The simulation result (GPP and LE) of CLM4.5 single point is much lower than the observational data!!!

[xxyii]

Dear all,
I tried to use CLM4.5 in CESM1.2.2 to perform a single-point simulation on the Qianyanzhou site in China in 2003-2005 year. Single-point surface data in 1850 and 2000 were made by ./mknoocnmap.pl, mkmapdata.sh,./gen_domain and ./mksurfdata.pl utilities. And the observed half-hour meteorological data from Global fluxnet (FULLSET Data Product - FLUXNET) were input as atmospheric forcing. I selected the CLM_USRDAT_NAME method in CLM4.5 for single-point simulation, and the compset I selected is I1PTCLM45CN （2000_DATM_1PT_CLM45%CN_SICE_SOCN_RTM_SGLC_SWAV .However, the obtained variables of interest(GPP, LE and H) are very different from the actual observations. I think the experimental data is real and reliable, and other related papers have used the same data source to simulate. The following is my meteorological data and the result file of the single point simulation.The time resolution of the output file is the day-scale mean.The atmospheric forcing data I made is stored in the attachment.



The results of the model simulation must be different from the observed data, but my simulation results are much smaller than the measured data. I have checked the process many times and I still don't know where the problem is. I sincerely hope someone can give me some suggestions and help.

XIuyi

 

[erik]

How did you spin the model up to start up your transient sequence of 2003-2005?

 

[xxyii]

How did you spin the model up to start up your transient sequence of 2003-2005?

I referd to the content in Chapter 4 of CLM4.5 in CESM1.2.0 User's Guide Documentation: Spinning up the CLM4.5 biogeochemistry (CLMBGC spinup). Because I chose the CN model, I only spin up to provide the initial files of the model for 650 years.

 

[xxyii]

I referd to the content in Chapter 4 of CLM4.5 in CESM1.2.0 User's Guide Documentation: Spinning up the CLM4.5 biogeochemistry (CLMBGC spinup). Because I chose the CN model, I only spin up to provide the initial files of the model for 650 years.
View attachment 1879

I cycled the atmospheric forcing data from 2003 to 2005, and spin up the initial files of the model for 650 years.

 

[oleson]

There could be a number of things going on here.
Check your surface dataset to see if the vegetation/soil types, etc. are specified in a manner consistent with your site (needleleaf evergreen forest?).
The incoming solar (FSDS) for this mid-latitude site (lat=27deg) seems a bit low to me (mid-day peak of 300 W/m2 in summer). Do you think it is correct?

 

[xxyii]

There could be a number of things going on here.
Check your surface dataset to see if the vegetation/soil types, etc. are specified in a manner consistent with your site (needleleaf evergreen forest?).
The incoming solar (FSDS) for this mid-latitude site (lat=27deg) seems a bit low to me (mid-day peak of 300 W/m2 in summer). Do you think it is correct?

Yes I checked my dataset, and the vegetation and soil types of suface data are basically consistent with my site real characteristics. As for the incoming solar(FSDS), it was downloaded from the global fluxnet website(https://fluxnet.org/data/download-data/), and the data quality is credible.

 

[oleson]

Ok, I didn't see that your plots are daily averages, so your peak solar of 300 makes more sense.
I assume sensible heat must overestimated since latent heat is underestimated?
How does your net radiation compare to observations?
Do you have plots of LAI, I assume it must be quite low compared to observations since GPP is so low...
Another thing to check is whether you are using present day CO2 forcing.
These are just basic checks that we do.

 

[xxyii]

Ok, I didn't see that your plots are daily averages, so your peak solar of 300 makes more sense.
I assume sensible heat must overestimated since latent heat is underestimated?
How does your net radiation compare to observations?
Do you have plots of LAI, I assume it must be quite low compared to observations since GPP is so low...
Another thing to check is whether you are using present day CO2 forcing.

These are just basic checks that we do.

The first figure below is a line graph of the daily simulated net radiation data (red) of CLM and the observed net radiation data (orange) of 2003.As you can see its simulation result is still underestimated. And the second figure is plot of daily LAI in 2003. In addition, I checked again I was using the current CO2 forcing and the value of CCSM_CO2_PPMV is 367.0. Is it possible that there is a problem elsewhere? I am really confused for this question for a long time.

Xiuyi

 

[oleson]

The very low net radiation may be an indication that the solar radiation forcing is not being treated properly by the model. It seems like the albedo must be very high and the surface is reflecting most of the incoming solar radiation. I would plot up the diurnal cycles of FSDS and FSR (at whatever time step you are running the model at). Maybe there is a mismatch between what the model and forcing data think the solar zenith angle is.

 

[xxyii]

The very low net radiation may be an indication that the solar radiation forcing is not being treated properly by the model. It seems like the albedo must be very high and the surface is reflecting most of the incoming solar radiation. I would plot up the diurnal cycles of FSDS and FSR (at whatever time step you are running the model at). Maybe there is a mismatch between what the model and forcing data think the solar zenith angle is.

Thanks for your patient reply. In the setting of my model atmospheric forcing data, as I have half-hour data, I only have two stream files.The following is a screenshot of the datm_atm_in setting in my CaseDocs folder. I don’t know if there is a problem. As you said, how can I check whether the sun zenith angle considered by the model and the forced data match? Can I give some more hints?

 

[oleson]

I would plot up the diurnal cycles of FSDS and FSR (at whatever time step you are running the model at). Check to see if the timing of the eflected solar is reasonable. For example, if the model receives solar radiation when it thinks the sun should be below the horizon then it will reflect all of that solar radiation. This would be an indication that the forcing and model times are not lined up properly.

 

[xxyii]

I would plot up the diurnal cycles of FSDS and FSR (at whatever time step you are running the model at). Check to see if the timing of the eflected solar is reasonable. For example, if the model receives solar radiation when it thinks the sun should be below the horizon then it will reflect all of that solar radiation. This would be an indication that the forcing and model times are not lined up properly.

I’m sorry to ask you again because some personal circumstances have passed for a long time.This question stiil puzzled me. I tried to draw a CLM4.5 simulation of the diurnal cycle of FSDS and FSR on January 1, 2003. The time resolution of the data is half an hour. It is found that from 10:30 in the morning, the solar radiation reflected on the ground is equal to the direct solar radiation. The same is true for other dates.

 

[xxyii]

I would plot up the diurnal cycles of FSDS and FSR (at whatever time step you are running the model at). Check to see if the timing of the eflected solar is reasonable. For example, if the model receives solar radiation when it thinks the sun should be below the horizon then it will reflect all of that solar radiation. This would be an indication that the forcing and model times are not lined up properly.

I also encountered the problem of time misalignment in the mode setting. I compared the FSDS data of a period of mode input and the FSDS data output of the model, and found that the output variable has a lag. The same problem exists for FLDS.



In addition, I would like to add the nc file information that I am forced to do. I don't know if there is a problem with the time setting. And the attachment is the nc file.


netcdf file:/C:/Users/wxy/Desktop/R_code/2003-01.nc {
dimensions:
scalar = 1;
lon = 1;
lat = 1;
time = 1488;
variables:
double EDGEE(scalar=1);
:units = "degrees E";
:long_name = "eastern edge in atmospheric data";

double EDGEN(scalar=1);
:units = "degrees N";
:long_name = "northern edge in atmospheric data";

double EDGES(scalar=1);
:units = "degrees N";
:long_name = "southern edge in atmospheric data";

double EDGEW(scalar=1);
:units = "degrees E";
:long_name = "western edge in atmospheric data";

double time(time=1488);
:units = "days since 2003-01-01 00:00:00";
:long_name = "Time axis";
:calendar = "noleap";

double FLDS(time=1488, lat=1, lon=1);
:units = "W/m2";
:long_name = "incident longwave (FLDS)";

double FSDS(time=1488, lat=1, lon=1);
:units = "W/m2";
:long_name = "incident solar (FSDS)";

double LATIXY(lat=1, lon=1);
:units = "degrees N";
:long_name = "latitude";

double LONGXY(lat=1, lon=1);
:units = "degrees E";
:long_name = "longitude";

double PRECTmms(time=1488, lat=1, lon=1);
:units = "mm/s";
:long_name = "precipitation (PRECTmms)";

double PSRF(time=1488, lat=1, lon=1);
:units = "Pa";
:long_name = "pressure at the lowest atm level (PSRF)";

double RH(time=1488, lat=1, lon=1);
:units = "%";
:long_name = "relative humidity at the lowest atm level (RH)";

double TBOT(time=1488, lat=1, lon=1);
:units = "k";
:long_name = "temperature at the lowest atm level (TBOT)";

double WIND(time=1488, lat=1, lon=1);
:units = "m/s";
:long_name = "wind at the lowest atm level (WIND)";

double ZBOT(time=1488, lat=1, lon=1);
:units = "m";
:long_name = "observation height";

// global attributes:
:site_location = "Latitude: 26.7414 Longitude: 115.0581 Elevation (masl): 23.";
}

 

[oleson]

The fact that you are reflecting all of the incoming solar radiation after 10:30am likely means that the model thinks it is nighttime. The model runs in GMT, maybe your forcing is in local time? You may need to shift your atmospheric forcing accordingly.
Regarding the lag issue, the first time slice of output from the model is actually an initialization step and not a full model time step. So you should compare the first time slice of input with the second time slice of output, etc.

 

[xxyii]

Dear Oleson，
I checked that the time of the forced data I entered was indeed in local time, and when I adjusted the input data and the compared observation data to the standard(GMT), I got the model simulation results that seemed to be normal! The following is the result of my re-simulation（H、LE and GPP）. The result has been running long ago, but due to network and other reasons, I cannot log in to the forum and let you know in time.
Thank you again for your patient answers. I now have a better understanding of the operation of the model.

Xiuyi

 

[oleson]

That looks more reasonable. Another thing to check would be the correlation coefficient between modeled and observed net radiation. Normally, the correlation will be well above 0.9.

 

[xxyii]

That looks more reasonable. Another thing to check would be the correlation coefficient between modeled and observed net radiation. Normally, the correlation will be well above 0.9.

I made a line graph and a scatter plot of the simulated net radiation and the observed net radiation data, and found that the correlation between the two is really very high, and R square reaches 0.97.

 

[oleson]

Great, that seems reasonable.