I've had a bit of a crack at attempting to replicate the nsb dark sky Eta Car results using the FITS survey data Gianluca sent me.
This is nsb:
This is Mellinger:
To within model uncertainty they don't seem that different, there's still one really bright pixel (although its position differs). But It seems like there's still some pointing issues, even though the Mellinger fits file includes WCS data. I think this confirms this star pointing issue might be a ctapipe thing:
Now there are a number of things I don't know how to handle/sort through:
The Mellinger data is in ADU in RGB channels, I've no idea how to convert this to nanoLamberts (maybe look at a bright reference star and compare)? So the Hz rates are likely off by a long way.
I've no idea how to handle the different exposure times (which is a mix of things for the Mellinger data), nsb was designed to replicate ns timescale images as opposed to highly calibrated, flat fielded stellar data. This might explain why the bright star is so dominant in the Mellinger data.
nsb Also has Krisciunas sky brightness data, which seems to have been explicitly calibrated out in the Mellinger data, hence why the mean value is much lower. But I'd still expect the contribution from stars in the Mellinger data to be a bit more prominent, perhaps this is because I averaged the R,G and B photutils results (it doesn't support multi-channel data).
Also the Mellinger data, whilst faster to initially process, almost crashes our (very RAM heavy) analysis machine every time I try to run it, probably because assigning 36000 pixels to an RA and DEC is a bit complex.
If anyone wants to take a look at the source code, it can be found in scripts/fromfits.py for nsb and scripts/fromwcs.py for the Mellinger data.
I've had a bit of a crack at attempting to replicate the nsb dark sky Eta Car results using the FITS survey data Gianluca sent me. This is nsb:
This is Mellinger:
To within model uncertainty they don't seem that different, there's still one really bright pixel (although its position differs). But It seems like there's still some pointing issues, even though the Mellinger fits file includes WCS data. I think this confirms this star pointing issue might be a ctapipe thing:
Now there are a number of things I don't know how to handle/sort through: The Mellinger data is in ADU in RGB channels, I've no idea how to convert this to nanoLamberts (maybe look at a bright reference star and compare)? So the Hz rates are likely off by a long way. I've no idea how to handle the different exposure times (which is a mix of things for the Mellinger data), nsb was designed to replicate ns timescale images as opposed to highly calibrated, flat fielded stellar data. This might explain why the bright star is so dominant in the Mellinger data. nsb Also has Krisciunas sky brightness data, which seems to have been explicitly calibrated out in the Mellinger data, hence why the mean value is much lower. But I'd still expect the contribution from stars in the Mellinger data to be a bit more prominent, perhaps this is because I averaged the R,G and B photutils results (it doesn't support multi-channel data). Also the Mellinger data, whilst faster to initially process, almost crashes our (very RAM heavy) analysis machine every time I try to run it, probably because assigning 36000 pixels to an RA and DEC is a bit complex.
If anyone wants to take a look at the source code, it can be found in scripts/fromfits.py for nsb and scripts/fromwcs.py for the Mellinger data.