OOS 25: Using Machine Learning to Quantify and Improve Earth System Predictions

Session Organizer(s)

• Forrest Hoffman

  Civil & Environmental Engineering, Oak Ridge National Laboratory
  Oak Ridge, Tennessee, United States

Moderator(s)

• Forrest Hoffman

  Civil & Environmental Engineering, Oak Ridge National Laboratory
  Oak Ridge, Tennessee, United States

Volunteer(s)

• Yanjun Yang

  University of Kentucky
  Lexington, Kentucky, United States

Predictions of environmental change are influenced by global carbon, water, and nutrient cycles; climate interactions; and feedbacks to the Earth system. Relevant processes operate at a large range of spatial and temporal scales and vary across terrestrial, coastal, and marine ecosystems. Feedbacks to the Earth system are driven by natural and anthropogenic disturbance agents, acceleration of nutrient and hydrological cycles, eutrophication, acidification, changes in climate and weather extremes, land cover and land use change, and potential climate intervention strategies. At the same time, increased emphasis on environmental sustainability and infrastructure resilience have driven dramatic increases in observational data and large-scale monitoring networks, exploitation of rapidly advancing computational capacity, and advanced artificial intelligence and machine learning methods applied to improving Earth system predictions. This session focuses on characterizing and reducing uncertainties of ecological feedback mechanisms that improve Earth system understanding and predictive ability through the development and application of data-driven approaches, including data mining, machine learning, data assimilation, and hybrid machine learning-/process-based Earth system modeling. Our ability to acquire, process, and store observational data and generate simulation output greatly exceeds our capacity to intelligently assimilate and apply those data to improve our knowledge of the dynamic Earth system. New methods and approaches are required to disentangle complex ecological interactions that have emergent impacts on the larger Earth system.

Presentations:

• ON DEMAND

  Bayesian and hybrid machine learning modeling for improving predictability of streamflow in data-scarce watersheds
  

  Presenting Author: Dan Lu – Computational Sciences and Engineering Division and Climate Change Science Institute, Oak Ridge National Laboratory
• ON DEMAND

  Combining remote sensing and machine learning to predict current and future vegetation community distributions on the Seward Peninsula, Alaska
  

  Presenting Author: Venkata Shashank Konduri – Northeastern University
• ON DEMAND

  Machine learning to investigate storage and dynamics of soil organic carbon
  

  Presenting Author: Umakant Mishra – Sandia National Laboratories
• ON DEMAND

  Machine learning to predict peatland greenhouse gas emissions
  

  Presenting Author: Yuanyuan Huang – CSIRO Oceans and Atmosphere
• ON DEMAND

  PROcess-guided deep learning and DAta-driven modelling (PRODA) to uncover key patterns and mechanisms in global soil carbon cycle
  

  Presenting Author: Feng Tao – Department of Earth System Science, Tsinghua University
• ON DEMAND

  Quantifying the drivers and predictability of seasonal changes in African fire
  

  Presenting Author: Jiafu Mao – Environmental Sciences Division and Climate Change Science Institute, Oak Ridge National Laboratory