stefan-it/Groundsource

---
license: cc-by-4.0
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: uuid
dtype: string
- name: area_km2
dtype: float64
- name: geometry
dtype: binary
- name: start_date
dtype: string
- name: end_date
dtype: string
- name: __index_level_0__
dtype: int64
splits:
- name: train
num_bytes: 893933735
num_examples: 2646302
download_size: 667210042
dataset_size: 893933735
---

# 🌊 Groundsource - A Dataset of Flood Events from News

This datasets hosts the Groundsource dataset from Google and mirrors it from [Zenodo](https://zenodo.org/records/18647054) to make it accessible within the Hugging Face awesome ecosystem.

> High-quality historical flood data is critical for disaster risk management, infrastructural planning,
and climate change attribution, however, existing global archives are constrained by sparse geograph-
ical coverage, coarse spatial resolution, or reliance on prolonged satellite observation. To address this
gap, we introduce Groundsource, an open-access global dataset comprising 2.6 million high-resolution
historical flood events, curated from the automated processing of over 5 million news articles across
more than 150 countries. Our methodology leverages Gemini large language models (LLMs) to sys-
tematically extract structured spatial and temporal data from unstructured journalistic text. Compre-
hensive technical validation demonstrates that the pipeline achieves an 82% practical precision rate in
manual evaluations. Furthermore, spatiotemporal matching against established external databases
reveals recall capturing 85% to 100% of severe flood events recorded in the Global Disaster Alert and
Coordination System (GDACS) between 2020 and 2026. By transforming unstructured global news
media into a structured, localized event archive, Groundsource provides a massive-scale, extensible
resource to support the training of predictive hydrological models, quantify historical exposure, and
advance global disaster research.

# Usage

To load and use the dataset:

```python
from datasets import load_dataset

ds = load_dataset("stefan-it/Groundsource")

print(ds)

# Outputs
#DatasetDict({
# train: Dataset({
# features: ['uuid', 'area_km2', 'geometry', 'start_date', 'end_date', '__index_level_0__'],
# num_rows: 2646302
# })
#})
```

# References

* [Introducing Groundsource - Google Blogpost](https://research.google/blog/introducing-groundsource-turning-news-reports-into-data-with-gemini/)
* [Groundsource Paper](https://eartharxiv.org/repository/view/12083/)
* [Zenodo Dataset](https://zenodo.org/records/18647054)

# Citation

Please make sure you cite the original dataset (taken from Zenodo):

```bibtex
@dataset{mayo_2026_18647054,
author = {Mayo, Rotem and
Zlydenko, Oleg and
Bootbool, Moral and
Fronman, Shmuel and
Gilon, Oren and
Hassidim, Avinatan and
Kratzert, Frederik and
Loike, Gila and
Matias, Yossi and
Nakar, Yonatan and
Nearing, Grey and
Sayag, Reuven and
Sicherman, Amitay and
Zemach, Ido and
Cohen, Deborah},
title = {Groundsource: A Dataset of Flood Events from News},
month = feb,
year = 2026,
publisher = {Zenodo},
doi = {10.5281/zenodo.18647054},
url = {https://doi.org/10.5281/zenodo.18647054},
}
```

license: CC BY 4.0
configs:
- 配置名称: default
数据文件:
- 划分: train
路径: data/train-*
dataset_info:
特征字段:
- 名称: uuid
数据类型: 字符串（string）
- 名称: area_km2
数据类型: float64
- 名称: geometry
数据类型: 二进制（binary）
- 名称: start_date
数据类型: 字符串（string）
- 名称: end_date
数据类型: 字符串（string）
- 名称: __index_level_0__
数据类型: 64位整数（int64）
数据集划分:
- 划分名称: train
字节数: 893933735
样本数: 2646302
下载大小: 667210042
数据集总大小: 893933735

# 🌊 地面源（Groundsource）—— 基于新闻的洪水事件数据集

本数据集托管谷歌发布的Groundsource数据集，并从[Zenodo](https://zenodo.org/records/18647054)镜像至Hugging Face生态，以便在该平台上获取。

> 高质量的历史洪水数据对于灾害风险管理、基础设施规划以及气候变化归因研究至关重要，但现有全球档案存在地理覆盖稀疏、空间分辨率粗糙，或依赖长期卫星观测等局限。为填补这一空白，我们推出Groundsource——一个开放获取的全球数据集，包含2646302条高分辨率历史洪水事件记录，其数据源自对150余个国家超500万篇新闻文章的自动化处理。我们的方法借助Gemini大语言模型（Large Language Models），从非结构化的新闻文本中系统性提取结构化的时空数据。全面的技术验证显示，该处理流程在人工评估中可达到82%的实际精准率。此外，与成熟外部数据库的时空匹配结果表明，其召回率可覆盖2020年至2026年全球灾害预警与协调系统（Global Disaster Alert and Coordination System，GDACS）记录中85%至100%的严重洪水事件。通过将非结构化的全球新闻媒体内容转化为结构化的本地化事件档案，Groundsource提供了一个大规模、可扩展的资源，可用于支持水文预测模型的训练、历史暴露量量化以及全球灾害研究的推进。

# 使用方法

加载并使用该数据集的代码如下：
python
from datasets import load_dataset

ds = load_dataset("stefan-it/Groundsource")

print(ds)

# 输出
#DatasetDict({
# train: Dataset({
# features: ['uuid', 'area_km2', 'geometry', 'start_date', 'end_date', '__index_level_0__'],
# num_rows: 2646302
# })
#})

# 参考文献

* [谷歌官方博客：Groundsource 正式发布](https://research.google/blog/introducing-groundsource-turning-news-reports-into-data-with-gemini/)
* [Groundsource 学术论文](https://eartharxiv.org/repository/view/12083/)
* [Zenodo 数据集存档](https://zenodo.org/records/18647054)

# 引用规范

请引用该数据集的原始版本（来自Zenodo）：
bibtex
@dataset{mayo_2026_18647054,
author = {Mayo, Rotem and
Zlydenko, Oleg and
Bootbool, Moral and
Fronman, Shmuel and
Gilon, Oren and
Hassidim, Avinatan and
Kratzert, Frederik and
Loike, Gila and
Matias, Yossi and
Nakar, Yonatan and
Nearing, Grey and
Sayag, Reuven and
Sicherman, Amitay and
Zemach, Ido and
Cohen, Deborah},
title = {Groundsource: A Dataset of Flood Events from News},
month = feb,
year = 2026,
publisher = {Zenodo},
doi = {10.5281/zenodo.18647054},
url = {https://doi.org/10.5281/zenodo.18647054},
}