Partial Recovery of Meter-Scale Surface Weather
Downscaling coarse-resolution weather forecasts to meter-scale resolution using deep learning
Partial Recovery of Meter-Scale Surface Weather
This work addresses the challenge of downscaling coarse-resolution weather forecasts to meter-scale resolution, enabling highly localized weather predictions for applications in agriculture, urban planning, and renewable energy.
The Challenge
Modern numerical weather prediction models typically operate at resolutions of several kilometers, which is insufficient for many practical applications requiring meter-scale precision.
Our Approach
We develop a deep learning-based super-resolution framework that:
- Leverages convolutional neural networks for hierarchical representations
- Incorporates topographic and land-use information
- Integrates physical constraints into the learning process
- Provides probabilistic predictions with confidence intervals