Presentations

Abstract

This study explores the potential of applying deep learning techniques, specifically super-resolution (SR), to generate high-resolution wind maps from those of low resolution. Traditionally, generating high-resolution maps used in wind energy assessments, such as mesoscale and microscale maps, has relied on computationally-intensive and often cost-prohibitive models like numerical weather prediction (NWP) and large eddy simulations (LES). These maps are critical for estimating wind energy production and optimizing turbine layouts. The research focuses on using SR to improve both mesoscale (1-3 km resolution) and microscale (30-100 m resolution) wind maps, which could significantly reduce the costs and computational resources associated with traditional methods. By employing a diffusion model, a type of deep learning algorithm often used in image generation, the study demonstrates how SR can upscale coarse data to finer resolutions by an 8x factor. The model was trained on wind speed data from different sources, including the Weather Research and Forecasting (WRF) model and the ERA5 reanalysis data. Two key applications are explored: enhancing global to mesoscale and mesoscale to microscale wind maps. Initial results show that the SR model effectively captures wind patterns and outperforms traditional downscaling methods. Validation against real-world wind speed observations in Texas demonstrates the model's promise, with the SR model showing better accuracy compared to existing techniques. The research highlights SR as a cost-effective and accessible solution for improving wind resource assessments, potentially supporting the broader adoption of renewable energy. Future work will refine the model further by integrating more atmospheric variables and features.