Staff Notes Daily Announcements

Interpretable Deep Learning for Severe Weather Prediction and Stochastic Parameterization

David John Gagne II
NCAR

The likelihood of tornadoes, hail, and severe winds is associated with the spatial structure of a thunderstorm and its near-storm environment. Existing statistical and machine learning modes for predicting these hazards, however, do not explicitly encode spatial information and focus on aggregate statistics. Deep learning models include convolutional spatial encoding layers that can learn storm structure. In this study, deep learning methods are evaluated for severe hail forecasting and are compared with less complex methods. An experiment is conducted with storm patches extracted from the NCAR convection-allowing ensemble to predict the probability of microphysics-derived severe hail over an hour from upper-air dynamic and thermodynamic fields. The convolutional neural network significantly outperforms the other models in Brier Skill Score. The convolutional neural network was interpreted to identify important input variables and storm structures. The neural network encoded physical storm features, such as wind shear, strong lapse rates, and favorable flow patterns for hail growth. 
 
Machine learning and deep learning models also have the potential to serve as sub-grid parameterizations for numerical weather and climate models. An initial feasibility study for this approach has been conducted with the Lorenz ‘96 chaotic dynamical system model. Generative adversarial networks (GANs), a technique for using one neural network to train another neural network to produce conditional samples of arbitrary distributions, are used to model the sub-grid forcing in the model. The GANs are compared with polynomial regression on both weather and climate time-scales. The GANs can approximate the true forcing distribution and with stochastic noise layers can produce ensemble error and spread that is competitive with the baseline model.

Thursday, May 31
11:00 a.m.-12:00 p.m.
FL2-1022

Posted by Jessa Johnson at ext. 2751, jessaj@ucar.edu

Thursday, May 17, 2018 to Thursday, May 31, 2018

The new Gmail interface will be enabled for our organization on Monday, May 21st. At that time staff can choose to enable the new Gmail interface in their Gmail settings; when the new Gmail is available, the first item under the Settings menu will be “Try the new UCAR Gmail”. If you do not see this menu item, it means the new Gmail is not available yet.

The new Gmail brings a new look and introduces new features. However, some announced features have still not been released by Google. Some new features should be used with caution as they may sound more secure than they really are. UCAR Cybersecurity Program Office (CPO) and the Google Admins would like to stress the following points about the new Gmail 'confidential mode':

- Confidential mode is not a viable replacement for PGP-style encryption. Staff who require secure messaging should continue to leverage PGP and other applications like Box for Gov and Acellion.
- Confidential mode unfortunately promotes clicking on links in email. Staff should always be cautious when doing that. If in doubt, always verify with the sender that they sent you a trusted link.
- Messages sent via confidential mode can be viewed in Google Vault even after they expire.

Once the new Gmail is enabled, staff will be able to switch back and forth between the new Gmail and the Classic Gmail interface. At some point in the future we expect the Classic Gmail will go away and all accounts will be automatically moved to the new Gmail by Google. Some helpful information about the new Gmail can be found by doing an internet search for "Learn what's new in Gmail" or at the following Google Support URL: https://support.google.com/mail/answer/7677724?utm_source=gsuite&utm_medium=blog&utm_campaign=new_gmail

Posted by CISL Help at ext. 2400, cislhelp@ucar.edu

Thursday, May 17, 2018 to Tuesday, May 22, 2018

A Statistical Approach towards improving Predictability of Cold Season Precipitation Type and Amount for the HRRR-TLE

Paul J. Roebber
Distinguished Professor, Atmospheric Sciences Group
Department of Mathematical Sciences
University of Wisconsin-Milwaukee
Milwaukee, Wisconsin

In this DTC-sponsored study we explore a method to improve the predictability of cold season precipitation type for the operational High-Resolution Rapid Refresh Model Time-Lagged Ensemble (HRRR-TLE) through application of unique postprocessing techniques. TLEs are a computationally efficient method for improving probabilistic forecasts as the differences between model runs can provide an additional measure of initial condition uncertainty. Here, given that precipitation will occur, we apply evolutionary programming using HRRR-TLE forecast inputs for each of the three time lags to construct logistic regression equations calculating the probability of rain (pRN), probability of mixed precipitation (pMX), probability of freezing rain (pZR), probability of ice pellets (pIP), and the probability of snow (pSN). These equations are derived for 5 regions from 100°W eastward across the CONUS. These probabilities are then bias corrected using a decaying average process and optimal weights for each time-lagged ensemble member are developed using Bayesian Model Combination (BMC). These forecasts provided enhanced probabilistic information for both the areal distribution of cold season precipitation and the timing and location for precipitation phase transitions.
 
Friday June 15, 2018
9:00-10:00
FL2-1001

Posted by Jessa Johnson at ext. 2751, jessaj@ucar.edu

Wednesday, May 16, 2018 to Friday, June 15, 2018

Let's call it a day on Friday June 15, 3:00 - 6:00 PM at the Foothills Lab Courtyard!

Please join the EAC and help kick the summer off with an all-staff party.

Appetizers and refreshments will be served. Bring your families and take some time to relax with your colleagues!

We look forward to seeing you there!

For more information, contact Toni Wallace at ext. 8716, twallace@ucar.edu

Posted by Toni Wallace at ext. 8716, twallace@ucar.edu

Wednesday, May 16, 2018 to Friday, June 15, 2018

Boulder’s Open Space and Mountain Parks (OSMP), together with the Boulder Fire-Rescue Department (BFRD), are planning an 85-acre prescribed burn on Shanahan Ridge this month. The burn area is southeast of the Mesa Lab.

The burn will take place some time between Wednesday, May 9, and the end of the month. The exact date will be based on wind and weather conditions, and firefighters will monitor the burn at all times until it is fully extinguished. Ignition could occur as early as 10:00 a.m. OSMP and BFRD will notify the community, including UCAR, prior to ignition.

Smoke from the burn area will likely be visible from the Mesa Lab and other parts of town. The burn area is over a mile south of the Mesa Lab and neither firefighters nor firefighting equipment are expected to be accessing the NCAR property during the burn.

For more information, please visit the City of Boulder web page for planned burns.

If you have questions or concerns, please feel free to contact Dave Maddy at maddy@ucar.edu or 303-497-1134.

Posted by Alexandra Branscombe at ext. 8609, abran@ucar.edu

Wednesday, May 9, 2018 to Thursday, May 31, 2018

Facing a medical challenge can be stressful enough without worrying whether you have the right diagnosis or the most effective treatment—especially when you consider that, according to the Institute of Medicine, it is likely that most of us will experience at least one diagnostic error in our lifetime.

UCAR is pleased to offer you the innovative medical benefit called Best Doctors.

Best Doctors is a better kind of second opinion service. It lets you get advice from the world’s leading physicians on everything from minor surgery and maternity challenges to more serious issues like cancer and heart disease. If you’re uncertain about a diagnosis or treatment, a Best Doctors clinician will take another look at your case and provide a confirmation or recommend a change.

What’s more: Best Doctors is 100% confidential and cost-free. It’s included in your benefits package and available at no charge to you and any of your dependents enrolled in the UCAR health plans – both Cigna and Kaiser throughout the United States. You don’t even have to make doctors’ appointments or travel. All of the Best Doctors services are provided over the phone or the Internet.

Please visit the Best Doctors website and the Best Doctors Guide for additional information.

Posted by Laurie Carr at ext. 8702, lcarr@ucar.edu

Tuesday, May 8, 2018 to Friday, May 25, 2018

CISL is asking for input from researchers in order to gauge their interest in Phase 6 of the Coupled Model Intercomparison Project (CMIP6) and planned use of CISL’s CMIP Analysis Platform.

NCAR data products are expected in mid-2018 and will be available to CMIP Analysis Platform users on the GLADE file system. As estimated time frames for availability of data from other modeling centers become known, CISL will publish that information for the community to help researchers make their plans. When other data sets are available, researchers who use the CMIP Analysis Platform will be able to submit requests to have the data sets they need added to GLADE.

Researchers who are or may be planning to use CMIP6 data are asked to respond to this survey by Friday, May 25, and are encouraged to share the survey with others.

Posted by B.J. Smith at ext. 1273, bjsmith@ucar.edu

Monday, May 7, 2018 to Friday, May 25, 2018

Parents’ Night Out is a great opportunity for you and your significant other to go out to dinner and a movie!

Contact Stephanie Ivancic, Director of KinderCare Learning Center at UCAR, to sign your child up on Friday, May 11 and/or Friday, May 25 from 6 pm - 8 pm ($30 for one child or $40 for two children) or 6 pm - 10 pm ($50 for one child and $60 for two children).

Children are welcome to wear their pajamas for a fun evening of movies and pizza! If your child has a video that he or she would like to share, bring it along! Please make sure it is labeled and let the fun begin! Siblings under the age of 7 are welcome, too.

We invite all UCAR employees to take advantage of this fun evening as your child does not need to be enrolled at the KinderCare Learning Center at UCAR to partake!

Please contact Stephanie Ivancic for enrollment as well as drop-in care rates. Stephanie’s email address is sivancic@cclc.com and her phone number is 303-443-5595.  The KinderCare Center is located at 3050 34th Street in Boulder.

Posted by Laurie Carr at ext. 8702, lcarr@ucar.edu

Monday, May 7, 2018 to Friday, May 25, 2018

TIAA Individual Counseling Sessions for May 2018 have been scheduled:

Tuesday, May 8, 2018
FL2 - Room 1002
Schedule Appointment

Thursday, May 24, 2018
RAF - Library
Schedule Appointment

SIGN UP TODAY: 
To schedule a counseling session: Click the link above or call their Scheduling and Service Group phone reservation center at: 1-800-732-8353. 

These sessions are in high demand, so if you have to miss your appointment, be sure to contact TIAA as soon as possible so they can fill your time slot.

Posted by Alyssa Fronk at ext. 8710, afronk@ucar.edu

Thursday, May 3, 2018 to Tuesday, May 22, 2018

April 30, 2018 - June 29, 2018 

Mixed Media Drawing and Watercolor by Ileana Barbu  

Ileana Barbu’s formations as a sculptor infused her artistic view with the fascination of three-dimensionality and the constant impulse of understanding forms in all angles. Drawing is an obsession for her – line and form in all their aspects have been a constant in her artwork. Her body of work can be viewed as a conglomerate of different directions, but she always returns back to line or linear compositions. Even when she “paints,” what really matters to her is the underlying structure. The paintings are, in fact, colored drawings. Ileana is fascinated with the human figure in all its aspects: physical, spiritual, natural, social and historical-temporal. Lately, she has been drawn to the images of trees – but also through the prism of human expression. Each “tree” is in fact another facet of herself.  

Mixed Media Painting by Frederick Pichon  

Frederick Pichon’s paintings reflect his personal interests in design, architecture, science and history. He is grounded in the pictorial tradition of Western art and perspective, but this body of work is contemporary. As a skilled draftsman, he loves precise drawing, but tries to be open to other stylistic expressions, so his work may straddle abstraction and representation, and incorporate materials, collages, computer images and more. His images express a position on man in the universe, both humanistic and technical to reflect on our condition. A good piece is a meditative work that opens on a different level of reality. 

Art Reception: Saturday May 5th from 4:00 pm to 6:00 pm in the Mesa Lab cafeteria.

All artists work will be on display in the UCAR Community Art Galleries at the NCAR Mesa Lab. 

Preview all artists work on the NCAR/UCAR Community Art Calendar http://scied.ucar.edu/exhibits/community-art-program

 

 

 

 


Posted by Audrey Lewis at ext. 2570, alewis@ucar.edu

Monday, April 30, 2018 to Friday, June 29, 2018

Pages