Subtract DC level from light curves?

[drphilmarshall]

At the August 25th Twinkles weekly meeting, we identified a function we need the Monitor to perform: according to @rbiswas4 it is standard practice to subtract off the flux due to the host galaxy from SN forced photometry light curves. To implement this, we could have the Monitor find the "DC level" of a light curve (far from the SN brightening and fading) and subtract it from the light curve (propagating errors appropriately). We could avoid this if sncosmo perfomed fits that included a constant background level - does it?

As the seeing varies from visit to visit, the amount of galaxy flux brought into the flux measurements will change - but perhaps this can be taken to be an additional noise term? Advice would be most welcome before we code this, @rbiswas4 @wmwv

[rbiswas4]

To implement this, we could have the Monitor find the "DC level" of a light curve (far from the SN brightening and fading) and subtract it from the light curve (propagating errors appropriately). We could avoid this if sncosmo perfomed fits that included a constant background level - does it?

It does not in the standard implementation

I was suggesting something different actually (simpler or more complicated).

• Do forced photometry at the location at all other times when the SN is not present, and average it to call the flux contribution of the galaxy, with errors.
• Subtract this number from what we are calling the light curve fluxes, as you would if you subtracted two estimators. This will lead to both smaller fluxes for the SN, and higher error bars.
• Do usual light curve fits

This is simpler because you directly estimate the galaxy flux without inferring it from measuring SN + galaxy flux and a model. It is more complicated in that you have to run the DM stack and find times where there are no SN (easy in reality, hard in that run).

As the seeing varies from visit to visit, the amount of galaxy flux brought into the flux measurements will change

Exactly. The flux contribution from the host depends somewhat on the observing conditions. This is why people want to try to model this using scene modeling for things like a final analysis. But this is completely lost in either the correction I suggested, or in the sncosmo fit method you proposed because it also assumes a constant background level.

As I was writing, I also think we can estimate the galaxy contribution (in a way not accessible to observers) by using the galaxy surface brightness and PSF in the relevant visit to calculate the galaxy contribution. (Is this already done in DM as model flux?) It would be a useful thing to have.

Also, I do think that the idea of implementing a fit for SN parameters and zero points simultaneously is useful and not too difficult (I even think this was done in some hack session but not taken to completion). Since you are in a astrohackweek next week, maybe this could be a project ? :)

[wmwv]

Image subtraction or scene modeling are the only ways to do this right. I don't think it's worth spending much effort trying to do this "right" by doing forced photometry on unsubtracted images. In particular if the goal is to do light-curve fits to recover parameters to try to learn how we're doing, I don't think we can learn that much quantitatively useful until we do either scene modeling or image subtraction.

[drphilmarshall]

OK: that says that maybe a simple DC level correction like the one Rahul suggested is all that is needed to make our current ForcedSource light curves correspond to the current industry standard - or would you say that the current industry standard is different?

I am all for pursuing image differencing, but we got talking in the meeting about fallback options, and what DM Level 2 analysis to do. If image differencing is not enabled until November, there could be something to be said for trying to do our current DM pipeline processing before then, to get it set up. What do you think?

On Thu, Aug 25, 2016 at 6:48 PM, Michael Wood-Vasey < notifications@github.com> wrote:

Image subtraction or scene modeling are the only ways to do this right. I don't think it's worth spending much effort trying to do this "right" by doing forced photometry on unsubtracted images. In particular if the goal is to do light-curve fits to recover parameters to try to learn how we're doing, I don't think we can learn that much quantitatively useful until we do either scene modeling or image subtraction.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/DarkEnergyScienceCollaboration/Monitor/issues/50#issuecomment-242600040, or mute the thread https://github.com/notifications/unsubscribe-auth/AArY9_fBlFEoroWmWAAZyUoMDovJ40l5ks5qjkXmgaJpZM4JtpYz .

[rbiswas4]

AFAIK the industry standard for publishing SN photometry is doing SMP (SDSS/SNLS/DES) went that route .... so I agree with @wmwv that doing forced photometry and galaxy subtractions will not be doing it 'right'. At the same time, forced photometry based results have been used before such a finalized version just to get analyses going. I would be very interested in working on a problem like SMP, but I don't know what the word on that is, and one where access to real data at high z might be important.

At the end of the day, we will be using a method which is more 'right' than what I described. So there is a sense in which that work would go to waste. So, it might be better to have me (since I was assigned this) try to push on diff im which will be useful to get going on Twinkles (for example it would be useful to do detections using difference images and comparing with out cheat mode).