BIGDATA: F: DKA: Collaborative Research: High-Dimensional Statistical Machine Learning for Spatio-Temporal Climate Data

点击翻译按钮获取中文摘要

While statistical machine learning has seen major advances over the past two decades, rigorous approaches for high-dimensional spatio-temporal scientific data analysis have not received as much attention. On the other hand, several core scientific areas, including climate science, ecology, environmental sciences, and neuroscience, are generating increasing amounts of high-resolution spatio-temporal data. It is vital to develop rigorous machine learning approaches for such complex high-dimensional spatio-temporal data in order for key scientific breakthroughs in these areas in the next few decades. The project contributes to these endeavors by focusing on two key technical and scientific areas: spatio-temporal big data analysis and climate science. The project systematically develops the statistical machine learning foundations for the analysis of large scale complex high-dimensional spatio-temporal data, and applies such advances to problems arising in climate science, where the total amount of data is set to cross an Exabyte (1 Exabyte = 1000 Petabytes) soon. The technical work in the project has three broad and interacting components: structured probabilistic graphical models for spatio-temporal data analysis, generalized graphical models for multivariate heavy tailed distributions, and physics-guided models with a richer class of structural constraints and capturing multi-scale phenomena. The project applies these technical advances to climate science, by generating climate projections at high-resolutions. Currently, the lack of requisite spatial resolution of current climate models makes automatic assessments of impacts, adaptation and vulnerability (IAV) difficult for a variety of sectors, including urban planning, freshwater resources, food security, energy, transportation systems, human health, and coastal systems.