Spatial Temperature Correction with a Vision Transformer
Using a vision transformer to correct synthetic temperature maps at finer resolution.
Why it matters
Spatial downscaling examines how a model represents finer detail from a coarse grid. This demonstration uses synthetic temperature maps rather than observed climate records.
My contribution
I built a vision-transformer demo that corrects and downscales synthetic temperature maps.
Result
The code and demo page show the workflow and synthetic outputs.
Tools
- Python
- PyTorch
- Vision transformer
- Spatial downscaling
These synthetic temperature maps are not validated heatwave forecasts or tests on observed climate records.
Project gallery
Synthetic demonstration figures; these do not show field performance. 4 images. Select an image to view it full size.
Technical details
Independent portfolio demonstration
A coarse temperature grid cannot show the same spatial detail as a finer grid.
- Create synthetic temperature maps.
- Convert an 8×8 input grid to a 32×32 output grid.
- Inspect the corrected maps.