10. Parameterization-specific Output

10.1. Overview

When used with UFS and the SCM, the CCPP offers the capability of outputting tendencies of temperature, zonal wind, meridional wind, ozone, and specific humidity produced by the parameterizations of selected suites. This capability is useful for understanding the behavior of the individual parameterizations in terms of magnitude and spatial distribution of tendencies, which can help model developers debug, refine, and tune their schemes.

The CCPP also enables outputting two-dimensional (2D) or three-dimensional (3D) arbitrary diagnostics from the parameterizations. This capability is targeted to model developers who may benefit from analyzing intermediate quantities computed in one or more parameterizations. One example of desirable diagnostic is tendencies from sub-processes within a parameterization, such as the tendencies from condensation, evaporation, sublimation, etc. from a microphysics parameterization. The output is done using CCPP-provided 2D- and 3D arrays, and the developer can fill positions 1, 2, .., N of the array. Important aspects of the implementation are that memory is only allocated for the necessary positions of the array and that all diagnostics are output on physics model levels. An extension to enable output on radiation levels may be considered in future implementations.

These capabilities have been tested and are expected to work with the following suites:

  • SCM: GFS_v16, GFS_v17_p8, RAP, RRFS_v1beta, WoFS, HRRR

  • ufs-weather-model (regional): GFS_v16, RRFS_v1beta, WoFS, HRRR

10.2. Tendencies

This section describes the tendencies available, how to set the model to prepare them and how to output them. It also contains a list of frequently-asked questions in Section 10.2.6.

10.2.1. Available Tendencies

The model can produce tendencies for temperature, wind, and all non-chemical tracers (see Table 10.1) for several different schemes. Not all schemes produce all tendencies. For example, the orographic and convective gravity wave drag (GWD) schemes produce tendencies of temperature and wind, but not of tracers. Similarly, only the planetary boundary layer (PBL), deep and shallow convection, and microphysics schemes produce specific humidity tendencies. Some PBL and convection schemes will have tendencies for tracers, and others won’t.

In addition to the tendencies from specific schemes, the output includes tendencies from all photochemical processes, all physics processes, and all non-physics processes (last three rows of Table 10.2). Examples of non-physical processes are dynamical core processes such as advection and nudging toward climatological fields.

In the supported suites, there are two types of schemes that produce ozone tendencies: PBL and ozone photochemistry. The total tendency produced by the ozone photochemistry scheme (NRL 2015 scheme) is subdivided by subprocesses: production and loss (combined as a single subprocess), quantity of ozone present in the column above a grid cell, influences from temperature, and influences from mixing ratio. For more information about the NRL 2015 ozone photochemistry scheme, consult the CCPP Scientific Documentation.

There are numerous tendencies in CCPP, and you need to know which ones exist for your configuration to enable them. The model will output a list of available tendencies for your configuration if you run with diagnostic tendencies enabled. To avoid overusing memory, you should enable just one tendency, which is available for all suites, the non-physics (ie. dynamics) tendency of temperature. Details of how to do this, and how to use the information, is below.

10.2.2. Enabling Tendencies

For performance reasons, the preparation of tendencies for output is off by default in the UFS and can be turned on via a set of namelist options. Since the SCM is not operational and has a relatively tiny memory footprint, these tendencies are turned on by default in the SCM.

There are three namelist variables associated with this capability: ldiag3d, qdiag3d, and dtend_select. These are set in the &gfs_physics_nml portion of the namelist file input.nml.

  • ldiag3d enables tendencies for state variables (horizontal wind and temperature)

  • qdiag3d enables tendencies for tracers; ldiag3d must also be enabled

  • dtend_select enables only a subset of the tendencies turned on by ldiag3d and qdiag3d

If dtend_select is not specified, the default is to select all tendencies enabled by the settings of ldiag3d and qdiag3d.

Note that there is a fourth namelist variable, lssav, associated with the output of parameterization-specific information. The value of lssav is overwritten to true in the code, so the value used in the namelist is irrelevant.

While the tendencies output by the SCM are instantaneous, the tendencies output by the UFS are averaged over the number of hours specified by the user in variable fhzero in the &gfs_physics_nml portion of the namelist file input.nml. Variable fhzero must be an integer (it cannot be zero).

This example namelist selects all tendencies from microphysics processes, and all tendencies of temperature. The naming convention for dtend_select is explained in the next section.

  ldiag3d = .true. ! enable basic diagnostics
  qdiag3d = .true. ! also enable tracer diagnostics
  dtend_select = 'dtend*mp', 'dtend_temp_*' ! Asterisks (*) and question marks (?) have the same meaning as shell globs
  ! The default for dtend_select is '*' which selects everything
  ! ... other namelist parameters ...

10.2.3. Tendency Names

Tendency variables follow the naming pattern below, which is used to enable calculation (input.nml) and output of the variable:


The dtend_ string stands for “diagnostic tendency” and is used to avoid variable name clashes. Replace variable with the short name of the tracer or state variable (see Table 10.1). Replace process with the short name of the process that is changing the variable (see Table 10.2). For example, microphysics (mp) temperature (temp) tendency is dtend_temp_mp.

The next section will tell you how to determine which tendency variables are available for your model.

Table 10.1 Non-chemical tracer and state variables with tendencies. The second column is the variable part of dtend_variable_process. The Index column is the first index of dtidx. Hence “X Wind” is at dtend(:,:,dtidx(index_of_x_wind,:)).


Short Name

Associated Namelist Variables

dtidx Index

Tendency Units





K s-1

X Wind




m s-2

Y Wind




m s-2

Water Vapor Specific Humidity




kg kg-1 s-1

Ozone Concentration




kg kg-1 s-1

Cloud Condensate or Liquid Water




kg kg-1 s-1

Ice Water




kg kg-1 s-1

Rain Water




kg kg-1 s-1

Snow Water




kg kg-1 s-1





kg kg-1 s-1

Cloud Amount




kg kg-1 s-1

Liquid Number Concentration




kg-1 s-1

Ice Number Concentration




kg-1 s-1

Rain Number Concentration




kg-1 s-1

Snow Number Concentration




kg-1 s-1

Graupel Number Concentration




kg-1 s-1

Turbulent Kinetic Energy




J s-1

Mass Weighted Rime Factor




kg kg-1 s-1

Number Concentration Of Water-Friendly Aerosols




kg-1 s-1

Number Concentration Of Ice-Friendly Aerosols




kg-1 s-1

Oxygen Ion Concentration




kg kg-1 s-1

Oxygen Concentration




kg kg-1 s-1

Table 10.2 Processes that can change non-chemical tracer and state variables. The third column is the process part of dtend_variable_process. The dtidx index is second index of dtidx, hence “Deep Convection” is at dtend(:,:,dtidx(:,index_of_process_dcnv)).


diag_table Module Name

Short Name

dtidx Index

Planetary Boundary Layer




Deep Convection




Shallow Convection








Convective Transport




Long Wave Radiation




Short Wave Radiation




Orographic Gravity Wave Drag




Rayleigh Damping




Convective Gravity Wave Drag




Production and Loss (Photochemical)




Ozone Mixing Ratio (Photochemical)




Temperature-Induced (Photochemical)




Overhead Ozone Column (Photochemical)




Sum of Photochemical Processes




Sum of Physics Processes (Including Photochemical)




Sum of Non-Physics Processes




10.2.4. Selecting Tendencies

With the many suites and many combinations of schemes, it is hard to say which variable/process combinations are available for your particular configuration. To find a list, enable diagnostics, but disable all tracer/process combinations except one:

  ldiag3d = .true. ! enable basic diagnostics
  qdiag3d = .true. ! also enable tracer diagnostics
  dtend_select = 'dtend_temp_nophys' ! All configurations have non-physics temperature tendencies
  ! ... other namelist parameters ...

After recompiling and running the model, you will see lines like this in the model’s standard output stream:

dtend selected: gfs_phys dtend_qv_mp = water vapor specific humidity tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_liq_wat_mp = cloud condensate (or liquid water) tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_rainwat_mp = rain water tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_ice_wat_mp = ice water tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_snowwat_mp = snow water tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_graupel_mp = graupel tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_cld_amt_mp = cloud amount integer tendency due to microphysics (kg kg-1 s-1)
dtend selected: gfs_phys dtend_temp_phys = temperature tendency due to physics (K s-1)
dtend selected: gfs_dyn dtend_temp_nophys = temperature tendency due to non-physics processes (K s-1)

There are three critical pieces of information in each line. Taking the third last line as an example,

  1. dtend_cld_amt_mp – this is the name of the variable in dtend_select; for the UFS, it is also the name of the variable in the diag_table

  2. gfs_phys – the diag_table module name (specific to the UFS, can be ignored for other models)

  3. “cloud amount integer tendency due to microphysics” – meaning of the variable

Note that the dtend_temp_nophys differs from the others in that it is in the gfs_dyn module instead of gfs_phys because it sums non-physics processes. This is only relevant for the UFS.

Now that you know what variables are available, you can choose which to enable:

  ldiag3d = .true. ! enable basic diagnostics
  qdiag3d = .true. ! also enable tracer diagnostics
  dtend_select = 'dtend*mp', 'dtend_temp_*' ! Asterisks (*) and question marks (?) have the same meaning as shell globs
  ! The default for dtend_select is '*' which selects everything
  ! ... other namelist parameters ...

Note that any combined tendencies, such as the total temperature tendency from physics (dtend_temp_phys), will only include other tendencies that were calculated. Hence, if you only calculate PBL and microphysics tendencies then your “total temperature tendency” will actually just be the total of PBL and microphysics.

The third step is to enable output of variables, which will be discussed in the next section.

10.2.5. Outputting Tendencies UFS

After enabling tendency calculation (using ldiag3d, qdiag3d, and diag_select), you must also enable output of those tendencies using the diag_table. Enter the new lines with the variables you want output. Continuing our example from before, this will enable output of some microphysics tracer tendencies, and the total tendencies of temperature:

"gfs_phys", "dtend_qv_mp",       "dtend_qv_mp",       "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_liq_wat_mp",  "dtend_liq_wat_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_rainwat_mp",  "dtend_rainwat_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_ice_wat_mp",  "dtend_ice_wat_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_snowwat_mp",  "dtend_snowwat_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_graupel_mp",  "dtend_graupel_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_cld_amt_mp",  "dtend_cld_amt_mp",  "fv3_history", "all", .false., "none", 2
"gfs_phys", "dtend_temp_phys",   "dtend_temp_phys",   "fv3_history", "all", .false., "none", 2
"gfs_dyn",  "dtend_temp_nophys", "dtend_temp_nophys", "fv3_history", "all", .false., "none", 2

Note that all tendencies, except non-physics tendencies, are in the gfs_phys diagnostic module. The non-physics tendencies are in the gfs_dyn module. This is reflected in the Table 10.2.

Note that some host models, such as the UFS, have a limit of how many fields can be output in a run. When outputting all tendencies, this limit may have to be increased. In the UFS, this limit is determined by variable max_output_fields in namelist section &diag_manager_nml in file input.nml.

Further documentation of the diag_table file can be found in the UFS Weather Model User’s Guide.

When the model completes, the fv3_history will contain these new variables. SCM

The default behavior of the SCM is to output instantaneous values of all tendency variables, and dtend_select is not recognized. Tendencies are computed in file scm_output.F90 in the subroutines output_init and output_append. If the values of ldiag3d or qdiag3d are set to false, the variables are still written to output but are given missing values.

10.2.6. FAQ What is the meaning of error message max_output_fields was exceeded?

If the limit to the number of output fields is exceeded, the job may fail with the following message:

FATAL from PE    24: diag_util_mod::init_output_field: max_output_fields =          300 exceeded.  Increase via diag_manager_nml

In this case, increase max_output_fields in input.nml:

    prepend_date = .F.
    max_output_fields = 600

10.2.7. Why did I run out of memory when outputting tendencies?

Trying to output all tendencies may cause memory problems. Use dtend_select and choose your output variables carefully!

10.2.8. Why did I get a runtime logic error when outputting tendencies?

Setting ldiag3d=F and qdiag3d=T will result in an error message:

Logic error in GFS_typedefs.F90: qdiag3d requires ldiag3d

If you want to output tracer tendencies, you must set both ldiag3d and qdiag3d to T. Then use diag_select to enable only the tendencies you want. Make sure your diag_table matches your choice of tendencies specified through diag_select.

10.2.9. Why are my tendencies zero, even though the model says they are supported for my configuration?

For total physics or total photochemistry tendencies, see the next question.

The tendencies will be zero if they are never calculated. Check that you enabled the tendencies with appropriate settings of ldiag3d, qdiag3d, and diag_select.

Another possibility is that the tendencies in question really are zero. The list of “available” tendencies is set at the model level, where the exact details of schemes and suites are not known. This can lead to some tendencies erroneously being listed as available. For example, some PBL schemes have ozone tendencies and some don’t, so some may have zero ozone tendencies. Also, some schemes don’t have tendencies of state variables or tracers. Instead, they modify different variables, which other schemes use to affect the state variables and tracers. Unfortunately, not all of the 3D variables in CCPP have diagnostic tendencies.

10.2.10. Why are my total physics or total photochemistry tendencies zero?

There are three likely reasons:

  • You forgot to enable calculation of physics tendencies. Make sure ldiag3d and qdiag3d are T, and make sure diag_select selects physics tendencies.

  • The suite did not enable the phys_tend scheme, which calculates the total physics and total photochemistry tendencies.

  • You did not enable calculation of the individual tendencies, such as ozone. The phys_tend sums those to make the total tendencies.

10.3. Output of Auxiliary Arrays from CCPP

The output of diagnostics from one or more parameterizations involves changes to the namelist and code changes in the parameterization(s) (to load the desirable information onto the CCPP-provided arrays and to add them to the subroutine arguments) and in the parameterization metadata descriptor file(s) (to provide metadata on the new subroutine arguments). In the UFS, the namelist is used to control the temporal averaging period. These code changes are intended to be used by scientists during the development process and are not intended to be incorporated into the authoritative code. Therefore, developers must remove any code related to these additional diagnostics before submitting a pull request to the ccpp-physics repository.

The auxiliary diagnostics from CCPP are output in arrays:

  • aux2d - auxiliary 2D array for outputting diagnostics

  • aux3d - auxiliary 3D array for outputting diagnostics

and dimensioned by:

  • naux2d - number of 2D auxiliary arrays to output for diagnostics

  • naux3d - number of 3D auxiliary arrays to output diagnostics

At runtime, these arrays will be written to the output files. Note that auxiliary arrays can be output from more than one parameterization in a given run.

The UFS and SCM already contain code to declare and initialize the arrays:

  • dimensions are declared and initialized in GFS_typedefs.F90

  • metadata for these arrays and dimensions are defined in GFS_typedefs.meta

  • arrays are populated in GFS_diagnostics.F90 (UFS) or scm_output.F90 (SCM)

The remainder of this section describes changes the developer needs to make in the physics code and in the host model control files to enable the capability. An example (Section 10.3.2) and FAQ (Section are also provided.

10.3.1. Enabling the capability Physics-side changes

In order to output auxiliary arrays, developers need to change at least the following two files within the physics (see also example in Section 10.3.2):

  • A CCPP entrypoint scheme (Fortran source code)
    • Add array(s) and its/their dimension(s) to the list of subroutine arguments

    • Declare array(s) with appropriate intent and dimension(s). Note that array(s) do not need to be allocated by the developer. This is done automatically in GFS_typedefs.F90.

    • Populate array(s) with desirable diagnostic for output

  • Associated CCPP metadata files for modified scheme(s)
    • Add entries for the array(s) and its/their dimension(s) and provide metadata Host-side changes UFS

For the UFS, developers have to change the following two files on the host side (also see example provided in Section 10.3.2)

  • Namelist file input.nml
    • Specify how many 2D and 3D arrays will be output using variables naux2d and naux3d in section &gfs_physics_nml, respectively. The maximum allowed number of arrays to output is 20 2D and 20 3D arrays.

    • Specify whether the output should be for instantaneous or time-averaged quantities using variables aux2d_time_avg and aux_3d_time_avg. These arrays are dimensioned naux2d and naux3d, respectively, and, if not specified in the namelist, take the default value F.

    • Specify the period of averaging for the arrays using variable fhzero (in hours).

  • File diag_table
    • Enable output of the arrays at runtime.

    • 2D and 3D arrays are written to the output files. SCM

Typically, in a 3D model, 2D arrays represent variables with two horizontal dimensions, e.g. x and y, whereas 3D arrays represent variables with all three spatial dimensions, e.g. x, y, and z. For the SCM, these arrays are implicitly 1D and 2D, respectively, where the “y” dimension is 1 and the “x” dimension represents the number of independent columns (typically also 1). For continuity with the UFS Atmosphere, the naming convention 2D and 3D are retained, however. With this understanding, the namelist files can be modified as in the UFS:

  • Namelist file input.nml
    • Specify how many 2D and 3D arrays will be output using variables naux2d and naux3d in section &gfs_physics_nml, respectively. The maximum allowed number of arrays to output is 20 2D and 20 3D arrays.

    • Unlike the UFS, only instantaneous values are output. Time-averaging can be done through post-processing the output. Therefore, the values of aux2d_time_avg and aux_3d_time_avg should not be changed from their default false values. As such, the namelist variable fhzero has no effect in the SCM.

10.3.2. Recompiling and Examples

The developer must recompile the code after making the source code changes to the CCPP scheme(s) and associated metadata files. Changes in the namelist and diag table can be made after compilation. At compile and runtime, the developer must pick suites that use the scheme from which output is desired.

An example for how to output auxiliary arrays is provided in the rest of this section. The lines that start with “+” represent lines that were added by the developer to output the diagnostic arrays. In this example, the developer modified the Grell-Freitas (GF) cumulus scheme to output two 2D arrays and one 3D array. The 2D arrays are aux_2d (:,1) and aux_2d(:,2); the 3D array is aux_3d(:,:,1). The 2D array aux2d(:,1) will be output with an averaging in time in the UFS, while the aux2d(:,2) and aux3d arrays will not be averaged.

In this example, the arrays are populated with bogus information just to demonstrate the capability. In reality, a developer would populate the array with the actual quantity for which output is desirable.

diff --git a/physics/cu_gf_driver.F90 b/physics/cu_gf_driver.F90
index 927b452..aed7348 100644
--- a/physics/cu_gf_driver.F90
+++ b/physics/cu_gf_driver.F90
@@ -76,7 +76,8 @@ contains
                flag_for_scnv_generic_tend,flag_for_dcnv_generic_tend,           &
                du3dt_SCNV,dv3dt_SCNV,dt3dt_SCNV,dq3dt_SCNV,                     &
                du3dt_DCNV,dv3dt_DCNV,dt3dt_DCNV,dq3dt_DCNV,                     &
-               ldiag3d,qdiag3d,qci_conv,errmsg,errflg)
+               ldiag3d,qdiag3d,qci_conv,errmsg,errflg,                          &
+               naux2d,naux3d,aux2d,aux3d)
       implicit none
       integer, parameter :: maxiens=1
@@ -137,6 +138,11 @@ contains
    integer, intent(in   ) :: imfshalcnv
    character(len=*), intent(out) :: errmsg
    integer,          intent(out) :: errflg
+   integer, intent(in) :: naux2d,naux3d
+   real(kind_phys), intent(inout) :: aux2d(:,:)
+   real(kind_phys), intent(inout) :: aux3d(:,:,:)
 !  define locally for now.
    integer, dimension(im),intent(inout) :: cactiv
    integer, dimension(im) :: k22_shallow,kbcon_shallow,ktop_shallow
@@ -199,6 +205,11 @@ contains
   ! initialize ccpp error handling variables
      errmsg = ''
      errflg = 0
+     aux2d(:,1) = aux2d(:,1) + 1
+     aux2d(:,2) = aux2d(:,2) + 2
+     aux3d(:,:,1) = aux3d(:,:,1) + 3
 ! Scale specific humidity to dry mixing ratio

The cu_gf_driver.meta file was modified accordingly:

diff --git a/physics/cu_gf_driver.meta b/physics/cu_gf_driver.meta
index 99e6ca6..a738721 100644
--- a/physics/cu_gf_driver.meta
+++ b/physics/cu_gf_driver.meta
@@ -476,3 +476,29 @@
   type = integer
   intent = out
+  standard_name = number_of_2d_auxiliary_arrays
+  long_name = number of 2d auxiliary arrays to output (for debugging)
+  units = count
+  dimensions = ()
+  type = integer
+  standard_name = number_of_3d_auxiliary_arrays
+  long_name = number of 3d auxiliary arrays to output (for debugging)
+  units = count
+  dimensions = ()
+  type = integer
+  standard_name = auxiliary_2d_arrays
+  long_name = auxiliary 2d arrays to output (for debugging)
+  units = none
+  dimensions = (horizontal_loop_extent,number_of_3d_auxiliary_arrays)
+  type = real
+  kind = kind_phys
+  standard_name = auxiliary_3d_arrays
+  long_name = auxiliary 3d arrays to output (for debugging)
+  units = none
+  dimensions = (horizontal_loop_extent,vertical_layer_dimension,number_of_3d_auxiliary_arrays)
+  type = real
+  kind = kind_phys

The following lines were added to the &gfs_physics_nml section of the namelist file input.nml:

naux2d         = 2
naux3d         = 1
aux2d_time_avg = .true., .false.

Recall that for the SCM, aux2d_time_avg should not be set to true in the namelist.

Lastly, the following lines were added to the diag_table for UFS:

# Auxiliary output
"gfs_phys",    "aux2d_01",     "aux2d_01",      "fv3_history2d",  "all",  .false.,  "none",  2
"gfs_phys",    "aux2d_02",     "aux2d_02",      "fv3_history2d",  "all",  .false.,  "none",  2
"gfs_phys",    "aux3d_01",     "aux3d_01",      "fv3_history",    "all",  .false.,  "none", FAQ How do I enable the output of diagnostic arrays from multiple parameterizations in a single run?

Suppose you want to output two 2D arrays from schemeA and two 2D arrays from schemeB. You should set the namelist to naux2d=4 and naux3d=0. In the code for schemeA, you should populate aux2d(:,1) and aux2d(:,2), while in the code for scheme B you should populate aux2d(:,3) and aux2d(:,4).