Dynamic Multivariate Continuous Data State-space Estimation for Agrometeorological Event Anticipation

Abstract

This paper describes the selection of a state-space estimation method for application to the emerging research domain of agrometeorology. The work comes from a wider geocomputational research programme that relates to climate and environment monitoring and subsequent data analysis. In particular, the data currently being collected refers to meso-micro climates in vineyards across eight countries. It is terrestrial in kind, being in the context of near-ground truth continuous data. The time-related nature of the data, being continuous across a geospatial plane, gives rise to the need for mathematical models that are intrinsically spatio-temporal and while effective in their robust adequacy, are also computationally efficient. State-space models are considered a class of model within the time-series literature but they have some uniquely distinguishing features for continuous multivariate data representation. Ensemble Kalman Filter models are Bayesian based estimators of multiple realisations of statespaces over time, so are proposed here as applicable to this analytical process domain.