EAGER: Collaborative Research: Learning Relations between Extreme Weather Events and Planet-Wide Environmental Trends

George Mason University, Fairfax VA

Investigators

Abstract

Extreme events, such as heat waves, cold spells, extreme precipitation, and severe storms, play a significant role in the loss of lives and damage to ecosystems and infrastructure, presenting fundamental challenges to sustainability. Under anticipated trends in planet-scale environmental trends, there is considerable uncertainty in the projected changes in the intensity, duration, and frequency of extreme events. Reducing these uncertainties is a grand challenge that will require substantial advances in both the environmental and data sciences. The proposed research seeks to advance both the environmental science that underpins predictions of extreme events, and the data science required to identify relations between variables in massive data sets. The results of this research will provide a basis for improving predictions of extreme events for use in sustainability planning. This project will educate and cross-train graduate students in both disciplines, allowing them to contribute to this new emerging field. The proposed research will also inform course development, and will be disseminated through tutorials, conferences, and seminars. The team's involvement with workshops, the GW Sustainability Institute, and GW Planet Forward will help to broaden the impact through public outreach. The proposed research will advance machine learning and statistical modeling of large-scale and regional events by: (1) using new tools in sparse regression in high dimensions, (2) identifying nonlinear relations in data, and (3) learning relations in spatiotemporal data that are non-stationary over space and time. The results of this research will advance understanding of extreme weather events and their relation to planet-wide environmental trends. Such relations will be learned by applying new statistical algorithms to analyze extensive climate model simulations which generate very large data sets. The findings will be validated against observations, and the learned relations will be compared between different models to assess consistency and robustness, and to validate models.