Skip to main content

Research lays groundwork for geospatial digital twins

ORNL researchers are establishing a digital thread of data, algorithms and workflows to produce a continuously updated model of earth systems.

Digital twins are exactly what they sound like: virtual models of physical reality that continuously update to reflect changes in the real world. They have already proven their unique value in manufacturing and other industries, where they can provide real-time updates on the functionality of equipment, down to the component level. At the Department of Energy’s Oak Ridge National Laboratory, David Page thinks digital twins could revolutionize Earth observation.

“With the vast amount of satellite imagery we have access to and the geospatial datasets ORNL produces, we already have the ground truth for geospatial digital twins,” said Page, who heads the lab’s Geographic Data Science Section. “That’s not the challenge.”

Instead, the problem Page wants to solve is creating an efficient “twinning rate,” or the time it takes for the digital model to update and accurately reflect the physical entity it represents. Page thinks ORNL, with its unmatched high-performance computing capabilities and geospatial data processing workflows, is the right place to reduce the twinning rate and deliver useful geospatial digital twins.

Getting there, though, will require Page and team to overcome several challenges — starting with an established industry mindset.

Traditionally, geospatial products such as maps, terrain models and processed imagery are produced via a stove-piped, sequential workflow. One team defines requirements for a specific product, another then collects the relevant data, a third processes that data, and so on until the final product is created. Unfortunately, these geospatial products are often single-use and ephemeral, quickly becoming outdated, irrelevant or simply deleted.

Page proposes a paradigm shift away from product silos and toward data interoperability. In his vision, geospatial digital twins are constantly updated, providing a foundational basis from which any product can be pulled as needed.

ORNL has been a pioneer in geospatial data sharing and interoperability since the late 1990s, when the LandScan project sought to build reliable global population models.

“They were focused on creating population models but realized they needed to first map building footprints,” Page said. “Traditionally, the building footprint map would have been disregarded once the final population model was produced, but the ORNL team realized that with the lab’s computing resources, we could keep that data and use it for a number of research needs.” (Learn more about LandScan and its development.)

Since then, ORNL’s geospatial scientists have continually collaborated across research groups, developing unique datasets — including datasets of critical infrastructure, population density and built environments — and capabilities, such as automated feature extraction, that inform and improve each other’s efforts along the way.

Even if everyone were to embrace Page’s vision of data interoperability, geospatial digital twins still face unique challenges related to scale, complexity, observation and measurement. In particular, they need to operate across regional and global scales, requiring massive amounts of data and the computational power to process it all quickly enough to provide an accurate picture of the physical world.

“The key enabler to all these digital twins is high-performance computing, which can increase the scale of the twins and greatly reduce the twinning rate,” Page said. “ORNL has been building world-leading supercomputers for years, and we have the expertise to integrate HPC with geospatial workflows and remote sensing.”

To eventually collapse the twinning rate from years to hours, ORNL researchers started with a project to develop 3D elevation models for the National Geospatial-Intelligence Agency. Today, the world’s most commonly used elevation — or topography — data comes from the February 2000 Shuttle Radar Topography Mission and the early 2000s Advanced Spaceborne Thermal Emission and Reflection Radiometer, or ASTER.

At best, the models are 15 years out of date — not exactly an effective twinning rate. Instead, Page and team have developed novel stereoscopic correlation algorithms that identify, rectify and marry multiple images of the same location — taken from different vantage points — to provide a 3D rendering of the Earth’s topology. Combining observed and measured data with ORNL’s HPC resources, they hope to provide near-constant updates — a digital twin of the planet’s terrain.

In doing so, they’ve established a digital thread that vastly reduces the number of steps requiring human input or action.

“The digital thread integrates the right data, algorithms and functions to automate as much of the workflow process as possible,” Page said. “If you can build an end-to-end solution with a good digital thread and an efficient twinning rate, the possibilities are nearly endless.”

Unlike the one-time datasets produced via the shuttle mission and ASTER models, ORNL’s digital twin of the Earth’s terrain could be used to generate as-needed geospatial products to support mobility, humanitarian assistance and disaster relief applications.

UT-Battelle manages ORNL for the Department of Energy’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit