Nudging E000 Experiments

Title:    YOTC Low Resolution Atmosphere Only Nudging


The CESM nudging formulation (V2) has been shown to effectively drive the model state toward observations. The goal of this study is to apply various combinations of U,V,T, and Q nudging using 2 years of YOTC data in order to evaluate their effect on AMWG diagnostic results.

Model Configuration:

Nudging utilizes YOTC data which has been interpolated to the prescribed 30 vertical levels with horizontal resolution ne30. The YOTC data covers the time period from May 1, 2009 thru April 30 2010.

The model is initialized with the YOTC data on May 1 2008 at 00Z.
Model Resolution:
ne30 / L30
Model Tag:
Starting Date:
5 / 1 / 2008
Length of Runs:
24 months
(Startup / Outputs)
 mfilt =1,4,8

FAMIPC5 runs thru 2012 require extended versions of some input files

  prescribed_ozone_datapath = '/glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/atm/cam/ozone'
  prescribed_ozone_file     = ''

 prescribed_volcaero_datapath = '/glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/atm/cam/volc'
  prescribed_volcaero_file    = ''

  bndtvghg  = '/glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/atm/cam/ggas/'

 tracer_cnst_datapath           = '/glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/atm/cam/chem/trop_mozart_aero/oxid'
 tracer_cnst_file               = ''
 ext_frc_specifier              = 'SO2         -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'bc_a1       -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'num_a1      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'num_a2      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'pom_a1      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'so4_a1      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'so4_a2      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/'
 srf_emis_specifier             = 'DMS       -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'SO2       -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'SOAG      -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'bc_a1     -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'num_a1    -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'num_a2    -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'pom_a1    -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'so4_a1    -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/',
  'so4_a2    -> /glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/camdata/aero_1850-2100-RCP45/'

  solar_data_file  = '/glade/p/work/patc/cesm1_1_0_rel06/inputdata30yr/atm/cam/solar/'

Experiment Configurations:
Namelist Parameters
Control Run:

Nudging Active but all
coefficients set equal to 0.0
  Nudge_Model        =.true.
  Nudge_Path         ='/glade/scratch/patc/inputdata/nudging/YOTC_ne30np4_005/'
  Nudge_Diag_Opt =0
  Nudge_Uprof  =0
  Nudge_Ucoef  =0.0
  Nudge_Vprof  =0
  Nudge_Vcoef  =0.0
  Nudge_Tprof  =0
  Nudge_Tcoef  =0.0
  Nudge_Qprof  =0
  Nudge_Qcoef  =0.0
  Nudge_PSprof =0
  Nudge_PScoef =0.00
  Nudge_Beg_Year =2008
  Nudge_Beg_Day  =1
  Nudge_End_Year =2100
  Nudge_End_Day  =30
Nudged Runs:

STRENGTH = [0. - 1.]
  Nudge_Model        =.true.
  Nudge_Path         ='/glade/scratch/patc/inputdata/nudging/YOTC_ne30np4_005/'
  Nudge_Diag_Opt =0
  Nudge_Uprof  =1
  Nudge_Ucoef  =STRENGTH
  Nudge_Vprof  =1
  Nudge_Vcoef  =STRENGTH
  Nudge_Tprof  =1
  Nudge_Tcoef  =STRENGTH
  Nudge_Qprof  =1
  Nudge_Qcoef  =STRENGTH
  Nudge_PSprof =0
  Nudge_PScoef =0.00
  Nudge_Beg_Year =2008
  Nudge_Beg_Day  =1
  Nudge_End_Year =2100
  Nudge_End_Day  =30


Unfortunately, the first 4 experiments were carried out before the case name nomenclature was established. The appropriate case names should be:
 E000  ==> E000.000
 E001  ==> E000.001
 E002  ==> E000.002
 E003  ==> E000.003
E004  ==> E000.004
Run Name
Control Run
Nudge U,V,T,Q
Nudge U,V      
Nudge T,Q      
Nudge U,V,Q   
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E004 E004

AMWG Diagnostics:

  • The current AMWG diagnostic script is diag121205.csh
  • Results from CAM-SE are regridded onto a regular grid with resolution 0.47x0.63
  • At this resolution, the script fails due to size constraints in the "Classic" version of netCDF. Rather than reduce the horizontal resolution to 0.9x1.25, the regridclimo.ncl script was modified to open the created climatology files using the "LargeFile" format
  • Because the 2 year record of YOTC data begins and ends in May, only 1 year of data is available for AMWG diagnostics.

E000 - OBS
E001 - E000
E002 - E000
E003 - E000

E001 - OBS
E002 - E001

E004 - E001

E002 - OBS

E003 - OBS

E004 - OBS


The AMWG results indicate that the temperature(T) and specific humidity(Q) nudging is having an undesirable effect on total precipitation.

Comparing the total annual precipitation (PRECT) for the case in which U,V,T, and Q are nudged against the case in which only U and V are nudged illustrates magnitude of the problem [E002-E001]. Though the differences are most evident in the tropical Pacific, the precipitation is reduced significantly everywhere. The case in which only T and Q are nudged indicates a similar decrease in precipitation values.

Precipitation Error
Nudging experiments to this point have never had such a detrimental effect on the CESM model results. There are several possible sources for this problem. First is that the nudging module has undergone substantial development and an error may have been introduced along the way. The second is that, due to the (1/Cp) error, these cases are the first in which temperature has been nudged significantly. So there may be some side effects to nudging T,Q values in CESM. Finally, the CESM model is always under development. There may be differences between cesm1_3_beta06 and the earlier tags, which cause CESM to be more sensitive to the nudging tendencies. 

The results for E004 in which {UVQ} are nudged, when compared to E001 indicate that while the {Q} nudging is contributing some to the reduction in mean precipitation, it is the additional {T} nudging that is acting to suppress precipitation.

E004-E001 Precip
The annual zonal mean precipitation profiles relative to GPCP for the control run (E000) and the experiments E001, E002, and E004 indicate the impact of the nudging forcing.

The {UV} nudging (E002) leads to better agreement, particularly near 30N and 30S. The addition of {Q} nudging in E004 produces higher precipitation values in the tropics and lower precipitation at high latitudes.

With fully nudged values of {UVTQ}, precipitation values are uniformly suppressed leading to unacceptable results.

Not shown are the results for nudging only {TQ} values (E003). This experiment also leads to erroneous precipitation values similar to those for E001.
In order to rule out code changes as the cause, go back to one of the preliminary runs using the CESM tag cesm_1_0_rel06.

The case  f.e11.FAMIPC5.ne30_ne30.JB_Nudge_YOTC_001, which had acceptable precipitation results, was configured the same as case E001 here. Since it is known that the temperature nudging was off by a factor of Cp, rather than nudging {U,V,T,Q}, this case actually only nudged {U,V,Q} with the {T} coefficient effectively 0.

So creating a new case with that tag in which the only change is to add Cp to the {T} nudging tendency should be equivalent to E001. If there is a similar problem with the precipitation, this will rule out code development as the cause and confirm that nudging {T} values introduces some problem in the CESM model.
Run Name
f.e11.FAMIPC5.ne30_ne30.JB_Nudge_YOTC_001 {U,V,Q}=1.0 {T}=(1/Cp)
f.e11.FAMIPC5.ne30_ne30.JB_Nudge_YOTC_004 {U,V,T,Q}=1.0

YOTC004 - YOTC001

The comparison of annual precipitation values between these two cases confirms that the problem is a result of nudging temperature values in CESM.

Before proceeding with further long-term nudging runs involving temperature, the origin of this effect must be investigated and fully understood.
OLD Precip Error

Nudging Development/Diagnostic Study #10:

Carry out 1 month diagnostic (A-series) runs to identify the origin of the precipitation errors. As a matter of habit, these studies are conducted using ERA-I nudging data for January 1980.

Finding #1:

           [INCOMPLETE - fill in summary of A-series results here.....]
      **************  ADD LINK TO A-Series Diagnostic Study #10   ***************

Finding #2:

           [INCOMPLETE - fill in summary of A-series results here.....]

Supplemental Experiments #1:


The results of the diagnostic study indicate that there is a ~1K temperature bias between ERA-I values and the {U,V,Q}-nudged model results. When temperature nudging is included to remove this bias, there is an anomalous decrease in precipitation values. The temperature differences, particularly in the tropical troposphere, appear to interfere with the convective parameterizations, leading to the decreased precipitation. Calculating the mean temperature difference between model results with and without {T}-nudging, and then subsequently applying bias adjustments to the ERA-I temperature values used for nudging, eliminated the precipitation errors.

There are two ways to look at this bias adjustment, the first is that you have 'built in' an erroneous temperature structure to accommodate the convective schemes rather than diagnosing their shortcomings. Which assumes that ERA-I values are the absolute truth and that clouds in the nudged model occur at the same space-time locations as clouds in the ERA-I reanalysis. The other is to consider that 1K is within the stated accuracy of ERA-I temperatures [VERIFY THAT!] and that the reanalysis model also has limitations due to it's convective parameterizations. So For now, we will move ahead by calculating and applying mean bias adjustments to the temperature nudging values.

In extending the results from the 1 month diagnostic runs to the 2 year YOTC data, there are several factors to consider:
  1. The test was conducted using 1 month of data, there may be annual variations in the bias values that need to be considered.
  2. Two tests of mean bias were done thus far, one using the time mean differences at each grid point, and the other using the time-zonal mean at each grid point using 10 degree latitude bands. Both of these should be tested for the 2 year YOTC runs to determine which is best.
  3. The mean temperature differences in the stratosphere and in polar regions do not affect convection/precipitation. Including them in the bias adjustments may have the undesirable effect of codifying temperature anomalies into the results that we would otherwise like to remove with nudging. Applying a spatial window to the bias adjustments may address this problem.
  4. Both {T} and {Q} are impacted by and affect the behavior of the convective parameterizations. While clearly anomalous results are not observed when nudging {Q}, it is possible that more subtle problems are being introduced.

Calculating Temperature Bias:

Using the model results for the 2 year YOTC experiments, the annual cycle fit of temperature differences using 6 harmonics is calculated at each ne30 grid point and model level. Only 3 of the 6 harmonics will likely be used.

: Program to read YOTC nudging results and calculate the annual cycle fits.
: Program to read in AC coefs, calculate zonal average values in 18 latitude bands.
AC_Dataset.F90 : Module to create/utilize the netcdf file of AC coefficients.
Lookat_AC_Bias.ncl : Program to graph AC fit coefficients.

Bias Datasets
Bias01 Bias values between {UVTQ}-{UV} nudging experiments.
Bias02 Zonal average of Bias01 values using 18 latitude bands.
Bias03 Bias values between {UVTQ}-{UVQ} nudging experiments.
Bias04 Zonal average of Bias03 values using 18 latitude bands.

The following table shows the Mean and first 3 harmonics of the Annual cycle at 3 model levels calculated from the E001, E002, and E004 experiments. Also shown are the zonal mean profiles calculated using 18 latitude bins.

Zonal Mean Profile
Model Level 25
Model Level 15
Model Level 5


For consistency, the incorrect experiment nomenclature will continue for the remainder of this study.
{T} Bias Adjustment
Run Name
          {UVTQ}-{UV  } (Bias01)
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E005 E005
{UVTQ}=1.0           {UVTQ}-{UVQ} (Bias03)
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E006 E006
{UVTQ}=1.0 Zonal {UVTQ}-{UV  } (Bias02)
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E007 E007
{UVTQ}=1.0 Zonal {UVTQ}-{UVQ} (Bias04)
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E008 E008

AMWG Diagnostics:

E005 - OBS
E005 - E002
E005 - E001
E006 - OBS
E006 - E002

E007 - OBS
E007 - E002

E008 - OBS
E008 - E002


In order to verify the application of the temperature bias in the model, compare the zonal mean temperature differences in E005-E002 and E002-E001 above. The results show that the added temperature Bias01 is achieving the desired result.

The following table contains the total precipitation compared with GPCP for the 4 bias adjustment experiments.

Finding: Time Mean Vs. Zonal-Time Mean
For each case, using zonal-time mean bias profiles rather than time mean profiles at each SE grid point does not have a significant impact on the precipitation results. Compare maps and zonal profiles for (E005 Vs. E007) and (E006 Vs. E008).

 Applying zonal-time mean bias values acts to balance the precipitation differences in the tropics/sub-tropics, leaving significant differences at all latitudes.

Finding: E005 Vs. GPCP Precipiation
Applying bias values relative to the {UV}-nudged case [BIAS01] leads to improved/increased annual total precipitation, with a global mean value of 2.10 (mm/day) compared to 1.58 (mm/day) for E001 above. This is too small compared to the GPCP mean of 2.67. As indicated in the zonal mean profiles, the time mean bias corrections to temperatures produces better agreement in the tropics, but at the expense of subtropical precipitation.

Finding: E005 Vs. TRMM (not shown)
Comparison of tropical precipitation with TRMM also shows better agreement, with a mean difference of only 0.01 (mm/day).

Finding: E006 Vs. GPCP
Applying bias values relative to the {UVQ}-nudged case [BIAS03] gives a better result with a global mean of 2.58 (mm/day), compared to 2.67 (mm/day) for GPCP. The zonal mean profile shows that the increase in the mean resulted from improved agreement in the extra-tropics, but also from larger overestimates of precipitation in the tropics.

Finding: E006 Vs. TRMM (not shown)
Comparisons of tropical precipitation rates with TRMM hi-lite the increase in the model results, with a mean of 3.15 (mm/day) compared to the TRMM mean of 2.72 (mm/day).

Annual Precipitation Vs. GPCP
Zonal Mean Profiles

Time Mean
Temperature Bias

    E001 -  E002

{UVTQ} - {UV}

Time Mean
Temperature Bias

    E001 -  E004
{UVTQ} - {UVQ}

Zonal-Time Mean
Temperature Bias

    E001 -  E002

{UVTQ} - {UV}
E007-Precip E007-PRECT-GPCP-Profile

Zonal-Time Mean
Temperature Bias

    E001 -  E004
{UVTQ} - {UVQ}
E008-Precip E008-PRECT-GPCP-Profile


Nudging {T} in addition to {UVQ} leads to unrealistic precipitation values. Diagnostic studies showed that in nudging experiments without temperature nudging, there was a nearly uniform 1K bias between model results and ERA-I temperatures. Nudging the temperatures into agreement with ERA-I, specifically in the troposphere, directly influenced the precipitation values. With the assumption that modifying the tropospheric temperature structure via nudging  was having a detrimental effect on the convective schemes, effort were made to preserve the temperature structure with the addition of bias values. Four methods for estimating these bias values were applied and tested. The results showed that the addition of the bias improved the agreement between resulting precipitation and GPCP values. However, none of these methods provided acceptable results, with the best nudged-model results being those for which only {UV} values were nudged.

Supplemental Experiments #2:


Since a problem with the application of temperature nudging in preliminary experiments was discovered and corrected, the experiments in this study have been focused on the addition of {T} nudging to cases in which {UVQ} values were being nudged. With the adverse effects the additional nudging had on precipitation values, assuming that it is the convective schemes that are being interfered with, efforts to bias adjust nudging temperatures were attempted. This left unexplored the possibility that the problem is not with the {T} nudging, but rather with the {Q} nudging. That the addition of {Q} nudging is not acting to improve results can be seen by comparing the zonal precipitation profiles for experiments E002 and E004.

The goal of these experiments is to evaluate the case of {UVT} nudging so that the only changes in water vapor values are due to advection and precipitation processes. Then to subsequently add nudging of {Q} at the lowest model level to this case in order to act as a source/sink of water vapor from the surface. 


Experiment Run Name Nickname
{UVT}=1.0 f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E009
{UVT}=1.0 {Qsfc}=1.0
    {T}=1.0 {Qsfc}=1.0
f.e13b6.FAMIPC5.ne30_ne30.NudgeV2_E011 E011

AMWG Diagnostics:

E009 - OBS
E009 - Exxx
E010 - OBS
E010 - Exxx
E011 - OBS
E011 - Exxx


Finding: E009 Vs. GPCP
The precipitation results are dramatically improved when applying only {UVT} nudging. The global mean precipitation is 2.86 (mm/day) compared to the GPCP value of 2.67 (mm/day), and there is clear agreement in the spatial distributions. The zonal profile indicates quite good agreement at all latitudes, with the overestimate of precipitation in the tropics significantly decreased compared to the E002 results.

Finding: E009 Vs. E004
Comparing the results for {UVT} nudging with those for {UVQ} nudging gives a clear indication that the poor results have actually emanated from the application of {Q} nudging rather than from the {T} nudging.

Finding: E010 Vs. GPCP
With the addition of {Q} nudging at the lowest model level, to act as a source/sink of water vapor, the precipitation results are further improved. Differences in the horizontal mean map are decreased and the model recovers the same global mean as GPCP. The maximum precipitation value in the zonal mean profile now agrees well with GPCP, though there is still an overestimate at the equator due to the higher maximum values there, particularly over central America.

Finding: E011 Vs. GPCP
For {T} only nudging, with {Q} nudged at the surface, there is a significant impact upon the tropical precipitation. A distinctly different pattern of precipitation is obtained in the tropics and the mean precipitation rate increases to 2.90 mm/day.  Most interesting is the improved agreement in the extra-tropical precipitation is in better agreement with the GPCP mean.

Annual Precipitation Vs. GPCP Zonal Mean Profiles

{UVT} Nudging

{UVT} Nudging
{Qsfc} Nudging

{T} Nudging
{Qsfc} Nudging


Nudging the model using only {UVT} values resulted in a dramatic improvement in precipitation values relative to GPCP and TRMM, establishing that the problem with previous nudging experiments emanated from the application of {Q} nudging. The most significant differences between the observational values and model results are in the amount of precipitation in the tropics. In hindsight, it now seems obvious that the artificial creation/removal of water vapor would be problematic for convective and precipitation processes. Since these are the only physical mechanisms by which water vapor can change.

Other than advection and precipitation, the only means by which water vapor should change in the atmosphere is via surface exchange. The addition of {Q} nudging at the surface represents a proxy for these sources/sinks of water vapor and led to a further improvement in the resulting precipitation values.

Nudging {T} only shows in improvement in precipitation values relative to GPCP at all latitudes except for 20S-20N. In the tropics, the precipitation results deviate from observations. This invites an experiment in which {UV} nudging is limited to the tropical latitudes.


From the AMWG diagnostic outputs, experiments E009 and E010 have been shown to produce representative results. In this section, carry out investigations utilizing the {h0/h1/h2} history output for those model runs.

Investigation 1: Systematic Annual Variations in Nudging Tendencies

Level 30 Nudge Q
Zonal Avg Nudge Q