Oversampling module results to satellite's pixels

[zxdawn]

Hi Kang @Kang-Sun-CfA ,

Your repository is very useful for resampling data.

Now, I'm trying to resample 3D model profiles to satellite pixels like TROPOMI, or reversely.
Does your method also work for this situation?

It seems you're updating the popy.py recently, is that the main script of oversampling now?

Thanks,
Xin

[Kang-Sun-CfA]

Hi Xin, Every 3D model would be different. I attempted to do that in popy.F_interp_profile, which is a wrapper around F_interp_gcrs. It only works for one version of GEOS-Chem. Feel free to add more. Kang

…

On Fri, Mar 20, 2020 at 10:14 AM Xin Zhang ***@***.***> wrote: Hi Kang @Kang-Sun-CfA <https://github.com/Kang-Sun-CfA> , Your repository is very useful for resampling data. Now, I'm trying to resample 3D model profiles to satellite pixels like TROPOMI, or reversely. Does your method also work for this situation? It seems you're updating the popy.py recently, is that the main script of oversampling now? Thanks, Xin — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGBHGQISIHDLHP5FYA6L353RIN233ANCNFSM4LQM3SAA> .

[zxdawn]

@Kang-Sun-CfA
Thanks, Kang!

I will take a look at that function later.
Hope I can make some contributions ;)

Regards,
Xin

[zxdawn]

@Kang-Sun-CfA
Hi Kang,

I read the pop.py today and have one question about the sounding_ps.
You use the surface pressure from reanalysis data to recalculate the pressures at each model level.

If we set different model levels, the Ap and Bp need to be changed in the F_interp_gcrs function, right?
So, why not using RegularGridInterpolator (the method of calculating the gc_gas_doy) to get pressures at the model levels over pixels?

Could tell me the advantage of using surface pressure from reanalysis data, instead of the pressure in the model?

Thanks,
Xin

[Kang-Sun-CfA]

Hi Xin, In this GEOS-Chem model I got from someone else, the pressure levels were not saved. The GC pressure levels are inherited from the reanalysis (MERRA-2 in this case), and they are terrain-following, i.e., the pressure levels are different for different surface pressure, not a uniform grid across the domain. Hence in this case I have to recover the pressure levels using MERRA-2 surface pressure and GC Ap/Bp. For many other models, the concentrations are defined at regular lat/lon/pressure grids, the interpolation would be easier. Kang

…

On Sat, Mar 21, 2020 at 7:40 AM Xin Zhang ***@***.***> wrote: @Kang-Sun-CfA <https://github.com/Kang-Sun-CfA> Hi Kang, I read the pop.py today and have one question about the sounding_ps. You use the surface pressure from reanalysis data to recalculate the pressures at each model level. If we set different model levels, the Ap and Bp need to be changed in the F_interp_gcrs function, right? So, why not using RegularGridInterpolator (the method of calculating the gc_gas_doy) to get pressures at the model levels over pixels? Could tell me the advantage of using surface pressure from reanalysis data, instead of the pressure in the model? Thanks, Xin — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGBHGQL6ULALZMZOINYJLDDRISRS5ANCNFSM4LQM3SAA> .

[zxdawn]

@Kang-Sun-CfA
Hi Kang,

Thank you for your explanation.
WRF-Chem is similar to GEOS-Chem. It supports the terrain following (TF) vertical coordinate and hybrid vertical coordinate (HVC). I've finished the interpolation draft for WRF-Chem using the pressure variable in the output file.

As the GEOS-Chem resolution is lower than that of OMI/TROPOMI data, it's OK to use the simple interpolation.
For the high resolution (<= 1km) model like WRF-Chem, is it better to use a weight function for resampling the simulation to pixels?

BTW, I checked the resampling method in the BEHR product. They use the tessellation approach as mentioned in your paper but without weight.

Since my purpose is replacing the a priori profiles in each pixel like that in the BEHR, is it necessary to use the physical resampling (which doesn't support resampling to pixels)? or just calculate the average of grids in pixels with weight based on overlapped area?
,
Xin

[zxdawn]

xarray just added the weighted function 3 days ago!
It may be useful to combine our own weights with xarray function.

BTW, I also used the interp function of xarray in interp_profile, like this:

sounding_profile = wc_gas_t.interp(west_east=x, south_north=y).transpose()
sounding_pEdge = wc_p_t.interp(west_east=x, south_north=y).transpose()

[Kang-Sun-CfA]

Hi Xin, If your model is <= 1km, you can average profiles on all model grids within one OMI pixel. For TROPOMI, you may want to use tessellation or my spatial response function to calculate the weight for each profile on the model grid. Kang

…

On Sun, Mar 22, 2020 at 1:35 AM Xin Zhang ***@***.***> wrote: @Kang-Sun-CfA <https://github.com/Kang-Sun-CfA> Hi Kang, Thank you for your explanation. WRF-Chem is similar to GEOS-Chem. It supports the terrain following (TF) vertical coordinate and hybrid vertical coordinate (HVC). I've finished the interpolation draft for WRF-Chem using the pressure variable in the output file. As the GEOS-Chem resolution is lower than that of OMI/TROPOMI data, it's OK to use the simple interpolation. For the high resolution (<= 1km) model like WRF-Chem, is it better to use a weight function for resampling the simulation to pixels? BTW, I checked the resampling method <https://github.com/CohenBerkeleyLab/BEHR-core-utils/blob/440e28e66911dc2ad0a18e525fb1796591750a59/AMF_tools/rProfile_WRF.m#L251> in the BEHR product. They use the tessellation approach as mentioned in your paper but without weight. Since my purpose is replacing the a priori profiles in each pixel like that in the BEHR, is it necessary to use the physical resampling (which doesn't support resampling to pixels)? or just calculate the average of grids in pixels with weight based on overlapped area? , Xin — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGBHGQI376W237Y272KOURTRIWPSBANCNFSM4LQM3SAA> .

[zxdawn]

@Kang-Sun-CfA
Hi Kang,

Thanks for your suggestion.

The resolution of my model is 1 km. I've tried the area_weighted in xESMF.
It works well: