Fix/gfs preprocessing

[AoufNihed]

GFS Preprocessing Fix: Longitude Range and Dimension Naming

Problem

The GFS data preprocessing had several issues:

1. Longitude range mismatch (-180° to 180° vs 0° to 360°)
2. Dimension naming conflict between 'variable' and 'channel'
3. Inefficient chunking strategy affecting performance

Solution

Implemented fixes in gfs_preprocessing.py and gfs_dataset.py:

1. Longitude Range Standardization:

   • Convert all longitudes to [0°, 360°) range
   • Added sorting to ensure consistent ordering
   • Updated UK region selection to handle wrap-around (350° to 2°)

2. Dimension Structure:

   • Standardized on 'channel' dimension name
   • Properly stacked variables into channel dimension
   • Added validation checks for required dimensions

3. Performance Optimization:

   • Implemented efficient chunking strategy
   • Set optimal chunk sizes for different dimensions
   • Added cleanup of existing chunk encoding

Testing

The changes can be verified by:

1. Loading preprocessed data and checking longitude range
2. Confirming dimension names and structure
3. Verifying UK region selection

[AoufNihed]

Hey @peterdudfield
GFS Preprocessing: Fix Longitude Range and Dimension Structure

• Fixed longitude range to [0°, 360°] for consistent UK region selection
• Standardized dimension naming to use 'channel' instead of 'variable'
• Optimized chunking for better performance
• Added data validation checks

Tested with sample GFS data and verified UK region selection works correctly.