Upcoming National ESPC-sponsored sessions

[AGU 2018] Improvements in synoptic, subseasonal to seasonal, and decadal projections through next-generation ocean model developments, observations, and coupled data assimilation

Session: Improvements in synoptic, subseasonal to seasonal, and decadal projections through next-generation ocean model developments, observations, and coupled data assimilation

Session Date/Time: (Oral session) 13 December 2018, 4 - 6 PM ET; (Poster session) 13 December 2018, 8 AM - 12:20 PM ET

Primary Convener: Jessie Carman, NOAA Office of Oceanic and Atmospheric Research 

Conveners: Luke P Van Roekel, Los Alamos National Laboratory, Santha Akella, NASA, and Bradford Johnson, Trivector Services/NOAA OAR

From weather to climate timescales, detailed atmospheric, oceanic, and land-surface predictions are critical to socioeconomic planning, mitigation, and emergency response. Earth System Models (ESMs) for synoptic, S2S, and decadal predictions depend on accurate ESM component initialization and accurate representations of critical ocean, sea ice, and wave processes. Higher resolutions and regional/component impacts on predictability complicate these dependencies; emerging high-performance computing (HPC) architectures require new algorithms for efficient system use. Implementing CDA in ESMs and improvements to component models, particularly at high resolution and in regionally refined configurations, represents a grand scientific, algorithmic, and computational challenge. Meeting these challenges efficiently within operational run-time constraints will require restructured ocean and atmosphere models that are performant on new HPC systems. This session seeks submissions discussing: novel theoretical and algorithmic techniques for CDA in ESMs, the use of CDA to understand ESMs biases, novel ocean model discretizations, and parameterizations of multiscale phenomena.

[AMS 2019] Assessing, defining, and communicating predictability on the subseasonal to seasonal timescale

Session: Assessing, defining, and communicating predictability on the subseasonal to seasonal timescale

Session Date/Time: 9 January 2019, 10:30 AM - 12 PM MT

Bradford Johnson, Trivector Services, Inc./NOAA/OAR, National ESPC, Silver Spring, MD; Jessie C. Carman, OAR, Silver Spring, MD and David McCarren, Navy/CNMOC, Silver Springs, MD

Various industries, agencies, and municipalities are increasingly dependent upon weather information beyond two weeks in efforts to leverage opportunities, mitigate socioeconomic impacts, and solidify the resilience of infrastructure. Traditionally, weather and climate information is relayed to end users in different manners. Weather forecasts under two weeks are usually presented in the form of diurnal extrema with a focus on relatively specific temporal windows for impact events, whereas, information on the seasonal or longer scale is shown relative to climatological averages and trends. While both of these techniques have useful components, neither has proven to be completely sufficient when coupled with current capabilities on the subseasonal to seasonal (S2S) timescale defined by the Weather Research and Forecasting Innovation Act of 2017 as two weeks to two years. Through the National Earth System Prediction Capability partnership is focused on improving predictive capability in S2S. Model developers seeking to meet this challenge can benefit from access to best practices in the dissemination of weather information. In the development of an effective communication framework, this session highlights current work focused on predictability in S2S through quantitative and qualitative approaches designed to meet user needs. The session seeks speakers with expertise in evaluation and definition of predictability on weather and climate scales with techniques geared to S2S portability as well as speakers with experience and perspectives in research to operations and operations to research.

[AMS 2019] Developing and Preparing Weather and Climate Models for Exascale

Session: Developing and Preparing Weather and Climate Models for Exascale

Session Date/Time: 7 January 2019, 2 - 4 PM MT

Mark W. Govett, NOAA/ESRL/GSD, Global Systems Division, Boulder, CO; David McCarren, Navy/CNMOC, Silver Springs, MD; John Michalakes, UCAR/NRL, MMD, Monterey, CA and Philip W. Jones, LANL, T-3, Los Alamos, NM

Continued advancement in weather and climate prediction models depend on (1) increasing spatial resolution, (2) increasing the number of ensemble members, (3) additional physical processes (eg. ocean, land, chemistry, etc) that were unaffordable decades ago, and (4) increasing the scale and accuracy of data assimilation to incorporate billions of observations collected by next-generation, high-resolution satellites, radars, and a myriad of in situ sensors worldwide. It will also require next generation exascale systems with at least 1000 times more computing power than used currently.

However, massive increases in computing power will not be sufficient to overcome significant obstacles in running such models on exascale systems.  Issues include under- or unexposed parallelism, load-imbalance, and critical dependence on scarce memory bandwidth (low computational intensity).  Most weather prediction models use less than 5 percent of the peak computational capabilities of today’s CPU chips, and less than 2 percent of the latest generation GPU and MIC chips. Future chips are expected to continue this downward trend of diminishing computational benefit.  Growing awareness of these problems led to formation of a multi-agency (NOAA, NASA, NCAR, DoE, DoD) working group in the Earth System Prediction Capability (ESPC), and spawned meetings on the topic with industry.

While hardware improvements will help, effective utilization of next generation systems with millions of compute cores will require adapting and rewriting applications to better prepare them for exascale. Weather and climate models are complex, multi-component models often designed separately, and then coupled or linked together. Lack of overall system design can limit parallelism, and make it difficult to understand, modify, test, and maintain them.  In addition, the scientific algorithms, grids, and formulations used are motivated by ease of development and scientific accuracy of the solution at the expense of computational efficiency, portability and maintainability. In the exascale era, a balance between accuracy and computational efficiency may be needed to further advance capabilities of our earth system models.

To address these challenges requires cooperation and collaboration between modeling teams, computational experts and software designers. We invite your submissions in a wide range of topics including model and assimilation development, algorithms, software design, languages, high performance I/O, performance portability, and computing hardware.