevant to decision-making. In particular, cheap, easy-to-use, microdrones can now capture high-resolution images of large swaths of vegetation with almost no legal or logistical constraints, making them universally accessible. However, AI and new algorithms are required to achieve the transformative potential of this technology for monitoring plant biodiversity. While impressive results have already been achieved with convolutional neural networks to classify, detect and segment plant species from high-resolution drone RGB imagery using visual features alone, smarter and better algorithms will need to leverage other key sources of information that are used by human botanists to efficiently and accurately visually identify plant species, such as the broader spatial, temporal and environmental context, as well as the co-occurrence of other species. The primary research objective of this project would be to develop these new AI algorithms that harness visual features as well as broader contextual information. Such research would also be an important step toward establishing a global plant biodiversity observation system based on satellite remote sensing, using AI algorithms trained with those higher-resolution biodiversity maps from drone imagery.

The ability to map plant biodiversity at high resolutions over large extents with drones and AI will benefit many sectors: land management agencies in charge of monitoring and preserving biodiversity, resource extraction sector (e.g. forestry, mining) who need to perform vegetation surveys for environmental assessments and/or ecosystem restoration, or precision agriculture (e.g. weed mapping). For example, we are working with MELCC to track plant biodiversity changes using drones and AI in their provincial biodiversity monitoring network, and they have expressed much interest in this approach to better characterize the vegetation within new protected areas, detect invasive species, and monitor populations of threatened species. MFFP would also benefit from improved tree species mapping, and they already invest heavily into human visual photo-interpretation of aerial imagery for forest species classification. MAPAQ could also be interested in this research for improved weed management. Hydro-Québec, which must manage vegetation within 34 000 km of power lines right-of-ways, would also benefit from improved plant species mapping, particularly large trees.

There would be potential to attract additional support for this project, either via government agencies (particularly MELCC and MFFP), or private companies (e.g. Hydro-Québec, HoromaAI). Recently, Microsoft AI for Earth has donated 1M$USD for a joint GEO BON–QCBS initiative (Laliberté is now co-director of QCBS) to improve biodiversity monitoring; they have expressed interest in supporting other transformative projects such as the one we propose here that could make excellent use of Microsoft’s Planetary Computer (either via cloud computing, human resources, access to Earth observation data to provide environmental contextual information, and/or funds). Such a collaboration with Microsoft’s Planetary Computer could help to operationalize and provide broad access to the research and tools we would be developing in this project. As lead PI of the CABO project and co-director of QCBS, Laliberté is already part of a team of a core team of researchers with much of the expertise required to make this project possible, including: Mickaël Germain (UdeS; AI in remote sensing, web-based GIS, active learning, decision systems), Anne Bruneau (UdeM, biodiversity information standards and data sharing), Margaret Kalacska (McGill; photogrammetry and image processing), and Nicholas Coops (UBC; fusion with satellite data for large-scale biodiversity mapping). The project would also benefit from establishing new collaborations, especially with other researchers within MILA and IVADO who specialize in computer vision, as well as other researchers interested in developing sub-projects about use cases of the proposed research (e.g. biodiversity conservation, vegetation management, forestry, precision agriculture, etc).

(Ajout 22/07) Deux applications bien précises, l’évaluation de la biodiversité dans un champ de fourrage permet de valider la valeur nutritive, donc la valeur de la biomasse à récolter. Dans un second temps, la biodiversité dans un champ de monoculture n’est pas une bonne chose. La capacité de détecter les mauvaises herbes tôt dans le cycle de croissance permet de prendre action rapidement pour éviter la contamination de la récolte et son déclassement.

Développer une plateforme d’annotation d’images géoréférencées de hautes résolutions, collaborative entre les humains et les modèles AI (feedback), ouverte et compétitive. Voir le projet AIDE sur github https://github.com/microsoft/aerial_wildlife_detection et les autres existantes collectées. https://vince7lf.github.io/assets/Outils%20annotation%20mask%20segmentation%202021%20Vincent%20Le%20Falher.pdf

Voir aussi : https://docs.google.com/document/d/1_JUZqxh3rtn6ZMa8X2Hn4G-X3GebpGS4KXQkvEGg9Po/edit?usp=sharing