Spatio-temporal statistics in Earth sciences

This paper looks at spatio‑temporal statistical modelling through the prism

of hierarchical statistical modelling, emphasising the role of dynamical temporal

models of spatial fields. A purely spatial dataset of soil mineralisation is used to

motivate the Spatial Random Effects statistical model, which is adept at handling

massive, nonstationary datasets. This is then generalised to a Spatio‑Temporal

Random Effects model, which is applied to a large global remote‑ sensing dataset of

aerosol optical depth (from the Multi‑angle Imaging SpectroRadiometer instrument

on the National Aeronautics and Space Administration’s Terra satellite). This results

in optimally interpolated and filtered estimates of the true underlying process, along

with a statistical measure of the estimates’ uncertainties. Finally, the problem of

combining data from multiple remote‑sensing instruments is discussed; the Spatial

Random Effects model is adapted to handle this problem.