O3B: Machine Learning


Moderator:
TBD

Advisors: (proposed)
Enrico Camporeale (CU/SWx TREC)
Gunes Baydin (Oxford University)


This cluster focuses on the research, development, and operational implementation of probabilistic machine learning systems for space weather forecasting.
 

Activities:

  • Development of large deep learning models for prediction of solar, magnetospheric, and Ionospheric-Thermospheric-Mesospheric (ITM) system states.
  • Development of community datasets and ML code frameworks for large-scale training sets with a focus on DA-enabled “reanalysis” runs of large-scale physics-based models. 
  • Establishment of common metrics of evaluation of ML model prediction skill using community-based standard test sets across all domains of space weather.
  • Interface with the Data Assimilation (DA) community to develop data-driven forecasting models with integrated DA systems. 
  • Work with instrument designers, data systems engineers, and storage and distribution engineers to develop requirements for NASA mission data to be “ML ready” early in Phase E operations.
  • Work with critical infrastructure operators and operational forecasters to understand the requirements for space weather product types, latency, accuracy, reliability, and other metrics of “end user success”.
  • Work with commercial storage, compute, and AI/ML framework providers to integrate space weather ML model development into current and future system plans.  


Action topics:

  • Improve space weather modeling and prediction capabilities through the implementation of machine learning.
  • Explore the synthesis of ML and DA methodologies (Cluster O3A activity) including purely data-driven forecast models with novel real-time data ingest and inference methodologies.
  • Work with the community to develop ML data storage formats, standard data sets, accessibility requirements, and training methodologies specific to space weather prediction problems.
  • Interface with the operational space weather forecasting enterprise to understand forecaster and end-user requirements, particularly in the field of LEO satellite navigation and collision avoidance. 
  • Facilitate the development of new instrumentation concepts for low-latency data suitable for data assimilation into existing and future forecasting models. 

 

Action Teams: 

O3B-1 SWORD – Space Weather Operational Readiness Development Center (Lead (proposed): Yang Chen, UMichigan)

O3B-2  TBD