A deep learning framework capable of simultaneously downscaling near-surface wind components and estimating hub-height wind fields has been developed to improve future wind power projections, according to a research study published in the Journal of Big Data. By resolving horizontal spatial downscaling and vertical wind height estimation concurrently, the framework aims to enhance the accuracy of wind energy resource assessments conducted in the context of global climate change.
Climate projections are widely utilized across the energy sector to project future wind power yields and assess the broader impacts of climate change on wind energy resources. Furthermore, standard climate model outputs deliver wind velocity components only at a near-surface height of 10 meters above the ground, rather than at the 100-meter hub heights where modern commercial wind turbines actually operate.
To overcome these persistent limitations, the authors developed a deep learning architecture that avoids empirical vertical extrapolation entirely.
The core of the proposed system is a ResUNet-based neural network architecture trained on coarsened ERA5 atmospheric reanalysis data covering the geographical region of Greece. This unified deep learning setup enables the model to take coarse atmospheric inputs and reconstruct high-resolution wind fields at a 0.25-degree resolution simultaneously at both the near-surface 10-meter level and the operational 100-meter hub-height level.
Validation of the ResUNet model demonstrated marked accuracy improvements over the traditional power-law baseline in estimating hub-height wind fields. The deep learning framework achieved a 24.38 percent reduction in root mean square error (RMSE) and a 42.78 percent reduction in mean absolute error (MAE) when estimating hub-height wind speeds. Published on November 11, 2026, in the Journal of Big Data, the study was shared as an open-access, peer-reviewed accepted manuscript subject to final production editing.
Subject of Research: Deep learning for simultaneous wind downscaling and hub-height estimation
Article Title: Joint near-surface wind downscaling and hub-height estimation for wind power projections
Article References: Moira, A., Karozis, S., Karakitsou, E., Zarikos, I., Gounaris, N., & Sfetsos, A. (2026). Joint near-surface wind downscaling and hub-height estimation for wind power projections. Journal of Big Data. https://doi.org/10.1186/s40537-026-01569-4
Image Credits: AI Generated
DOI: 10.1186/s40537-026-01569-4
Keywords: deep learning, ResUNet, wind power projections, hub-height estimation, near-surface wind downscaling, ERA5 reanalysis, wind energy resources, climate projections, wind power law, vertical extrapolation, wind speed modeling, elevation mapping, land use data, Greece, spatial resolution, root mean square error, mean absolute error
News Source: Scienmag.
