reconsider cross correlation step
Opened this issue · 1 comments
jradavenport commented
We currently do a correlation of a flare model with the detrended data, essentially a match-filter for better detecting the flares.
I think this is a good idea but:
- the model may need rescaling to better preserve flux
- we maybe should use a longer timescale for the model flare
- @ekaterinailin has indicated it may not be useful for the Long Cadence data.
Discuss!
ekaterinailin commented
I have noticed that cross-correlation with the inverse model filter, i.e. a convolution seems to me more like what we want. Here's what I mean on a synthetic example:
On a real light curve (green and red dots are flare candidate data points detected by Appaloosa, red ones are later rejected as systematics):
With cross-correlation:
With convolution:
Question: Should Appaloosa use convolution instead of cross-correlation? Or do I misunderstand the purpose of this step?