Researchers at the U.S. Department of Energy’s (DOE’s) National Renewable Energy Laboratory (NREL) have developed a faster, more accurate way to enhance climate data by incorporating adversarial training into machine learning.

Source: U.S. Department of Energy’s (DOE’s) National Renewable Energy Laboratory (NREL)Source: U.S. Department of Energy’s (DOE’s) National Renewable Energy Laboratory (NREL)Adversarial training involves having neural networks compete with each other to generate more realistic data. According to NREL, the researchers trained two networks, "one to recognize physical characteristics of high-resolution solar irradiance and wind velocity data and one to insert those characteristics into the coarse data." The networks get better at distinguishing between real and fake inputs over time and produce more realistic data. In this case, the result produced an additional 2,500 pixels for every original pixel.

Prior to this, machine learning techniques have been used to enhance the coarse data through super-resolution (sharpening an image by adding pixels), but this is the first time adversarial training has been used to super-resolve climate data. This technique saves computing time, data storage costs and makes the data more accessible, said Karen Stengel, an NREL graduate intern specializing in machine learning.

“To be able to enhance the spatial and temporal resolution of climate forecasts hugely impacts not only energy planning, but agriculture, transportation, and so much more,” said Ryan King, a senior computational scientist at NREL who specializes in physics-informed deep learning.

Accurate, high-resolution climate forecasts are crucial for predicting wind, cloud, rain and sea current variations for renewable energies. Short-term forecasts inform operational decision-making, medium-term forecasts are used for scheduling and resource allocation and long-term forecasts affect infrastructure planning and policy. NREL's approach can be applied to regional- and global-scale climate scenarios.

An article detailing this approach,“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.