A Bayesian Spatial-Temporal Functional Model for Data with Block Structure and Repeated Measures

TL;DR

This paper introduces a Bayesian spatio-temporal functional regression model with block structure and repeated measures, effectively capturing complex climate data patterns and improving precipitation estimation in Brazil.

Contribution

It develops a novel Bayesian model incorporating block structures, functional predictors, and random effects to analyze complex spatio-temporal climate data with repeated measures.

Findings

Model performs well in simulation studies.

Effectively estimates precipitation from real data.

Captures complex spatial-temporal-block relationships.

Abstract

The analysis of spatio-temporal data has been the object of research in several areas of knowledge. One of the main objectives of such research is the need to evaluate the behavior of climate effects in certain regions across a period of time. When certain climate patterns appear for several days or even weeks, causing the areas affected by them to have the same kind of weather for an extended period of time, the use of blocks for these phenomena may be a good strategy. Additionally, having repeated measures for observations within blocks helps to control for differences between observations, thus gaining more statistical power. In view of these perspectives, this study presents a spatio-temporal regression model with block structure with repeated measures incorporating as predictors functional variables of fixed and random nature. To accommodate complex spatial, temporal and block structures, functional components based on random effects were considered in addition to the class Mat\'ern covariance structure, which was responsible to account for spatial covariance. This work is motivated by a precipitation dataset collected monthly from various meteorological stations in Goi\'as State, Brazil, covering the years 1980 to 2001 (21 years). In this framework, spatial effects are represented by individual meteorological stations, temporal effects by months, block effects by climate patterns, and repeated measures by the years within those patterns. The proposed model demonstrated promising results in simulation studies and effectively estimated precipitation using the available data.