How adversarial machine learning can lead to far better climate data
- By Susan Miller
- Jul 08, 2020
Researchers at the Department of Energy’s National Renewable Energy Laboratory (NREL) are using a machine learning technique called adversarial training to quickly enhance the resolution of climate data up to 50 times. The accurate, high-resolution data is then better suited for assessing renewable energy resources, like wind and solar power, whose efficiency varies by time and location.
The current lack of up-to-date, high-resolution data for running different renewable energy scenarios has been a major challenge for energy resilience planning. It makes it difficult to conduct day-to-day decision-making; develop medium-term weather forecasts for scheduling and resource allocations and build long-term climate forecasts for infrastructure planning and policymaking.
Machine learning has been used to enhance coarse data through a process called super resolution -- sharpening a fuzzy image by adding pixels -- but until now, adversarial training techniques had not been used to “super-resolve” or improve the resolution climate data.
Adversarial training improves the accuracy of an algorithm by teaching it to recognize bad or adversarial input. It can also improve “the performance of neural networks by having models compete with one another to generate new, more realistic data,” lab officials said.
In this case, one model was trained to recognize physical characteristics of high-resolution solar irradiance and wind velocity data, and the other was trained to insert those characteristics into the coarse data. Over time, the networks produced more realistic data and were better able to distinguish between real and fake inputs. In the end, “the inferred high-resolution fields are robust, physically consistent with the properties of atmospheric turbulence and solar irradiation,” the researchers wrote in their paper.
Eventually, the NREL researchers were able to add 2,500 pixels for every original pixel, lab officials said.
“By using adversarial training -- as opposed to the traditional numerical approach to climate forecasts, which can involve solving many physics equations -- it saves computing time, data storage costs, and makes high-resolution climate data more accessible,” said Karen Stengel, an NREL graduate intern who specializes in machine learning.
This approach can be applied to a wide range of climate scenarios from regional to global scales, changing the paradigm for climate model forecasting, the researchers said.
The paper, “Adversarial super-resolution of climatological wind and solar data,” appears in the journal Proceedings of the National Academy of Sciences of the United States of America. Software for the trained networks as well as some example data are available on GitHub.
Susan Miller is executive editor at GCN.
Over a career spent in tech media, Miller has worked in editorial, print production and online, starting on the copy desk at IDG’s ComputerWorld, moving to print production for Federal Computer Week and later helping launch websites and email newsletter delivery for FCW. After a turn at Virginia’s Center for Innovative Technology, where she worked to promote technology-based economic development, she rejoined what was to become 1105 Media in 2004, eventually managing content and production for all the company's government-focused websites. Miller shifted back to editorial in 2012, when she began working with GCN.
Miller has a BA and MA from West Chester University and did Ph.D. work in English at the University of Delaware.
Connect with Susan at [email protected] or @sjaymiller.