Climate–Energy Extremes Analysis Codebase

This repository contains the code used in the manuscript:

Climate change drives imbalance spikes, generation droughts, and chronic seasonal synchronization in renewable power systems.

The study quantifies multi-timescale energy-system extremes driven by physical climate variability and change, using data-driven methods and generative learning models.

---

How to cite

If you use this repository, please cite the accompanying manuscript (Currently Under Review in Science Series Journals) and include a link to this repository and the commit hash (or a tagged release).

---

Overview

A conceptual illustration of the study workflow:

---

System requirements

Operating systems

Tested on:

• Linux (Ubuntu)
• macOS

Windows should also work for most analyses, but paths may require minor adjustments.

Software

• Python 3.10+ recommended
• JupyterLab (recommended) or Jupyter Notebook

Typical Python dependencies

Exact dependencies depend on which modules you run (plotting, ML training, geospatial, etc.). The code uses standard scientific Python tooling such as:

• numpy, pandas, scipy, xarray, netCDF4 / h5py
• matplotlib (and/or plotly) for figures
• scikit-learn
• torch (PyTorch) for train_srgan/ and train_unet/
• tqdm

Recommended for reproducibility: add a pinned environment file at the repo root (e.g., requirements.txt or environment.yml) and record the OS + Python version used for the final paper runs.

Hardware

• No GPU is required to run most analysis + plotting code once inputs are prepared.
• GPU recommended only if you re-train SRGAN / U-Net from scratch.
• For full-scale analysis on the 1 TB dataset, we recommend ≥16–32 GB RAM and sufficient local storage.

---

Data

Google Drive bundle (required for full reproduction)

The full analysis depends on a Google Drive dataset (~1 TB):
https://drive.google.com/drive/folders/1nR9cPL55tvpurUy_4ExEhLzU9bbjwgeF?usp=sharing

It includes:

• 2018–2022 hourly data (renewable generation, load, meteorology, hydrology)
• County-level hourly loads (2018)
• Future weather projections (CMIP6-based inputs)
• AI model training checkpoints (where applicable)

Recommended local layout

After download, place the Drive contents into a single directory and point the code to it via an environment variable:

export CLIMATE_EXTREMES_DATA="/path/to/drive_bundle"

If a module/script expects a specific subfolder layout, document it in that module’s header or in a short CONFIG.md.

Data sources

Upstream datasets and preprocessing are described in the paper and Supplementary Information (SI). Some inputs were obtained using:

• Herbie (HRRR reanalysis): https://github.com/blaylockbk/Herbie
• AWS CLI (AWS archives): https://aws.amazon.com/cli/
• PyDrive (Google Drive utilities): https://pythonhosted.org/PyDrive/

Third-party terms: Some raw inputs may have separate terms-of-use. Users are responsible for complying with upstream dataset licenses/requirements.

---

Running the code

Quick start: generate paper/SI figures

1. Install dependencies (see above).
2. Download the required subset (or full) data bundle and set CLIMATE_EXTREMES_DATA.
3. Launch Jupyter:

jupyter lab

4. Run the notebooks/scripts in plot_examples/.

Expected outputs

• Figures are typically written to figs/ (or a script-specific output directory).
• Intermediate tables/arrays may be written alongside each module or under a dedicated outputs folder.

Typical runtime (ballpark)

• Figure generation using prepared inputs: minutes to ~1 hour per full figure set, depending on I/O and compute.
• Model training from scratch (SRGAN/U-Net): hours to days (GPU-dependent).

Re-training models (optional)

• train_srgan/: weather super-resolution SRGAN training scripts/configs
• train_unet/: U-Net training for renewable prediction tasks
• model/: shared architectures and loss functions

If you only need to reproduce the paper’s plots/results, you typically do not need to re-train models if checkpoints are provided in the data bundle.

---

Repository structure

Folder	Description
county/	County-level metadata (region mappings, FIPS codes, etc.)
figs/	Figures used in the paper (and default output location for plots)
future/	Code for future energy/climate scenario projections (e.g., 2030–2050)
hydro/	Hydropower-related analysis (including SI components)
load/	Load data processing (commercial/residential/industrial/transport)
model/	Shared model architectures and loss functions
plot_examples/	Scripts/notebooks to generate main-text and SI figures
train_srgan/	SRGAN training scripts/configs
train_unet/	U-Net model training scripts/configs
utils/	Shared utilities (loading, normalization, evaluation, plotting helpers)

---

How to cite

Please cite the accompanying manuscript. For software citation, include:

• the GitHub repository URL, and
• the specific commit hash (or a tagged release)

---

Contact

For questions, reproducibility issues, or missing-path documentation, please open a GitHub issue in this repository.