WRF-SUEWS evaluation in Shanghai case about wrf-suews HOT 62 CLOSED

from wrf-suews.

Building heights based on gis (provided by my colleague) are put into wrfinput. It looks like:

I run the model for 13 days from 2013-08-06 to 2013-08-19 and there was no rain during this period. Both the observation and model output are at 30-min interval. Here are some results (LH may have some abnormal values):

from wrf-suews.

Good job!

from wrf-suews.

For QE obs, better to remove the outliers.

from wrf-suews.

Also, provide some statistics, e.g. MAE, MBE, to show the performance.

from wrf-suews.

Yes, I will try tomorrow. I am new to data visualization using python, so it takes a long time to achieve these.

from wrf-suews.

Figs above are updated and T2 and QF are plotted.

from wrf-suews.

from wrf-suews.

from wrf-suews.

@suegrimmond : I think he updated the results above for the newest version. I think it would be good to have the both previous results and the new version.
@zhenkunl : are these updated version after updating the land cover from data?

from wrf-suews.

The modelled T2 – is that diagnosed by SUEWS or WRF/original?

By SUEWS.

@zhenkunl: let's focus on the SEB results first; the diagnostics (e.g., T2) make less sense to see at this stage as the bias may be attributable to various factors.

from wrf-suews.

@hamidrezaomidvar I haven't done a new run and I just dropped the abnormal values in observations.
@sunt05 What does SEB mean here?

from wrf-suews.

SEB: surface energy balance.

from wrf-suews.

@hamidrezaomidvar @sunt05 when I run offline SUEWS for Shanghai to spin up, it is possible I only load the initial conditions if I change the path_runcontrol in file wrfinput-processor/param_extractor_SuPy/getting_SUEWS_params.py as df_forcing keep the values from the sample data.

  print('Initializing SUEWS variables.....')
  df_state_init, df_forcing = sp.load_SampleData()

  path_runcontrol = Path('./run_London') / 'RunControl.nml'
  df_state_init = sp.init_supy(path_runcontrol)

I think I should add

grid = df_state_init.index[0]
df_forcing = sp.load_forcing_grid(path_runcontrol, grid)

It that right?

from wrf-suews.

from wrf-suews.

The land use information are updated and I rerun the model for three days. Here are the flux components:

from wrf-suews.

@zhenkunl What are the LANDUSEF_SUEWS for the pixel that you are comparing the results?

from wrf-suews.

sorry for late reply. LANDUSEF_SUEWS of XJH site are [0.5, 0.5, 0. , 0. , 0. , 0. , 0. ]. I can't get more accurate land use info.

from wrf-suews.

It seems WRF-SUEWS has some problem with goddard lw/sw radiation scheme. It had run for hours but crashed. The error logs are:

running goddard lw radiation
running goddard sw radiation
d01 2013-08-11_06:00:00  Input data is acceptable to use:
d01 2013-08-11_12:00:00  Input data is acceptable to use:
d01 2013-08-11_18:00:00  Input data is acceptable to use:
d01 2013-08-12_00:00:00  Input data is acceptable to use:
d01 2013-08-12_06:00:00  Input data is acceptable to use:
forrtl: error (69): process interrupted (SIGINT)
Image              PC                Routine            Line        Source
wrf.exe            0000000003447861  Unknown               Unknown  Unknown
wrf.exe            0000000003445FB7  Unknown               Unknown  Unknown
wrf.exe            00000000034011C4  Unknown               Unknown  Unknown

from wrf-suews.

If I have high resolution land use categories data based on SUEWS (i.e. seven types), how can I put it into WRF-SUEWS without reclassification to the MODIS 21-category?

from wrf-suews.

20190719 meeting

• radiation issue
  @zhenkunl to check the popular WRF physics configurations used in SIMS.

• land cover

(@hamidrezaomidvar to modify this ) #53

@zhenkunl: to provide an up-to-date building footprint data to @hamidrezaomidvar.

from wrf-suews.

I have run a test run with both ra_lw_physics and ra_sw_physics equal 24, here are the results:

More work need to be done to trace the error sources.

from wrf-suews.

Do we know if there has been any evaluations in the past by others of the WRF modelling Kdown in Shanghai? If not there may be something interesting there to have a look at. If yes, please can we determine what was concluded, settings etc

from wrf-suews.

I talked with one of postdocs who work with Baode Chen today. She told me their BL scheme has not been released by WRF but it is been preparing. She has lots of modelling experiences with cloud and aerosol. We discussed the kdown bias in WRF-SUEWS. Unfortunately, She didn't compare the radiation results before. She said kdown may be reduced when accounting for aerosol in WRF, however, the gap between model and obs seems too large even if we change radiation scheme or add aerosol in WRF. Also she provided the wrfout data of one case which start on 20160719. The kdown is as below.

The maximum is close to 1000w/m2 in the figure. So I think I should double check the radiation observations are reliable or not first.

from wrf-suews.

I have run another case with all the settings same as previous one except using Noah land-surface model. The following figure shows the comparison of SWDOWN. The top is based on WRF-Noah, and the bottom WRF-SUEWS.
We can see the magnitudes are similar, so we can conclude that WRF may overestimate SWDOWN over Shanghai area, and it can not be improved by land surface model.

from wrf-suews.

Sorry I can't find valid observations of SWDOWN, though I download the ERA5 hourly reanalyses. Here is the ERA5 surface solar radiation downwards.

from wrf-suews.

Please put these, ERA-5, WRF simulations and observations, together for a comparison.

from wrf-suews.

from wrf-suews.

It looks WRF does overestimate Kd, even compared with ERA-5.

@zhenkunl for your WRF simulation, which forcing did you use? FNL or ERA-5?

from wrf-suews.

from wrf-suews.

I have run cases with one domain around center Shanghai (9x9 grids and 9km resolution) for quick test. First, I ran WRF-SUEWS with original Modis landuse fraction (refer to as Modis Landuse), then a sensitive experiment with modification of the landuse fraction according to Ao's paper. Specifically, I set urban fraction of all the grids which exceed 90% to 85% and it was split into 62% paved and 23% buildings, and the other five types have values of 3%, 1%, 10%, 0, 1% respectively (refer to as Ao's Landuse). The landusef distribution of urban type are shown as below.

from wrf-suews.

I compared the flux and T2 of the two cases.


The differences of corresponding flux component are very small. It shows the bias between the model and observation can not be reduced by more accurate land use. Also I have tested other shortwave radiation schemes, the results are similar.

from wrf-suews.

I ran another case using the original WRF with Noah LSM as well. Since there are no QN, KUP and LUP in wrfout, I only compared the other components.

from wrf-suews.

It is strange that the phase of anthropogenic heat flux of WRF-SUEWS is not consistent with observation. Here observation means LUCY output from Ao.

from wrf-suews.

@zhenkunl Is the time zone correct? the WRF outputs are in UTC. Also check the time saving in namelist.suews

from wrf-suews.

I changed timezone to 8 in module_sf_suews.F, but the result are the same as before. Any ideas?

from wrf-suews.

just looked at the code, I think this is a bug/caveat: timezone is only used in calculation of sun positions but nowhere else. so for QF calculations, the UTC time, rather than local time, is still used.
Let me think how can we fix this.

from wrf-suews.

from wrf-suews.

Good idea for the workaround! @hamidrezaomidvar

from wrf-suews.

Some further thoughts on this:

1. as WRF uses UTC while SUEWS uses local time, we may change this at the interface: SuMin.
2. as only human activity related profiles are impacted by this issue, we can adjust the profiles during SUEWS_init at the WRF site; while users can still provide profiles in LST, which might be more intuitive.

from wrf-suews.

from wrf-suews.

I think we just need to pass timezone to sue_init and rotate the profile that is written in LOcal time
…

I think this would be a better solution.

from wrf-suews.

I have rotated the AHProf_24hr, then QH looks fine.

So the sensible flux becomes reasonable as well.

from wrf-suews.

from wrf-suews.

Could be due to daylight saving related calculations.

from wrf-suews.

from wrf-suews.

@sunt05 I have done a case with setting all the urban grids to 100% cropland. Here are the results.

from wrf-suews.

But China doesn’t have it – unless it is turned on when it should not be! Best wishes Sue

Yep, seems to be the case: daylight saving magically happened.

@zhenkunl, you may need to check how the DLS-related parameters are set in your simulation.

from wrf-suews.

But why LH is so low?

from wrf-suews.

@sunt05 @hamidrezaomidvar Could you tell me if the sequence of AHProf_24hr means UTC time or local time, or which time representation is expected in AHProf_24hr? I just rotate the array similar to cshift(AHProf_24hr, 8)

from wrf-suews.

1. Looks like there is a timing problem in Lup also
2. Is QF= gone to 0 ? Please can we plot this each time. (if not- we need to modify SUEWS to check that population data /QF is sensible given the land surface choices (i.e. if 100 % veg or water etc--t there should not be a large population -- or there should be some comment put out.

from wrf-suews.

@sunt05 @hamidrezaomidvar Could you tell me if the sequence of AHProf_24hr means UTC time or local time, or which time representation is expected in AHProf_24hr? I just rotate the array similar to cshift(AHProf_24hr, 8)

Please can you check what your DLS settings are

from wrf-suews.

startdls = 180
enddls = 181

from wrf-suews.

@suegrimmond Here is QF

from wrf-suews.

from wrf-suews.

I agree with you. I should change the land use fraction before scaling

from wrf-suews.

In the preprocessing, some of the variables are urban related meaning they can scale or not scale based on the urban fraction (note sure which one was the case, please check) so for example population is scaled based on the urban fraction(unless we give the inputs the real population data as the case in London). So you want to make sure if you give one value to population to all grids, they are actually scaled. You can do this in the json file listing all the variables in the preprocessing ( Never thought I will send an email related to WRF-SUEWS in skies ) ...

@hamidrezaomidvar one more evidence the coupling really works! Re land-atmosphere interactions due to WRF-SUEWS!

from wrf-suews.

This time I modified urban to cropland first, then scaled all the variables. So QF=0 during the whole period, but the latent heat flux is still close to zero.

from wrf-suews.

So what is the soil moisture?
Please can we plot the SMD

from wrf-suews.

As far as I know, SUEWS doesn't use SMOIS in the wrfinput, but use SOILMOIST_SUEWS which is assigned values by ourselves instead. This is timeseries of moisture of grass, which has a 100% fraction in my case.

from wrf-suews.

from wrf-suews.

Given the status of this discussion, I see two specific and impending issues to resolve:

1. overestimated Kdown by WRF in Shanghai;
2. bias in simulated QF and QH fluxes due to incorrect timing in profiles of human activity set in WRF-SUEWS.
   So I'll close and reference this in other more specific threads.

from wrf-suews.