Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

McGill University

$44,000.00

Spatio-temporal models for highly structured point-referenced and area-level data

EF 2024-25

Résumé
Research questions in the fields of environmental science, ecology, epidemiology, and hydrology commonly involve observations that have attached a location in space and an instant in time. Observations made at neighboring locations and closely adjacent points in time tend to be highly correlated. In recent years, it has become more common to observe heavy tailed processes (e.g., temperature measurements) such that usual normal assumptions, even after a transformation of the data, are not suitable. My long-term research programme focuses on developing statistical methods for spatio-temporal processes that, besides capturing the complex correlation structure of the data, also accommodate observations that fall within the tail of the distribution without relying on arbitrary transformations of the data. I will focus on univariate and multivariate processes observed either at fixed locations that vary continuously across a region of interest (geostatistics) or at disjoint areas that together form a region of interest (area-level). The short-term goals of my research involve proposing a multivariate spatio-temporal model which builds upon my previous work on non-Gaussian univariate dynamical modelling for processes observed at fixed locations of a region. In addition to relaxing the gaussianity assumption, the proposed approach extends the well known class of multivariate dynamic linear models by accommodating heavy-tailed observations without resorting to any transformation of the data. Theoretical properties of the proposed model will be derived. For processes observed at the areal level, I plan to develop models based on my recent contributions to heavy-tailed processes for geostatistical data. This will be done by assuming the latent effects of Poisson-lognormal mixture models to follow a multivariate scale-mixture structure, wherein the variance of the latent process changes across areas and allows for outlier identification. For all developed approaches, inference will be performed under the Bayesian paradigm, which usually relies on Markov chain Monte Carlo (MCMC) methods to sample from the resultant posterior distribution. I will also develop alternative approximating methods (e.g., variational Bayes, sequential inferential approaches based on information geometry) to perform inference more efficiently on these new class of models. As we observe the impact of climate change on environmental (e.g., temperature levels) and epidemiological (e.g., disease counts) processes, I expect this research program to impact on how statisticians model processes that more commonly result in discrepant observations, while naturally accounting for their complex structures and providing realistic uncertainty quantification. All software code and R packages developed during this research will be made publicly available to facilitate the dissemination of the proposed models and their implementation by a wide audience.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Quebec, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2024-04312
Name nom
Schmidt, Alexandra
Department département
Epidemiology, Biostatistics and Occupational Health
Organization id
61
Institution établissement
McGill University
Province en
Québec
Province fr
Québec
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
44000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1002
Area of application group en
Northern development
Area of application group fr
Développement du nord
Area of application en
Environment
Area of application fr
Environnement
Research subject code
3007
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Stochastic processes
Research subject fr
Processus stochastiques
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Spatio-temporal models for highly structured point-referenced and area-level data
Keywords
bayesian inference; conditional autoregressive models; dynamic linear models; geostatistics; linear model of coregionalization
Application summary
Research questions in the fields of environmental science, ecology, epidemiology, and hydrology commonly involve observations that have attached a location in space and an instant in time. Observations made at neighboring locations and closely adjacent points in time tend to be highly correlated. In recent years, it has become more common to observe heavy tailed processes (e.g., temperature measurements) such that usual normal assumptions, even after a transformation of the data, are not suitable. My long-term research programme focuses on developing statistical methods for spatio-temporal processes that, besides capturing the complex correlation structure of the data, also accommodate observations that fall within the tail of the distribution without relying on arbitrary transformations of the data. I will focus on univariate and multivariate processes observed either at fixed locations that vary continuously across a region of interest (geostatistics) or at disjoint areas that together form a region of interest (area-level). The short-term goals of my research involve proposing a multivariate spatio-temporal model which builds upon my previous work on non-Gaussian univariate dynamical modelling for processes observed at fixed locations of a region. In addition to relaxing the gaussianity assumption, the proposed approach extends the well known class of multivariate dynamic linear models by accommodating heavy-tailed observations without resorting to any transformation of the data. Theoretical properties of the proposed model will be derived. For processes observed at the areal level, I plan to develop models based on my recent contributions to heavy-tailed processes for geostatistical data. This will be done by assuming the latent effects of Poisson-lognormal mixture models to follow a multivariate scale-mixture structure, wherein the variance of the latent process changes across areas and allows for outlier identification. For all developed approaches, inference will be performed under the Bayesian paradigm, which usually relies on Markov chain Monte Carlo (MCMC) methods to sample from the resultant posterior distribution. I will also develop alternative approximating methods (e.g., variational Bayes, sequential inferential approaches based on information geometry) to perform inference more efficiently on these new class of models. As we observe the impact of climate change on environmental (e.g., temperature levels) and epidemiological (e.g., disease counts) processes, I expect this research program to impact on how statisticians model processes that more commonly result in discrepant observations, while naturally accounting for their complex structures and providing realistic uncertainty quantification. All software code and R packages developed during this research will be made publicly available to facilitate the dissemination of the proposed models and their implementation by a wide audience.