Make a system for generating fake / "synthetic" / "simulated" K2 data

[gully]

Description

You only understand your data to the extent that you understand its generating process. The generative process of Kepler/K2 photometric time series data includes both instrumental and scientific artifacts. In this research project, we would develop a system for generating synthetic K2 lightcurves that can be used to test various de-trending algorithms. This framework will quantify the extent-to-which interesting astronomical signals are degraded, and will inform how algorithms developed for planet searches can be adapted for recovery of transient supernovae events and young star modulation.

Benefits

• We can test various algorithms for how much they recover/degrade known astrophysical signals
• Develop better understanding of our numerous noise sources.
• Possibly enable forward modeling methods for pixel-level data

Costs

• You have to consider a lot of sources of noise in order the synthetic data to be useful; some artifacts are too subtle or unknown to include in the generative model. Enumerating and quantifying all of these sources would be difficult.

[mirca]

A simplified API for this could be:

lc = synthetic_lightcurve(shot_noise, readout_noise, total_flux, pixel_response_function, 
                          inter_pixel_flat_field, motion_profile)

Where shot_noise could be a realization of a Poisson process or an array with the mean levels at each pixel and time measurement; readout_noise (same as shot_noise but zero mean Gaussian process); total_flux total flux (or counts) of the star; pixel_response_function (we could use the Kepler PRF model here); inter_pixel_flat_field a meaningful floating point number for each pixel; motion_profile motion of the spacecraft over time.

[JeffLCoughlin]

I like the idea of a fully synthetic curve.

It would also be great if we were able to create, for real, existing light curves:

• Inversion function (real simple, basically multiply by -1)
• Scrambling function (lot of possible ways to scramble data, for Kepler and K2, so maybe have user specify input time boundaries and final ordering.)
• Transit injection (specify period, epoch, depth, and width maybe. Can get complicated if wanting to depend on stellar parameters and LD coeffs, etc.)