Problems encountered in conducting the 1850-2100 simulation

[zjulin]

Hi, I am a novice to CESM. I am using CESM2.1.0 and several problems were encountered.

First, I know that "Pre-industrial" means all the forcings (greenhouse gases, ozone, aerosols, etc.) are fixed at their 1850 values. So what is the meaning of 2000: present day? Is it the variables above fixed at their 2000 values? If so, what about HIST? Is it the variables values from historical real observations?

Second, since I want to conduct a simulation in 1850-2100, so which compsets should I choose? I checked out B1850, BHIST and BRCP85C5L45BGC compsets, and I found differences in their *_in file, such as atm_in and pop_in. In my opinion, I should use BHIST compset and set STON_N=250, STOP_OPTION=nyears and RUN_STARTDATE=1850-01-01. Do I need to download extra inputdata in addition to the files required by default for the BHIST compset?
Additionally, if there's something wrong with me, please tell me which compset should I choose and how to set env_run.xml.

Last is extra questions. If I want to change my simulation year to 1800-2100, I think I cannot use only one compset because I found BHIST inputdata is always started from 1850. So should I run the first 50 years with B1850 and then run from 1850 to 2100 utilizing the restart file before with BHIST compset? Again, I hope to correct my mistakes.

Sincerely hope to get your advice.
Best wishes!

 

[katec]

Hi there, so it sounds like you have already started to look at the namelists and the forcing datasets that are described for each of these runs. Typically, the B1850 run uses "pre industrial" forcing, so the files reflect what is going on around 1850, and the forcing doesn't change. Similarly, the F2000 compsets and the like reflect forcing from around the year 2000 and do not change. The BHIST compset has access to forcing files that reflect the observations at a specific year or time between 1850 and 2014. So, you can use this compset to run the model for multiple years with changing forcing along the way.

The RCP or SSP compsets have access to forcing files that reflect a projected climate state between 2015 and 2100 (typically, we have some SSPs that go on to 2300).

For a run between 1850 and 2100 you will need to break it into the two parts. The first from 1850-2014 using the BHIST forcing and the second from 2015 to 2100 using whichever future projection forcing you want for your experiment.

Your env_run settings are close to right, except that you can't do it all in one block (ie, STOP_N=250) for two reasons. One is that you will need to switch from observed to projected forcing (change the compset) the other is that on most machines you will have a limited time for your job. You should choose a STOP_N number of years that will fit within the wall clock limits on your computer. So, if you have a 12 hour limit, you should set STOP_N=5 years because that will likely finish in the 12 hours. Then set RESUBMIT=32 (32*5=160 years) to resubmit in 5 year chunks and from 1850 to 2010.

To run from 1800 is more complicated because I don't think we have data for the actual years between 1800 and 1850. You may need to create that forcing data yourself, or find somebody else who has done a similar run.

 

[zjulin]

Hello katec! Thank you very much for your kind reply! But at the same time I have new questions.

You said 'the second from 2015 to 2100 using whichever future projection foricing you want for your experiment', but I found only 2 compsets called BRCP85C5L45BGC and BRCP85L45BGCR in CESM2.1. I can't seem to find the SSP and RCP compsets you mentioned. So can I only use command './create_newcase --case xxx --compset BRCP85C5L45BGC --res xxx --machine xxx --run-unsupported' to create simulation 2015-2100? If the command is true, what about other SSP or RCP? And if the command is false, I hope you can correct me.

In addition, when setting env_run.xml, could I change REST_N or OPTION to match settings you mentioned, such as setting REST_N=5 and REST_OPTION=nyears to avoid possible hpc errors?

Once again, thanks for your advice and hope to get your guidance.
Best wishes!

 

[zjulin]

Hi there, so it sounds like you have already started to look at the namelists and the forcing datasets that are described for each of these runs. Typically, the B1850 run uses "pre industrial" forcing, so the files reflect what is going on around 1850, and the forcing doesn't change. Similarly, the F2000 compsets and the like reflect forcing from around the year 2000 and do not change. The BHIST compset has access to forcing files that reflect the observations at a specific year or time between 1850 and 2014. So, you can use this compset to run the model for multiple years with changing forcing along the way.

The RCP or SSP compsets have access to forcing files that reflect a projected climate state between 2015 and 2100 (typically, we have some SSPs that go on to 2300).

For a run between 1850 and 2100 you will need to break it into the two parts. The first from 1850-2014 using the BHIST forcing and the second from 2015 to 2100 using whichever future projection forcing you want for your experiment.

Your env_run settings are close to right, except that you can't do it all in one block (ie, STOP_N=250) for two reasons. One is that you will need to switch from observed to projected forcing (change the compset) the other is that on most machines you will have a limited time for your job. You should choose a STOP_N number of years that will fit within the wall clock limits on your computer. So, if you have a 12 hour limit, you should set STOP_N=5 years because that will likely finish in the 12 hours. Then set RESUBMIT=32 (32*5=160 years) to resubmit in 5 year chunks and from 1850 to 2010.

To run from 1800 is more complicated because I don't think we have data for the actual years between 1800 and 1850. You may need to create that forcing data yourself, or find somebody else who has done a similar run.

Hello katec! Thank you very much for your kind reply! But at the same time I have new questions.

You said 'the second from 2015 to 2100 using whichever future projection foricing you want for your experiment', but I found only 2 compsets called BRCP85C5L45BGC and BRCP85L45BGCR in CESM2.1. I can't seem to find the SSP and RCP compsets you mentioned. So can I only use command './create_newcase --case xxx --compset BRCP85C5L45BGC --res xxx --machine xxx --run-unsupported' to create simulation 2015-2100? If the command is true, what about other SSP or RCP? And if the command is false, I hope you can correct me.

In addition, when setting env_run.xml, could I change REST_N or OPTION to match settings you mentioned, such as setting REST_N=5 and REST_OPTION=nyears to avoid possible hpc errors?

Once again, thanks for your advice and hope to get your guidance.
Best wishes!

 

[katec]

HI there, so most of the SSP compsets from the CMIP6 experiments are available in CESM 2.1.3 and higher versions. You should always be using the most recent release version of the model series for machine and compiler support. You can see the SSP experiments that are available at this site:

CESM2.1.3 Component Sets Definitions

Component Sets Definitions for CESM2.1.3

docs.cesm.ucar.edu

And type "BSSP" into the search bar.

Your create_newcase command for the SSP runs looks fine to me, but you will probably want to specify a Start date in the case once you have it created. Like most other CESM experiments, there will likely be some configuration you want to do in the case.

For your last question, if you haven't changed REST_N or REST_OPTION in the case, then it should automatically set itself to whatever you make STOP_N and STOP_OPTION. But if you want to be extra careful, you can make these changes explicitly, and it will be fine too. So, yes, feel free to set REST as well to avoid hpc errors.

 

[zjulin]

HI there, so most of the SSP compsets from the CMIP6 experiments are available in CESM 2.1.3 and higher versions. You should always be using the most recent release version of the model series for machine and compiler support. You can see the SSP experiments that are available at this site:

CESM2.1.3 Component Sets Definitions

Component Sets Definitions for CESM2.1.3

docs.cesm.ucar.edu

And type "BSSP" into the search bar.

Your create_newcase command for the SSP runs looks fine to me, but you will probably want to specify a Start date in the case once you have it created. Like most other CESM experiments, there will likely be some configuration you want to do in the case.

For your last question, if you haven't changed REST_N or REST_OPTION in the case, then it should automatically set itself to whatever you make STOP_N and STOP_OPTION. But if you want to be extra careful, you can make these changes explicitly, and it will be fine too. So, yes, feel free to set REST as well to avoid hpc errors.

Thanks for your advice! I have solved these problems

 

[xiangli]

Hi there, so it sounds like you have already started to look at the namelists and the forcing datasets that are described for each of these runs. Typically, the B1850 run uses "pre industrial" forcing, so the files reflect what is going on around 1850, and the forcing doesn't change. Similarly, the F2000 compsets and the like reflect forcing from around the year 2000 and do not change. The BHIST compset has access to forcing files that reflect the observations at a specific year or time between 1850 and 2014. So, you can use this compset to run the model for multiple years with changing forcing along the way.

The RCP or SSP compsets have access to forcing files that reflect a projected climate state between 2015 and 2100 (typically, we have some SSPs that go on to 2300).

For a run between 1850 and 2100 you will need to break it into the two parts. The first from 1850-2014 using the BHIST forcing and the second from 2015 to 2100 using whichever future projection forcing you want for your experiment.

Your env_run settings are close to right, except that you can't do it all in one block (ie, STOP_N=250) for two reasons. One is that you will need to switch from observed to projected forcing (change the compset) the other is that on most machines you will have a limited time for your job. You should choose a STOP_N number of years that will fit within the wall clock limits on your computer. So, if you have a 12 hour limit, you should set STOP_N=5 years because that will likely finish in the 12 hours. Then set RESUBMIT=32 (32*5=160 years) to resubmit in 5 year chunks and from 1850 to 2010.

To run from 1800 is more complicated because I don't think we have data for the actual years between 1800 and 1850. You may need to create that forcing data yourself, or find somebody else who has done a similar run.

Hi Katec,

Thanks for your detailed reply! I learned a lot from that.

I'm planning to run a fully-coupled simulation with fixed present-day forcing, so maybe a B2000 compset will work for me. However, I did not find the B2000 compset. Do we have this kind of compset? If not, how can I do this simulation properly?

Looking forward to your suggestions!

Thanks,
Xiang

 

[slevis]

In /cime/scripts, typing "./query_config --compsets" shows that there isn't a B2000 compset.

I'm quickly brainstorming the following:
- In the above result from "./query_config --compsets" you will see a "long" version of each compset. You can use the long instead of the short version, and you may also define a long version of your own. Based on B1850, B2000 may be:
2000_CAM60_CLM50%BGC-CROP_CICE_POP2%ECO_MOSART_CISM2%GRIS-NOEVOLVE_WW3_BGC%BDRD
- If this does not work, then you could create a B1850 case and manually change all year-1850 references to 2000. To do this, you may find guidance by also creating an F2000 and/or I2000 case and seeing how these differ from the B1850.

 

[xiangli]

In /cime/scripts, typing "./query_config --compsets" shows that there isn't a B2000 compset.

I'm quickly brainstorming the following:
- In the above result from "./query_config --compsets" you will see a "long" version of each compset. You can use the long instead of the short version, and you may also define a long version of your own. Based on B1850, B2000 may be:
2000_CAM60_CLM50%BGC-CROP_CICE_POP2%ECO_MOSART_CISM2%GRIS-NOEVOLVE_WW3_BGC%BDRD
- If this does not work, then you could create a B1850 case and manually change all year-1850 references to 2000. To do this, you may find guidance by also creating an F2000 and/or I2000 case and seeing how these differ from the B1850.

Hi slevis,

Great idea! I successfully created and submitted to run this case: ./create_newcase --case /work/xl468/cesm2.1/scratch/case/test_B2000_240305 --res f19_g17 --compset 2000_CAM60_CLM50%BGC-CROP_CICE_POP2%ECO%ABIO-DIC_MOSART_CISM2%NOEVOLVE_WW3_BGC%BDRD --mach duke --run-unsupported. The only difference between this long name and that of a B1850 case is the "2000" at the beginning ("1850" for B1850). In addition, I also had to add --run-unsupported in order to create this case.

In comparison, I also created a B1850 case: ./create_newcase --case /work/xl468/cesm2.1/scratch/case/test_B1850_240305 --res f19_g17 --compset 1850_CAM60_CLM50%BGC-CROP_CICE_POP2%ECO%ABIO-DIC_MOSART_CISM2%NOEVOLVE_WW3_BGC%BDRD --mach duke.

It turned out that the "B2000" case did have changed files. For example, I compared their atm_in:





lnd_in:



drv_in:



Does that mean there is "B2000" compset in CESM which is not scientifically supported though?

I'm also wondering if I could continue to create and run B2000 cases in this way.

Thanks,
Xiang

 

[slevis]

This means that you can use long names to define compsets that do not have abbreviated names. I'm not sure whether there's a limit to the range of compsets that you could define this way.

The answer to your last question depends on you as a researcher. Does this compset do what you want for your research?

 

[xiangli]

This means that you can use long names to define compsets that do not have abbreviated names. I'm not sure whether there's a limit to the range of compsets that you could define this way.

The answer to your last question depends on you as a researcher. Does this compset do what you want for your research?

Hi slevis,

Yes, this compset, with forcing fields fixed at 2000, is exactly what I need for my research.

So I'll use the compset this way!

Thanks,
Xiang

 

[Ceil]

Hi there, so it sounds like you have already started to look at the namelists and the forcing datasets that are described for each of these runs. Typically, the B1850 run uses "pre industrial" forcing, so the files reflect what is going on around 1850, and the forcing doesn't change. Similarly, the F2000 compsets and the like reflect forcing from around the year 2000 and do not change. The BHIST compset has access to forcing files that reflect the observations at a specific year or time between 1850 and 2014. So, you can use this compset to run the model for multiple years with changing forcing along the way.

The RCP or SSP compsets have access to forcing files that reflect a projected climate state between 2015 and 2100 (typically, we have some SSPs that go on to 2300).

For a run between 1850 and 2100 you will need to break it into the two parts. The first from 1850-2014 using the BHIST forcing and the second from 2015 to 2100 using whichever future projection forcing you want for your experiment.

Your env_run settings are close to right, except that you can't do it all in one block (ie, STOP_N=250) for two reasons. One is that you will need to switch from observed to projected forcing (change the compset) the other is that on most machines you will have a limited time for your job. You should choose a STOP_N number of years that will fit within the wall clock limits on your computer. So, if you have a 12 hour limit, you should set STOP_N=5 years because that will likely finish in the 12 hours. Then set RESUBMIT=32 (32*5=160 years) to resubmit in 5 year chunks and from 1850 to 2010.

To run from 1800 is more complicated because I don't think we have data for the actual years between 1800 and 1850. You may need to create that forcing data yourself, or find somebody else who has done a similar run.

Hi katec, many thanks for your detailed information! I also have a question.

To get simulations that closer to observations, is BHIST better than FHIST or FHIST_BGC? I used the FHIST (with the default model setting) to get some simulations, but it seems there are large biases against observations. My purpose is to do model validation using field observations.

In addition, I also noticed that there are different values for setting CO2 as below:
co2_ppmv:
Valid Values ['any real']
Possible Default Values is 379.0 for: {'sim_year': '1000'}
is 336.6 for: {'sim_year': '1979'}
is 379.0 for: {'sim_year': '2000'}
is 388.8 for: {'sim_year': '2010'}
is 397.5 for: {'sim_year': '2015'}
is 284.7 for: {'sim_year': '1850'}
is 284.7 for: {'sim_year': 'PtVg'}
Should I manually change the values of CCSM_CO2_PPMV? e.g., when I run the period during 1979-1999, I should execute ./xmlchange RUN_STARTDATE=1979; ./xmlchange CCSM_CO2_PPMV=336.6, but when I run the period during 2000-2009, I should execute ./xmlchange RUN_STARTDATE=2000; ./xmlchange CCSM_CO2_PPMV=379.0.
Or the model can self-adjust the values of co2 during the running, so I do not need to do the above modification (i.e., I just directly submit the whole period during 1979-2009, and there is no need to cut the period into two parts and submit twice).

I am looking forward your early reply. Many thanks!

 

[slevis]

This should begin to answer your question...
In your FHIST or BHIST case, look in the env_run.xml:
The variable CCSM_CO2_PPMV says that it sets the CO2 for CLM if CLM_CO2_TYPE is constant or for POP if OCN_CO2_TYPE is constant.

The main difference between BHIST and FHIST is that BHIST has all components active (including ocean and sea-ice), while FHIST does not have active ocean and sea-ice. If you found large biases with FHIST, you may wish to experiment with BHIST, but I would not assume ahead of time whether it would come closer or not to observations.

 

[Ceil]

This should begin to answer your question...
In your FHIST or BHIST case, look in the env_run.xml:
The variable CCSM_CO2_PPMV says that it sets the CO2 for CLM if CLM_CO2_TYPE is constant or for POP if OCN_CO2_TYPE is constant.

The main difference between BHIST and FHIST is that BHIST has all components active (including ocean and sea-ice), while FHIST does not have active ocean and sea-ice. If you found large biases with FHIST, you may wish to experiment with BHIST, but I would not assume ahead of time whether it would come closer or not to observations.

Hi slevis,

Many thanks for your reply! They are very useful for me to understand the model.
I have another questions about the comparisons between the simulations and observations.


I used FHIST to run the model. But the running was always killed when the model outputed AF_FHIST_f09_cesm213.clm2.h0.2015-01-03-00000.nc

The error messages show as below,
Opened existing file
/cluster/work/users/inputdata/atm/cam/ozone_strataero/ozone_strataero_WACC
M_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180924.nc 2949120
Opened existing file
/cluster/work/users/inputdata/atm/cam/ozone_strataero/ozone_strataero_WACC
M_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180925.nc 2359296
ERROR:
GETFIL: FAILED to get /cluster/work/users/inputdata/atm/cam/ozone_strataer
o/ozone_strataero_WACCM_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180926.nc
ERROR:
GETFIL: FAILED to get /cluster/work/users/inputdata/atm/cam/ozone_strataer
o/ozone_strataero_WACCM_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180926.nc
ERROR:
GETFIL: FAILED to get /cluster/work/users/inputdata/atm/cam/ozone_strataer
o/ozone_strataero_WACCM_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180926.nc

Only ozone_strataero_WACCM_L70_zm5day_18500101-20150103_CMIP6ensAvg_c180923.nc is availble from https://svn-ccsm-inputdata.cgd.ucar.edu/trunk/inputdata/atm/cam/ozone_strataero/. To test the model, I copied this file to another two existing files shown above (i.e., *c180924.nc, *c180925.nc).

It seems impossible to have outputs after the date 2015-01-03. Because the used inputdata show the date 18500101-20150103. I wonder how to get the outputs after 2015, should I use ssp?


Another related question is, if I use F2010climo to obtain the outputs during 1979-2023, does the period 1979-2023 represent the years in the real world? Can we validate these outputs against field observations? Because we also can get the outputs after 2023, such as 2024 and 2029. It seems doing forecasts using F2010climo.

 

[slevis]

I think you are correct: You should continue the simulation as an ssp in order to access the input data past 2015.

I consider "climo" to mean climatology, so I'm pretty sure that the output will not represent real years in the real world.