Introduction

This HydroShare Resource provides the codes, processed data, and results for the following study under review at JAMES: "Parsimonious and Transferrable Parameterization of Reservoir Operations: A Modular Approach for Large-Scale Modeling" by Donghui Li and Gabriele Villarini, in which we developed a MOdular Data-driven Reservoir Operation Model (MODROM) that can be implemented for large-scale hydrological and water resource modeling.

The files are organized as below: - readme.md: this markdown file. - code: the folder of codes for raw data retrieval, process, and analysis. - data: the folder of processed data. - results: the folder of results.

code

• benchmark.ipynb: this notebook contains the scripts to implement the benchmarking reservoir operation models compared in the study.
• fit_modrom.ipynb: this notebook contains the scripts to calibrate our MODROM with all available operation data.
• parameter_generalization.ipynb: this notebook contains the scripts to train the random forest model to transfer calibrated MODORM parameters from data-rich reservoirs to data-scarce reservoirs using static reservoir characteristics.
• gdrom_rules.py: this python script contains the "if-else" operation rules for 452 large reservoirs across the CONUS, which is used to run the GDROM, one of the benchmarking models.
• istarf.R: this R script contains the implementation of ISTARF, one of the benchmarking models.

data

• daily_records/: the folder of daily operation data.
• istarf_simulation/: the folder of ISTARF simulation, including both calibration and generalization results.
• classification_data.csv: the file of features and targets used to train the random forest model that determines which module to apply.
• GRAND_ID through USE_NAVI are obtained from GRanD, representing different reservoir characteristics
• ppt: 30-year normal precipitation obtained from PRISM.
• mode: module ID at the given month for the reservoir.
• sin_month and cos_month: harmonic function of month.
• regression_data_mode2.csv: the file of features and targets used to train the random forest model that determines the parameter value for each applied module 2 (note that module 1 parameters are implicitly contained in this as linear regression can collapse to constant value).
• regression_data_mode3.csv: the file of features and targets used to train the random forest model that determines the parameter value for each applied module 3.
• reservoir_metadata.csv: the file of reservoir metadata.
• reservoir_norm_info.json: the file of normalization information (maximum value of inflow, storage, and outflow, in acft/day) for each reservoir.

results

• calibration_metrics.csv: the calibration KGE for each reservoir.
• generalization_metrics.csv: the generalization KGE for each reservoir, including the statistics out of the 100 independent random train-test splits.
• bootstrap_metrics.csv: the KGE from the boostrap validation.
• calibrated_parameters.json: the calibrated parameters for each reservoir.
• benchmark_metrics.csv: the benchmark KGE for each reservoir across benchmarking reservoir operation models.
• final_clf.pkl: the pickle file of trained classification random forest model that is used to determine which module to apply.
• final_reg_mode2.pkl: the pickle file of trained regression random forest model that is used to determine parameter values for module 2.
• final_reg_mode3.pkl: the pickle file of trained regression random forest model that is used to determine parameter values for module 3.