Modelling observed data from a latent stochastic spatial process

Room 3.08 of DMat at UMinho | - | 15:00


Raquel Menezes

CMAT - University of Minho


Abstract :: Geostatistical models become important when data is collected from different locations in space, and the variable of interest can (in theory) be measured at any location in the study area. One should assume an underlying spatial stochastic process indexed in a continuous domain, and spatial correlation must be taken into account. These models can be extended to include time, if one has data collected over space and time. In Portugal, the spatial distribution and abundance of several commercial fish species is mostly unknown, and there are many open questions about fishing sustainability. Geostatistical models, relying on information from scientific surveys or commercial fisheries, become useful tools for the assessment of distribution species. Fishery-dependent data present advantages, namely easier to obtain and better coverage of the time dimension, but it leads to domain representation issues. The fishermen movements are guided by some prior-knowledge of the places where it is expected to find the target species, thus the sampled data do not equally represent the study area. This is coined preferential sampling [2]. In this talk, after the introduction of background concepts in spatial statistics, the differences between “preferential" and “clustering" sampling design issues are emphasized. The model in [2] is briefly presented, for which the locations of the observed points are assumed to be informative, ie, the presence of preferential sampling is assumed. Recent contributions that aim to overcome the computational limitations of the previous model are discussed [1], [3], [5], which allow creating the appropriate context to apply this model to large volumes of data. This research falls within the scope of FCT project PTDC/MAT-STA/28243/2017 - Prefferential, with a strong collaboration from the Department of the Sea and Marine Resources of the Portuguese Institute of Sea and Atmosphere (IPMA).

[1] Diggle P. and Giorgi E., Model-based geostatistics for global public health: Methods and applications, Chapman and Hall/CRC Press, 2019.

[2] Diggle P., Menezes R. and Su T.L., Geostatistical Inference under Preferential Sampling (with discussion), Journal of Royal Statistics Society, series C, 59(2), 191-232, 2010.

[3] Dinsdale D. and Salibian-Barrera M., Methods for preferential sampling in geostatistics, Journal of the Royal Statistical Society, Series C (Applied Statistics), 68(1), 2019.

[4] Lindgren F., Rue H. and Lindstrom J. (2011). An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 73, 423-498, 2011.

[5] Monteiro A., Menezes R. and Silva M.E., Modelling preferential sampling in time, Boletin de Estadstica e Investigacon Operativa, 35(3), 180-196, 2019.

[6] Zuur A., Ieno E. and Saveliev A., Beginner's Guide to Spatial, Temporal and Spatial-Temporal Ecological Data Analysis with R-INLA, Highland Statistics Ltd., 2017.


Seminar Room of DMat-UMinho (3.08), and via zoom at

Seminar for the Doctoral Program in Applied Mathematics (MAP-PDMA Seminar)