Eventos

Workshop on spatio temporal modelling - Pamplona

Fecha: Mar, Jun 25 - Jue, Jun 27 2024

Hora: 9:15 - 19:00

Ubicación: Aulario de la Universidad Pública de Navarra - Pamplona (Spain)

Ponentes: More information on the programme

To enroll in the workshop, please fill out the following form.
Deadline: May 31, 2024
https://forms.gle/jTj2wuzKAVwNeZLE9

The Workshop will consist of a Course on Geostatistics, a Workshop on Areal Data, and a Seminar on Point Processes

Course Description:

This course aims at giving an introduction to spatial modelling of point referenced data under the Bayesian paradigm. Topics that will be discussed include an introduction to Bayesian inference; Gaussian processes; Stationarity and Isotropy; Geometric anisotropy; Variogram; Correlation Functions; Bayesian kriging; Bayesian kriging in non-normal models. The last part of the course will point out some current topics of research in the area, including large spatial data and spatio-temporal models. All the theory presented will be followed by examples with real data analysis using packages (e.g., Nimble and Stan) in R.

Reference material

All course material (slides and R codes) used during the lectures will be made available to attendees. The slides are based on the following references:

• Banerjee, S., Carlin, B. P. and Gelfand, A. E. (2004) Hierarchical modeling and analysis for spatial data CRC Press/Chapman Hall.
• Bivand, R. S., Pebesma, E. and Gómez-Rubio, V. (2013) Applied Spatial Data Analysis with R. Springer, New York, USA.
• Diggle, P.J. and Ribeiro Jr., P.J. (2007) Model-based Geostatistics (Springer Series in Statistics).
• Wikle, C. K., Zammit-Mangion, A., Cressies, N. (2019) Spatio-Temporal Statistics with R. Chapman & Hall/CRC. Free download from here.

Workshop abstract:

Several statistical models and computational methods have emerged in the disease mapping literature, aiming to derive smoothed risk (or rates) estimates for areal data by integrating spatial and/or spatio-temporal dependence structures. However, the development of scalable models for the analysis of high-dimensional count data remains limited. The R package bigDM addresses this gap by implementing a range of univariate and multivariate scalable Bayesian models, using a "divide-and-conquer" approach. It relies on the well-known INLA (integrated nested Laplace approximation) technique for approximate Bayesian inference in latent Gaussian models.

Seminar abstract:

The talk introduces statistical approaches for understanding the temporal and spatial dynamics of infectious diseases, particularly focusing on Covid-19. It details a non-stationary spatio-temporal point process, using a neural network-based kernel to capture spatial triggering effects. Exogenous influences from city landmarks are considered, and mechanistic models provide data-driven forms for spatio-temporal intensity functions. Cluster models for identifying unknown parents are proposed, and a method to evaluate spread direction and velocities is presented using a growth differential equation.

Crime science analyzes diverse crime data, using statistical models to detect crime generators, identify factors attracting/inhibiting crimes in a spatio-temporal region. Methods address data dimensionality, employing AI. Two key probabilistic models involve log-Gaussian Cox processes for forecasting crime risk in city subregions and stochastic models with differential equations governing crime spread.