SPARTA 1591 - Development of Space-Time Random Fields based on Local Interaction Models
The integration, efficient processing, and accurate visualization of spatiotemporal information present scientific challenges with significant societal benefits, such as the use of information from monitoring sensor networks for environmental protection and policy making. The mathematical framework for processing spatiotemporal data is the theory of space-time random fields. Current methods are limited in their flexibility and due to significant computational costs for large datasets.
The main objective of SPARTA is to develop a novel framework for stochastic space-time analysis based on statistical physics concepts. We will pursue this objective by developing random fields based on effective energy functions with local interactions, guided by our successful development of Spartan spatial random fields.
Expected benefits from SPARTA include (i) development of novel, flexible space-time covariance models (ii) computational efficiency (reduced memory requirements and numerical complexity) (iii) ability to process very large datasets and large simulation grids (iv) cross-fertilization with research fields that require efficient computation of large covariance matrices (v) generation of parameter inference, interpolation, prediction and simulation algorithms for use in intelligent mapping systems.
The SPARTA advances will provide spatiotemporal methods with strong potential for fast and unsupervised operation; e.g., they will permit near-real-time, accurate generation of maps for atmospheric pollutants, radioactivity dose rates, groundwater table level and quality, and the evolution of epidemics. Potential application fields include environmental monitoring, remote sensing, geographic information systems, brain mapping, machine learning, epidemiology, data assimilation, signal and image processing, hydrology, meteorology, mineral reserves estimation, and oil reservoir simulation.