Data processing

Our climate projections start from the original simulations of the Coupled Model Intercomparison Project (CMIP6), the simulations behind the IPCC’s sixth assessment report. They are bias-adjusted and downscaled to 0.1° against the ERA5-Land reanalysis, checked, then delivered in the form a project needs. The chain is described in two peer-reviewed data papers (Noël et al., 2021 and 2022), and its downscaling method was tested in a “perfect model” experiment (Vrac et al., 2025).

1. Sourcing

The climate simulations are daily runs of the five CMIP6 models selected by the ISIMIP project: GFDL-ESM4, IPSL-CM6A-LR, MPI-ESM1-2-HR, MRI-ESM2-0 and UKESM1-0-LL. They cover the historical period (1951–2014) and three scenarios to 2100: SSP1-2.6, SSP2-4.5 and SSP5-8.5. They come from the Copernicus Climate Change Service and the Earth System Grid Federation (ESGF). The models’ own resolution is between about 0.9° and 2.5°.

The reference is the ERA5-Land reanalysis from ECMWF, also distributed by the Copernicus Climate Change Service, at 0.1° resolution. Daily values are computed from its hourly data.

2. Remapping

Each model’s simulations are interpolated onto the ERA5-Land grid (0.1° × 0.1°), so that model and reference can be compared grid cell by grid cell. Temperature is interpolated bilinearly; precipitation conservatively, in steps through intermediate resolutions. Remapping adds no information: it prepares the data for downscaling.

3. Downscaling

Bias adjustment and downscaling are done in one step with the CDF-t method (Cumulative Distribution Function transform). For each grid cell, CDF-t relates the distribution of the model’s daily values to that of ERA5-Land over the calibration period, 1981–2010, and applies this relation to the whole simulation, 1951–2100. Unlike simple quantile mapping, it first estimates how the model’s distribution changes in the future, so the climate change signal of the model is kept.

• Continuous series. The simulation is processed in 20-year windows, each downscaling the 10 years at its centre, so there are no jumps between periods.
• Extremes. The few most extreme values at each end of the distribution are adjusted with the mean correction of their neighbours (Lanzante et al., 2019), so they cannot become outliers.
• Precipitation. Dry days are handled with the Singularity Stochastic Removal method (Vrac et al., 2016), which keeps the number of rainy days realistic.

In a perfect-model test over Europe, CDF-t kept the climate change signal in the means, variability and extremes of 2071–2100, where simple quantile mapping did not (Vrac et al., 2025). CDF-t adjusts one variable at a time: the day-to-day sequence and the links between variables come from the climate model. See Limitations.

4. Checks

• Standardisation. Files and metadata are rewritten to the climate community’s standards: the Climate and Forecast (CF) metadata conventions, and a Data Reference Syntax for adjusted projections (Levavasseur and Noël, 2021).
• Quality control. We check compliance with these standards, data consistency and metadata, then look for outlier values in the downscaled data.

5. Delivery

The downscaled data are daily, global over land, at 0.1°. We deliver them in the form a project needs:

• In time: daily values, or means over any period, such as monthly, annual or 30-year means.
• In space: sites, cities, countries or the global grid.
• Formats: NetCDF, CSV, GeoTIFF or shapefile.

References

• Vrac, M., Loukos, H., Noël, T., & Defrance, D. (2025). Should we use quantile-mapping-based methods in a climate change context? A “perfect model” experiment. Climate, 13(7), 137. https://doi.org/10.3390/cli13070137
• Noël, T., Loukos, H., Defrance, D., Vrac, M., & Levavasseur, G. (2022). Extending the global high-resolution downscaled projections dataset to include CMIP6 projections at increased resolution coherent with the ERA5-Land reanalysis. Data in Brief, 45, 108669. https://doi.org/10.1016/j.dib.2022.108669
• Levavasseur, G., & Noël, T. (2021). Data Reference Syntax (DRS) for bias-adjusted C3S-CMIP5 simulations. https://doi.org/10.31223/X5389H
• Noël, T., Loukos, H., Defrance, D., Vrac, M., & Levavasseur, G. (2021). A high-resolution downscaled CMIP5 projections dataset of essential surface climate variables over the globe coherent with the ERA5 reanalysis for climate change impact assessments. Data in Brief, 35, 106900. https://doi.org/10.1016/j.dib.2021.106900
• Lanzante, J. R., Nath, M. J., Whitlock, C. E., Dixon, K. W., & Adams-Smith, D. (2019). Evaluation and improvement of tail behaviour in the cumulative distribution function transform downscaling method. International Journal of Climatology, 39, 2449–2460. https://doi.org/10.1002/joc.5964
• Vrac, M., Noël, T., & Vautard, R. (2016). Bias correction of precipitation through Singularity Stochastic Removal: because occurrences matter. Journal of Geophysical Research: Atmospheres, 121, 5237–5258. https://doi.org/10.1002/2015JD024511