Participation in the VALPRED Workshop
In December 2024, Professor Johan Segers and four PhD students from our team attended the VALPRED workshop in Aussois, France. This workshop is renowned for fostering insightful discussions on the validation of forecasting methods and related topics, offering a platform for cutting-edge research in predictive modeling.
The workshop provided an invaluable opportunity for our team to engage with an international network of researchers, share knowledge, and gain new perspectives on challenges and methodologies in the field.
Mini-Course by Professor Johan Segers
Professor Johan Segers delivered a mini-course on Multivariate Generalized Pareto Distributions, a topic central to the modeling of multivariate extremes. His lecture was highly appreciated for its clarity and depth, contributing significantly to the workshop’s academic discourse.
Abstract:
The peaks-over-threshold approach for univariate extremes stipulates that the excess of a random variable over a high threshold can be modeled by the two-parameter family of generalized Pareto distributions. In statistical practice, the generalized Pareto distribution is fitted to the observed excesses of a variable over a high threshold, and the fitted model is used as a basis for extrapolation.
For multivariate extremes, the same strategy can be followed using multivariate generalized Pareto distributions. The difference is now that these distributions no longer comprise a parametric family.
In this mini-course, we discovered various surprising properties and representations of multivariate generalized Pareto distributions that facilitate working with them for modeling multivariate extremes.
Poster Presentation by Alexandre Tytgat
PhD student Alexandre Tytgat presented a poster titled “Antarctic Sea Ice: Recent Variability and Trends”, which sparked interest and engaging discussions among the workshop participants.
Abstract:
The Antarctic sea ice extent (SIE) has exhibited unusual variability in recent years, including record lows in annual minima observed in 2017, 2022, 2023, and 2024. This marks a sharp contrast with the sea ice extent evolution between 1979 and 2015, during which yearly mean sea ice extent slightly increased despite global warming.
Such events raise critical questions about their probabilities under current and future conditions, as well as the potential factors driving such events. To address these questions, we use block-minima models to analyze and interpret the yearly minima of the Antarctic SIE. Additionally, we extend this approach to the five regional basins that constitute the Antarctic, aiming to shed some light on their contributions to the recent extreme events.
Our results suggest that the trend in SIE yearly minima is best captured by a piecewise linear model for the mean, and an increasing scale parameter over time, and that the recent record lows, though unprecedented, were not highly improbable given recent trends. The regional analysis indicates that the Ross Sea, Weddell Sea, and Western Pacific Ocean basins are the primary contributors to the recent decline in total SIE.