ZTF vs PS1

ZTF filters do not match PS1 filters:
A comparison of ZTF filter transmission with PS1 and SDSS (Ngeow et al. 2019).

In fact, there is a ~15% difference in ZTF g-r colours and PS1 g-r colours (~5% in g, and ~10% in r). The effect of these differences becomes obvious for red sources in ZTF r-band. Here we see strong colour dependent residuals due to differences in wavelength response. In bulk, the difference between these two systems is accounted for, on the quadrant level, by the calibration colour coefficient. However, since there is only a single photometric calibration per quadrant, any variations within a quadrant will not be accounted for.

For example, if the degree of reddening varies across a quadrant (this can be > 1 magnitude in low Galactic latitude fields), then the extinction of PS1 calibrator stars will not match that for their ZTF counterparts. This will produce spatially (reddening) dependent residuals in photometric calibrations. Reddening very strongly drives the values of colour coefficents required to match PS1 in r-band. Thus, even at very low levels of reddening can result in poor calibration.

The wavelength response of the individual ZTF CCDs is incompletely known. Nevertheless, it is known that transmission varies strongly within CCDs because of thickness variations, and between CCDs because of differing average QE and varying AR coatings. Additionally, there could be some slight non-uniformity among filter coatings.

Current analysis suggests that each g-band quadrant has a different colour coefficient. While in r-band the variations are mainly just and offset between the central AR-coated CCDs and the outer ones. The results for i-band are unknown. Nevertheless, just considering g and r-band, there are potentially at least 50 different reddening coefficients.

Variations in atmopsheric conditions would not be a significant problem if these variation were slow and predictable. However, this is clearly very often not the case since ZTF observes whenever possible. Variations in transmission are even seen within individual quadrants and thus cannot be accounted for with the current single quadrant level calibrations.

Overall, failure to account for the extinction colour term can lead to a small bulk residual effect. But since this is small, it is not always obvious.

The amount of extinction likely varies slightly between CCDs due to known variations in the filter transmission and CCD QE + AR coating, and hence the response function. Considering this:
=> 48 slightly different extinction coefficients are expected for to the three ZTF filters and 16 CCDs.

There are also clear issues on the edges of fields. These are due to both, varying degrees of scattered light, as well as poorly constrained PSF solutions (average PSF fit reduced chisq values are large along edges). Additionally, early ZTF data might show a larger PSF dependence due to poorer focus calibration.

For median absolute photometric calibration, these problems are taken into account to some extent by the spatial structure map. However, variations in this structure over time are expected since ZTF flatfields show significant variations with time.

Our approach to separate instrumental variation from atmospheric ones has been to model the underlying dependencies of the ZPs.

The magnitude bias has been corrected for bright sources, but may be field dependent and is not corrected for faint objects due unquantified Eddington bias. No corrections have been attempted in i-band.

Current flatfield analysis results.

Individual frames and fields have offsets due to a combination of different effects.

Half the data taken between mid-Dec 2019 and early Jan 2020 was reduced with bad flatfields due to a CCD readout issue. The photometry will have been affected.

Current overscan analysis results.

The g-band data appears much more susceptible to observing conditions than r-band. In g-band, there are variations in photometry with skylevel - number calibrator stars - colour coefficient and airmass in g-band. In r-band calibration dependence on airmass and number of calibrators. The residuals in r-band exhibit a complex dependency on ZP and color coefficient (related to field reddening).

The g-band skylevel problem is largely due to a lack of calibration stars within sparse fields. The importance of the number of calibrators on the ZP uncertainty is clear. High skylevels and few calibrators lead to calibrations that are systematically wrong and thus can be corrected to some extent. These effects has been quantified. However, the current solution based on bright stars (mag < 18.5, high S/N photometry) does not appear ideal in g, since there appears to be a magnitude dependency.
There is very little skylevel effect in r-band. However, there are still dependencies on the number of calibrator stars and airmass. Both g-band and r-band show a similar strong dependence on the number sources within a field.

Fringing interference still present to some extent in i-band. No work has been carried out i-band biases.

Skylevel variations with the images are caused by variations in the transparency due to partial coverage of clouds. Gradients are also present due to the moon and other light sources.

These intra-quadrant effect are not corrected in the calibration process. The lack of calibrators make spatial corrections difficult in sparse fields. A different approach may be necessary in such cases.

Corrections for fields offsets have been applied to ZTF PSF photometry, but these may be incorrect since the average correction varies with the number of frames observed.