Staff Notes Daily Announcements

We will have a table set up near the cafeteria at ML on Friday, January 24, where you will be able to purchase a selection of drinkware with the NCAR pastel print (check, cash, or discretionary account key only).

• Friday, January 24, 11:30-1:30 Mesa Lab Cafeteria

Other opportunities to purchase items:

• Friday, January 31, 11:30-1:30 Center Green Cafeteria
• Friday, February 7, 111:30-1:30 Foothills Lab Cafeteria

If you are at a campus without a cafeteria, please email or call Peggy Stevens (x8601) in UCAR Communications to discuss how best to serve your campus. We are also offering staff the opportunity to order:

• LL Bean vests branded with the NCAR|UCAR seal

All items sold will be offered at cost and will not financially benefit NCAR or UCAR. We will not facilitate the return of any items. If you have questions, comments, or suggestions, please let Peggy Stevens or Tracy Baker know.

CISL Visitor Program (CVP) Seminar - On the way to robust methods for training Artificial Neural Networks for time integrating PDEs

Speaker: Dr. Martin Schreiber, Technical University of Munich

Time integration of PDEs to simulate the atmosphere is a non-trivial task once considering wallclock time as well as accuracy constraints. There are claims about machine learning with neural networks being able to approximate high-dimensional spaces as well as non-linearities due to the activation functions used. This motivates an investigation of their suitability to overcome time step restrictions in the context of dynamical cores. 

This work targets exploiting neural networks for temporal integration of PDEs using large time step sizes beyond the CFL condition. We design the underlying neural network with an inspiration from exponential integrators: Instead of using arbitrary off-the-shelf neural networks in a blackbox fashion, we will design neural networks to reflect as a first step the underlying linear terms.First, purely linear one-dimensional problems will be discussed and problems of high errors when using state-of-the-art gradient-based optimizers, which interestingly already exists for linear optimization of neural networks. As a first step towards a robust training, this will be overcome with a reformulation to a linear optimization problem and using a preconditioned conjugate gradient solvers as an optimizer. As a second part, possible extensions to non-linear problems will be discussed. 

Biography:

Dr. Martin Schreiber started his doctorate in 2010 at the Technical University of Munich where he developed new and cutting edge algorithms for dynamical adaptive meshes leading to certain quasi-optimal properties and highly efficient parallelization on shared as well as distributed memory systems, hence being ready for future (exascale) HPC architectures. He has investigated and collaborated on resource-aware programming on all levels of computer hardware and software development. This has led to various multi-disciplinary publications. In 2015 he was appointed as a proleptic lecturer at the University of Exeter in the Department of Mathematics. Here, he initialized collaborations across a wide field of areas and also represented the university in the OpenPOWER group. In 2018 he rejoined TUM as a researcher and lecturer. His main research focus is on novel time integration methods, including parallel-in-time methods, for large-scale massively parallel high-performance computing architectures enhancing climate and weather simulations.

Date: Tuesday, January 28, 2020 Time: 2:00 p.m. – 3:00 p.m. Location: Mesa Lab, Chapman Room

The High Performance Storage System (HPSS) tape archive that CISL manages became read-only for most users on Monday, January 20. The only exceptions are for a small number of curated collections such as the Research Data Archive. This is a preliminary step toward HPSS decommissioning in 2021.

While writing HPSS files is no longer possible, users can still read their existing HPSS files and perform most common metadata operations on their HPSS holdings, including deleting, renaming, and moving files. Those who have not already done so should begin deleting files that are no longer needed and moving other data to alternative storage systems.

CISL has published documentation and training for users on recommended processes for identifying and organizing HPSS holdings; copying files that need to be preserved to another storage resource; and deleting files that are no longer needed. Follow these links:

Migrating files from HPSS - documentation

Training video and slides

Additional CISL Daily Bulletin announcements will follow as new tools, training, and expanded documentation become available to support the data migration. In the meantime, please contact CISL for advice on individual workflows and storage options.

This week summary: ~400 appearances in the news media with ~$400,000 in equivalent advertising value

Notable news stories

An EPIC way to improve weather forecasts (UCAR, Antonio Busalacchi)

Boston Globe

*******

Australia's Wildfires Are Releasing Vast Amounts Of Carbon (ACOM, Rebecca Buchholtz)

NPR

*This story was republished in dozens of radio outlets.

*******

2019 Was a Record Year for Ocean Temperatures, Data Show (CGD, Kevin Trenberth)

New York Times

*Multiple outlets have also published articles about this study, including The Guardian and CNN.

*******

Firenadoes, ember attacks and megafires: Australia is seeing sci-fi weather (MMM, Janice Coen)

Washington Post

*******

Colorado has seen the largest increase in natural disasters, study says (RAL, Jason Knievel)

Loveland Reporter-Herald

*******

What Is Ozone, and Why Are Colorado’s Levels So Bad? (ACOM, Gabi Pfister and Frank Flocke)

5280 Magazine

*******

House panel approves space weather bill (UCAR, Antonio Busalacchi)

Space News

*******

Social media highlight

Social media posts from the AMS meeting in Boston, including tweets that highlight booth talks and swag, have been viewed by more than 12,000 users.