---
license: cc-by-4.0
tags:
- croissant
- weather-forecasting
- extreme-weather
- deep-learning
- high-resolution


---

# HR-Extreme Dataset

## Overview
HR-Extreme is a high-resolution dataset designed to evaluate the performance of state-of-the-art models in predicting extreme weather events. The dataset contains 17 types of extreme weather events from 2020, based on High-Resolution Rapid Refresh (HRRR) data. It is intended for researchers in weather forecasting, encompassing both physical and deep learning methods.
[Github Link](github_link: https://github.com/HuskyNian/HR-Extreme)
## Dataset Structure
The dataset is divided into two main folders:
- `202001_202006`: Contains data from January 2020 to June 2020.
- `202007_202012`: Contains data from July 2020 to December 2020.

Each folder stores the dataset in the WebDataset format, following Hugging Face's recommendations. Every 10 `.npz` files are aggregated into a single `.tar` file, named sequentially as `i.tar` (e.g., `0001.tar`).

## Usage
To construct the dataset, use the provided scripts in the GitHub repository. The main script, `make_datasetall.py`, generates an index file for the dataset:
```bash
python make_datasetall.py 20200101 20200630