When is the meeting?	20-May-2022 at 10.00 - 11.00
Where is the meeting?	Zoom video-conference https://uu-se.zoom.us/j/67222760675
What is the meeting about?	Filling gaps in our knowledge about biodiversity using a combination of AI and metabarcoding
Who is the responsible presenter?	Tobias Andermann, new DDLS fellow Biodiversity and evolution

Abstract: Biodiversity is distributed unevenly across the planet, with some areas having a naturally low species diversity (e.g., Scandinavia), while other areas are very species-rich (e.g., the Amazon rainforest). These patterns often follow environmental or other abiotic gradients, such as the latitudinal gradient, with species richness in most taxa generally increasing toward the equator. However, on a more local scale, the drivers of species-richness are more complex and vary substantially between taxa. Classic approaches of modeling the spatial distribution of species diversity usually require available geographic information for each species belonging to the group of question. This bottleneck hampers the modeling of biodiversity patterns beyond a very few well studied groups such as mammals, birds, reptiles, and amphibians. In a recent study, we developed a neural network model that can learn from available point estimates of species diversity from inventories of a defined vegetation plot. The trained model can be used to predict species diversity for any given point in space, at an adjustable spatial resolution. In this approach species diversity is modeled as a feature of the landscape, rather than the sum of individual species distributions, making the estimation of individual species ranges obsolete and avoiding the previous bottleneck. I plan to further develop this approach and use species lists (or lists of undefinable ASVs) from environmental DNA sampling sites to train similar models to be able to extrapolate species diversity of groups that we currently have no knowledge of. I further plan to increase the spatial resolution of these models by implementing remote sensing data, such as satellite images and lidar point clouds.
The neural network approach is described in our recently published article: https://www.frontiersin.org/articles/10.3389/fpls.2022.839407/full

Browser not supported