Bayesian Spatio-Temporal Time Series Model for Rice and Cassava Yields in Thailand

Abstract

ABSTRACT A linear mixed model (LMM) with trend and spatial effects to forecast rice and cassava yields in Thailand is proposed. It is a modification of our previous model, which was a multivariate conditional auto regressive model (MCAR) for spatial time series data without trend. An MCAR is assumed to account for the spatial effects and a linear trend is assumed for temporal effects. A Bayesian method is adopted for parameter estimation via Gibbs sampling in Markov chain Monte Carlo (MCMC). The model is applied to the monthly spatio-temporal rice and cassava yield data, which have been extracted from the Office of Agricultural Economics, Ministry of Agriculture and Cooperatives of Thailand. Using the mean absolute error criterion (MAE), the results show that the proposed model has a better performance in most provinces in the fitting part, and all provinces in the validation part compared to the exponential smoothing (ES) with trend (Holt ES) and the MCAR from our previous study.   Keyword: Bayesian linear mixed models, Multivariate conditional auto regressive model (MCAR), Rice and cassava yields, Spatio-temporal data, Time series data