toolevalxm/ClimateAI-ERA5-Downscale

ClimateAI: Climate Intelligence Framework

📑 Paper    |    🌐 Project Page    |    💾 Released Resources    |    📦 Repo

This is the resource page of our resources collection on Huggingface, we highlight your current position with a blue block.

Dataset

Dataset	Link
ClimateAI-ERA5-Downscale	🤗

Please also check the raw data after our processing if you are interested: [WeatherML/ClimateAI-ERA5-Raw](https://huggingface.co/datasets/WeatherML/ClimateAI-ERA5-Raw).

Models

Base Model / Training	ClimateAI		ClimateAI++
	Stage 1	Stage 2	Stage 1	Stage 2
Pangu-Weather Base	🤗	🤗	🤗	🤗
GraphCast 1deg	🤗	🤗	🤗	🤗
FourCastNet v2	🤗	🤗	🤗	🤗

Introduction

While having high-resolution climate reanalysis data theoretically enables accurate weather predictions, two challenges arise: 1) The computational cost of running full physics simulations at high resolution is prohibitive; 2) Traditional numerical weather prediction models are constrained by simplified parameterizations and lack the adaptability of ML-based approaches. Thus, we adopt a fully transformer-based approach for downscaling climate data using Pangu-Weather architecture, as it has demonstrated state-of-the-art performance on medium-range forecasting while being computationally efficient.

*Due to data licensing constraints from ECMWF, we only release the ERA5-Downscale subset (this page) of the full dataset.

License

The license for this dataset is CC-BY-NC-SA-4.0.