k0stal/sisr-downscaling ? reverse-engineered prompt

Build me a small research repo for downscaling climate projections to 1 km using single image super resolution.

I want notebooks for exploring the ReKIS climate data and preparing it for model training, plus a second notebook for evaluating trained models on climate indicators at weather station locations. The goal is to train on high resolution observational data, then apply the model to low resolution climate projections and compare the results against the original low resolution baseline.

Also add a script that can generate a training run with chosen model settings and hyperparameters, and a few Python training scripts that actually run the model training. Keep it easy to reproduce and focused on daily mean temperature for now. If you need current setup details for the model code or data loading, look up the latest docs online and wire it in cleanly.