PrExDA: Predicting Extremes by Data-Driven Analytics

Extreme events and associated hazards in natural, commercial, and security systems underlie the most devastating catastrophes. The societal risks arising from extreme events rank high both in terms of likelihood and impact. The increasing impact of disasters and the consequent risks have led worldwide efforts to develop new approaches to mitigate the impacts and develop strategies for resilience. The worldwide community of experts and practitioners in academia, government, industry and non-profits have been developing initiatives for collaboration, knowledge sharing, and innovation in identifying and assessing disaster risk. Fundamental to such strategies is the harnessing of the comprehensive data of extreme events underlying the natural and anthropogenic hazards and enhancing the predictive capability.

Improving the predictive capability for extreme events is a critical scientific need in order to achieve effective disaster risk assessment and resilience. Disaster risk is a product of three main factors: the likelihood of the underlaying events, vulnerability of the system and consequences therein. Among these factors the understanding and modeling of extreme events yield the probabilities of their occurrences. A key objective of modeling and prediction of extreme events is to provide reliable estimates of their probabilities, which are used in risk assessment. The probability estimates are usually based on the first-principles of the specific phenomena and because of the complex set of processes interacting across scales in space and time and prediction of extremes based on this approach are continuing challenges. The data-driven models are based on extracting the leading features inherent in the data, independent of modeling assumptions, and provide a framework for predicting extreme events.

The low-probability, high-impact nature of extreme events makes the development of their predictive capability a continuing imperative. The data-driven analytics, with its system-based approach, provides a natural framework to harness the unprecedented and massive data of multiple systems to yield improved predictive capability for extreme events.

Extreme events are an emergent property of many complex, nonlinear systems in which interdependent components and their interactions lead to a competition between organized and irregular behavior. Recently the modeling of such phenomena from the complex systems perspective using observational data have enabled development of models, without a priori modeling assumptions, and have shown high predictive capabilities. Thus complexity science provides a transdisciplinary framework for integrating the theories, models and data across disciplines for convergence research of extreme events.

The workshop will target four deliverables: 1) bringing together teams from different sectors and communities and form cohorts with common goals and objectives; 2) identify data availability and data needs for different systems; 3) proof-of-concept of integrative data-driven modeling of extreme events in multiple systems; and 4) proto-type prediction system for forecasting and risk assessment.

Surja Sharma (Chair)
University of Maryland, College Park, MD
Michael Bonadonna
NOAA / FCMSSR, SWORM
Dilly Devi
Adla Research, Bethesda, MD
Jan Eichner
Munich RE, Munich, Germany
Eugenia Kalnay
University of Maryland, College Park, MD
Venkat Krishnamurthy
George Mason University, Fairfax, VA
Tony Mannucci
Jet Propulsion Laboratory, CalTech, Pasadena, CA
Raj Pandya
American Geophysical Union, Washington, DC
Xi Shao
NOAA STAR and University of Maryland, College Park, MD
Jeffrey Smith
NASDAQ Research, Rockville, MD
Nicholas Watkins
London School of Economics, London, UK

The NSF Convergence Accelerator is a new organizational structure at NSF to accelerate the transition of convergence research into practice. The Convergence Accelerator brings teams together in a cohort, with time-limited tracks that focus on grand challenges of national importance that require a convergence research approach requiring the merging of ideas, approaches and technologies from widely diverse fields of knowledge. Convergence Accelerator teams are required to be interdisciplinary in their composition and research approach and expected to leverage multi-sector partnerships across academia, industry, government, non-profit and other sectors. FY 2019 was the pilot year for the NSF Convergence Accelerator program. The CA program is structured to include a Phase I (9 months) for Planningand a Phase II (2 years) for Implementation. Each year, the CA releases a solicitation which includes the description of the tracks for that year.

NSF Convergence Accelerator RFI Workshops. The NSF Convergence Accelerator Office invites a set of “Request for Information (RFI)” workshops each year on specific topics that have the potential to be considered as new Convergence Accelerator Tracks in the following, or subsequent, years.

For FY 2019, the first year of the program, tracks were selected based on NSF's Big Ideas -- with Track A on Open Knowledge Networks and Track B on AI and Future Jobs and the National Talent Ecosystem.

For FY 2020, tracks were selected based on RFI Workshops conducted in the September-November 2019 time frame, resulting in Track C: Quantum Computing and Track D: Enabling AI Innovation via Data and Model Sharing.

For FY 2021, workshops are being conducted in the September-October 2020 timeframe, to help define CA track for the FY 2021 solicitation/competition.

Workshop Expectations. The NSF CA RFI Workshops are expected to generate “crisp” reports, immediately after the meeting, incorporating all the material needed by NSF to determine whether a particular topic is suitable as a CA Track for the next fiscal year.

Workshop organizers and participants should enter the meeting knowing that NSF is seeking actionable outcomes from the meeting that would assist NSF in making its track selection decision.


The CA Program Structure. Once tracks are selected, NSF will issue a combined solicitation for Phase I and Phase II of the CA program for the next fiscal year.

Pre-proposals. A 2-page pre-proposal is required from everyone interested in submitting a Phase I full proposal. Pre-proposals are reviewed and only those that are invited are allowed to submit a full proposal for Phase I.

Phase I full proposals. In addition to intellectual merit and broader impact, Phase I proposals are judged on key CA-specific criteria including, relevance to the Track topic; convergence research (deep multidisciplinary partnerships); multi-sector partnerships across academia, industry, government, non-profits, etc.; and a clear set of deliverables by the end of Phase II. In FY19 and FY20, Phase I project budgets could be up to $1M for 9 months.

Phase I execution. During the 9 months of Phase I, projects are expected to solidify their efforts in identifying key users and user communities for their idea; firming up multi-sector partnerships; and defining and developing prototypes to help clarify Phase II deliverables. Projects members are expected to undergo a Convergence Accelerator Curriculum that includes modules on design thinking and user-centered design, team science, and engagement with domain experts from industry and elsewhere for their respective tracks.

Phase II proposals. At the end of Phase I, all Phase I projects are eligible to submit a Phase II Implementation proposal for up to $5M for 2 years (up to $3M in Year 1 and up to $2M in Year 2). In addition to the regular NSF review panel, Phase II projects are required to make a 10-minute pitch to a separate Pitch Panel, consisting of experts who are not from academia—they may be from industry, government, non-profit sector, etc. The Phase II funding decision is based on the outcomes from both reviews. The number of Phase II projects funded depends on the quality of the proposals (as judged by outcomes from both the review panels) and availability of funding.

Phase II Execution. Project are expected to work on their deliverables over the 2 years of Phase II. Project progress will be reviewed every quarter, along with a formal review process at the end of Year 1 to determine the funding for Year 2.


Workshop

The Workshop was a virtual meeting held 30 Sep - 2 Oct, 2020. Connection information was provided to registered participants.

9 Dec 2020: 2020 Workshop Preliminary Report (pdf)

30 Sep 2020: Schedule (pdf)

All times are US Eastern (EDT, UTC-4)

Session A

  • 09:00 am Surja Sharma, University of Maryland
    Welcome, Workshop Plan and Tasks
     
  • 09:10 am Chaitan Baru, NSF Office of Integrated Activities
    NSF Convergence Accelerator Program
     
  • 09:30 am Nick Watkins, London School of Economics
    From Rocket Science to Anomalous Time Series: Concepts, history, applications and inference
     
  • 10:00 am Jan Eichner, Munich RE
    (Re-)Insurance's view on extreme events and how they are managed
     
  • 10:30 am Dolores Knipp, University of Colorado
    A Historical Perspective on Space Weather Effects on Communication & Navigation Signals
     
  • 11:00 am Misha Sitnov, Johns Hopkins University Applied Physics Lab
    Empirical reconstruction of extreme geomagnetic storms: Breaking the data paucity curse
     
  • 11:30 am Misha Balikhin, University of Sheffield
    NARMAX modelling and forecasting with multiple data sets
     
  • 12:00 pm Discussion
     
  • 12:15 pm Lunch Break
     

Session B

  • 01:00 pm Surja Sharma, University of Maryland
    Prediction and Predictability of Complex Systems
     
  • 01:30 pm Juan Valdivia, University of Chile
    Combined System Science and Machine learning for space physics
     
  • 02:00 pm V. Krishnamurthy, George Mason University
    Prediction of intraseasonal climate and extreme events
     
  • 02:30 pm Discussion: Complex Systems Framework for Modeling and Prediction
     
  • 03:00 pm Adjourn
     

Session A

  • 09:00 am M S Santhanam, IISER Pune
    Extreme events in correlated series and on complex networks
     
  • 09:30 am Leon Wei, University of Sheffield
    Data-Driven Modelling and Prediction using Transparent, Interpretable and Parsimonious Machine Learning
     
  • 10:00 am Reinaldo Rosa, INPE, Sao Jose dos Campos
    Modeling and predicting extreme events from p-model and RNN-LSTM: limitations and perspectives
     
  • 10:30 am Ian Richardson, NASA GSFC and University of Maryland
    Solar wind drivers of extreme space weather
     
  • 11:00 am Lauren Orr, University of Warwick
    Directed network modelling of geomagnetic activity
     
  • 11:30 am Simon Wing, JHU Applied Physics Lab
    Using information theory to improve predictive modeling
     
  • 12:00 pm Lunch Break
     

Session B

  • 01:00 pm Eugenia Kalnay, University of Maryland
    How can we improve Predictability in Earth System Models (not just for Climate)?
     
  • 01:30 pm Erin Lynch, NOAA STAR and University of Maryland
    Ensemble forecasting of extreme events
     
  • 02:00 pm Eviatar Bach, University of Maryland
    Overcoming the curse of dimensionality: Combining data-driven forecasting with physical models for Earth system prediction
     
  • 02:30 pm Discussion
     
  • 03:00 pm Adjourn
     

Session A

  • 09:00 am Xi Shao, NOAA STAR and University of Maryland
    NOAA NPP VIIRS imaging data of natural hazards
     
  • 09:30 am Dimitris Vassiliadis, NOAA
    NOAA Spacecraft for solar wind monitoring
     
  • 10:00 am Raj Pandya, American Geophysical Union
    Using community priorities to guide actionable science: Examples from Thriving Earth Exchange
     
  • 10:30 am Surja Sharma, University of Maryland
    Integrating Modeling, Prediction and Predictability
     
  • 11:00 am Discussion: Ideas and concepts for convergence ecosystems
     
  • 11:30 am Framework for partnerships
     
  • 12:00 pm Adjourn
     

Participation

The workshop is open to participation by interested professionals in academia, government, industry, non-profit and other organization. Participants from underrepresented minority groups will be encouraged. Please send a brief expression of interest, including your contact details, professional experience, and areas of interest to the workshop organizers at exeventswx@gmail.com.

Code of Conduct

The organizers of the workshop are committed to providing an environment free from all forms of discrimination, harassment, and retaliation. Recognizing the need for an atmosphere that encourages the free expression and exchange of ideas, the workshop is dedicated to the philosophy of equality of opportunity and treatment for all members, regardless of gender, gender identity or expression, race, color, national or ethnic origin, religion or religious belief, age, marital status, sexual orientation, disabilities, veteran status, or any other reason not related to professional merit. Harassment, sexual or otherwise, is a form of misconduct that undermines the integrity of the workshop.

The workshop will adopt a code-of-conduct to foster harassment-free environment that encourages the free expression and exchange of ideas. The policies and requirements of the University of Maryland for a harassment-free environment will be followed at the workshop and related activities. The commitment of the University of Maryland in fostering harassment-free environment is supported by the Office of Civil Rights and Sexual Misconduct (OCSRM) and these policies and guide-lines are accessible at https://www.ocrsm.umd.edu/about/index.html. The Department of Astronomy at the University provides support in the implementation of harassment-free environment and the resources and policies including reporting processes, accessible at https://www.astro.umd.edu/EDI/EDIResourcePage.html#Harassment.

The workshop organizers will adopt procedure of reporting complaints and follow-up action consistent with the University of Maryland policy and procedures detailed at the UMD OCSRM website. Any individual participating in the workshop who believes that he or she or they has been subjected to harassment should submit the incident for review. All complaints will be treated seriously and be investigated promptly. Confidentiality will be honored to the extent permitted as long as the rights of others are not compromised. The reporting options will include online, telephone and email. The follow-up process will be designed, consistent with the University of Maryland policy, to be fair and avoid potential conflicts of interest.