This morning Benedetta Ciarli (MPA) gave an interesting talk about cosmic reionisation, modelling they're doing and observations with LOFAR, where they might glimpse the 21 cm line forest.
Two take-home messages were:
Later on I was solving a problem that preoccupies me these weeks ahead of the thesis comittee meeting: automated photometry of CALIFA (actually, any SDSS) galaxies, using growth curve analysis for flux extraction. The field stars and other irrelevant objects were already masked by N., but the holes left in the image arrays have to be interpolated, and that's not a trivial task, apparently. Python's interp2d fails on arrays this big (standard SDSS frames, I won't do manual cutouts), besides, I'm not sure it is suitable for this goal, as well as all similar python/numpy routines.
My current approach is finding contiguous bad (=masked) regions in the image array, forming smaller sub-arrays and filling the masked regions within them with the mean of a circular aperture around. That's obviously a naive approach to take -- once I get it running, I'll be able to do more realistic interpolation, e.g. using the galaxy's Sersic index.
It was an interesting detour through numpy array masking, motion detection, image processing and acquainting myself with the slice object. I kind of enjoy diving into those gory details instead of using some nice high level routine -- I'm reinventing the wheel, obviously, but that's more scientific than using black box methods.
Finding contiguous regions within an array, apparently, is not that trivial as I first expected, but the solution I used was really helpful -- stackoverflow rocks as usual.
Two take-home messages were:
- Helium fraction must be included in reionisation simulations
- _If_ (apparently, it's a big if) LOFAR finds a high-redshift radio-loud quasar, they might see the absorbtion features along the line of sight, and obtain better constraints on reionisation parameters.
Later on I was solving a problem that preoccupies me these weeks ahead of the thesis comittee meeting: automated photometry of CALIFA (actually, any SDSS) galaxies, using growth curve analysis for flux extraction. The field stars and other irrelevant objects were already masked by N., but the holes left in the image arrays have to be interpolated, and that's not a trivial task, apparently. Python's interp2d fails on arrays this big (standard SDSS frames, I won't do manual cutouts), besides, I'm not sure it is suitable for this goal, as well as all similar python/numpy routines.
My current approach is finding contiguous bad (=masked) regions in the image array, forming smaller sub-arrays and filling the masked regions within them with the mean of a circular aperture around. That's obviously a naive approach to take -- once I get it running, I'll be able to do more realistic interpolation, e.g. using the galaxy's Sersic index.
It was an interesting detour through numpy array masking, motion detection, image processing and acquainting myself with the slice object. I kind of enjoy diving into those gory details instead of using some nice high level routine -- I'm reinventing the wheel, obviously, but that's more scientific than using black box methods.
Finding contiguous regions within an array, apparently, is not that trivial as I first expected, but the solution I used was really helpful -- stackoverflow rocks as usual.
No comments:
Post a Comment