Machine Learning in Montpellier, Theory & Practice
The acceleration of human-mediated species transfers between biogeographical regions increases the frequency of biological invasions, with significant costs to the environment and our societies. Spatial dynamic models based on established and interpretable ecological mechanisms are a crucial tool for understanding and anticipating biological invasions. However, we lack standardized spatio-temporal data to measure species life history traits that determine these mechanisms, such as their species fecundity, mortality, movement strategies and speed, and their dependence on life stages or the environment. Since my postdoc at Stellenbosch University, I have been working to integrate available massive but heterogeneous biodiversity observations to estimate species’ life history traits using Bayesian state-space models, including observation models adapted to the many sampling biases that affect data distribution relative to actual population distribution. I will present a proof of concept of the approach where we integrated presence-only data from three datasets to reconstruct the past spatial dynamics of an invasive bush in South Africa, as well as general limitations to introduce ongoing and future work.
