The Gaia spectrophotometric standard stars survey - III. Short-term variability monitoring

We present the results of the short-term constancy monitoring of candidate Gaia Spectrophotometric Standard Stars (SPSS). We obtained time series of typically 1.24 hour - with sampling periods from 1-3 min to a few hours, depending on the case - to monitor the constancy of our candidate SPSS down to 10 mmag, as required for the calibration of Gaia photometric data. We monitored 162 out of a total of 212 SPSS candidates. The observing campaign started in 2006 and finished in 2015, using 143 observing nights on nine different instruments covering both hemispheres. Using differential photometry techniques, we built light curves with a typical precision of 4 mmag, depending on the data quality. As a result of our constancy assessment, 150 SPSS candidates were validated against short term variability, and only 12 were rejected because of variability including some widely used flux standards such as BD+174708, SA 105-448, 1740346, and HD 37725.

. Observation diary of the constancy monitoring campaign. The columns contain: (1) the instrument used; (2) the telescope; (3) the observing site; (4) the campaign time span; (5) the number of nights awarded to the short-term monitoring campaigns (within brackets the number of useful nights); (6) the number of observed time series; (7) the number of monitored SPSS. to 1050 nm and peaking around 500-600 nm. The blue (BP) and red (RP) spectro-photometers will provide dispersed images with 20< λ/∆λ <100 over the spectral ranges 330-680 nm and 640-1050 nm, respectively. The final conversion of internally-calibrated G instrumental magnitudes and BP/RP instrumental fluxes into physical units requires an external absolute flux scale, that our team is in charge of providing . Ideally, the Gaia spectrophotometric standard stars (SPSS) grid should comprise of the order of 200 SPSS in the range 9 V 15 mag, properly distributed in the sky to be observed by Gaia as many times as possible. The mission requirement is to calibrate Gaia data with an accuracy of a few percent (1-3%) with respect to Vega (Bohlin 2007).
The obvious and fundamental requirement for a SPSS is that its magnitude (flux) is constant. Since only a few of our SPSS candidates have accurate enough photometry in the literature and have been monitored for variability in the past, it is very important to perform repeated and accurate observations of our targets in order to detect variability larger than 0.01 mag (if any), both intrinsic (e.g., pulsations) or extrinsic (i.e., binarity), before we invest a large amount of resources and observing time in getting their absolute SEDs. In addition, we remind that even stars used for years as spectrophotometric standards have been found to vary when dedicated studies have been performed (see, e.g., G24-9, that has been found to be an eclipsing binary, Landolt & Uomoto 2007).
Additionally, a comprehensive search of the literature for each candidate analyzed in this research was conducted. Most of our SPSS are white-dwarfs (WD hereafter) and hot subdwarfs, the remaining are dwarf/giant stars covering different spectral types, including cool stars of late spectral type up to M. WDs may show variability with (multiple) periods from about 1 to 20 min and amplitudes from about 1-2% up to 30%. We have tried to exclude stars within the instability strips for white dwarfs of type ZZ Ceti, DQV, V777 Her or GW Vir (see Castanheira et al. 2007;Althaus et al. 2010, and references therein). However, in many cases the existing information is not sufficient (or sufficiently accurate) to firmly establish the constant nature of a given WD. Hence, many of our WD SPSS candidates needed to be monitored. Similar considerations are valid for hot subdwarfs (Kilkenny 2007).
Redder stars are also often variable: K stars have shown variability of 5-10% with periods of the order of days to tens of days (Eyer & Grenon 1997); M stars can vary, for example, because of flares. In addition, binary systems are frequent and eclipsing binaries can be found at all spectral types. The periods of the known variables span from a few hours to hundreds of days, most of them having P ∼ 1-10 days (Dvorak 2004).
The main use of the SPSS grid is of course the absolute calibration of Gaia spectrophotometric data. Nevertheless, such a large grid (more than 200 stars) represents an unprecedented catalogue of spectro-photometric standard stars, characterized by high precision and accuracy, full sky coverage, including stars spanning a wide range of spectral types, and with spectra covering a large wavelength range. For comparison, CALSPEC 2 (Bohlin 2007), containing the composite stellar spectra which are the fundamental flux standards for HST calibrations, consists of about 90 stars: our SPSS grid will be more than two times larger, and it will be about two times larger than the Stritzinger et al. (2005) catalogue. Our sample will be comparable to recent catalogues, such as the STIS Next Generation Spectral Library Version 2 3 which, in any case, does not meet as well all Gaia requirements (such as, for example, the magnitude range). This paper is the third of a series (see Pancino et al. 2012;Altavilla et al. 2015, hereafter Paper I and Paper II, respectively), and presents the results of our short-term constancy monitoring campaign. It is organized as follows: the observation strategy and data reduction are briefly described in Section 2; Section 3 describes the light curves production and our quality control (QC) procedures, and describes our criteria for constancy assessment; the results of the shortterm constancy monitoring campaigns are presented and discussed in Section 4. We summarize our results and present our conclusions in Section 5.

OBSERVATIONS AND DATA REDUCTION
The various observing campaigns for the absolute flux calibration of the Gaia spectrophotometry (Paper I), are both spectroscopic and photometric. At survey completion, the SPSS observing campaings amounted to 515 observing nights, at eight different telescopes and nine instruments (see Table 1), with 58% useful data, and the rest lost for bad weather, technical problems, or other reasons. The global campaign produced approximately 102 500 frames to be quality controlled, reduced, and analyzed. The survey started in August 2006 and ended in July 2015.

Photometry observations
Photometry observations were performed in two different flavours: night points or time series. The night point can be absolute or relative (depending on the sky conditions), and is formed by at least 9 consecutive frames acquired in the Johnson-Cousins B, V and R filters (and sometimes also I and U). The study of SPSS absolute night points will be presented in a future dedicated paper (see Paper I for more details about the observing strategy details).
A time series contains at least 30 consecutive exposures in one filter (normally in the bluest available) covering approximately 1-2 hours, in order to monitor the short-term photometric constancy for each SPSS candidate. The actual duration of the 162 time series presented in this paper is illustrated in Figure 1. We will focus here on the analysis of time series, in order to validate our SPSS candidates against short-term variability phenomena. Data coming from the night points obtained close to a time series were sometimes included, when useful and appropriate. We also use our preliminary results on the absolute night points to further investigate the long-term constancy of our SPSS candidates (see Section. 4).
The candidate SPSS were prioritized according to available literature data, placing them more or less close to the main instability strips and variability regions across the parameter space. They were then observed, subject to the scheduling allocation of various observatories, and to visibility and weather constraints. The short-term constancy monitoring campaign shared the allocated time with the absolute photometry and the spectroscopy campaigns -approximately 143 nights were dedicated to it -and as a re-sult was not always performed in optimal conditions. Consequently, not all candidate SPSS were observed, and some were observed with non-optimal data 4 . In addition to our study, Gaia observations themselves will further check variability, because each SPSS will be observed tens of times. Hidden, nearby companions may be discovered from space. These stars will simply be eliminated from the SPSS grid, which is built with some redundancy.
As can be seen from Table 1, each SPSS candidate was observed multiple times (on average, roughly 3 series per SPSS), either to repeat a non-optimal series, or to observe a subset of SPSS with different telescopes to compare the results. Here we present only the best time series obtained for each star.

Data reduction
To ensure that the maximum quality could be obtained from the SPSS photometric observations, a careful data reduction protocol ) was implemented, following an initial assessment (Marinoni 2011;Marinoni et al. 2013) of all the instrumental effects that can have an impact on the photometry precision and accuracy. The methods and results of such instrumental effects study were presented in their final form in Paper II. Particularly relevant for the present paper are the characterizations of the minimum acceptable exposure time, the CCD linearity study, and the automated quality assessment and stability monitoring of the calibration frames (on timescales as long as 9 years in some cases).
The data reduction methods were fairly standard but, due to the large amount of data collected, an automated IRAF-based 5 pipeline was built in order to take care of both the quality control of the images and the removal of all the instrumental signatures. We applied the usual detrending steps including dark removal (for REM and MEIA), bias and overscan (when available) correction, flat fielding, bad pixel correction, and fringing correction when relevant.

Data availability
The raw and reduced data are stored at ASDC (ASI Science Data Center 6 ) in a dedicated database that will contain all the data products of the SPSS campaign and will be opened to the public with the first SPSS public data release in the 4 Of the 212 initial candidates reported in Paper I, for 43 the kind of monitoring described in the present paper was not considered necessary because, according to the available literature data the SPSS candidate was not close to any of the instability regions. Of the remaining list of 169 prioritized candidates for short term monitoring, 162 could be observed with the described strategy and for 7 we did not obtain usable data. 5 IRAF is the Image Reduction and Analysis Facility, a general purpose software system for the reduction and analysis of astronomical data. IRAF is written and supported by the IRAF programming group at the National Optical Astronomy Observatories (NOAO) in Tucson, Arizona. NOAO is operated by the Association of Universities for Research in Astronomy (AURA), Inc. under cooperative agreement with the National Science Foundation. 6 http://www.asdc.asi.it/ near future. Additionally, the light curves presented in this paper will be published in the electronic version of the Journal and in the CDS Vizier service 7 , with the form illustrated in Table 2.

LIGHT CURVES
We describe here the procedure to obtain relative light curves from the aperture magnitudes measurements on the single frames, the QC process, and the adopted criteria for constancy assessment.

Magnitudes
were measured with SExtractor (Bertin & Arnouts 1996), a simple and powerful tool to perform reliable photometric measurements. We used the variety of flags and parameters output by the code, to write the semi-automated procedures for the QC of our large data set. 7 http://vizier.u-strasbg.fr/ Our observed fields are not crowded, and we aimed at very high precision (10 mmag or better), therefore we performed aperture photometry. We were free to choose an aperture large enough to avoid significant light losses, therefore we generally used six times the FWHM (Full Width at Half Maximum) that granted light losses well below 1%.
The SExtractor catalogues were cross-matched with the catalogue cross-correlation software CataPack 8 , and in particular with the CataXcorr routine, to cross-identify the SPSS candidate and the comparison stars in each frame of a time series. A final catalogue was then created with the CataComb routine, containing the aperture magnitudes of the target SPSS and of suitable comparison stars in the field, that should be visible in all the frames included in a series. We selected at least two bright stars around the SPSS, and we monitored the magnitude difference between the target and the comparison stars, as described more in detail in Section 3.3. If one of the comparison stars was found to be variable, it was rejected from the adopted set of comparison stars.
The errors on the differential SPSS magnitudes (δmag) -based on the formal SExtractor errors on the aperture magnitudes for each epoch -are displayed in Figure 2. The histogram contains 9377 single-epoch catalogues, used to build the best light curves presented in this paper. As can be seen, the typical error is 0.002 mag, with some measurements reaching as high as 0.01 mag and more, but also many going as low as 0.0005 mag.
Instrumental effects were investigated in detail in Paper II, as mentioned previously, and taken into account as explained there. Briefly, we decided to exclude from the present analysis all the images affected by low S/N, saturation, non-uniform CCD illumination or significant geometric distortions. A few dubious cases in which atmospheric conditions were not optimal are discussed in Section 4. Moreover, the use of strictly differential photometry ensures that all remaining systematic or instrumental error soruces are beaten below the 1% level -which is our requirement for Gaia -and therefore were not explicitly considered in the error computation.

Quality control
Because we required a high precision in the final light curves, great care was taken in the selection of appropriate frames and time series for the analysis, with three levels of automated QC. At the star level, both the SPSS and the candidate comparison stars measurements had to satisfy a set of criteria: SNR>100, no saturation (using the measurements presented in Paper II), non-distorted PSF, seeing below 5", and no bad pixels within the selected aperture. At the frame level, we rejected frames in which the SPSS did not pass the star level QC, or where there were less than two comparison stars passing the star level QC. Finally, at the series level, we only accepted time series with at least 30 exposures passing the frame level QC, and lasting in total at least one hour.
The QC procedure issues warnings that can be examined later to assess the causes. In fact, our QC criteria are quite restrictive, and in some cases we were forced to accept some series that did not actually pass all of the QC chain. For example, when no better series existed for a particular SPSS, we accepted series slightly shorter than 30 points, or with one comparison star only, or with a large spread in the average comparison star magnitudes. We used these less precise data anyway, to put some constraints on the maximum variation amplitude allowed by the data, on the tested timespan of 1-3 hours, as further discussed in Sections 3.4 and 4.
The actual duration of the 162 time series presented in this paper is illustrated in Figure 1. It can be seen that a few light curves shorter than 1 hour were included for lack of better data, while the typical duration is 1.26 h, with a few curves lasting 4 h or more.

Light curves production
Using the catalogues that passed the QC criteria already described in Section 3.2, we built the light curves in two main steps. In the first step we created a super-comparison light curve using all the chosen comparison stars. At each epoch, the average flux of the chosen comparison stars was computed and converted into an average epoch magnitude. For each star, the difference between its epoch magnitude and the average epoch magnitude above was then used to build a light curve. The comparison light curves were reported to the same zeropoint using the difference between the median magnitude of each curve with that of a chosen comparison curve, usually the one of the star with the highest S/N ratio (see bottom panel of Figure 3). The super-comparison curve (magenta line in Figure 3) was then created as the average of normalized light curves for all the chosen comparison stars.
As a second step, the SPSS light curve was computed, normalized to zero. This was achieved by first computing the SPSS normalized light curve in the same way as for the comparison stars (black filled circles in the bottom panel of Figure 3) and then by subtracting the super-comparison curve from it (top panel of Figure 3).
A number of useful quantities were also computed, along with a checkplot that summarizes all relevant information (Figure 3). The quantities that were used for the light curve QC and constancy assessment are the following: • σprec is the average standard deviation of the reference In the top panel the zeroed light curve for the SPSS is plotted, along with some useful quantities for light curve validation and SPSS constancy assessment (see text for more details). The dotted lines mark the ±1, 2, and 3 σprec thresholds.
stars and tracks the precision to which the magnitude of constant stars is reproduced over the whole series; • σSPSS is the standard deviation of the SPSS light curve; • the ratio σSPSS/σprec is an indication of variability: when this ratio is above one, the variation detected in the SPSS light curve is larger than the curve precision; there is a significant indication of variability when the ratio becomes higher than 3, for example; • F±1σ is the fraction of SPSS light curve points lying within ±1σprec; similarly to the ratio described above, this is another way of searching for variability: when the fraction is significantly below 68%, the distribution of points is not normal; • finally, (P −N )/T is a fractional indicator of asymmetry in the distribution of points, where P is the number of points above zero, N below zero, and T the total number of points in the curve.
After light curve production, a simple QC criterion is applied: σprec 10 mmag. Ideally, constancy assessment is relevant for Gaia only when based on light curves passing this criterion. In practice, as mentioned previously, for some SPSS we also accepted curves that did not pass the criterion, for lack of better data, because we assumed that even some less strict indication of the SPSS nature was better than no indication at all. The distribution of σprec for the 162 light curves presented in this paper is shown in Figure 4, where it can be noticed that 7 light curves failing this QC Figure 4. Histogram of σprec of the 162 light curves presented in this paper. The vertical dashed line marks our criterion of 0.01 mag. As can be seen, the typical curve precision is of 0.004 mag and there are only 7 light curves that do not match our criterion, and that were included for lack of better data.
step were included. We note that the typical precision of our light curves is comparable to that of recent studies for the search of exoplanets from space (Nascimbeni et al. 2012) or from the ground (Klagyivik et al. 2016).
An output file containing the resulting (relative) light curve for each time series of each SPSS was produced and archived at ASDC. The best light curve for each SPSS, presented in this paper, is also published electronically in the form presented in Table 2.

Constancy assessment
The simplest form of variable star analysis is the inspection of the shape of the light curve, and the time and magnitude of maximum and minimum, if any. The term range is here adopted to denote the difference between maximum and minimum (if present), or the maximum magnitude excursion observed in the curve if no clear maximum and minimum are detected. The term amplitude (A) is used to denote the half range, as in the coefficient of a sine or cosine function.
For the light curves passing our QC criterion (σprec 10 mmag), we contemplate two cases: (i) a candidate SPSS is validated as constant over the sampled periods if no coherent pattern is present in the light curve and if σSPSS/σprec 1.5 9 ; in that case, even if the star varies, its amplitude is A< σSPSS ± σprec and it is accepted as an SPSS; (ii) a candidate SPSS is judged variable and rejected as an SPSS if a coherent pattern is evident; in this case, we define Amax = max(LCmax, |LCmin|), where LCmax and LCmin are the maximim and minimum magnidude differences measured on the LC, and if Amax > 3σprec the pattern is significant; the estimated amplitude of the variation is: • if no clear maximum or minimum is visible in the curve, the period of the variation is most probably longer than the time series, and we assumed that A>Amax ± σprec; • if only the maximum or minimum is visible in the curve, the period of the variation is again most probably longer than the time series, but we assumed that A Amax ± σprec; • if both the maximum and minimum are visible, we assumed A=Amax ± σprec, but of course this is only valid over the sampled periods.
As can be seen, the criteria can lead to dubious cases, for which additional checks have to be performed and a discussion of special cases can be found in Section 4. For the curves that do not pass the QC because σprec > 10 mmag or because they failed any other of the QC levels described previously, we still attempted a constancy assessment, although with less stringent results: (i) if no clear pattern was present in the light curve, and if the various indicators described in the previous section were compatible with a normal distribution, we validated the SPSS candidate as likely constant and kept it in the SPSS sample; in this case, we set the allowed amplitude of the variation, if present, as A < 3 σSPSS ± σprec; (ii) if some coherent pattern was present, the star was judged suspect variable; in that case the amplitude was assigned with the same criteria used for variable stars and the candidate SPSS was removed from the canditade SPSS list.

Additional criteria
We originally planned and started an additional constancy monitoring campaign on longer timescales (see Paper I) with the goal of observing each SPSS candidate 4 times per year, Figure 6. The best light curve for each monitored SPSS candidate is displayed in each panel using the same magnitude scale, when possible. The figure is available in its entirety in the electronic version of the Journal. Here we show some example of constant stars (the first four panels), suspect variable stars (panels five to eight, these stars are discussed in Section 4.4), and confirmed variable stars (the remaining panels, these sars are discussed in Section 4.5). This is the final zeroed light curve as in the top panel of Figure 3. The differential magnitudes are in the sense SPSS minus supercomparison (see text for more details), i.e., negative values mean that the SPSS becomes brighter. The ±1, 2, and 3 σprec thresholds are marked by colored dotted lines in each panel.
for 3 years. However, because we mostly obtained the observing time each semester at six different facilities with normal time applications 10 it was not practically possible to obtain regularly the required sampling. We therefore stopped our long-term campaign, also because it became apparent that Gaia itself will be able to perform this kind of monitoring excellently on the SPSS sample.
We instead used the absolute night points (see Section 2) of our absolute photometry campaign, that are presently being analyzed (a paper is in preparation). We 10 With the exceptions of: NTT for which we obtained once the ESO large programme status; Calar Alto for which we obtained a few years of granted time; and TNG for which we obtained a medium-sized programme allocation once.
have at least three independent night points per SPSS candidate, and in some cases they cover up to 6 years or more for a single SPSS. Therefore, in some difficult cases we could use absolute night points to investigate suspicious trends found in our short-term light curves or claims of long-term variability found in the literature. Some examples of this kind of additional checks can be found in the discussions in the next section. Additionally, a thorough literature search for each of the candidate SPSS presented here was carried out.

RESULTS
We were able to perform the constancy assessment on a total of 162 SPSS candidates. As a result, 150 SPSS candidates were validated against short-term variability, and only 12 8 S. Marinoni et al. were rejected (8 being clearly identified as variable stars, and 4 being classified as suspect variable stars).
We needed to either reject or retain each SPSS, therefore for some uncertain cases we had to take decisions, motivated by more than the simple criteria described in the previous section. We discuss in the following the most difficult cases, both to illustrate the assessment procedure, and to provide useful information for future studies of these stars. The best light curves obtained for each SPSS are displayed in Figure 6, and the outcome of the assessment procedure is summarized in Table 3, along with some other useful information.

Control cases
The first three SPSS candidates are the Pillars, three pure Hydrogen WDs adopted by Bohlin, Colina & Finley (1995) as fundamental calibrators (see also Paper I). We expect these extremely well studied stars to be constant, and therefore we observed them as control cases, to verify that with our criteria they turned out as constant stars. Indeed, even for GD 153 (SPSS 003) that was observed in a windy night with high and variable seeing and some veils, we found σprec=0.005 mag and no sign of variability. We conclude that even in the presence of veils our method can be applied.

Constant stars
Of the 126 candidate SPSS that successfully passed all the QC criteria and were judged constant, five deserve to be discussed.
A slight trend was observed for GRW+705824 (SPSS 015), especially towards the end of the light curve (see Figure 6). Even if the trend was within ±5 mmag, this could be an indication of longer term variability. We note that veils appeared during the light curve, and the seeing got worse towards the end of the series. In spite of all these effects, the curve is of very high quality (σprec=0.0017 mag). Only recently, Bohlin & Landolt (2015) found some weak indication of a 0.004 mag/yr, variability after monitoring it from 1986 to 1991. We thus checked our absolute night points, spanning 3 years (from 2009 to 2012), and we found no significant variation within 0.0065 mag.
Similarly, the well studied Feige 110 (SPSS 023) shows a weak trend along the time series, still contained within ±5 mmag, possibly because the sky was veiled during the curve acquisition. In the literature, no previous detection of variability was found. Our absolute night points show large variations in the B and R bands over one year, but there appear to be problems in the data quality and analysis for this particular set. On the other hand, the corresponding V magnitude appears constant within 0.007 mag. Therefore, we accept this SPSS candidate.
Another similar case is 1812095 (SPSS 037), with a weak trend contained in the ±5 mmag range, clear sky observations, and no sign of variability in the literature. Our absolute photometry data contain only one reliable night point so we cannot use it as an additional constraint.
A different problem was apparent for WD 2028+390 (SPSS 203), where the only available curve was probably interrupted for the nitrogen refilling of the instrument and therefore a small jump appears between the two branches of the light curve. Nevertheless, the SPSS candidate appears constant within 4 mmag and the curve is of good quality.
Finally, SDSS J125716+220059 (SPSS 355) shows a trend, entirely contained within ±10 mmag, that is most probably caused by the high (from ≃5 to 7") and variable seeing, with thin clouds, during observations. No other curves were obtained for SPSS 355. No sign of variability was found in the literature, and our absolute photometric measurements are always contained within a few mmag. So we decided to retain this SPSS candidate.

Likely constant stars
We retained as likely constant 24 SPSS candidates that formally failed some QC level, or did not completely fulfil our constancy criteria.
One example of likely constant stars are SPSS candidates for which only one of the field comparison stars passed all QC steps, generally because the other comparison stars were too faint (S/N ratio below 100). In most cases, including one or more faint reference stars did not improve significantly the quality of the curve, while in a few cases, including a second comparison stars with S/N ratio below 100 provided better results. Examples of these SPSS are: . For all these stars we did not detect any clear sign of variability and no variability indication was found in the literature or in our absolute photometry night points. Of these, SPSS 121 was found to be a magnetic star (Aznar Cuadrado et al. 2004) in a wide binary system (Koester et al. 2009;Zuckerman 2014), SPSS 129 is not a white dwarf being a G8IV star (Kharchenko& Roeser 2009), and SPSS 134 is suspected to be member of a wide binary system (Zapatero Osorio & Martín 2004). These findings do not pose significant danger to the use of these candidate SPSS, so they were retained in the Gaia SPSS grid.
Another example concerns time series having too few points (less than 30) or lasting less than one hour, like: . Each star suffered from various other drawbacks like low S/N ratio in the SPSS as well, or veils and other weather problems. In any case, the remaining points generally showed no clear sign of variability and no variability was found in the literature or in our absolute photometry. Zapatero Osorio & Martín (2004) noted that SPSS 140 is part of a wide binary system. SPSS 254 was also reported as non-variable by McGraw (1977), Dolez, Vauclair, & Koester (1991), and Gianninas, Bergeron, & Fontaine (2005).
One star, WD 2216-657 (SPSS 276), did not pass the criteria because the S/N ratio of the SPSS was lower than 100 in the only curve available. However, the curve precision Gaia SPSS variability monitoring 9 was acceptable (σprec=0.008 mag) and no sign of coherent trends or variability was detected. Our absolute photometry is stable within 0.015 mag and no indication of variability was found in the literature either.
Finally, three stars showed coherent patterns in their light curves, although they roughly remained within the actual variability limit of 0.01 mag. The first one, LTT 9239 (SPSS 021), a widely used flux standard, was observed in good conditions and our absolute photometry did not show any significant variation. The second, WD 0943+441 (SPSS 180), was considered non-variable by Kepler et al. (1995) who found it constant within 0.0051 mag. The third WD 0954-710 (SPSS 181), was observed with ROSS2 when there was a problem with the dichroic, no sign of variability was found in the literature, and our absolute photometry is all contained within 0.004 mag over 6 years.

Suspect variable stars
Similarly to the likely constant case, some candidates SPSS were classified as suspect variable stars either because the data did not formally pass the QC but some significant trends were revealed, or because we identified significant trends, but without finding a definite maximum or minimum in the light curve. Four candidate SPSS were judged suspect variables and removed from the SPSS candidate list.
The first one is BD+284211 (SPSS 018), a star present in the Landolt & Uomoto (2007) catalogue and a widely used CALSPEC standard (Bohlin 2007). The trend we observe has an amplitude of almost 0.010 mag in the B band, and although this trend is still formally within 3 σprec, we clearly do not sample the entire period: the variation is therefore likely to exceed our criteria. In addition, there are hints in the literature that this star might be peculiar in many respects. It has a 5 mag fainter red companion at about 3" (Massey & Gronwall 1990) as shown also by our data in Figure 7, and it has variable emission in Hα with emission lines (Herbig 1999;Latour et al. 2013). It is a suspect binary based on its measured infrared excess (Ulla & Thejll 1998) and has X-ray emission (La Palombara et al. 2014;Latour et al. 2015).
The second one is WD 1327-083 (SPSS 131). This star was classified as non variable by Giovannini et al. (1998). Nevertheless, we observe a clear variation of about 0.015 mag in a high quality curve lasting more than 3 h and our absolute photometry data show a variation of 0.04 mag in B band in slightly less than one year. The period observed in the light curve is too long to be compatible with the typical periods of pulsations for ZZ Ceti stars: perhaps SPSS 131 could be a multiple object.
The spectroscopic binary G 190-15 (SPSS 152) was classified as single-lined spectroscopic binary by Latham et al. (2002) and later as double-lined specrtoscopic binary by Halbwachs et al. (2014). The level of optical variability was uncertain in the past. Our best light curve shows a possible trend that is contained within ±5 mmag and could thus in principle be ignored as done for a few stars in the previous sections. The curve quality was quite good. However, our absolute photometry, spanning 3 months, showed a variation of 0.03 mag in B, that is above our limit of 0.01 mag.
The last star, WD 0009+501 (SPSS 216), is a magnetic WD, the first one discovered below 100 kGauss (McCook & Sion 1999). While we could find no study reporting light curves or variability detections, we observe a variation of about 0.013 mag in a high-quality light curve, lasting approximately 2.5 hours. The first part of the light curve is quite flat and then the curve declines, so we could not detect any clear maximum or minimum. No apparent technical problems could explain the trend, but we prefer to avoid using this star in the Gaia SPSS grid.

Confirmed variable stars
Eight of the SPSS candidates in our original list (reported in Paper I) are now known to be variable and were rejected from the candidate SPSS list.
The first one is BD+174708 (SPSS 020), a well known and widely used CALSPEC standard that was also selected as target by the ACCESS rocket mission (Kaiser et al. 2014). It was shown to brighten by 0.04 mag over a period of 5 years by Bohlin & Landolt (2015). Our absolute photometry data also show a variation of 0.03 magnitudes in the B, V, and R filters over a period of 7 years. Even if our short-term data do not show any variation in the monitoring period of 1 hour, we clearly cannot use it as a Gaia SPSS.
SA 105-448 (SPSS 029) is another widely used CAL-SPEC standard and one of the photometric standard stars in the Landolt (1983) catalogue. Our best light curve shows a very definite variability pattern with an amplitude of 0.034±0.005 mag in B band, compatible with δ Scuti type pulsations. We do see in the LC one minimum and one maximum separated by roughly 0.5 hours. Among the numerous studies employing this star as a standard, we could find no indication of variability. Unfortunately we have only one absolute night point of non-optimal quality. We found data for this star in NSVS (Northern Sky Variability Survey, Woźniak et al. 2004). A preliminary analysis of the NSVS data shows a variation with a period of roughly one hour and amplitude ≃0.03 mag in the V band.
Another star that was already shown to be variable (see also Paper I) is 1740346 (SPSS 034). We publish here a longer light curve with respect to that paper (6 hours), showing various maxima and minima. Its maximum amplitude is 0.015±0.002 mag in the B band, with various periods and amplitudes. It was proposed to be a δ Scuti star by Marinoni (2011).
Unluckily, another candidate SPSS that was selected to be in common with the targets of the ACCESS mission also turned out to be variable: HD 37725 (SPSS 051). Its light curve shows an unmistakable variation over a timescale of roughly half an hour or less, and a maximum amplitude of 0.015±0.003 mag in B band. Our light curve covers almost two hours and a clear global brightening in the variation pattern is visible. The star is an A3 V and there is evidence of multiple periods, therefore it is likely a δ Scuti star.
Among the several SPSS selected from the SEGUE (The Sloan Extension for Galactic Understanding and Evolution, Beers et al. 2004) sample to fill in a few under-represented spectral types, four showed clear signs of variability, with amplitudes ranging from 0.018±0.001 to 0.079±0.003 mag. The first one, SDSS J164024.18+240214.9 (SPSS 316), was classified as a candidate blue horizontal branch star by Xue et al. (2008), but not confirmed by spectroscopy. Due to the evidence of multiple periods in the light curve, it could be a variable of the δ Scuti type. For this star, we found some data in NSVS, CSS (Drake et al. 2009, Catalina Sky Survey), and WISE (Cutri et al. 2013, Wide-field Infrared Survey Explorer). However the data have not sufficient quality to draw any firm conclusion.
The second one, SDSS J224204.16+132028.6 (SPSS 318), was classified as a blue straggler by Xue et al. (2008). Again its light curve is compatible with a δ Scuti family pulsator with a clear minimum and maximum separated by ≃40 minutes. From a preliminary analysis of data available in CSS, we confirm a variation with an amplitude of roughly 0.05 mag and a period of ≃50 minutes.
The last one, SDSS J000752.22+143024.7 (SPSS 325), has no specific mention of variability in the literature, but it has low amplitude pulsation and a period compatible with those of a δ Scuti star.
Finally, we note that -even if we did not obtain any useful data -SDSS J204738.19-063213.1 (SPSS 317) was reported to be an RR Lyrae star by Abbas et al. (2014). Therefore this SPSS was excluded from our candidate list.

Variable comparison stars
Among the field stars that could be used as comparison, we found three that were variable.
The first one was initially chosen as comparison star for the light curve of WD 1659-531 (SPSS 196), as shown in the first panel of Figure 8, and later rejected because of its clear variability pattern. Its light curve is shown in the upper panel of Figure 9. It was identified as 2MASS J17025022-5315592, but no further information was found in the liter-ature. In our light curve, lasting 1.2 hours, we could not observe any clear maximum or minimum but the range of the variation is at least 0.1 mag in B band.
The second one was a comparison star for WD 0123-262 (SPSS 220), with its position and light curve reported in Figures 8 and 9, respectively. The star was identified as 2MASS J01252029-2602141, but also in this case no further information was found in the literature. We observe a variation of about 0.1 mag in B band, with a possible double minimum towards the end of the curve. The CSS data show a typical RR-Lyrae light curve with an amplitude of ≃0.5 mag and a period of ≃0.57 days. The double minimum we observed in our data is caused by the typical luminosity bump associated with the early shock phase in RRab variables.
The last one was one of the comparison stars for the light curve of WD 0859-039 (SPSS 307), as shown in Figures 8 and 9. It was identified as 2MASS J09015726-0411010, an eclipsing binary of W Ursae Majoris type, with a period of 0.420775 days by the NSVS. In our curve we observed a variation of about 0.25 mag in B band.

CONCLUSIONS
We presented the results of our extensive short-term (1-2 hrs, see also Figure 1) constancy monitoring campaign of 162 candidate SPSS from the list reported in Paper I, that needed this kind of monitoring. A total of approximately 143 observing nights -on a grand-total of more than 515 dedicated to the Gaia SPSS grid ground-based campaigns -were dedicated to the short-term constancy monitoring. We present in this paper the best available curve for each of the monitored SPSS candidates, but on average we obtained approximately 3 curves for each candidate SPSS. The typical precision of the light curves was 4 mmag (see also Figure 4).
We presented and discussed the criteria adopted to assess the constancy or variability of the candidate SPSS. When the candidate SPSS were judged constant or likely constant, because no clear variability pattern was present and the maximum amplitude allowed by our data was estimated as 10 mmag or less, they were retained in the Gaia SPSS candidate list. Otherwise, the candidate SPSS were rejected from the list. In all doubtful cases, we made extensive use of the available literature data and of our absolute photometry measurements (that will be published in a forthcoming paper) to reach a decision.
In conclusion, of the 162 monitored SPSS candidates, 150 were validated against short-term variability, and only 12 were rejected: eight were found to be variable stars, and four were classified as suspected variables. Among the confirmed or suspected variable stars, there are some widely used flux standards from the CALSPEC set, i.e., BD+174708, SA 105-448, 1740346, and HD 37725. Table 3. Results. The columns contain: (1) the SPSS internal ID; (2) the name of the star; (3) and (4) the star coordinates compiled in Paper I; (5) the spectral type compiled in Paper I; (6) the instrument used; (7) photometric band of the best observed time series; (8) the sky quality during the acquisition of our best time series; (9) indicates whether the following column is an upper ( ) or lower ( ) limit or a determination (=) of the amplitude; (10) the minimum tested amplitude in case of constant stars, or the maximum observed amplitude in case of variable stars; (11) the precision of the light curve used (see text); (12) the number of comparison stars used for the SPSS light curve production; (13) specifies whether the star will be used as a Gaia SPSS; (14) any additional comment. The table is available in its entirety in the electronic version of the Journal and at CDS. Here we show a portion to illustrate its contents. This paper has been typeset from a T E X/ L A T E X file prepared by the author.