Separable spatio-temporal kriging for fast virtual sensing

TL;DR

This paper introduces a separable spatio-temporal kriging method that significantly reduces computational costs, enabling fast virtual sensing in large environmental datasets by decoupling spatial and temporal modeling.

Contribution

It proposes a novel separability assumption in spatio-temporal kriging that simplifies computations and allows independent modeling of space and time, facilitating scalable environmental monitoring.

Findings

Reduces kriging computational complexity for large datasets.

Enables decentralized forecasting at individual sensors.

Maintains accuracy with simplified models in tall datasets.

Abstract

Environmental monitoring is a task that requires to surrogate system-wide information with limited sensor readings. Under the proximity principle, an environmental monitoring system can be based on the virtual sensing logic and then rely on distance-based prediction methods, such as $k$-nearest-neighbors, inverse distance weighted regression and spatio-temporal kriging. The last one is cumbersome with large datasets, but we show that a suitable separability assumption reduces its computational cost to an extent broader than considered insofar. Only spatial interpolation needs to be performed in a centralized way, while forecasting can be delegated to each sensor. This simplification is mostly related to the fact that two separate models are involved, one in time and one in the space domain. Any of the two models can be replaced without re-estimating the other under a composite likelihood approach. Moreover, the use of convenient spatial and temporal models eases up computation. We show that this perspective on kriging allows to perform virtual sensing even in the case of tall datasets.