WEATHER-5K: A Large-scale Global Station Weather Dataset Towards Comprehensive Time-series Forecasting Benchmark

全球站点天气预报（GSWF）对于各个领域都非常重要，包括航空、农业、能源和灾害准备。深度学习的最新进展通过优化基于公共气象数据的模型，显著提高了天气预报的准确性。然而，现有的GSWF优化和基准数据集仍存在显著限制，如数据规模较小、时间覆盖有限以及缺乏全面的变量。这些缺点阻碍了它们有效地反映当前预测方法的基准，并未能支持实际天气预报的实际需求。为了解决这些挑战，我们提出了WEATHER-5K数据集。该数据集包括来自全球5,672个气象站的全面数据集，覆盖了10年的时间段，每小时一个间隔。它包括多个关键天气元素，为预测提供了更可靠和可解释的资源。此外，我们的WEATHER-5K数据集可以作为全面评估现有知名预测模型的基准，超越GSWF方法，支持未来时间序列研究挑战和机遇。该数据集和基准实现可在以下网址公开获取：https://github.com/taohan10200/WEATHER-5K。

WEATHER-5K dataset addresses the limitations of existing public datasets for Global Station Weather Forecasting (GSWF) optimization and benchmarking, which fail to reflect the benchmarks of current forecasting methods and support the real needs of operational weather forecasting.

The WEATHER-5K dataset comprises a comprehensive collection of data from 5,672 weather stations worldwide, spanning a 10-year period with one-hour intervals, including multiple crucial weather elements, providing a more reliable and interpretable resource for forecasting. It can serve as a benchmark for comprehensively evaluating existing well-known forecasting models and support future time-series research challenges and opportunities.

The WEATHER-5K dataset is publicly available and can be used to improve the accuracy of weather predictions for various sectors. The dataset is a significant improvement over existing public datasets, with a larger size, longer temporal coverage, and more comprehensive variables. The dataset and benchmark implementation are publicly available on GitHub. The paper also provides experimental results and comparisons with existing methods.

Recent advancements in deep learning have significantly improved the accuracy of weather predictions. Other related studies include 'Deep Learning for Precipitation Nowcasting: A Benchmark and A New Model' and 'A Review of Recent Advances in Deep Learning for Time Series Forecasting'.