BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000713Z
LOCATION:405-406-407
DTSTART;TZID=America/Denver:20231116T133000
DTEND;TZID=America/Denver:20231116T140000
UID:submissions.supercomputing.org_SC23_sess175_pap105@linklings.com
SUMMARY:Rapid Simulations of Atmospheric Data Assimilation of Hourly-Scale
  Phenomena with Modern Neural Networks
DESCRIPTION:Yiyuan Li, Xiting Ju, Yi Xiao, Qilong Jia, and Yongxiao Zhou (
 Tsinghua University, China); Simeng Qian (National Supercomputing Center i
 n Wuxi); Rongfen Lin (National Research Center of Parallel Computer Engine
 ering and Technology, China); Bin Yang (Tsinghua University, China); Shupe
 ng Shi (National Supercomputing Center in Wuxi); Xin Liu, Jie Gao, Zhen Wa
 ng, Sha Liu, Jian Tan, and Xuan Wang (National Research Center of Parallel
  Computer Engineering and Technology, China); Zhengding Hu (University of 
 Science and Technology of China); Limin Yan (Beijing Sankuai Online Techno
 logy Co, Ltd; National Supercomputing Center in Wuxi); and Wei Xue (Tsingh
 ua University, China; Department of Computer Technology and Application, Q
 inghai University)\n\nAtmospheric data assimilation is essential for numer
 ical weather prediction. Ensemble-based data assimilation connects multipl
 e instances of atmospheric model through Kalman-filter-based algorithm, wh
 ich is regarded as a challenging computing task today. In this work, we pr
 esent our efforts to build a fast, low-cost, and scalable atmospheric data
  assimilation prototype for the new-generation Sunway supercomputer, inclu
 ding (1) A UNet-neural-network-based surrogate model for atmospheric dynam
 ic simulation to generate all the background ensemble with both satisfacto
 ry accuracy and reasonable robustness; (2) Batched LETKF with an efficient
  eigenvalue decomposition implementation and a data staging strategy to co
 ver the observation IO time ; (3) A framework able to flexibly deploy the 
 components, thus available to reach the maximum resource efficiency. Exper
 imental evaluations show that our AI-integrated ensemble data assimilation
  prototype can finish hour-cycle assimilation in minutes, keep linear scal
 ability and save an order of magnitude of computing resources compared wit
 h the traditional scientific method.\n\nTag: Artificial Intelligence/Machi
 ne Learning, Applications, Modeling and Simulation, State of the Practice\
 n\nRegistration Category: Tech Program Reg Pass\n\nSession Chair: Wei Xu (
 Brookhaven National Laboratory)\n\n
END:VEVENT
END:VCALENDAR