Sidebar
Summary of email discussion in November
I tried to summarize the number of cases and to estimate the CPU time:
Surface albedo
- 3 albedo values
0.05, 0.1, 0.2 for visible range
0.05, 0.2, 0.5 for O2A band
GEO
SZA: 20,40,60°, SAA: -90,-45,0,45,90°, VZA: 58.3°, VAA: 196.3° In total: 3*3*1*5*1=45 combinations
For the test case (~a quarter of the full domain) I required ~30 hours on one CPU to calculate full high resolution spectra and layer AMF in visible. Only layer-AMF at 1 wavelength require ~5 hours on one CPU.
⇒ estimated CPU hours for GEO (full visible spectrum, 500 spectral points, layer AMF): 45 cases *30 hours * 4 (full domain) = 5400 hours, with 500 CPUS: 10.8 hours (doable)
As discussed, memory requirements for the data become large with full spectral resolution, so I suggest to calculate full spectra only for one reference case (e.g. albedo 0.2, SZA 40°, SAA 0°, VZA 58.3° and VAA 196.3°). Retrieval accuracy can be assessed using layer-AMF.
⇒Only layer-AMF at one wavelength require only 1.8 hours on cluster
For O2A-band we still need to investigate, which spectral resolution is required. As suggested by Huan, it might be sufficient to calculate only 2 wavelengths (absorbing and non-absorbing). We have a good absorption parameterization which gives the same result as averaged over a line-by-line simulations (averaged over e.g. 1nm using a box-type response function). If it turns out that we could use only the 2 wavelengths (or 3?), we would need at maximum 5 hours for O2A band simulations on the cluster. For full line-by-line spectra (0.005nm spectral resolution) a lot more (number of spectral points:2000).
LEO
SZA: 20,40,60°, SAA= 13,353°, VZA=0,20,60°, VAA=109.5, 281.7° 3*3*3*2*2=108 combinations
Computational time will be a factor of about 2.4 higher than for GEO case, still doable.
Next steps
1. Run retrieval on test case (visible range, data provided on my website) and check whether this is ok. Huan, when do you have time for this? Before I start with more simulations I would like to know whether everything is correct.
Done.
2. Decide on cases which require full spectral resolution + layer-AMF or only layer-AMF. Is it ok to calculate only for the one reference case (see above) the full spectrum? I think that we agreed that this is ok.
For all cases full spectra are simulated in the VIS.
3. Investigate and decide on spectral resolution in O2A-band. Huan, could you define 3 wavelengths which you think would be useful so that we can test, whether these are sufficient or whether we need full spectral resolution. Then I would run the test case for the full spectral resolution and the 3 wavelength bands, respectively.
Simulations for 3 wavelengths bands done, investigation ongoing (see below).
Apart from this: Run the 1D cloud case for more geometries so that we better understand the dependence of the results on scattering angle. This data has been provided.
Done.
Retrieval results from Huan
