We believe that open data are a treasure trove of knowledge that has not been entirely used yet, and require ecologists to get curious about tools and practices from the field of data science. We explore applications of machine learning and deep learning to biodiversity.
We are working on understanding the type of problems that we can solve without having to do new sampling, and how to handle errors and uncertainty.
As a bonus, we use the tools we develop to improve the monitoring and reporting abilities of various stakeholders. One of our key question related to this topic at the moment is to examine the “suitability” of existing data to address a given question. We work on measures to identify areas in which information is low, which should be sampled with a higher priority.
Addressing new questions often requires to develop new tools. We develop statistical and mathematical approaches, implement them, and release them as free and open-source software to make analyses reproducible and reliable. A lot of our work uses Julia, a high-performance language for numerical computing. We are developing a few of our own packages, and contribute to the EcoJulia project on GitHub.
We lead the development of the mangal.io project, an open database of species interactions and associated packages. We think a lot about what the best practices for scientific software should be, and do a lot of training. We sincerely believe that good science requires good tools, and we want to help everyone build and use them.