Hybrid Statistical/Machine Learning Deep Dynamical Spatio-Temporal Models for Evaluating Climate Impacts

• Activity Number: 151 - Beyond the VAR: Advances in Spatial and Spatio-Temporal Modeling for Climate and Environmental Data
• Type: Invited
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Christopher K. Wikle*
• Companies: University of Missouri
• Keywords: spatio-temporal; deep models; dynamic; reservoir computing; climate impacts

Abstract

Climate impacts a wide range of socio-demographic, biological, and physical processes across multiple spatial and temporal scales of variability. Most impact studies focus primarily on summary measures that do not explicitly account for the fact that impacts are driven by multiscale dynamical processes. Unfortunately, we often do not have complete understanding of the mechanistic components of such processes. This talk will focus on methods that can efficiently model such processes and provide a more realistic sense of potential impacts and their uncertainties across multiple time and spatial scales. In particular, it will consider recent advances in reservoir computing methods that have been adopted in a spatio-temporal statistical framework.