Improving tropical tree biodiversity mapping from drone imagery to support the scientific study and conservation of tropical forests

Rubisco AI is a startup that aims at monitoring forests worldwide to study their biodiversity using high resolution drop imagery. One of their core applications is to monitor tropical rainforest where tree identification and mapping is currently a very hard task: a single local biologist expert can only identify ~20 individual trees per day due to the difficulty of the task. By contrast, Rubisco AIs technology could speed this up by at least 1000 x. Current field-based tropical tree inventories are a bottleneck to the overall understanding of tropical rainforests and their use for local communities, and how to protect forests from human activity and climate change. This project will leverage the momentum that both the industrial and academic partners have obtained by being part of the winning team in the XPRIZE Rainforest competition this past year. We developed a pipeline to go from high resolution drone imagery of forests to different forms of annotation, from bounding boxes of trees to segmentation maps of different species. One of the key issues we have encountered is that it is extremely challenging to obtain ground truth labels for less known tree species. A performant tropical rainforest tree segmentation and species identification AI pipeline would have a large number of applications to improve monitoring to help mitigate deforestation, biodiversity loss and climate change in those crucial ecosystems that hold most of the Earths terrestrial carbon and biodiversity. This will help to improve tree species distribution models, forests, biodiversity mapping, new species discovery, improved quantification of above-ground biomass/carbon stock and better atmospheric carbon flux predictions. It would also empower local communities as they could quickly and reliably monitor the forest ecosystems they manage and care for. Overall, this would be highly relevant to Rubisco AI and our overall canopy tree mapping solution using drone imagery and AI.

Faculty Supervisor:

Christopher J. Pal

Student:

Partner:

Rubisco AI

Discipline:

Computer science

Sector:

Information and cultural industries

University:

Université de Montréal

Program: