This seminar presents a comprehensive Bayesian hierarchical framework for modeling daily temperatures, integrating mean, quantile, and record-breaking analyses across space and time. We first introduce a spatio-temporal mean model for daily maximum temperatures, incorporating two temporal scales, explicit autoregressive dependence, fixed effects, and multiple random effects to capture spatial variability. Building on this, we develop a mixed effects quantile regression model with asymmetric Laplace errors to explore climate change across quantiles, enabling marginal quantile inference from conditional autoregressive structures and revealing pronounced spatial-quantile heterogeneity in climate signals. Next, we introduce a new framework for analyzing high-temperature events, proposing hierarchical models for both univariate and bivariate record-breaking temperatures. The univariate model employs a logistic regression formulation with an explicit long-term trend and strong daily spatial random effects to analyze the occurrence of calendar-day records across years, allowing inference on the number, spatial distribution, and temporal evolution of record-breaking events under climate change. The bivariate model uses a probit regression to jointly model maximum and minimum temperature records, capturing spatial and temporal dependence through anisotropic coregionalized Gaussian processes and revealing correlated yet distinct patterns of record-breaking behavior. All models leverage Gaussian latent representations for closed form Gibbs sampling and provide spatial predictions at unobserved locations. Applications to long-term datasets from Spain illustrate trends in daily temperatures, quantile-specific climate change, and the increasing occurrence of record-breaking temperatures, offering new tools for the statistical analysis of environmental extremes.
11 Novembre 2025, ore 12
Jorge Castillo-Mateo
Department of Statistical Methods and IUMA, University of Zaragoza
In person: Room 34 (4th floor) building CU002 Scienze Statistiche
Webinar: https://uniroma1.zoom.us/j/83625004899?pwd=bXCtz0mp759PUh2lkqT0BUoVa0Uegg.1
ID riunione: 836 2500 4899
Passcode: 123456