Adversarial super-resolution of climatological wind and solar data

Area of Science:

• Climate Science
• Renewable Energy Systems
• Artificial Intelligence

Background:

• High-resolution climate data is crucial for energy resource planning but current global climate models lack necessary spatiotemporal detail.
• Existing models struggle to provide the resolution required for accurate assessment of wind and solar energy potential.

Purpose of the Study:

• To develop and validate a deep learning approach for enhancing the resolution of climate model outputs for renewable energy assessment.
• To improve the spatiotemporal accuracy of wind velocity and solar irradiance data from global climate models.

Main Methods:

• An adversarial deep learning approach was employed to super-resolve climate model outputs.
• Adversarial training was used to enhance both physical accuracy and perceptual quality of the super-resolved data.
• A fully convolutional architecture allowed for training on small domains and evaluation on arbitrarily-sized inputs.

Main Results:

• Achieved up to a [Formula: see text] resolution enhancement for wind and solar data.
• Validated that the inferred fields are robust to input noise and retain large-scale consistency.
• Demonstrated correct small-scale properties of atmospheric turbulent flow and solar irradiance in the enhanced data.

Conclusions:

• The developed deep learning method significantly enhances climate data resolution for renewable energy applications.
• The super-resolved data provides crucial insights for policy makers regarding future energy resources and infrastructure.
• The approach is scalable to global-level analysis using climate scenario data.