| NSF Org: |
AGS Division of Atmospheric and Geospace Sciences |
| Recipient: |
|
| Initial Amendment Date: | November 3, 2017 |
| Latest Amendment Date: | November 3, 2017 |
| Award Number: | 1802627 |
| Award Instrument: | Standard Grant |
| Program Manager: |
Jielun Sun
AGS Division of Atmospheric and Geospace Sciences GEO Directorate for Geosciences |
| Start Date: | January 15, 2018 |
| End Date: | December 31, 2019 (Estimated) |
| Total Intended Award Amount: | $168,517.00 |
| Total Awarded Amount to Date: | $168,517.00 |
| Funds Obligated to Date: |
|
| History of Investigator: |
|
| Recipient Sponsored Research Office: |
660 PARRINGTON OVAL RM 301 NORMAN OK US 73019-3003 (405)325-4757 |
| Sponsor Congressional District: |
|
| Primary Place of Performance: |
OK US 73019-9705 |
| Primary Place of
Performance Congressional District: |
|
| Unique Entity Identifier (UEI): |
|
| Parent UEI: |
|
| NSF Program(s): | Physical & Dynamic Meteorology |
| Primary Program Source: |
|
| Program Reference Code(s): |
|
| Program Element Code(s): |
|
| Award Agency Code: | 4900 |
| Fund Agency Code: | 4900 |
| Assistance Listing Number(s): | 47.050 |
ABSTRACT
This study seeks to apply novel data science techniques (such as tree-based classification models and deep learning) to four-dimensional (4D) weather radar observations of thunderstorm dynamics to enable identification of storms capable of producing tornadoes up to an hour prior to tornadogenesis. Real-time severe storm prediction is a challenging task that currently requires a human forecaster with a thorough understanding of the dynamics and current state of the atmosphere. This study will develop and apply data science techniques to four-dimensional radar data from severe storms throughout the continental U.S. with the goal of identifying critical spatiotemporal relationships that can improve the understanding and prediction of tornadoes. The long-term goal will be to develop techniques to fundamentally improve our understanding of severe storms in general (including hail, wind, and tornadoes) by analyzing the new knowledge identified by the data science models.
This study seeks to advance the scientific knowledge of tornadogenesis by identifying novel precursors to tornadoes in two unique 4D weather radar datasets. Data science has the potential to advance knowledge by processing and objectively evaluating a large amount of data in a relatively short period of time. This provides a mechanism by which large, complicated meteorological datasets can be assessed for their predictive capability or alternative applications without the need for time consuming subjective evaluation. The methods developed will enable others to evaluate existing Earth system data to a spatiotemporal extent that is not possible with established approaches. The application of data science techniques to a novel domain will require the development of new techniques focusing on spatiotemporal 4D weather radar data.
PUBLICATIONS PRODUCED AS A RESULT OF THIS RESEARCH
Note:
When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external
site maintained by the publisher. Some full text articles may not yet be available without a
charge during the embargo (administrative interval).
Some links on this page may take you to non-federal websites. Their policies may differ from
this site.
PROJECT OUTCOMES REPORT
Disclaimer
This Project Outcomes Report for the General Public is displayed verbatim as submitted by the Principal Investigator (PI) for this award. Any opinions, findings, and conclusions or recommendations expressed in this Report are those of the PI and do not necessarily reflect the views of the National Science Foundation; NSF has not approved or endorsed its content.
In “EAGER: Improving our Understanding of Supercell Storms through Data Science”, we used data science/artificial intelligence methods to improve the prediction and understanding of tornadoes. The goal of this work was to improve the physical understanding of the formation of supercell storms with a specific focus on those that generate tornadoes. Our approach was to develop and apply artificial intelligence techniques, specifically deep learning, to the task of improving tornado prediction and then to develop novel interpretation methods to examine what the methods learned in order to identify new knowledge. The deep learning methods used data available for real-time forecasting, including weather radar and environmental soundings. We compared the use of 3D radar data with 2D data and also compared two different sources of radar data.
We demonstrated that deep learning could be used to significantly improve tornado prediction, at least in hindcasting, where we apply the techniques to storms from the past. Future work will examine the use of the techniques in real-time. Since the main focus of the work was to improve our scientific understanding of supercells and tornadoes, we focused much of the work on developing novel model interpretation and visualization techniques to enable atmospheric scientists to peer inside the “black box” of deep learning algorithms. The methods that we developed will be applicable to many physical science domains beyond the atmospheric sciences. The machine learning interpretation results revealed that differences in storm structure and rotation can be used to predict tornado occurrence at longer long lead times than current approaches. The models developed can be refined in future work to potentially make significant contributions to tornado prediction and understanding.
Last Modified: 02/07/2020
Modified by: Amy Mcgovern
Please report errors in award information by writing to: awardsearch@nsf.gov.
