FaIR calibration data

This dataset contains the full data, input scripts and produced output data for the AR6-consistent calibration of FaIRv2.1.0.

The zipfile contains everything, allowing you perform your own analysis. The GitHub version contains enough for "bare bones" reproducibility, including downloading of external datasets and generation of intermediate files. 

The CSV file is the output most people will probably care about, containing the constrained, calibrated parameter set.

FaIR v2.1.0

**Update September 2023: a recent change in xarray breaks fair-2.1-0. Use v2.1.1. Calibrations produced for v2.1.0 here are still applicable.**

Obtainable from https://pypi.org/project/fair/

From the command line:

pip install fair==2.1.1

Calibration v1.1

• 1.5 million prior ensemble
• Climate response calibrated on 49 abrupt-4xCO2 experiments from CMIP6 and sampled using correlated kernel density estimates
• Methane lifetime calibrated on 4 AerChemMIP experiments for 1850 and 2014 (Thornhill et al. 2021a, 2021b). Unlike other variables which are sampled around some prior uncertainty, only the best estimate historical calibration is used. The base (1750) lifetime has been fixed and consistently used across projections.
• Carbon cycle uses the parameters from Leach et al. 2021 calibrated for FaIR 2.0.0 using 11 C4MIP models.
• Aerosol cloud interactions depend on SO2, BC and OC, using calibrations from Smith et al. 2021a (developed for AR6) using 13 RFMIP and AerChemMIP models, with the APRP code fixed by Mark Zelinka (see Taylor et al. 2007; Zelinka et al., 2014), with a prior of -2 to 0 W/m2.
• Aerosol radiation interactions use prior values from AR6 Ch6, with a factor of two uncertainty for each species and a prior in the range of -0.6 to 0.0.
• Ozone uses the same coefficeints as AR6 (Smith et al. 2021b), with the code now brought inside the pipeline.
• NOx emissions are updated from RCMIP to fix a unit conversion issue, where biomass burning emissions from GFED are in units of NO and everything else (from CEDS) is in NO2. FaIR uses NO2 units consistently. In RCMIP v5.1.0 (Nicholls and Lewis 2021), the conversion was not done. When using this calibration for future scenarios, it is important that this is carefully considered.
• Effective radaitive forcing uncertainty follows the distributions in AR6.

AR6_updated (v1.1.0)

• 1001-member posterior (deliberately chosen).
• Emissions and concentrations from RCMIP (i.e. based on CMIP6), but with NOx updated as above
• Temperature from AR6 WG1 (1850-2020, mean of 4 datasets), constrained using ssp245 projections beyond 2014.
• ssp245 projections for 2081-2100.
• Ocean heat content from AR6 WG1 (1971-2018), linear.
• two step constraining procedure used: first RMSE of less than 0.16K, then 6-variable distribution fitting.
• Aerosol ERF, ERFari and ERFaci as in AR6 WG1

Performance relative to AR6 assessed ranges

*likely (66%) ranges, otherwise very likely (95%) ranges.

Note: TCRE and AF are very poor constraints, taken only from model evidence, so we should not put too much weight on them.

References

• Leach et al. 2021: https://doi.org/10.5194/gmd-14-3007-2021
• Nicholls and Lewis 2021: https://doi.org/10.5281/zenodo.4589756
• Smith et al. 2021a: https://doi.org/10.1029/2020JD033622
• Smith et al. 2021b: https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_FGD_Chapter07_SM.pdf
• Taylor et al. 2007: https://doi.org/10.1175/JCLI4143.1
• Thornhill et al. 2021a: https://doi.org/10.5194/acp-21-853-2021
• Thornhill et al. 2021b: https://doi.org/10.5194/acp-21-1105-2021
• Zelinka et al. 2014: https://doi.org/10.1002/2014JD021710