york-stsci
commented
6 years ago Note that the simulation number above is no longer accessible, as it was not performed on any of the new servers, and the old servers have been taken offline.
Please confirm that what you're doing is the following:
- Creating a STIPS simulation with a supplied input file
- Adding multiple exposures to that simulation with the same exposure time, the same position, and the same filters
- Combining those exposures via some tool (as yet unspecified)
If so, please let me know how you're combining such simulations. If not, please let me know what you are in fact doing.
Note that, if you were using the "co-added exposures" input box on the old web server, that particular STIPS feature was never in a working condition, and making it a working feature was considerably beyond the scope of STIPS, so it has been removed from current versions and will not return. It is known to provide erroneous output, there is no way to fix it, and there is no available workaround. It was added for a specific reason at the request of the (at the time) STIPS project management, and likely should have been removed immediately thereafter.
rubab1
This issue has not been tested against the bugfix branch and is being documented here for keeping track of as other fixes are implemented. Will test locally after #2 & #8 are resolved, and post update here.
"When combining multiple exposures for same total integration time, a larger number of shorter exposures lead to higher SNR than smaller number of longer exposures, which is opposite to what the WFIRST ETC says should happen.
ETC: https://wfirst.ipac.caltech.edu/sims/ETC.html
Simulation number: sim148358135117
(Here, to achieve 108ks total integration, we are using 140/70/35 exposures each 720s/1440s/2880s seconds long.)
According to the headers, the noise residuals (including read noise) added after multiplying by total integration time are essentially the same in each case, but the constant added per pixel before multiplying by exposure time is larger for longer individual exposures (0.063564 / 0.089893 / 0.127128 ... scaled by sqrt(exp_time) ). This would make sense as in "sky being higher" in longer individual exposures, but then that'd get multiplied by a function of the number of exposures as well. Inspection of the simulated images show that the math is indeed what the header says it is. The result is that 2880sx35 has lower SNR (lower depth) than 720sx140."