Extrapolation of extreme covariates with an application to environmental data (Viviana Carcaiso)
We propose methods to enhance the predictive performance of generalized additive models (GAMs) in the context of covariate extrapolation, where predictions rely on covariates beyond their observed range. When using predictive models such as GAMs, shifts in the covariate distribution between training and prediction datasets can occur. Ignoring this issue may lead to inaccurate predictions near the extremes of the covariate distributions. For example, this problem is particularly critical in climate-change scenarios, where covariates simulated from climate models are likely to contain more extreme conditions. Our approach integrates GAMs for the bulk of covariate distributions with asymptotic models from multivariate extreme value theory at high covariate values. We consider binary responses based on a latent variable assumption, but also continuous responses. For large values of the covariates, the latent variable or continuous response is assumed to depend linearly on the covariates with an additive error term. The distribution of the error terms is characterized by the link function of the regression model. In applications to ecological and climate data, we explore how the new method can improve predictions, using environmental and meteorological variables.
Quantitative frameworks for atmospheric pollen analysis and forecasting: Integrating physics-based and data-driven modeling (Antonio Spanu)
This seminar outlines the quantitative framework used for the analysis and forecasting of atmospheric pollen. The discussion begins with the critical assessment of time-series data acquisition and their inherent limitations. The core focus is on pollen modeling, contrasting physics-based process-driven dispersion models with predictive data-driven techniques (ML/statistics). Finally, we present an application that utilizes atmospheric back-trajectories and multivariate statistics to define a pollen distance metric between different urban areas. This robust methodological approach offers potential for wider application in problems concerning ecological diversity.
