Implement lunar model

[dkirkby]

The IDL quicksim already has an implementation of Krisciunas & Schaefer (1991) in desimodel/pro/lunarmodel.pro. This issue is to implement the same functionality here as an intermediate (and more urgent) step towards #2.

[dkirkby]

@crockosi @sbailey I am working on this now (finally!)

My plan is to:

• Use the same high-resolution KPNO extinction table for the moon as we use for sources (Duplicate KPNO extinction tables desimodel#8)
• Update the solar spectrum from Wehrli 1985 to ASTM E-490-00 from 2000. (Although your code describes sky/solarspec.txt as 2000 ASTM, the values are actually from Wehril according to this spreadsheet).
• Calculate the V-band extinction coefficient directly from the input extinction table and solar spectrum, instead of treating it like a separate parameter. Hopefully this gives the same answer for KPNO but would be useful to allow other extinction tables to be simulated.
• Add the lunar scattering surface brightness to the dark sky tabulated in spectra/spec-sky.dat.

Any feedback is welcome.

[dkirkby]

On the last point above, I implemented the V-band extinction coefficient calculation as:

kV = V* - V

where V is the Bessell V-band magnitude of the un-extincted solar spectrum and V* is calculated with the solar spectrum multiplied by (i.e., airmass 1):

10 ** (-k(lambda) / 2.5)

Using the KPNO zenith extinction table, I calculate kV = 0.1515, compared with the value 0.162 that is hardcoded in lunarmodel.pro. I am calculating kV correctly? Should I worry about the 7% discrepancy between these values?

[dkirkby]

I am mostly done with this now. Does this plot of scattered moon V-band surface brightness look reasonable?

This is for a moon phase of 25% (zero is full) at zenith angle 70 deg. I used kV=0.162 to match the IDL code.

[moustakas]

What an awesome plot!

[dkirkby]

Awesome, but wrong :-( After fixing a little bug, it looks quite different:

[dkirkby]

This plot shows the dark sky (green), scattered moon (blue) and extinction coef (red) for a 25% moon at a zenith angle of 70 deg (same as the plot above):

I am planning to add the green and blue surface brightness values in the simulation, which assumes that our KPNO "dark sky" spectrum has no scattered moonlight. However, I put the moon below the horizon in the default config, so that only the green curve is used (i.e., no change from the current simulations).

[crockosi]

This is even more awesome! I spot-checked it using the old IDL version of
the code and I get numbers that as as close as I can read off that
colorbar. So this looks great.

Thanks, David. I'll run a bunch of my other tests once this is merged.

--Connie

On Mon, Feb 22, 2016 at 5:32 PM, dkirkby notifications@github.com wrote:

Awesome, but wrong :-( After fixing a little bug, it looks quite different:

[image: lunar]
https://cloud.githubusercontent.com/assets/185007/13238743/132602ce-d98a-11e5-82d4-dd84b257948f.png

—
Reply to this email directly or view it on GitHub
#9 (comment).

[sbailey]

Thanks for adding this. I defer to Connie for validating the results, but she sounds happy.

Could you add a code snippet showing how to use this for simulating a series of spectra under different lunar conditions? In particular, if I create an Atmosphere giving it a Moon object, can I then directly call moon.update() to alter how the moon appears to the Atmosphere? Or do I create a new Atmosphere with that updated moon before simulating more spectra?

Please add a docstring for Atmosphere.set_condition(). I think that gets a name from the constructor surface_brightness_dict, but the chain wasn't obvious without reading code. I think these do provide flexible development/testing alternatives to the Moon model, so I think it is worth keeping these two methods for similar functionality.

[dkirkby]

@sbailey I made the changes you suggested. In particular, there is now an example of changing moon parameters on the fly in the docs:

http://specsim.readthedocs.org/en/ks_moon/config.html#atmosphere

I also added moon parameters to the quickspecsim command-line script:

% quickspecsim -h
...
  --moon-phase P        moon phase between 0 (full) and 1 (new) (default:
                        None)
  --moon-zenith Z       zenith angle of the moon in degrees (>90 is below the
                        horizon) (default: None)
  --moon-separation S   opening angle between moon and this observation in
                        degrees (default: None)

[djschlegel]

When one has both moon and clouds, the sky brightness can increase by quite a bit more.
Attached is a plot of the sky brightness with our last night observing with the Mayall on Feb 19.
We were observing 50 deg from a 94% full moon for most of the night, with the moon setting
before the last exposures. Note the (z-band) sky brightness could increase by 1.5 mag from
clouds. The continuum sky brightness would be increasing even more than that.
How should that be incorporated in the models?

[dkirkby]

For comparison, the moon-only specsim predictions for those conditions are 18.8 mag/arcsec2 in z-band with the moon below the horizon (compared with ~19.0 in your graph, if you used the same units), but the typical increase at 50 deg from a 94% full moon is only ~0.8 mag/arcsec2.

In the limit of single scattering, only the cloud along the boresight should matter. If that's a reasonable approximation, then perhaps thin cloud could be included in the KS1991 model as an additive correction to the extinction.

[moustakas]

Not sure if this should be a new or separate issue, but @djschlegel has posted some relevant data on the sky brightness as a function of moon angle and moon phase from the SDSS that specsim's model could be compared against. The relevant thread was sent to desi-bgs@ and is archived here:
https://desi.lbl.gov/trac/tracmailman/browser/private/desi-bgs/2016-April/000371.html

This issue is also relevant to the bright-time simulations open issue desihub/desisim#72

[dkirkby]

Thanks for flagging this John. I will take a look and then create a new issue.