experimental libRadtran multipart transmittance#833
Conversation
|
Update: Using the same extraterrestrial solar irradiance data as the emulator improves things. |
| prebuilt_lut_path=prebuilt_lut, | ||
| inversion_windows=INVERSION_WINDOWS, | ||
| multipart_transmittance=multipart_transmittance, | ||
| modtran_path=paths.modtran_path, |
There was a problem hiding this comment.
The modtran_path arg is a bit overloaded now. We could alternatively add a libradtran_path. Although, I think it could be nice (not in this PR), to have simply an rtm_path or engine_path, with logic after to decide if its an emulator, modtran, etc. I had something written here, but then realized this would break all of our workflows and tests.. just something to mull over.
|
Checked this thoroughly last week and this is looking quite promising. There are differences with the 6c emulator, but seem to be centered around aerosol profile and optical properties selection. To help make this more customizable, I've added 9 new options to Additionally, on the delta-M errors, internally libRadtran will never allow DISORT to run radiance quantities with less than 16 streams. Therefore, any errors associatedwill likely be small. To showcase this, I ran a sort of validation run with 32streams and found minimal differences in the transmittances. 
AERFRAC_2-0.4002_H2OSTR-0.1000_relative_azimuth-179.9883_surface_elevation_km-2.6968
|
|
@unbohn, ready for your review -thank you! |
|
Set to come after #848 |
|
Thanks, @brentwilder! This is very useful. |
This assumes that any delta-M scaling effects are relatively small for 8 streams (default in modtran template). Preliminary comparisons to 6c-emulator at least for spaceborne looks fairly reasonable.
Here is a test for AOT=0.2 ; water vapor =1.0 g cm-2 (note this is in units of rdn for rhoatm and the coupled terms). Most of the residual in the forward simulation is via the rhoatm interestingly.
--
However, there must be a user error on my end of setting up irradiance, because using the same source file as sRTMnet (
tsis_f0_0p5.txt) is yielding me different raw irradiance values convolved to sensor grid. This may be influencing the discrepency above with therhoatm.--
Other smaller differences in the VNIR I hypothesize could possibly due to 1) different default aerosol settings in libRadtran, 2) unresolved delta-M errors. Although, if it were delta-M errors alone, I would suspect the transmittance would be slightly increased for the direct terms as a result. However, it is the diffuse terms that have slightly higher radiance.
--
Finally, here is some spectra solved by an EMIT scene for a reference. These are snow and veg pixels sampled at random.
--
More systematically for the scene, here is average residuals. Notably most of the wavelengths even at the 5-95 percentiles fall within a range of -0.05 to 0.05 residual. I think finding my error with the
solar_irrwill likely reduce this residual for the 400-500 nm range.