How to use my atm foring datasets to replace the default GSWP3 ?

[jack]

If you believe that the forcing data is setup correctly, then I wouldn't recommend moving it around just to get model output variables to match up with whatever in-situ data you might have. There could be many reasons why model output, e.g., TG doesn't agree with in-situ data (e.g., surface characterization, deficiencies in parameterizations). I assume you are accounting for any difference in time zone between the in-situ and model data.

oleson, thank you for your help . There are four questions. (1) Is there any configuration of greenwich (e.g., true or false) in CLM model? (2) The in-situ LST was calculated based on atmospheric downward longwave radiation, surface upward longwave radiation and ground emissivity, If I want to compare model output 0.5hrly LST with in-situ LST, which output variable (e.g. TG and TSKIN) should I adopt ? (3) According to your opinion, I doubt the LST peak difference (1-1.5hr) between model output TSKIN and in-situ LST may be due to time zone and greenwich time configuration. My study area is located at 100.4°E (local time=UTC+6.7), the model forcing data is UTC time, in-situ data is recorded with Beijing time (i.e., site time =UTC +8), so I compared the first value output from the model（00:00 UTC）with the 8:00 recorded by the site, then I found the model outpu TSKIN was 1-1.5 hours later than in-situ LST. Do you have any suggsetion? This problem has troubled me for a long time (see the screenshot1 and 2 below) (4) For 0.5 hourly model output, I found the model output TBOT and FSDS（ bilinear interpolation）were always 0.5 hour later than the input TBOT and FSDS, thus resulting 0.5/1 hour later of TG/TSKIN than in-situ LST. In addition, when the interpolation method of solar was changed to coszen, the output TG/TSKIN was found 1/1.5 hour later than in-situ LST. In principle, the model output should not change input FSDS and TBOT. Do you know why? (see the screenshot3(FSDS) and screenshot4 (TBOT) below).
note: my forcing data Precip: first time stamp is 0 and the offset = -5400. Solar: first time stamp is 0 and the offset = -5400. TPQW: first time stamp is 0 and the offset = 0. CLDAS LST is an LST(TG) product with 1hr resolution.

 

[oleson]

(1) No.
(2) I've answered this question before in your other posts, but I'd recommend using TSKIN because it includes vegetation temperature whereas TG is just ground (soil) temperature.
(3) Just checking. I don't have any more suggestions.
(4) It is not recommended to set the interpolation of solar radiation to anything other than coszen when the forcing data is at a coarser time resolution than the model time step. The coszen interpolation is designed to distribute all of the incoming solar radiation from the forcing file according to the model's solar zenith angle calculation. Other methods of interpolation may result in "lost" solar radiation at the beginning and/or end of the day.

 

[jack]

From the User's Guide (1.2.4. Customizing the DATM namelist — ctsm CTSM master documentation)

For coszen the time-stamps of the data should correspond to the beginning of the interval the data is measured for. Either make sure the time-stamps on the datafiles is set this way, or use the offset described above to set it.
For nearest and linear the time-stamps of the data should correspond to the middle of the interval the data is measured for. Either make sure the time-stamps on the datafiles is set this way, or use the offset described above to set it.

My interpretation of this, applied to your data, is that since solar is a mean of -1.5 to 1.5 hours, either set the first time stamp to -5400/86400 = -0.0625 with offset = 0, or set the first time stamp to 0 and the offset = -5400.
For the other variables (TPQWL and precip), the first time stamp should be set to the middle of the interval the data is measured for. If not, then the offset can be used to move the data forward or backward. For example, if precipitation is the average of 0 to 3 hours, then the first time stamp should be 5400/86400 = 0.0625. For instantaneous data, I think the time stamps would be set to the time the data is valid for. E.g., if the first value of temperature is instantaneous at 1.5 hours, then set the first time sample to 0.0625 and offset = 0.

thanks,oleson. The last check, the TPQWL is instantaneous data at UTC 00:00 (i.e. the first value is UTC 0), so set the first time sample to 0 and offset = 0 is the right configuration? and I wanna ask that does preciptation influence the calculation of TG or TSKIN?

 

[jack]

Just one thing makes me very confused ： I found the peak of model output FSDS is UTC 6:00, however the peak of model output TG and TSKIN is 1.5 hour later than FSDS, i.e. at UTC 07:30, can you give some advice?

 

[jack]

oleson, thank you for your help . There are four questions. (1) Is there any configuration of greenwich (e.g., true or false) in CLM model? (2) The in-situ LST was calculated based on atmospheric downward longwave radiation, surface upward longwave radiation and ground emissivity, If I want to compare model output 0.5hrly LST with in-situ LST, which output variable (e.g. TG and TSKIN) should I adopt ? (3) According to your opinion, I doubt the LST peak difference (1-1.5hr) between model output TSKIN and in-situ LST may be due to time zone and greenwich time configuration. My study area is located at 100.4°E (local time=UTC+6.7), the model forcing data is UTC time, in-situ data is recorded with Beijing time (i.e., site time =UTC +8), so I compared the first value output from the model（00:00 UTC）with the 8:00 recorded by the site, then I found the model outpu TSKIN was 1-1.5 hours later than in-situ LST. Do you have any suggsetion? This problem has troubled me for a long time (see the screenshot1 and 2 below) (4) For 0.5 hourly model output, I found the model output TBOT and FSDS（ bilinear interpolation）were always 0.5 hour later than the input TBOT and FSDS, thus resulting 0.5/1 hour later of TG/TSKIN than in-situ LST. In addition, when the interpolation method of solar was changed to coszen, the output TG/TSKIN was found 1/1.5 hour later than in-situ LST. In principle, the model output should not change input FSDS and TBOT. Do you know why? (see the screenshot3(FSDS) and screenshot4 (TBOT) below).
note: my forcing data Precip: first time stamp is 0 and the offset = -5400. Solar: first time stamp is 0 and the offset = -5400. TPQW: first time stamp is 0 and the offset = 0. CLDAS LST is an LST(TG) product with 1hr resolution.

just looked at the picture of FSDS, It peaked at UTC 6:00, but the TG/TSKIN peaked at UTC 7:00/7:30

 

[oleson]

Seems reasonable regarding TPQWL.
Precipitation affects TG/TSKIN indirectly in that it will affect soil moisture and hence the heat capacity of the soil.
For some perspective on the relationship between FSDS and TG/TSKIN see:

Solar radiation is the main cause for diurnal variations on land. Looking at this slightly differently than how it is normally done helps to better understand observations and evaluate models of the land surface

My former postdoc, Maik Renner, just got his paper published in the Journal of Hydrometeorology, in which he evaluated the performance of common land surface models at the diurnal time scale using …

earthsystem.org

 

[jack]

Seems reasonable regarding TPQWL.
Precipitation affects TG/TSKIN indirectly in that it will affect soil moisture and hence the heat capacity of the soil.
For some perspective on the relationship between FSDS and TG/TSKIN see:

Solar radiation is the main cause for diurnal variations on land. Looking at this slightly differently than how it is normally done helps to better understand observations and evaluate models of the land surface

My former postdoc, Maik Renner, just got his paper published in the Journal of Hydrometeorology, in which he evaluated the performance of common land surface models at the diurnal time scale using …

earthsystem.org

thanks

 

[jack]

One more quesition, if my atmospheric forcing is with 1hr resolution (all of the variables are instantaneous data starting from UTC 0)
(1) What DATM_MODE should be set ? CLM_1PT or the default GSWP3 with interpolation method of all data "nearest"?
(2) I set all the first time stamp to 0 and the offset = 0 of precip, solar and TPQWL, is that right?
(3) Which time step does the CLM model start reading data from? 0 or 1？
In other word, will nearest or bilinear interpolation method change the time line of input atmospheric forcing? For example, the maximum of input TBOT or FSDS is at UTC 06:00, after interpolation, will the maximum of model output TBOT or FSDS occurs at UTC 07:00?
Thanks for your kind help in advance.

 

[jack]

One more quesition, if my atmospheric forcing is with 1hr resolution (all of the variables are instantaneous data starting from UTC 0)
(1) What DATM_MODE should be set ? CLM_1PT or the default GSWP3 with interpolation method of all data "nearest"?
(2) I set all the first time stamp to 0 and the offset = 0 of precip, solar and TPQWL, is that right?
(3) Which time step does the CLM model start reading data from? 0 or 1？
In other word, will nearest or bilinear interpolation method change the time line of input atmospheric forcing? For example, the maximum of input TBOT or FSDS is at UTC 06:00, after interpolation, will the maximum of model output TBOT or FSDS occurs at UTC 07:00?
Thanks for your kind help in advance.

do you have any opinion?