ecRad Overview#

The NWP physics package of ICON uses the ecRad radiation scheme developed by ECMWF. This page aims to described the configuration and input options for ecRad that are available in ICON. For a more detailed and complete description of the radiation scheme itself is provided in the ECMWF Confluence Wiki. The original source code is publicly available on GitHub, please note that ICON uses a modified version of this code.

Reduced Radiation Grid#

Radiation is one of the computationally expensive physical parameterizations. There are several possibilities to decrease the computational cost of radiation, e.g. by reducing the temporal, spatial or spectral resolution. By activating lredgrid_phys=.true. and specifying the corresponding grid file with radiation_grid_filename, the radiation is calculated on a one grid level coarser domain to reduce the computational cost of the radiation by about a factor of 4.

We highly recommend activating the reduced radiation grid option for the following reasons:

It is computationally cheaper by a factor of 4, usually without a degradation of the results. Since the radiation is treated as a slow physics process, it is not called every time step anyways. A coarser horizontal grid thus fits better to the advective time scale.
Only for the reduced radiation grid, there is an additional option named latm_above_top. This option adds an extra layer at the top to account for the incoming long-wave radiation. This reduces the biases at the model top significantly.
For global domains, there is a load balancing for sunlit and shadowed parts of the earth for the reduced grid.

For a more detailed description of the reduced radiation grid implementation, see ICON Tutorial.

Radiation output variables#

These lists should provide a quick overview on output variables for radiation. The variables are defined in mo_nwp_phy_state. Diagnostic ouput variables are often computed in mo_nwp_diagnosis. The diagnostic derived type is defined in mo_nwp_phy_types.

Net fluxes are defined as downward positive.

Shortwave#

Some shortwave output variables exist without and with orographic shading (1) and with slope-dependent and orographic shading (2). The variables ending on either _os or _tan_os are defined depending on the namelist parameter islope_rad(dom) in the radiation_nml.

icon variable name	grib2 name¹	direction	specifics	acc.²	location	Description
asob_s	asob_s	net		x	surface	Surface net solar radiation since model start
asob_t	asob_t	net		x	TOA	TOA net solar radiation since model start
asobclr_s	asob_s_cs	net	clear-sky	x	surface	Clear-sky surface net solar radiation since model start
asod_s	asod_s	down		x	surface	Surface down solar rad. since model start
asod_s_os		down		x	surface (1)	Surface down solar rad. incl. orographic shading since model start
asod_s_tan_os		down		x	surface (2)	Surface down solar rad. incl. slope-dependent and orographic shading since model start
asod_t	asod_t	down		x	TOA	Top down solar radiation
asodifd_s	aswdif_s	down	diffuse	x	surface	Surface down solar diff. rad. since model start
asodifu_s	asdifu_s	up	diffuse	x	surface	Surface up solar diff. rad. since model start
asodifu_s_os		up	diffuse	x	surface (1)	Surface up solar diff. incl. orographic shading since model start
asodifu_s_tan_os		up	diffuse	x	surface (2)	Surface up solar diff. incl. slope-dependent and orographic shading since model start
asodird_s	aswdir_s	down	direct	x	surface	Surface down solar direct rad. since model start
asodird_s_os		down	direct	x	surface (1)	Surface down solar direct rad. incl. orographic shading since model start
asodird_s_tan_os		down	direct	x	surface (2)	Surface down solar direct rad. incl. slope-dependent and orographic shading since model start
asou_t	uswrf	up		x	TOA	Top up solar radiation since model start
sob_s_t_*³	sobs_rad	net			surface, tile	tile-based shortwave net flux at surface
sob_s	sobs_rad	net			surface	shortwave net flux at surface
sob_s_os		net			surface (1)	shortwave net flux at surface incl. orographic shading
sob_s_tan_os		net			surface (2)	shortwave net flux at surface incl. slope-dependent and orgraphic shading
sob_t	sobt_rad	net			TOA	shortwave net flux at TOA
sobclr_s	sobs_rad	net	clear-sky		surface	net shortwave clear-sky flux at surface
sod_t	sodt_rad	down			surface	downward shortwave flux at TOA
sodifd_s	swdifds_rad	down	diffuse		surface	shortwave diffuse downward flux at surface
sou_s	swdifus_rad	up			surface	shortwave upward flux at surface
sou_s_os		up			surface (1)	shortwave upward flux at surface incl. orographic shading
sou_s_tan_os		up			surface (2)	shortwave upward flux at surface incl. slope-dependent and orographic shading
sou_t	uswrf	up			TOA	shortwave upward flux at TOA
swflx_dn_clr	-	down	clear-sky		3d	shortwave downward clear-sky flux
swflx_dn	-	down			3d	shortwave downward flux
swflx_up_clr	-	up	clear_sky		3d	shortwave upward clear-sky flux
swflx_up	-	up			3d	shortwave upward flux
trsolall	-	net	transmissivity		3d	shortwave net transmissivity

Thermal #

icon variable name	grib2 name¹	direction	specifics	acc.²	location	Description
athb_s	athb_s	net		x	surface	surface net thermal radiation since model start
athb_t	athb_t	net		x	TOA	TOA net thermal radiation since model start
athbclr_s	athb_s_cs	net	clear-sky	x	surface	clear-sky surface net thermal radiation since model start
athd_s	athd_s	down		x	surface	Surface down thermal radiation since model start
athu_s	athu_s	up		x	surface	Surface up thermal radiation since model start
lwflx_dn_clr	-	down	clear-sky		3d	longwave downward clear-sky flux
lwflx_dn	-	down			3d	longwave downward flux
lwflx_up_clr	-	up	clear-sky		3d	longwave upward clear-sky flux
lwflx_up	-	up			3d	longwave upward flux
lwflxall	nlwrf	net			3d	longwave net flux
thb_s_t_*	thbs_rad	net			surface	tile-based longwave net flux at surface
thb_s	thbs_rad	net			surface	longwave net flux at surface
thb_t	thbt_rad	net			TOA	thermal net flux at TOA
thbclr_s	thbt_rad_cs	net	clear-sky		surface	net longwave clear-sky flux at surface
thu_s	thus_rad	up			surface	longwave upward flux at surface

Diagnostic for bands PAR, VIS, NIR#

The bands for diagnostic output are:

PAR : photosynthetically active flux (\(400 - 700 nm\))
VIS : visible (\(300 - 700 nm\))
NIR : near-infrared (\(0.7 - 5 \mu m\))

icon variable name	grib2 name¹	direction	specifics	acc.²	location	Description
aswflx_par_sfc	apab_s	down		x	surface	Downward PAR flux
aswflx_par_sfc_tan_os		down		x	surface (2)	Downward PAR flux incl. slope-dependent and orographic shading
fr_nir_sfc_diff	-	down	diffuse fraction		surface	diffuse fraction of downward near-infrared flux at surface
fr_par_sfc_diff	-	down	diffuse fraction		surface	diffuse fraction of downward photosynthetically active flux at surface
fr_vis_sfc_diff	-	down	diffuse fraction		surface	diffuse fraction of downward visible flux at surface
swflx_nir_sfc		down			surface	downward near-infrared flux at surface
swflx_par_sfc		down			surface	downward photosynthetically active flux at surface
swflx_par_sfc_tan_os		down			surface (2)	downward photosynthetically active flux at surface incl. slope-dependent and orographic shading
swflx_vis_sfc		down			surface	downward visible flux at surface

The grib2 names in the table refer to the short names resulting from the DWD grib definition files.
Output variables starting with the letter a are likely “accumulated” since model start. Depending on the switch lflux_avg, they contain either averages since model start (lflux_avg=.true.) or accumulated values.
The asterisk stands for the number of the surface tile.

Gas Input Options#

There are multiple options for the specification of several components of the gaseous composition of the atmosphere available. The corresponding namelist parameters are irad_h2o for water vapor, irad_o3 for ozone, irad_co2 for carbon dioxide, irad_n2o for nitrous oxide, irad_ch4 for methane, irad_o2 for oxygen, irad_cfc11 for trichlorofluoromethane and irad_cfc12 for dichlorodifluoromethane.

External specification#

For all of the above described gases, the option -1 (e.g., irad_h2o=-1) allows for an external specification of the gaseous concentrations. A variable <gas>rad_ext (e.g., h2orad_ext) is created which can be filled with mass mixing ratios (\(kg\,kg^{-1}\)) from an external source, for example via the Community Interface ComIn. There is no cross-check that the arrays contain meaningful values. This is left to the user.

Aerosol Input Options#

Tegen climatology#

Climatological aerosol based on the Tegen et al. 1997 climatology can be selected by choosing irad_aero=6. This options has the following characteristics:

Optical thicknesses at the wavelength 550 nm of the 5 species Sea Salt, Soil Dust, Sulfate, Organic Carbon and Black Carbon are provided in the external parameter file.
The annual cycle is considered by providing monthly data which is linearly interpolated inside ICON to the target date.
The original data is vertically integrated optical thickness. For the use in ecRad, an exponentially decaying, normalized vertical profile is added by ICON.
The target variables optical thickness (SW/LW), single scattering albedo (SW) and asymmetry parameter (SW) at the radiation wavelength bands are derived based on lookup tables in the ICON code.

Simplified Prognostic Aerosol Module Prog2DAero#

See here for further information.

CAMS climatology or CAMS forecast aerosol#

ICON currently supports the use of either the CAMS 49R2 aerosol climatology, or CAMS forecast aerosol for direction aerosol-radiation interactions. Aerosol-cloud interactions using CAMS are not yet supported.

Using the 49R2 CAMS climatology#

ICON supports use of the recent (December 2024) CAMS climatology, version 49R2. This option is activated with the namelist parameter irad_aero=7.

Older versions (43R3) are no longer supported. Aerosol mixing ratios are supplied monthly on 21 pressure surfaces. Aerosol species affected by human activity have an additional dimension epoch covering thirteen 5-year long periods from 1955 to 2015. The original climatology file can be downloaded from this ECMWF webpage.

At this point in time, ICON does not support the new epoch dimension of the climatology (i.e. anthropogenic change of aerosol over 5+ year periods). The ICON repository contains the script make_camsclim_onICONgrid.sh to extract the latest (2015) epoch from the original data file and interpolate the climatology onto an ICON grid of the user’s choice. Installation of CDO, NCO and python3 (numpy, xarray) tools is required to run this script.

Using CAMS forecasts#

ICON can also use CAMS forecast aerosol fields on 137 model levels. This option is activated with the namelist parameter irad_aero=8.

CAMS forecast aerosol fields can be retrieved from ECMWF via MARS request. The ICON repository contains the script make_camsforc_onICONgrid.sh which then interpolates the CAMS forecast aerosol onto an ICON grid of the user’s choice. The script header contains more information on how to retrieve CAMS forecast aerosol from MARS. CDO, python3 and ecmwf-toolbox are required to run this script.

Information relevant to both CAMS climatology and forecast#

A note on remapping to the ICON grid#

CDO remapping is used for the interpolation. None of the available remapping options are perfect:

Conservative remapping (remapcon) leads to visible ‘squares’ corresponding to the original, coarser CAMS climatology in the interpolated fields, which translates into visible ‘square’ shapes in the clear-sky radiation also. This is obviously undesireable.
Bicubic remapping (remapbic) leads to ‘overshooting’ features around high orography (Himalayas, Andes). Also undesireable.
The option considered to be best at the moment (and implemented by default) is bilinear remapping (remapbil), which produces a reasonably smooth field without obvious overshooting.

However, the ‘best’ option may depend on the application: for regional simulations away from steep orography, remapbic may be more advantageous (smoother).

Vertical interpolation onto ICON model levels#

ICON interpolates from the original CAMS vertical levels (21 pressure surfaces for climatology, 137 IFS model levels for forecast) to the ICON vertical model levels. Because the horizontal resolution of the original CAMS files can be much lower than the ICON resolution, the difference in surface pressure between ICON and CAMS can be in excess of 200hPa around steep orography. A straight interpolation between pressure levels would therefore neglect a significant amount of aerosol mass if the lowest 200hPa of the CAMS profile were ignored. To avoid this, the CAMS profile is re-distributed between the ICON surface pressure and top of the atmosphere before interpolation, conserving total aerosol mass in the column.

Differences to previous CAMS versions#

The previous CAMS climatology 43R3 provided aerosol as layer mass. This climatology now also exists (v2) in a format providing mixing ratios instead. Since layer mass will no longer be used from the CAMS side, reading in layer mass is no longer supported by ICON.

The 43R3 CAMS climatology also had some unrealistic features, such as high dust accumulations in the stratosphere. These artefacts have now been removed in the 49R2 climatology. Since it is unlikely that the older, flawed climatology will continue to be used, the option to read in the 43R3 climatology is also no longer supported. However, it is easy to make ICON compatible with this older version again, if so desired, by changing the number of CAMS levels (nlev_cams) in mo_reader_cams.f90 from 21 back to 60, and using the more recent mixing ratio version of the 43R4 CAMS climatology which can be obtained from ECMWF.

Appropriate aerosol optical properties#

The new 49R2 CAMS climatology was created using an updated version (CY48R1) of the CAMS aerosol model, which contains some significant changes relative to the older version used to create the 43R3 climatology. Therefore, the new climatology should be used with an appropriate set of optical properties! A tabulated list of the appropriate optical properties for each version of the climatology/forecasts can be found here.

For this implementation, the appropriate aerosol optical properties have been pre-selected for use with the CAMS 49R2 climatology. The pre-selected default for the CAMS forecasts is set to the most recent CAMS IFS cycle 49R1. If working with older CAMS forecasts, the user has to adapt the selected properties in mo_nwp_ecrad_init.f90 (line 342 and following) according to the table linked above.

Known limitations

Known limitations of the new 49R2 climatology include that the “far field” aerosol such as in the Arctic is too low. The IFS is run with an additional artificial small constant background term to get the best results.

Cloud droplet number concentration (cdnc)#

Climatological data of cloud droplet number concentration from external parameter file can be used in ICON when namelist parameter icpl_aero_gscp is set to 3. The external parameter file must then contain the field cdnc and can be generated with Extpar. When using the external cdnc, it is advisable to set:

icpl_aero_conv=1: simple coupling between auto-conversion (in convection scheme) and aerosol

The external climatological cdnc can be scaled when setting namelist parameter scale_cdnc_mode to 1 or 2, and providing the necessary input file for the Simple Plume model in the run directory:

scale_cdnc_mode = 0 (default): no scaling, the cdnc used by ICON will be the same regardless of the simulation year.
scale_cdnc_mode = 1: apply year-dependent scaling of cdnc using the scale factor from Simple Plume model (e.g. for experiments in historical period after 1850 or climate projections).
scale_cdnc_mode = 2: apply constant scaling of cdnc to year 1850 (e.g. for pre-industrial experiment)

Condensate heterogeneity - the FSD parameter#

ICON predicts one value for the condensate mixing ratios of liquid and ice for each grid box. This is supplemented by a cloud fraction from the diagnostic cloud scheme. Because the amount of radiation reflected or absorbed depends non-linearly on condensate amount, it matters how this condensate is distributed within each grid box. The default assumption is that the condensate amount is distributed within the cloudy part of the grid box with the functional shape of a Gamma distribution. The distribution average corresponds to the predicted grid box condensate amount, while the width of the distribution is given by the “fractional standard deviation” (FSD) parameter, which is defined as

\[FSD=\frac{standard\, deviation}{ mean}\]

By default, ICON assumes that FSD=1 everywhere, i.e. the normalised width of the condensate distribution is the same everywhere. However, observations show that this is not the case. Condensate is distributed more homogeneously in stratiform clouds compared to cumuliform clouds. Also, grid boxes containing cloud edges (i.e. not overcast, with a cloud fraction < 1) have wider condensate distributions than overcast grid boxes, because cloud edges naturally contain less condensate than the cloud interior.

To account for this effect, a regime-dependent parameterization for the FSD parameter can be used by setting the namelist parameter in the radiation_nml:

lcalculate_fsd = .true.

This parameterization is based on the publications Ahlgrimm et al. 2016 and Ahlgrimm et al. 2017, with some minor modifications documented in the ICON code. Broadly, the effect of using this parameterization is to make clouds appear more reflective in areas dominated by stratiform clouds (e.g. stratocumulus regions, extratropics) and less reflective in regions dominated by more convective cloud (e.g. trade cumulus regions, tropics).

Two additional parameters may be set in the radiation_nml:

fsd_background = 1 is the default value used when the parameterization is switched off entirely, or in cloud-free regions of the model atmosphere when the parameterization is active. When radiation is calculated on the reduced grid, the interpolation from the full grid to the reduced grid may interpolate between cloudy and cloud-free grid points, which therefore must be assigned a valid FSD value.

fsd_gridlen = 80 is the assumed horizontal grid spacing of the ICON grid (by default set to 80km). Observations show that the unresolved condensate heterogeneity that must be parameterized should reduce as the resolution of the model increases. This means the parameterized FSD parameter becomes smaller (clouds become more homogeneous) for a finer grid resolution. However, ICON has been operating with a fixed FSD value of 1 at all resolutions for years, and produces a resolution-independent top-of-the-atmosphere radiation balance with this fixed value. Replacing this fixed value with a resolution-dependent FSD value (in the absence of compensating changes elsewhere) would produce a resolution-dependent TOA radiation balance, which is not desirable. Therefore, for the time being, it is recommended to use the fixed gridlength value of 80km at all resolutions, as this produces FSD values that average out to approximately 1 globally, maintaining the usual TOA radiation balance.

Lastly, it should be mentioned that the FSD calculation for liquid clouds depends on the cloud fraction: High cloud fraction is a proxy for stratiform clouds, which are assigned lower FSD values. In cases where the model predicts an incorrect cloud fraction (e.g. prediction cloud fractions <50% in stratocumulus regions), the error in the cloud radiative effect may be enhanced when using the FSD parameterization. In this example, the parameterization would make clouds with fraction <50% less reflective in an area where cloud cover (and therefore cloud radiative effect) is already too low.

Single precision#

The ecRad radiation scheme supports single-precision computation, which improves performance and reduces memory usage. This approach is accurate enough for most applications. It is especially beneficial for high-resolution simulations or large ensemble runs, where computational efficiency is crucial. You can enable single-precision computation via the appropriate configure option. Refer to the output of ./configure --help for details.

Please note that the quality of this feature has not been evaluated for global applications (i.e., altitudes higher than 25 km). Users are therefore advised to run a benchmark simulation before using this combination.

Glossary of Namelist Parameters#

Operational NWP setting marked by

lredgrid_phys#: (&grid_nml) If set to .TRUE. radiation is calculated on a coarser grid (i.e. one grid level coarser).
radiation_grid_filename#: (&grid_nml) Filename of the grid to be used for the radiation model. Must only be specified for the base domain, since for child domains the grid of the respective parent domain serves as radiation grid. An empty string is required, if radiation is computed on the full (non-reduced) grid.
latm_above_top#: (&nwp_phy_nml) Adds an extra layer at the model top to account for the incoming long-wave radiation if set to .TRUE..
lflux_avg#: (&io_nml) If .true., radiative fluxes are averaged since model start instead of accumulated. Default: .true.
irad_aero#: (&radiation_nml) Specify aerosol input for radiation. 0: None, 3: externally specified (e.g. Community Interface (ComIn)) 6: Tegen climatology, 7: CAMS 3D climatology, 8: CAMS 3D forecasted, 9: Aerosols & Reactive Trace gases (ART), 12: tropospheric Kinne climatology (constant in time), 13: tropospheric Kinne climatology (time-dependent), 14: volcanic stratospheric aerosols for CMIP6 (time dependent), 15: combination of 13 and 14, 18: tropospheric natural Kinne climatology + volcanic stratospheric aerosols + anthropogenic ‘simple plumes’ (time-dependent), 19: as 18 without volcanic stratospheric aerosols
lcalculate_fsd#: (&radiation_nml) Main switch to activate regime-dependent FSD parameterization (Default: .FALSE.)
fsd_background#: (&radiation_nml) Background value for assumed horizontal grid spacing in FSD parameterization.
fsd_gridlen#: (&radiation_nml) Value for assumed horizontal grid spacing in FSD parameterization.