Current Research Projects
Project: Deep learning for operational identification and prediction of synoptic-scale fronts
Funded Period: August 2020 - July 2023
Project Details
This project develops an automated machine learning (ML) approach for frontal analysis that will be used in a human-in-the-loop manner by forecasters. ML will provide a first-guess map for warm, cold, occluded, and stationary fronts, as well as drylines over the region covered by the Unified Surface Analysis, which is the Northern Hemisphere from 130E eastward to 10E from the Equator to 80N. The human forecasters will make use of the first-guess map to significantly reduce the forecaster time needed to produce the final map. Deep learning has recently emerged as a high-performing machine learning algorithm in a variety of fields. It is particularly well designed for tasks where the input is either images, or gridded fields. The approach will use model data and observations to create the frontal analysis, both of which are gridded fields. Performance of these methods have previously demonstrated that deep learning can be used to identify cold fronts and warm fronts in a highly skilled manner (Lagerquist et al., 2019a,b), as well as drylines (Clark et al. 2015).
Personnel Involved
Dr. Amy McGovern, Principal Investigator (University of Oklahoma)
Dr. John T. Allen, Co-Principal Investigator
Andrew Justin, MS Student in Meteorology, Formerly Undergraduate Research Assistant (OU)
Elizabeth Wawrzyniak, Undergraduate Research Assistant (CMU)
Colin Willingham, Undergraduate Research Assistant (OU)
Project Deliverables & Outcomes
Work on this project has focused on the development of the new multi-front type datasets from the Unified analysis in a form suitable for upgrades to the modeling framework and a climatology of drylines. Preliminary work has been presented at the AMS Annual Meeting by a Masters student supported by the project (Woodward et al. 2021). We have developed a four-class model that is capable of predicting warm, cold, occluded and stationary fronts using both a 2D and 3D framework (Justin et al. 2021, 2022). This has recently been published in the paper Justin et al. (2023), which provides evaluation of the performance for the WPC and Unified Surface Analysis domain. This model has been tested in an prototype environment with a set of operational forecasters at NOAA (RL5). We have transferred this approach to GFS and GDAS analysis and forecasts, completed model testing, and the system is currently implemented in the NOAA operational environment via the Google Cloud - this is now being using in NAWIPS operationally (RL7). Work is ongoing with developing the dryline model.
Publications
Justin+*, A. D., Willingham+, C., McGovern, A., and J. T. Allen, 2023: Toward Operational Real-time Identification of Frontal Boundaries Using Machine Learning. In Press, Artificial Intelligence for the Earth Systems. doi: 10.1175/AIES-D-22-0052.1
Presentations
Justin*, A. D., McGovern, A., and J. T. Allen, 2023: Operational Analysis of Frontal Boundaries using U-Nets. 22nd Conference on Artificial Intelligence for Environmental Science, x103rd Annual Meeting of the American Meteorological Society, Denver, Colorado. AMS Student Presentation Award Winner (1st Place)
Justin#, A., C., Willingham#, A. McGovern and J. Allen, 2022: Toward Operational Real-Time Identification of Frontal Boundaries Using Machine Learning: A 3D Model.21st Conference on Artificial Intelligence for Environmental Science, 102nd Annual Meeting of the American Meteorological Society, Virtual.
Justin#, A., C., Willingham#, A. McGovern and J. Allen, 2021: Toward Operational Real-time Identification of Frontal Boundaries Using Machine Learning. 3rd NOAA Workshop on Leveraging AI in Environmental Sciences, NOAA, Boulder, CO.
Woodward, A.*, McGovern, A., Allen J. T., Basara, J., 2021:Frontal Identification using a Machine Learning Model. 20th Conference on Artificial Intelligence for Environmental Science, 101st AMS Annual Meeting, Virtual.
All images and content on this website are used with permission or ©2019 John T. Allen
