CORDEX-ML-Bench: A benchmarking dataset for data-driven regional climate downscaling.

CORDEX-ML-Bench Dataset

CORDEX ML-Bench is a benchmark designed to evaluate the performance of machine learning–based climate downscaling models across different regions covering both the standard (perfect prognosis ESD) and emulation climate downscaling approaches. It defines standardized training and test experiments assessing various downscaling challenges along with the corresponding datasets from Regional Climate Models (RCMs) driven by different Global Climate Models (GCMs). 

This repository provides the training and testing datasets for CORDEX-ML-Bench. For detailed information on data loading, model training, and evaluation strategies, visit: 📦 WCRP-CORDEX/ml-benchmark. 

Overview

The CORDEX-ML-Bench Dataset provides a standardized benchmark for evaluating machine learning approaches to climate downscaling. The dataset is publicly available on Zenodo as zip files (~10 GB per domain). The dataset spans three regions, each with identical domain sizes (same number of grid boxes). Each domain includes data from one Regional Climate Model (RCM) driven by two Global Climate Models (GCMs): one for training and testing, and another exclusively for testing transferability.

Predictors (coarse-resolution ~200km, 16×16 grid):There are 16 predictor variables in total. 

Atmospheric variables at 850 hPa, 700 hPa, 500 hPa (~200km) :

u - zonal wind component

v - meridional wind component

q - specific humidity

t - temperature

z - geopotential height

Static field: Orography (topography; ~10km)

Predictands (high-resolution ~10km, 128×128 grid):

Temperature (tasmax)

Precipitation (pr)

🌍 Geographic Domains

Domain
Resolution
Target Variables
Target Grid Size
Predictor Variables
Predictor Grid Size
Static Fields

New Zealand (NZ)
0.11°
Tasmax, Pr
128 × 128
u, v, q, t, z at 850, 700, 500 hPa (15 variables)
16 x 16 (2°)
Orography (128 x 128; 0.11°)

Europe (ALPS)
0.11°
Tasmax, Pr
128 × 128
u, v, q, t, z at 850, 700, 500 hPa
16 x 16 (2°)
Orography

South Africa

(SA)

0.10°
Tasmax, Pr
128 × 128
u, v, q, t, z at 850, 700, 500 hPa 
16 x 16 (2°)
Orography

New Zealand (NZ) – 0.11° resolution

RCM: CCAM (CMIP6-downscaled)

GCM 1 (train/test): ACCESS-CM2_r4i1p1f1 (historical and ssp370)

GCM 2 (test only): EC-Earth3_r1i1p1f1 (historical and ssp370)

Grid: Regular lon/lat

South Africa (SA) – 0.10° resolution

RCM: CCAM (CMIP6-downscaled)

GCM 1 (train/test): ACCESS-CM2_r4i1p1f1 (historical and ssp370)

GCM 2 (test only): NorESM2-MM_r1i1p1f1 (historical and ssp370)

Grid: Regular lon/lat

Europe (ALPS) – 0.11° resolution

RCM: Aladin63 (CORDEX-CMIP5)

GCM 1 (train/test): CNRM-CM5 (historical and rcp85)

GCM 2 (test only): MPI-ESM-LR (historical and rcp85)

Grid: Lambert Conformal Conic projection

Data Description

Training Data

Includes predictors and predictands for two benchmark experiments:

ESD Pseudo-Reality: Standard empirical statistical downscaling

Emulator Hist+Future: Physical emulation approach

Test Data

Includes predictors and predictands for three time periods (historical, mid-century, end-century) with:

Perfect predictors: Upscaled from RCM (as in training)

Imperfect predictors: From driving GCM

📁 File Structure

Domain/├── train/│   ├── ESD_pseudo-reality/│   │   ├── predictors/│   │   │   ├── {GCM}_1961-1980.nc│   │   │   └── static.nc│   │   └── target/│   │       └── pr_tasmax_{GCM}_1961-1980.nc│   └── Emulator_hist_future/│       ├── predictors/│       │   ├── {GCM}_1961-1980_2080-2099.nc│       │   └── static.nc│       └── target/│           └── pr_tasmax_{GCM}_1961-1980_2080-2099.nc└── test/    ├── historical/    │   ├── predictors/    │   │   ├── perfect/    │   │   │   ├── {GCM1}_1981-2000.nc    │   │   │   └── {GCM2}_1981-2000.nc    │   │   └── imperfect/    │   │       ├── {GCM1}_1981-2000.nc    │   │       └── {GCM2}_1981-2000.nc    │   └── target/    │           ├── pr_tasmax_{GCM1}_1981-2000.nc    │           └── pr_tasmax_{GCM2}_1981-2000.nc    ├── mid_century/    |   ├── predictors/    |   │   ├── perfect/    |   │   └── imperfect/    |   └── target/    └── end_century/        ├── predictors/        │   ├── perfect/        │   └── imperfect/        └── target/

Usage

For detailed instructions on downloading and using the data, please refer to notebooks in the Github:  https://github.com/WCRP-CORDEX/ml-benchmark/tree/main. 

data_download.ipynb – Download instructions

experiments.ipynb – Data walkthrough and experiment configuration

Citation

Rampal, N., González-Abad, J., Gibson, P., Engelbrecht, F., Steinkopf, J., & Hardy, C. (2025). CORDEX-ML-Bench: A benchmarking dataset for data-driven regional climate downscaling. Zenodo. https://doi.org/10.5281/zenodo.17957264

Data Preprocessing

Region-specific preprocessing information:

NZ Domain: nram812/CORDEXBench-nzdomain-preprocessing

ALPS Domain: jgonzalezab/CORDEXBench-alpsdomain-preprocessing