A Computationally Efficient Method for Fitting Big Data Spatio-Temporal Log-Gaussian Cox Processes

• Activity Number: 274 - Statistical Analysis of Large Time Series of Remotely Sensed Environmental Measurements: Tales of the Collision Between Statistics, Remote Sensing and Geo-Computation
• Type: Topic Contributed
• Date/Time: Tuesday, August 4, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Megan Gelsinger*
• Keywords: big data; spatio-temporal; EM; computational efficiency

Abstract

In this work we propose a new computationally efficient method for fitting big data spatio-temporal log-Gaussian Cox processes (LGCPs). Large spatio-temporal data sets are ubiquitous today, appearing in fields such as finance, epidemiology and meteorology. Unfortunately, computer power has not kept up with the pace of data collection, requiring researchers to develop new computationally efficient ways of fitting standard models, such as LGCPs, to big data. There exist several methods for fitting big spatial/spatio-temporal LGCP models; however, for reasons of approachability, interpretability, and scalability, we felt the field was still lacking a high-performing method. To this end, we propose a computationally efficient procedure for fitting big data LGCPs which leverages the concepts of EM and Laplacian approximations to make for a faster and more user-friendly procedure compared to our competitors’.