Revisiting urban heat indices in Switzerland using low-cost measurement networks

Materials to reproduce the results of the article "Revisiting urban heat indices in Switzerland using low-cost measurement networks" (arXiv:2606.09364).

Pre-trained bias correction model weights are available on the Hugging Face Hub.

Requirements

Install pixi. All other dependencies are managed automatically by pixi.

Instructions to reproduce

The workflow is managed with Snakemake and executes Jupyter notebooks via papermill.

Note: Before running the pipeline, download the URS dataset zip from BORIS and place it in data/raw/. Snakemake will extract it automatically.

To reproduce all results:

pixi run snakemake results --cores 1

Here is a schematic overview of the pipeline:

flowchart LR
    raw_lcd(["LCD raw data\n(per network)"])
    raw_parallel(["Intercomparison\nmeasurements"])
    extents(["Spatial extents\n(per city)"])

    subgraph meteo ["1. Meteorological data"]
        get_aws["get-aws-data\n(per city)"]
        get_lcd["get-lcd-data\n(per network)"]
    end

    subgraph bias ["2. Bias correction"]
        agreement["agreement-metrics"]
        train["train-bias-correction"]
        apply["apply-bias-correction\n(per network)"]
    end

    subgraph indices ["3. Heat indices"]
        heat["heat-indices"]
    end

    raw_lcd --> get_lcd
    raw_parallel --> agreement
    raw_parallel --> train
    extents --> get_aws
    extents --> get_lcd
    get_aws --> apply
    get_lcd --> apply
    train --> apply
    get_aws --> heat
    get_lcd --> heat
    apply --> heat

    heat --> results(["Results"])
    train --> results
    agreement --> results

Loading

Citation

@misc{bosch2026revisiting,
      title={Revisiting urban heat indices in Switzerland using low-cost measurement networks},
      author={Martí Bosch and Moritz Burger},
      year={2026},
      eprint={2606.09364},
      archivePrefix={arXiv},
      primaryClass={physics.ao-ph},
      url={https://arxiv.org/abs/2606.09364},
}

Acknowledgments

• Based on the cookiecutter-data-snake 🐍 template for reproducible data science.