Schedule for: 17w5153 - Inferential Challenges for Large Spatio-Temporal Data Structures

Meet me near the Tim Horton's at 11.45 if you'd like a ride.

https://goo.gl/maps/Tn28osn85Pn

I'll update this page if my plane is late. Send me an email if you're in the airport but can't find me.

This bus banffairporter.com/rates is CAD 64 one way, get a 15% discount by entering the promo code "birs" at the payment screen. Bus leaves at 10:00 11:00 12:30 13:30 14:30 15:30 16:30 18:30 20:30 22:30

www.brewster.ca/brewster-express/schedule-locations is CAD 69 one way leaves at 08:30 09:30 11:30 13:00 14:00 15:30 17:15 18:30 22:00

A buffet dinner is served daily between 5:30pm and 7:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

A buffet dinner is served daily between 5:30pm and 7:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

In order to adapt to a changing climate, policymakers need information about what to expect for the climate system. Typically local information about certain aspects of the climate system comes from regional climate models as well as from observational records. A regional climate model is a downscaled global circulation model, a mathematical model that describes, using partial differential equations, the temporal evolution of climate, oceans, atmosphere, ice, and land-use processes over a gridded spatial domain of interest. An important problem is understand how well regional models can reproduce observed climate variables. Using two motivating analyses based on data from the Swedish Meteorological and Hydrological Institute, I will discuss different spatio-temporal modeling strategies that can be used to assess regional climate models using observational data. I will also outline the associated statistical and computational challenges in building hierarchical models using data sources with varying spatial and temporal support. My intention is to motivate a dialogue about the broader challenges underlying spatio-temporal climate model assessment. This talk is based on joint research with Peter Guttorp and Veronica Berrocal.

Non-Gaussian spatial data arise in a number of disciplines. Examples include spatial data on disease incidences (counts), and satellite images of ice sheets (presence-absence). Spatial generalized linear mixed models (SGLMMs), which build on latent Gaussian processes or Gaussian Markov random fields, are convenient and flexible models for such data and are used widely in mainstream statistics and other disciplines. For high-dimensional data, SGLMMs present significant computational challenges due to the large number of dependent spatial random effects. Furthermore, spatial confounding makes the regression coefficients challenging to interpret. I will discuss projection-based approaches that reparameterize and reduce the number of random effects in SGLMMs, thereby improving the efficiency of Markov chain Monte Carlo (MCMC) algorithms for inference. Our approach also addresses spatial confounding issues. This talk is based on joint work with Yawen Guan (SAMSI) and John Hughes (U of Colorado-Denver).

Meet in foyer of TCPL to participate in the BIRS group photo. The photograph will be taken outdoors, so dress appropriately for the weather. Please don't be late, or you might not be in the official group photo!

Advances in geo-spatial technologies have created data-rich environments which provide extraordinary opportunities to understand the complexity of large and spatially indexed data with rich and complex spatial models. Spatial process models for analyzing geostatistical data often entail computations that become prohibitive as the number of spatial locations becomes large. We propose a divide-and-conquer strategy within the Bayesian paradigm to achieve massive scalability for spatial process models. We partition the data into subsets, implement a Bayesian spatial process model to analyze data in each subset and then obtain approximate posterior inference for the entire dataset by optimally combining the individual posterior distributions from each subset. Importantly, we offer full posterior predictive inference on the residual spatial surface as well as on the outcome at arbitrary locations, and full posterior inference on model parameters. We call this approach ``Distributed Kriging" (DISK). The approach has the major advantage of employing embarrasingly parallel computation and not having to store the entire data in one processor, this leads to massive scalability. Though the framework is general in nature and essentially applicable to any spatial model, the present article carefully demonstrate its performance with the stationary Gaussian process model and nonstationary modified predictive process model. The approach is intuitive, easy to implement and leads to accurate characterization of spatial processes with massive datasets. Moreover, the present article rigorously develops theoretical results justifying optimal performance of the proposed approach. We further illustrate its significantly superior inferential and predictive ability in comparison with the state-of-the-art competitors using different simulation experiments and a geostatistical analysis of the Pacific Ocean sea surface temperature data.

Gaussian Cox process (LGCPs) are a type of inhomogeneous Poisson point process where the log intensity surface is a GP. A point process approach is useful when each observation is indexed to a particular point in space and time. This is in contrast to the common area-level approach in epidemiology wherein observations and risk factors are summarised over several small regions (e.g., counties or local authorities). The spatially-continuous approach inherent in LGCPs naturally accommodates risk factors measured on different spatio-temporal units and avoids some forms of ecological bias. However, we still face many computational and interpretation issues with these models. In this talk I compare maximum likelihood and Bayesian techniques for estimating systematic trends in the spatio-temporal risk surface as well as the latent GP and discuss the strengths and weakness of the existing computational tools for fitting LGCPs. Finally I will use a spatio-temporal LGCP to investigate the roles of environmental and socio-economic risk-factors in the incidence of campylobacter (a common bacterial case of food borne disease) in the UK.

Many people living in low- and middle-income countries are not covered by civil registration and vital statistics systems. Consequently, household sample surveys with complex designs are often used to estimate health and population indicators. In this talk I will describe both discrete and continuous spatial models with time modeled discretely. Topics that will be touched upon include: how to account for the sampling design; the simultaneous use of both point- and area-level data; how to make adjustments for HIV epidemics; the inclusion of so-called indirect data; and covariate modeling. The modeling of under-5 mortality in Kenya is used to motivate and illustrate the issues raised. Data come from a variety of sources including Demographic and Health Surveys conducted over the period 1991–2010. Collaborators include: Sam Clark, Andrea Riebler, Geir-Arne Fuglstad, Jessica Godwin and Katie Wilson.

In statistical ecology a specific data structure, e.g. resulting from a common survey method, often motivates both statistical methodology and software development. The specific survey method as well as the statistical methodology adapt to the practicality of data collection but do not directly reflect the underlying ecological process of interest. This results in highly specialised modelling approaches and, at the same time, little exchange among the developers of the different strands of methodology. In this talk we discuss how thinking in terms of the processes that we would like to model rather than thinking in terms of the survey method can yield a flexible class of models. Specifically, the ecological processes of interest here are the structures formed by individuals in space and time, reflecting the individualsâĂŹ interaction among each other and with the environment.

The EUSTACE project will give publicly available daily estimates of surface air temperature since 1850 across the globe for the first time by combining surface and satellite data using novel statistical techniques." To fulfil this ambitious mission, a spatio-temporal multiscale statistical Gaussian random field model is constructed, with a hierarchy of spatio-temporal dependence structures, ranging from weather on a daily timescale to climate on a multidecadal timescale. Data from weather stations, ships, and satellites are combined in a hierarchical structure with individual measurement error models, and transformations of the latent random fields to allow joint estimation of current and past temperature fields across the globe. Connections between SPDEs and Markov random fields are used to obtain sparse matrices for the practical computation of point estimates, uncertainty estimates, and posterior samples. The extreme size of the problem necessitates the use of iterative solvers, which requires using the multiscale structure of the model to design an effective preconditioner, leveraging domain specific knowledge, traditional statistical techniques, and modern numerical methods.

5-day workshop participants are welcome to use BIRS facilities (BIRS Coffee Lounge, TCPL and Reading Room) until 3 pm on Friday, although participants are still required to checkout of the guest rooms by 12 noon.

I will post a sign-up sheet, everyone is welcome to join me