Wind Power in a Warming World: Diffusion-Based Downscaling of CMIP6 Climate Projections for Future Energy Planning

The rapid expansion of wind power is essential for global decarbonization, yet the industry relies on historical wind data for site selection and financing based on expected power generation — an approach that is becoming unreliable in our warming world. To bridge this gap, we propose a novel deep learning framework leveraging Denoising Diffusion Probabilistic Models (DDPMs) to bias-correct and downscale coarse-resolution CMIP6 climate projections. Our model generates 5km wind speed and power projections across Europe by training on historical wind data from ERA (low resolution) and CERRA (high resolution) while integrating remotely sensed terrain variables across spatial scales. Validated against the historical record, our approach surpasses traditional downscaling methods, improving bias correction and spatial fidelity. Applying this technique to NASA-NEX-GDDP CMIP6 climate simulations, we generate high-resolution projections of wind power density under SSP5-8.5, offering a valuable dataset for optimizing wind energy deployment.