Flux-anomaly-forced model intercomparison project (FAFMIP)


FAFMIP is an atmosphere-ocean general circulation model intercomparison project of CMIP6, initiated in 2016. Its purpose is to investigate the model spread in AOGCM projections of ocean climate change forced by CO2 increase. This spread is due to the different AOGCMs' differing simulations of regional ocean density and circulation changes, especially in high latitudes and the North Atlantic (Yin, 2012; Bouttes et al., 2012; IPCC AR5 WG1 chapter 13, Church et al., 2013; Slangen et al, 2014; IPCC AR6 WG1 chapter 9, Fox-Kemper et al., 2021). The scientific aims of FAFMIP are as follows:

FAFMIP defines a set of experiments, which are to be carried out with AOGCMs in the CMIP6 design but which have also been done with OGCMs. The FAFMIP experiments are aimed at increased physical understanding. They are not themselves policy-relevant scenarios, but obviously the uncertainties in projection of global and regional sea level and AMOC change are of great policy relevance. See below for more information on the experiments and diagnostics.

The outcomes of FAFMIP are documented in several published papers (see bibliography). The main scientific findings (new discoveries or confirmation of previous ones) from investigations enabled by FAFMIP experiments or diagnostics are as follows. All these results apply to CO₂-induced climate change.

The CMIP6 FAFMIP steering committee consider that the analyses that have been completed largely address the aims of the project. Repeating the same experiments with AOGCMs of the same kind in CMIP7 would probably not be scientifically so productive. Therefore FAFMIP has not been proposed as a CMIP7 project.

Bibliography of FAFMIP publications

The published articles listed below make use of FAFMIP experiments or diagnostics. Please let us know of any which should be added. FAFMIP has an automatically updating Google Scholar publication list (thanks to Catia Domingues), which also includes conference abstracts, datasets and documents on the web.

Records of the CMIP6 FAFMIP project

We held the following FAFMIP meetings:

FAFMIP used to maintain an email list for discussion among participants and other interested parties. Owing to changes in mailing-list software at Reading, the list has become inaccessible.

The rationale and progress with the project are further described in these documents:

The members of the FAFMIP steering committee for CMIP6 were Jonathan Gregory, Stephen Griffies, Johann Jungclaus, Oleg Saenko, Detlef Stammer and Laure Zanna.

FAFMIP experiments

In the FAFMIP experiments, a prescribed set of surface flux perturbations are applied to the ocean water surface. These perturbations are obtained from the ensemble-mean changes simulated at time time of doubled CO₂ by CMIP5 AOGCMs under the 1pctCO2 scenario, so they are representative of projected anthropogenic climate change. The experiments are defined by

All the experiments add perturbations to the surface fluxes computed by the AOGCM (like a flux adjustment). The perturbative fluxes depend on the time within the year but are the same in every year. All the experiments are 70 years long, and should branch from the standard CMIP DECK pre-industrial control (piControl). The best point to branch is the same point as the 1% CO₂ experiment (1pctCO2), with which FAFMIP results are compared. All the FAFMIP experiments have pre-industrial atmospheric composition and all other forcing agents as in piControl.

Input files

The surface flux perturbations are supplied as climatological monthly means in the netCDF files below, conforming to the CF metadata convention. The data variables in these files are dimensioned (longitude,latitude,time) in Fortran order, (time,latitude,longitude) in CDL. The time dimension has size 12, for months from January to December. The data can be regarded as applying at the middle of the month and it is recommended to interpolate linearly between them in time to obtain updates at the atmosphere-ocean coupling interval, as for e.g. AMIP simulations. The spatial resolution is 1 degree. The longitude dimension has size 360, with points running eastward starting from 0.5degE, and the latitude dimension has size 180, with points running northward starting from 89.5S. The current versions of these files were made available in August 2015.

These fields are the time-mean of the difference of 1pctCO2 in years 61-80 from the corresponding time-mean of piControl in the ensemble mean of 13 CMIP5 AOGCMs viz. CNRM-CM5 CSIRO-Mk3-6-0 CanESM2 GFDL-ESM2G HadGEM2-ES MIROC-ESM MIROC5 MPI-ESM-LR MPI-ESM-MR MPI-ESM-P MRI-CGCM3 NorESM1-ME NorESM1-M. This was the set for which all the required diagnostics were available.

Ocean process tendency diagnostics

For the scientific analyses of FAFMIP experiments, it is very valuable to have 3D ocean diagnostics of temperature and salinity tendencies (∂T/∂t and ∂S/∂t) due to the individual physical processes which modify the state (advection, diffusion, etc.). These diagnostics were recommended by the CLIVAR Ocean Model Development Panel, and are described in Appendix L of the paper by Griffies et al. (2016) in Geoscientific Model Development, on the OMIP contribution to CMIP6. For FAFMIP they are requested for the portion of the DECK piControl which is parallel to the FAFMIP experiments, and for the DECK idealised climate change experiments abrupt4xCO2 and 1pctCO2, as well as the FAFMIP experiments themselves.

The request to CMIP6 is for annual means of the d/dt diagnostics as priority 1, because groups might not feel it was practical to save monthly means. Monthly means are useful for studying unforced variability and are therefore requested at priority 2.

Please note when setting up your CMIP6 experiments that the CMIP request should always be included when working out the diagnostic list for a given experiment. The FAFMIP request alone is for the diagnostics specific to FAFMIP (mostly the tendency diagnostics), not the core diagnostics (most of the things you expect to use). The core request is included by default in the output of the drq utility.

Ocean surface flux diagnostics

It is useful to have the net heat flux hfds and net water flux wfo into the ocean water (these are standard CMIP diagnostics) in all experiments for FAFMIP. Although the FAF experiments do not have modified CO₂, the surface fluxes can be indirectly affected by climate change due to the perturbations applied. The standard CMIP definitions for these diagnostics are unfortunately inconsistent regarding "flux adjustment", in that wfo should include the FAFMIP water flux perturbation, but hfds should not include the FAFMIP heat flux perturbation.

Methods of implementation for the FAFMIP experiments

Method for the tier-1 faf-stress experiment

Interpolate the perturbative stress components to your own ocean grid, and add them to the momentum balance of the ocean water surface. They should not enter the sea-ice momentum balance, although presumably the sea-ice velocity will be indirectly affected. There should be no perturbation applied to any turbulent mixing scheme that depends on the windstress; the idea is that the perturbation is just to the momentum balance of the ocean.

Method for the tier-1 faf-heat experiment

The redistributed heat tracer is the passive tracer Tc of Bouttes et al. (2014, 10.1007/s00382-013-1973-8), and the added heat tracer is their passive tracer Ta. The FAFMIP design paper (Gregory et al., 2016) comments on the implementation of the faf-heat experiment:
Careful formulation is required to ensure that Q (the atmosphere-ocean heat flux computed by the model, not including the heat flux perturbation) is applied in the same way to T (denoted θ in the paper i.e. the "real" ocean temperature) and TR (the redistributed ocean temperature), and some differences may be unavoidable, depending on model formulation. In particular, absorption of solar radiation should occur with the same vertical profile for both (assuming that some of it penetrates the top layer), and the same heat flux should be applied to both of them for evaporation and precipitation (if the sensible heat content of these water fluxes is considered in the model). If the same amount of heat is extracted from both tracers for frazil sea-ice formation, T may sometimes fall below freezing point, requiring special treatment of the equation of state; on the other hand if T and TR are separately kept above freezing, there will be a difference in the heat fluxes implied. It may be useful to check the implementation of TR in the model with an experiment in which F=0, which should reproduce the piControl experiment.
Different choices have been made in the various models. Steve Griffies, Mike Winton and Bill Hurlin have compiled notes on how the faf-heat experiment has been implemented in GFDL-ESM2M. In their model, they calculated sea-ice (frazil) formation separately for T and TR, meaning that the net heat flux applied to the two tracers is not exactly the same. In CanESM2, the heat flux is the same for the two tracers; T is allowed to fall below freezing point, and the equation of state has been modified to deal with this. In HadCM3, the heat flux is the same, and there is an extra implicit vertical mixing scheme (already included in the model) to prevent freezing when sea-ice formation does not occur. The preliminary experiments reported in the FAFMIP design paper show that TA and TR combine nearly linearly to give T. On the basis of this result, Oleg Saenko simplified the faf-heat implementation in CanESM5, by omitting the redistributed heat tracer TR. Instead, TR for coupling to the atmosphere and sea-ice models is calculated as T-TA.

Method for the tier-1 faf-heat-NA50pct experiment and the tier-2 faf-heat-NA0pct experiment

These experiments follow exactly the same method as faf-heat, the only difference being in the choice of the surface heat flux input file.

Method for the tier-1 faf-water experiment

Interpolate the prescribed water flux perturbation fields to your ocean grid, conserving the ocean area integral as far as possible. For example, the CDO remapcon tool could do it. Note that the fields have missing data over land. For comparative analysis it is useful if you could save the fields you actually apply on your grid. The area-weighted average (over the non-missing area i.e. the ocean) of the annual mean of the water flux field is very small, only 7.19009e-08 kg m-2 s-1, which is two orders of magnitude smaller than the spatial standard deviation. This is because water does not accumulate outside the ocean, in general, and must indicate that water is reasonably accurately conserved in the CMIP5 model mean. The fraction of non-missing area in the world is 0.730988 and the world area-integral of the field is 2.68102e+07 kg s-1.

Method for the tier-2 faf-passiveheat experiment

This experiment should be identical in evolution to the piControl, with the addition of a passive tracer. Hence if the passive tracer can be implemented in the piControl, there is no need to do this experiment separately.

Method for the tier-2 faf-all experiment

In this experiment, the same changes to tracers as in faf-heat should be implemented, and the perturbative fluxes of faf-stress, faf-heat and faf-water should all be applied.


Jonathan Gregory