Advanced Computational Tools
Machine Learning, Exascale Computing and Data Cyberinfrastructure Positioned to Transform Prediction of Earth System Behavior
Our ability to collect and create diverse Earth and environmental data – at scales of seconds to decades, and from microns to thousands of kilometers – far outpaces our ability to assimilate it, much less improve our predictive understanding of complex, multi-scale Earth system behavior. These data sample hydrological, biological, geochemical, geological, atmospheric, and geomechanical properties using a variety of platforms, including omic’, wellbore, fiber optic, surface geophysical, and UAV and satellite approaches. In the past several years, EESA scientists have launched several new projects and have greatly advanced data archiving, modeling, and machine learning (ML) tools to advance predictive understanding using such diverse data.
Examples aligned with EESA’s ‘Scale-aware data and simulation capability’ Cross-Cutting platform include mechanistic subsurface simulation capabilities that can take advantage of emerging exascale computers and that use adaptive mesh refinement methods to ‘telescope’ into a system where higher modeling resolution is required, as well as cyberinfrastructure to enable long-term preservation and public dissemination of multiscale, diverse data, along with rapid analysis and synthesis capabilities.
Examples aligned with EESA’s new Cross-cutting Technology of Machine Learning include methods to rapidly characterize the organization of a landscape, provide near-real time indications of hazards, and downscale forcing data. ML has also been used to advance hybrid modeling approaches that take advantage of both mechanistic models and data-driven approaches to extract insights and improve prediction of complex phenomena using diverse datasets. EESA scientists have developed and tested these approaches over the last few years for problems ranging from detection of extreme events, to quantifying ecosystem and watershed hydrobiogeochemical behavior to advancing understanding of subsurface geomechanics and flow. During 2019, EESA scientists co-chaired each of the four DOE ‘AI for Science’ Townhall Earth and Environmental sessions, which were held across different regions of the U.S.
Photo Credit: Berkeley Lab
Recent science & program advances
ML approaches to:
- Predicted groundwater fluctuations using weather predictions
- Generated difficult-to-observe watershed model inputs from sparse, proxy atmospheric observations
- Characterized the organization of permafrost systems and their associated capacity for carbon fluxes
- Produced precision analytics of extreme events, including tropical cyclones and atmospheric rivers
- Solved data challenges associated with ultra-dense fiber optic monitoring systems
- Development of TECA (Toolkit for Extreme Climate Analytics) framework
Relevant Projects
- Exasheds: First systematic approach to advance watershed prediction using ML and exascale technologies
- ESS-DIVE: Preserving, expanding access to, and improving usability of critical data generated through DOE environmental system science projects
- BASIN 3D: Data broker for enabling synthesis of watershed time-series observations, tools for semi-automated QA/QC, and interactive visualizations
- IDEAS (Interoperable Design of Extreme-Scale Application Software)
- CASCADE: Calibrated & Systematic Characterization, Attribute and Detection of Extremes
- Exascale Computing Project: Earthquakes
- Exascale Computing Project: Subsurface Geochemistry
- ALTEMIS: Advanced Long-Term Environmental Monitoring Systems
- SMART: AI for Fossil Energy
- DOE Early Career Project: Data Integration, Analytical, and Machine-Learning Framework to Understand an Predict the Resilience of Watershed Water Quality
- Watershed Function SFA: Scale-Aware and Functional Zone Approaches, BASIN-3D data integration broker and end-end cyberinfrastructure for managing sensor network data
- AR1K: Co-benefits of soil carbon sequestration for agriculture, healthy soils, and climate using AI
- SMARTFARM: ML approaches to advance carbon accounting for agriculture
- RESONANTHPC: HPC-Enabled pre- and post-processing with Jupyter (partnership with Berkeley Lab and Kitware)
Partners
EESA benefits from rich partnerships with our collaborators and sponsors. See project & program links above for more information.
Publication Highlights
Surrogate Optimization of Deep Neural Networks for Groundwater Prediction, Muller et al., 2020.
In situ monitoring of groundwater contamination using Kalman Filter, Schmidt et al., 2018.
End-to-End system for acquisition, management and integration of diverse data, Varadharajan et al., 2019.
Launching an accessible archive of environmental data, Varadharajan et al. 2019.
Emerging technologies and radical collaboration to advance predictive understanding of watershed hydrobiogeochemistry, Hubbard et al., 2020.
Plant species distribution estimated using hyperspectral data, Falco et al., 2020.
Wavelet-based local mesh refinement for rainfall–runoff simulations. Journal of Hydroinformatics, Özgen-Xian et al., 2020.
Sequential Imputation of Missing Spatio-Temporal Precipitation Data Using Random Forests. Front. Water 2:20, Mital et al., 2020.
News Coverage
Using ML to estimate COVID-19s Seasonal Cycle
Fiber Optics & Machine Learning to Advance Safe and Renewable Energy
Advancing Watershed System Understanding through ML and Exascale
Assessing Regional Earthquake Risks in the Age of Exascale
Supercomputing potential impacts of Quake by Building Location and Size
To pump or not to pump: New tool will help water managers make smarter decisions
Machine Learning provides early warning system for tracking contaminant plumes
Berkeley Lab-developed Digital Library is a Game Changer for Environmental Research
A U.S. Department of Energy National Laboratory Managed by the University of California
Lawrence Berkeley National Laboratory · Earth and Environmental Sciences Area · Privacy & Security Notice