Date and time: Wednesday 19th November 2-3pm (UK)
In-person location: 1-19 Torrington Place, B09
Zoom link: contact stats-seminars-join@ucl.ac.uk
––––––––––––––––––––––––––––––––––––––––-
Title: Gaussian process models for pollution in rivers
Abstract: The impact of human activity on the quality of surface waters, particularly rivers, has recently received considerable attention. Statistical models aimed at characterizing the spatiotemporal distribution of biological and chemical indicators across river networks must address several unique challenges not typically encountered in standard spatial modelling. In this seminar, I will present a class of spatiotemporal models designed for multivariate data sampled on river networks. These models build upon the framework introduced by Ver Hoef et al. (2006), which replaces traditional Euclidean distances with stream distance metrics to better represent spatial relationships in river systems. We extend this approach to incorporate temporal dynamics and multivariate dependencies, while also addressing practical aspects of water quality monitoring such as censoring limits and intermittent sampling. Finally, I will introduce a variational inference framework for fitting these models efficiently and share results from simulation experiments comparing our proposed approach with more conventional Gaussian Process models.
For up to date details of upcoming seminars, please see the seminar webpage:
https://www.ucl.ac.uk/statistics/seminar
Recordings of previous seminars are available on our YouTube channel:
https://www.youtube.com/@UCLStatisticalScienceseminars