Computationally efficient estimation of high-dimension autoregressive models : with application to air pollution in Malta

Abstract

The modelling and analysis of spatiotemporal behaviour is receiving wide-spread attention due to its applicability to various scientific fields such as the mapping of the electrical activity in the human brain, the spatial spread of pandemics and the diffusion of hazardous pollutants. Nevertheless, due to the complexity of the dynamics describing these systems and the vast datasets of the measurements involved, efficient computational methods are required to obtain representative mathematical descriptions of such behaviour. In this work, a computationally efficient method for the estimation of heterogeneous spatio-temporal autoregressive models is proposed and tested on a dataset of air pollutants measured over the Maltese islands. Results will highlight the computation advantages of the proposed methodology and the accuracy of the predictions obtained through the estimated model.