In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance

This repository contains all code, data, and analyses for the paper "In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance" published in WRR.

Contents of this repository

• website/ -- This folder contains the code to build and run the study website that we used to collect hydrograph ratings.
• rmh-meta-stats.ipynb -- This Jupyter notebook contains statistics about participation and demographic data.
• rmh-stats.ipynb -- This Jupyter notebook contains the analyses of model ranking.
• rmh-classifier-metrics.ipynb -- This Jupyter notebook contains the code to train a Random Forest on classifying rating outcomes.
• rmh-metrics-vs-hydrographs.ipynb -- This Jupyter notebook contains the code to compare a model trained on metrics vs. a model trained on raw hydrographs.
• rmh-cycle-analyses.ipynb -- This Jupyter notebook contains the consistency analyses.
• data/ -- This folder contains all data used in the study, as well as csv files with the collected ratings from study phases 1 and 2.

The simulated and observed hydrographs used in this study are from the "The Great Lakes Runoff Intercomparison Project Phase 4: the Great Lakes (GRIP-GL)". The full data repository for GRIP-GL is available here.

Contact

Martin Gauch: gauch (at) ml.jku.at

Citation

@article{gauch2023metrics,
    author = {Gauch, Martin and Kratzert, Frederik and Gilon, Oren and Gupta, Hoshin and Mai, Juliane and Nearing, Grey and Tolson, Bryan and Hochreiter, Sepp and Klotz, Daniel},
    title = {In Defense of Metrics: Metrics Sufficiently Encode Typical Human Preferences Regarding Hydrological Model Performance},
    journal = {Water Resources Research},
    year = {2023},
    volume = {59},
    number = {6},
    pages = {e2022WR033918},
    url = {https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2022WR033918},
    doi = {10.1029/2022WR033918}
}