Tobias Andermann: Spatial biodiversity modeling with remote sensing and AI

Convolutional Neural Networks (CNNs) allow the seamless integration of multiple complex and heterogenous data types, making them very suitable tools for modeling biodiversity. Here we apply such CNN models to learn the complex correlations between biodiversity and the multidimensional biotic and abiotic matrix that the species community interacts with and exists within. 

The proof-of-concept model is trained on polygons with high and low biodiversity values distributed across Sweden. The model learns how to correlate biodiversity value with the multitude of spatial predictors used in the model, including data products from remote sensing techniques. This allows us to produce continuous heat-maps (rasters) of biodiversity value across the entire country of Sweden at a 10x10m spatial resolution, providing estimates at a spatial scale that is useful for conservation planning, biodiversity offset evaluation, and simulation-based (in silico) impact assessments of infrastructure projects and other anthropogenic landscape modifications. 

In our ongoing work we are combining this model architecture with high- resolution biodiversity data generated with environmental DNA, improving the predictive power and taxonomic resolution of these models. With increasing biodiversity data becoming available across the world, these models can eventually be applied on a global scale, constituting computational tools for standardized and automated high-resolution biodiversity predictions and impact assessments.

Tobias Andermann is a biodiversity researcher dedicated to providing data and computational tools for combating the global biodiversity crisis. He holds a position as assistant professor at Uppsala University and is a research fellow of the SciLifeLab and Wallenberg Data-Driven Life Science program. His group, the Biodiversity Data Lab, is working on the intersection of molecular biology, spatial ecology, and machine learning, with the mission to provide a more comprehensive view on the distribution of biodiversity, including hidden diversity of inconspicuous and even undescribed species through the use of environmental DNA.