We have now in excess of 100,000 images fully processed into our data base. A huge THANK YOU to everyone who is providing the data, and a special thanks to the team that is doing a large proportion of the processing and quality control. They are currently working their way through a large backlog of about 2500 images of the final target region (Gaia19eyy) for which we recently reinstated our reference catalogues after they were lost in the hard disk failures last winter.
In the plot above we show a part of the light curve of the namesake of this target region, the source Gaia19eyy. This is a regularly (almost periodic) outbursting Be-type star. We have published a detailed analysis of this and another similar object in the 5th paper of our series of papers (check them out here). We are only showing a fraction of the data uploaded, and will post the entire dataset for the last year when it is fully processed.
The focus of this post is to have a look at the error bars of the photometry data points. All the data shown is from the same observatory. However, for the fist half the error bars are much higher than for the second half. So what is going on?
First we need to understand how the error bars are determined. We have a deep reference image for each filter and field. In this case these are images obtained from stacking all data taken of this field with this telescope in September 2023. In this deep image we measure the apparent magnitudes of all stars. When we calibrate a new image, we determine the off-set between the magnitudes of the stars in that image and the reference image and correct for this. This off-set is slightly different from star to star. The scatter of this off-set from the mean, as a function of the magnitude, is then used as the error bar for the magnitude measurements.
The observatory did replace their camera from a CCD camera to a new CMOS detector in May this year. And if one converts the Julian Date on the x-axis where the change in error bars occurs one finds that this change happened between May 18 and 23. Thus, the later images are taken with exactly the same instrumentation as the reference images, while the first part of the data uses a different detector – note the filters have not changed. The new CMOS detector has a different detection efficiency as a function of colour than the CCD, and hence the scatter increases. This is especially apparent in the V-filter, where the data with the new CMOS detector have typical uncertainties of 2.4 percent. The data taken with the CCD have uncertainties of 6 percent.
However, this does not mean the data taken with the CCD is more noisy. Note we have an additional calibration step (described in the second of our papers) that corrects for this colour dependence. When applied to these data, it will reduce the uncertainties to the same level and adjust for any systematic off-sets. Note there seems to be a small systematic step in the brightness of the star before and after the change of instrumentation, which will be removed by this procedure.