The First Post-Kepler Brightness Dips of KIC 8462852

KIC 8462852 is a northern hemisphere target ($\delta =+44^\circ$) with V = 11.7 mag. From May to September, the star is typically available above airmass 2 for ∼4–8 hr at northern hemisphere observatories, with a decreasing window of visibility until the end of December. In this paper, we present photometric data from several observatories, as described briefly here and in the Appendix.

Regular photometric monitoring in multiple filters of KIC 8462852 started in 2016 March with the Las Cumbres Observatory (LCOGT) 0.4 m telescope network, which consists of telescopes at two northern hemisphere sites: TFN (Canary Islands, Spain) and OGG (Hawaii, USA).¹²⁸ On JD 2,457,892 (UT 2017 May 18; UT dates are used throughout this paper), a drop in brightness was claimed as significant from both TFN and OGG measurements. Observations acquired at Fairborn Observatory corroborated this drop in brightness, and an alert for triggered observations was immediately executed. In response to the alert, we acquired additional photometric observations from the Calvin, Master-II, Wise, Joan Oró (TJO), Trappist-N, Thatcher, NITES, and Gran Telescopio Canarias (GTC) telescopes. We showcase this first event observed in 2017 May ("Elsie"; Section 2.1.1) observed by each observatory in Figure 1, each point of the time series being a nightly average of the differential magnitude in the specified band. Since most observatories did not have monitoring before the event, we normalize each data set assuming that the stellar flux was at the "true" stellar level just after the first event (JD 2,457,900) to ensure consistency. Note that the analysis carried out in Section 3.1 to determine colors of the dip uses the LCOGT data sets. These data are normalized using before dip data where the longer baseline improves the normalization (details in Section 3.1).

Zoom In Zoom Out Reset image size

2.1.1. The Elsie Dip Family

In Figure 2, we show the full LCOGT 0.4 m time series from 2017 May through December in the ${r}^{{\prime} }$ band. The LCOGT observations are a product of a successful crowd-funding effort through Kickstarter.¹²⁹ As a part of the campaign's design, those who supported the project were eligible to nominate and then vote on the name of the dips. Four major dipping events happened over this time period. The first, which we named "Elsie,"¹³⁰ reaches a minimum around UT 2017 May 19 (JD 2,457,893). Its full duration spans the course of ∼5 days. The temporal gradient of the dip ingress is much sharper than the egress, where the recovery appears to hesitate, making an asymmetric profile overall. Hereafter, we refer to the full period of variability observed in 2017 as the Elsie dip family.

Zoom In Zoom Out Reset image size

A few weeks after the end of Elsie, the second event, "Celeste,"¹³¹ began to make an appearance, reaching its deepest point around UT 2017 June 18 (JD 2,457,922). Celeste lasted for a couple of weeks, having a slow decline to its minimum, but then a much more rapid post-minimum rise up. The rapid egress was stunted about two-thirds of the way up, however, and the flux remained ∼0.5% below normal brightness levels. This 30-day-long depression after Celeste was never given a name, since it was unsubstantial compared to the first two events. We do, however, identify this region to be a statistically significant detection of a ∼0.5% decrease in brightness, which was also independently observed at several other observatories (E. Bodman et al. 2018, in preparation).

The third event, "Skara Brae,"¹³² began to appear just when the month-long depression after Celeste showed possible signs of recovery. Skara Brae behaved very similarly to Celeste in a mirror-image profile, with a ∼1% depth in ${r}^{{\prime} }$-band lasting on the order of two weeks. In the middle of Skara Brae (on UT 2017 August 8; JD 2,457,974) there was an additional narrow (few hour) ∼2% dip. We did not observe the ingress of this dip (which could have lasted up to 7 hr) because this region fell within a gap between observatories, but observations from both LCOGT and TJO caught the rapid return to the level of the broad dip over a 4 hr period. This event was the first and only significant short-timescale ($\lt 1$ day) variability observed in the 2017 Elsie family of dips (Figure 2, open circles). We show a close-up of this feature in Figure 3. While this is a seemingly drastic change in brightness within the Elsie family, we find that the dip gradient here (∼0.2% hr⁻¹, or ∼5% day⁻¹) is not extreme compared to the dip gradients observed by Kepler, which can be up to a factor of 10 steeper (Boyajian et al. 2016).

Zoom In Zoom Out Reset image size

This smaller gradient could indicate that the current dips are less optically thick than those with steeper gradients by Kepler, which would be consistent with the smaller dip depths, although other geometric factors may also play a role. In any case, the Kepler observations still provide the most stringent constraint on the orbit of the parent body (or bodies) from consideration of the light-curve gradient (Boyajian et al. 2016). We also find that the duration of this event is no shorter than the shortest seen by Kepler, which was 0.4 days. Finally, we note that Skara Brae's full contour—a symmetrical broad shallow dip containing a brief deep dip within—is unique.¹³³ This allows us to compare its appearance with the dips observed by Kepler, where only a couple were at a similar level of 1%–2%. We find the Kepler dip at D1540 (Boyajian et al. 2016) being remarkably similar in depth, shape, and duration (see Figure 3 for a direct comparison).

"Angkor,"¹³⁴ the fourth significant dip in the complex, appeared two weeks after Skara Brae, reaching its deepest depth around UT 2017 September 9 (∼JD 2,458,006). The Angkor ingress exhibited the most rapid, sustained (∼4 days long) flux gradient observed for the star from the ground to date, and the egress was similarly rapid, though it remained at a depth of ∼0.5% below normal for about a week before brightening to slightly above normal levels. The post-Angkor monitoring data rose up ∼0.5% compared to pre-Elsie levels (dotted line, Figure 2). Around the end of 2017 October (JD 2,458,050) this brightening trend reversed, returning back to unity by mid-December (∼JD 2,458,100).

Over a six-month period, beginning 2017 May 20, we acquired low-resolution spectra with the Kast Double Spectrograph, mounted on the 3 m Shane telescope (Miller & Stone 2013) at the Lick Observatory, as well as from the DEep Imaging Multi-Object Spectrograph (DEIMOS; Faber et al. 2003) and the Low Resolution Imaging Spectrometer (LRIS) using the Keck telescopes. All of the Lick and Keck low-resolution spectra are broadly consistent with the object being a main-sequence F star; a future paper will present them in greater detail (M. Martínez González 2018, in preparation). A total of eight high-resolution spectra were acquired with the HIRES spectrograph on Keck I. Five of the eight were taken prior (circa 2015, 2016), and the other three were taken during the Elsie event. First, we checked for radial velocity (RV) variations. We fine-tuned the wavelength solution using the numerous telluric lines in the vicinity of Na i D using the telluric spectrum from the Wallace et al. (2011) solar atlas, applied barycentric velocity corrections calculated with the IDL code BARYCORR (Wright & Eastman 2014), verified that the ISM lines remained fixed, and measured RV offsets between each out of dip spectrum along with their coadded average and the three in-dip spectra, resulting in ${\rm{\Delta }}\mathrm{RV}=0.38\,\pm 0.35\,\mathrm{km}\,{{\rm{s}}}^{-1}$. We are unable to discern RV variability with the present quality of the wavelength solutions during these epochs, limiting the presence of anything larger than a gas giant within 0.1 au.

The ISM imprints absorption lines in spectra at Ca ii (H and K at 3968.469 Å and 3933.663 Å), Na i (D2 at 5889.951 Å, D1 at 5895.924 Å), and K i (7698.964 Å); unfortunately, our in-dip spectra do not reach K i. Close-up views of the Ca ii K and Na i D1 spectral regions are shown in Figure 4, along with their residuals (prior minus during Elsie). The foreground ISM can be described by three Gaussians (the profile of the redder of the two apparent components is asymmetric and is clearly blended with at least one additional component). We find no significant variations in the depths of these ISM lines (i.e., differences in equivalent widths of the ISM lines are <1%), following the procedures outlined by Wright & Sigurðsson (2016) and Curtis (2017). The amplitude of the residuals at these ISM lines is on par with variability due to the S/N, and it will be challenging to discern real changes in interstellar absorption at or below this level during future events without higher-quality spectra (or larger dips).

Zoom In Zoom Out Reset image size

NEOWISE observations of KIC 8462852 in the W1 (3.4 μm) and W2 (4.6 μm) bands were acquired a week before Elsie, and serendipitously on JD 2,457,892 (around the time of the largest optical flux decrease for Elsie) with the spacecraft's occasional toggle to avoid the Moon. All NEOWISE observations of KIC 8462852 were extracted without any cuts on flag parameters. Inspecting each detection, we see that all observations had ph_qual ratings of "A" in both bands, and none of the detections were affected by latent images, other persistence features, or diffraction spikes as would be indicated by the "ccflags" parameter (Cutri et al. 2015). We show this temporal series in Figure 5. We find no detectable change in brightness with the observations taken during Elsie compared to the week prior in either of the NEOWISE bands. Work is currently underway analyzing near-IR measurements (D. Clemens et al. 2018, in preparation) and additional mid-IR measurements (H. Y. A. Meng et al. 2018, in preparation) taken throughout the Elsie family of dips to put additional constraints on variability after this first event.

Zoom In Zoom Out Reset image size

Evidence for the scattering of stellar light by circumstellar dust can be probed with polarimetry. Spectropolarimetry of KIC 8462852 was obtained with the Kast spectropolarimeter on the 3 m Shane reflector at the Lick Observatory (see Mauerhan et al. 2014 for a description of the Kast instrument) at eight epochs: 2017 May 20.4; June 2.3, 21.4, 27.4; July 1.4, 16.4, 17.4; and August 1.4. On each night, the source was observed 18 minutes for each Stokes parameter Q and U, for a total of 36 minutes on-source integration time. The polarization P and position angle θ were derived according to the methodology described by Mauerhan et al. (2014). On every night, two strongly polarized standard stars (Hiltner 960 and HD 204827) and one weakly polarized star (HD 212311) were observed for calibration purposes. HD 204827, in particular, was used to calibrate the position angle on the sky.

Over the course of our eight epochs, KIC 8462852 exhibited an average (standard deviation) polarization of $P=0.46(0.05) \%$ and position angle $\theta =91\buildrel{\circ}\over{.} 0(2\buildrel{\circ}\over{.} 6)$ in the V band (5050–5950 Å). The statistical uncertainty in P was typically 0.01% on a given epoch, but in our experience with Kast the systematic uncertainties are usually at least 0.1%–0.15%. Over the same period, the average (standard deviation) V-band polarization of our polarized standard Hiltner 960 is $P=5.58(0.18) \%$, $\theta =54\buildrel{\circ}\over{.} 9(0\buildrel{\circ}\over{.} 7)$, consistent with published values (Schmidt et al. 1992). The weakly polarized star HD 212311 exhibited $P=0.13(0.05) \%$, within 2σ of the Schmidt et al. (1992) value. The results are illustrated in Figure 6, and indicate that KIC 8462852 did not vary significantly in polarization, relative to our calibration stars.

Zoom In Zoom Out Reset image size

The polarization characteristics of KIC 8462852 appear to be consistent with interstellar polarization, presumably induced by the dichroic absorption of light by nonspherical dust grains oriented along the Galactic magnetic field. Figure 6 illustrates the polarization and position-angle spectra of the source, along with the F5 V star HD 191098 (V = 10.09 mag; $0\buildrel{\circ}\over{.} 2$ from KIC 8462852 on the sky), which we used as a probe of the approximate interstellar polarization on 2017 June 2; the spectroscopic parallax distance of this star (M_V = 3.5 mag) is ∼201 pc, which is not ideal, since KIC 8462852 lies at a much greater estimated distance of 391 pc. Nonetheless, HD 191098 can provide a useful probing of interstellar polarization if the majority of intervening material is closer to Earth than both sources. The spectropolarimetric data for both KIC 8462852 and HD 191098 follow the wavelength-dependent "Serkowski" form Serkowski et al. (1975) that is expected for interstellar polarization from Galactic dust (where the peak polarization is typically seen near 5300–5400 Å). HD 191098 exhibits a V-band polarization of $P\approx 0.40 \%$, $\theta \approx 78^\circ$, only slightly lower than that of KIC 8462852 in magnitude and offset by $\sim 14^\circ$ in position angle. Moreover, we examined the polarization of three stars from the catalog of Heiles (2000) that lie within 2° of KIC 8462852 on the sky: HD 190149, HD 191423, and HD 191546. The average (standard deviation) polarization and position angle exhibited by these stars is $P=0.47(0.36) \%$ and 113°(18°). This is also comparable to our measurements of KIC 8462852.

The polarization we measure appears to be significantly lower than the broadband imaging polarimetry values that Steele et al. (2018) have reported with the RINGO3 instrument (Słowikowska et al. 2016) over a similar time period. For example, over 2017 May through August, they reported a nonvariable source ($\lt 0.2 \%$ fluctuation) with average polarization and position angle of $P=1.2 \% (0.2 \% )$, $\theta =72^\circ (6^\circ )$ for the b* band (3500–6400 Å); $P=0.6 \% (0.1 \% )$, $\theta =74^\circ (6^\circ )$ for the g* band (6500–7600 Å); and $P=0.6 \% (0.1 \% )$, $\theta =73^\circ (6^\circ )$ for the r* band (7700–10000 Å). However, the level of discrepancy between these measurements and ours is not particularly surprising, given the large systematic uncertainties associated with the RINGO3 instrument (Słowikowska et al. 2016). Regardless of systematics, we reiterate that the results here and those from Steele et al. (2018) both show that the object has polarization properties typical of its location on the sky, and limit any change in optical polarization between in and out of dip epochs of $\lt 0.05$% (Kast) and $\lt 0.1 \% \mbox{--}0.2 \%$ (RINGO3). Finally, we note that analyses of near-IR polarimetry taken during the Celeste event are underway and should be able to provide additional constraints for the system and its environment (D. Clemens et al. 2018, in preparation).

The Allen Telescope (SETI Institute) performed a series of KIC 8462852 radio observations during the Elsie dimming event. These were subsequent to the observations in 2015 (Harp et al. 2016). From 2017 July 8 through August 8, 1022 separate 92 s observations were performed that resulted in scanning 1419–4303 MHz for radio signals. No narrowband radio signals were found at a level of 180–300 Jy in a 1 Hz channel, or medium-band signals above 10 Jy in a 100 kHz channel, which would correspond to transmitters at the distance of KIC 8462852 having effective isotropic radiated power (EIRP) of $(4\mbox{--}7)\times {10}^{15}$ W and 10¹⁹ W for the narrowband and moderate-band observations. While orders of magnitude more powerful than the planetary radar transmitter at Arecibo Observatory, $2\times {10}^{13}$ W EIRP, the actual power requirements would be greatly reduced if emissions were beamed directly at Earth.

In the discussion presented here, we limit the range of our analysis to the first event, Elsie; the analysis of the remaining Elsie dip family will be presented by E. Bodman et al. (2018, in preparation). We also assume that the differential fluxes found in each filter are measures of the additional opacity causing the dips, not the total color of all the material along the line of sight (i.e., we apply no correction in the baseline flux for the longer-term brightness variations). For the following discussion, we define depth ratios, $B/{i}^{{\prime} }$ and ${r}^{{\prime} }/{i}^{{\prime} }$, as the ratio of the dip depths measured in normalized flux in the B to ${i}^{{\prime} }$ filters and ${r}^{{\prime} }$ to ${i}^{{\prime} }$ filters, respectively.

Continuous time-series coverage throughout the Elsie event shows the largest depth of 2.5% in the B band on JD 2,557,893. With monitoring across the many observatories, we obtained a nearly continuous cadence time series during Elsie, and thus can rule out a missed detection of a deep, short-lived dip, with the longest data gap occurring daily from 15 to 19 UT (a duration of ∼0.2 days). We fit the LCOGT multiband light curve with an $N+8$ parameter model, where N is the number of days over which the fitted data are taken. The unbinned data for each site are used for the fit.

A single depth ratio for each filter set ($B/{i}^{{\prime} }$ and ${r}^{{\prime} }/{i}^{{\prime} }$) is assumed for the entire dip. We include normalizations (n) of each telescope in each filter as free parameters, resulting in a total of six normalizations, to ensure that the different sites are normalized in a consistent manner. Each day has a depth (d_i) in the ${i}^{{\prime} }$ filter fit to the data binned in that day. When d_i is fit, it is compared to the ${i}^{{\prime} }$ band for each applicable site with normalization, and to the B and ${r}^{{\prime} }$ data after multiplied by the depth ratio (R) and normalization, e.g., ${F}_{\mathrm{Filter}}={n}_{\mathrm{Filter}}(1-{d}_{i}\times {R}_{\mathrm{Filter}})$. This allows for the color ratios, the normalizations, and the depths to be fit simultaneously. In order to prevent the normalizations from wandering too far from 1.0, we include 10 days of pre-Elsie data, where the depths are set to 0. To overcome degeneracies in the fit, we explored the parameter space with the Markov Chain Monte Carlo method using the "emcee" package (Foreman-Mackey et al. 2013). We used flat priors over ranges 0.5–5, 0.99–1.01, and 0–0.2 for the depth ratios, normalizations, and depths (respectively), with the results of a least-squares fit as the starting point. The results of the Bayesian fit are shown in Figure 7, and the depth ratios are $B/{i}^{{\prime} }=1.94\pm 0.06$ and ${r}^{{\prime} }/{i}^{{\prime} }=1.31\pm 0.04$. The ${r}^{{\prime} }/{i}^{{\prime} }$ ratio measured at the Calvin Observatory (1.29) is in agreement with the LCOGT value. The extinction due to optically thin dust is typically characterized by a higher extinction at bluer wavelengths such as is observed for KIC 8462852 during the Elsie dip. The ratios observed indicate dust grain sizes larger than are seen for interstellar dust and are more likely to arise from circumstellar dust (Savage & Mathis 1979; Meng et al. 2017).

Zoom In Zoom Out Reset image size

Under the assumption that the dimming is caused by intervening dust, the multicolor observations taken during Elsie may be used to put some constraints on the particle size and chemical composition. We therefore compared the measured amount of attenuation in different filters found in the previous section, $B/{i}^{{\prime} }=1.94\pm 0.06$ and ${r}^{{\prime} }/{i}^{{\prime} }=1.31\,\pm 0.04$, with the theoretical ratios of dust opacities for optically thin distributions at those wavelengths. The opacities of grains for different chemical compositions and different particle sizes are taken from the tables of Budaj et al. (2015). These opacities assume homogeneous spherical dust grains with a relatively narrow Deirmendjian particle size distribution (FWHM ≈ 0.4 dex; Deirmendjian 1964), as well as effective wavelengths of B, ${r}^{{\prime} }$, and ${i}^{{\prime} }$ of 0.4361, 0.6215, and 0.7545 μm, respectively. We explore the signatures of a few refractory dust species and water ice. Silicates are perhaps the most important refractory species. They are divided into the family of pyroxenes and olivines. Optical properties of silicates are quite sensitive to the amount of iron in the mineral and that is why we explore iron free and iron rich cases. Namely, for pyroxenes (Mg_xFe${}_{1-x}$SiO₃), we consider the two cases: a mineral with zero iron content (x = 1), which is called enstatite and a pyroxene with 60/40 iron/magnesium ratio. The refractive indexes of these pyroxenes are taken from Dorschner et al. (1995). Olivines (Mg_2yFe${}_{2-2y}$SiO₄) may also have different iron content. Again we consider the mineral with zero iron content (y = 1) called forsterite, and use its refractive index from Jäger et al. (2003). We also consider an olivine with 50/50 iron to magnesium ratio, taking its index of refraction from Dorschner et al. (1995). Alumina (Al₂O₃) is one of the most refractory species, which might be present in such environments and we take the complex refractive index for γ alumina from Koike et al. (1995). The refractive index for iron is taken from Johnson & Christy (1974). Those of carbon are from Jager et al. (1998), and they assume carbon at high temperature (1000 °C). Lastly, the complex index of refraction for water ice is taken from Warren & Brandt (2008).

The comparison between the observed and theoretical opacity ratios for different modal size (i.e., radius) of the grains and their chemical composition is shown in Figure 8. We find that most of the silicates and alumina should have particle sizes of about 0.1–0.2 μm. If the dust were composed solely from iron, it would have significantly smaller particles of about 0.04–0.06 μm. Carbon would require even smaller particles, $\lt 0.06$ μm. On the other hand, water ice would require 0.2–0.3 μm particles. We conclude that regardless of composition, the dust is $\ll 1$ μm in size and optically thin, which is consistent with the predictions in Figure 3 of Wyatt et al. (2018). Circumstellar dust that is this small would be strongly affected by radiation forces, putting it on an unbound orbit as soon as it was created. Thus, if it is circumstellar, the dust causing these short-duration dips must have been created recently, since radiation forces would cause it to spread from its point of origin.

Zoom In Zoom Out Reset image size

Regular monitoring of KIC 8462852 started in 2016 March with the Las Cumbres Observatory (LCOGT) 0.4 m telescope network, which has identical robotic capabilities to the LCOGT 1 m and 2 m telescope networks (Brown et al. 2013). The northern hemisphere 0.4 m network currently consists of telescopes at three sites: TFN (Canary Islands, Spain), OGG (Hawaii, USA), and since 2017 November, ELP (Texas, USA). The scheduled LCOGT requests consisted of Johnson B and Sloan $r^{\prime}$ and $i^{\prime}$ images, taking two exposures per sequence with a cadence of ∼30 minutes. On JD 2,457,892 (UT 2017 May 18; UT dates are used throughout this paper), a drop in brightness was claimed as significant at more than one site. We then increased the priority and cadence of the LCOGT ${Br}^{\prime} i^{\prime}$ sequence, and submitted additional ${Br}^{\prime} i^{\prime}$ sequence requests for coverage on the LCOGT 1 m telescopes in Texas, USA (ELP) and California, USA (SQA) and on the 2 m telescope in Hawaii, USA (OGG2). Data are automatically processed by LCOGT servers using BANZAI¹⁴¹ and transferred to local machines where we perform differential photometry for each telescope and filter image stacks using AstroimageJ (Collins et al. 2017).

Two hundred and thirty-one nightly observations of KIC 8462852 were acquired from 2016 April 15 to June 21 and again from 2016 November 12 to 2017 July 1 with the Tennessee State University Celestron 14 inch (C14) automated imaging telescope (AIT) at Fairborn Observatory. See Sing et al. (2015) for a brief description of the C14 operation and data analysis. We saw no evidence for variability in the first observing interval to a limit of a few millimagnitudes. KIC 8462852 likewise appeared to be constant in the second interval until 2017 May 18, when the C14 revealed the star had dipped in brightness around a percent. With the Fairborn and LCOGT data both showing signs of the start of a dip in brightness, an alert for triggered observations was executed.

Observations of KIC 8462852 were also made with Calvin College's remotely operated telescope in Rehoboth, NM (Molnar et al. 2017). Images in Sloan ${g}^{{\prime} }{r}^{{\prime} }{i}^{{\prime} }{z}^{{\prime} }$ filters were taken on 10 nights, on April 26 and then for a string of 9 nights beginning May 20. We used MaxIm¹⁴² to perform differential photometry.

KIC 8462852 was observed with the Joan Oró robotic 0.8 m telescope (TJO) at the Montsec Astronomical Observatory (Catalonia). The star was regularly monitored from 2017 March 14 and the priority and cadence of the observations was increased once the drop in brightness was detected. Sequences of five images in the Johnson R filter were then obtained one to three times per night and automatically processed with the TJO reduction pipeline (Colome & Ribas 2006). Differential photometry was performed using AstroimageJ.

The 0.6 m robotic telescope TRAPPIST-North¹⁴³ (Gillon et al. 2011; Jehin et al. 2011) observed KIC 8462852 starting 2017 May 20. Each night of observation consisted of exposures of 13 and 15 s gathered within an "I + z" filter and Johnson V filter, respectively. Data reduction was done automatically and consisted of the calibration (bias, dark, and flat field) and alignment of the images, and then of the extraction of the fluxes of selected stars by aperture photometry with the DAOPHOT software (Stetson 1987).

Observations at Thacher Observatory, on the campus of The Thacher School in Ojai, CA, began in the spring of 2017 in the Johnson V band. A variable number of observations of KIC 8462852 were performed nightly depending on the weather and telescope demand. Relative fluxes were obtained with aperture photometry and the ratio between KIC 8462852 and the sum of the seven reference stars are averaged to produce one measurement per night using a custom pipeline.

Johnson-Bessel BVRI images of KIC 8462852 were acquired with the NITES telescope (McCormac et al. 2014) on La Palma. The data were reduced in Python with CCDPROC (Craig et al. 2015) using a master bias, dark, and flat. A total of 5, 5, 4, and 2 nonvariable comparison stars were used for each of the B, V, R, and I filters, and aperture photometry was extracted using SEP (Bertin & Arnouts 1996; Barbary 2016).

We obtained 60 s exposure photometry using the SBIG STX-16803 CCD mounted on the 1 m telescope of the Wise Observatory in Israel, on a total of 15 nights between 2017 May 19 and June 21. We alternated between the B, V, ${r}^{{\prime} }$, ${i}^{{\prime} }$, and ${z}^{{\prime} }$ filters with 3 × 3 binning. The images were bias, dark, and flat-field corrected using IRAF.¹⁴⁴ Aperture photometry was performed using AstroimageJ.

Spectrophotometric observations were obtained with the OSIRIS long-slit spectrograph on the 10.5 m Gran Telescopio Canarias (GTC). Between 2017 May 17 and September 9, 13 pointings were performed, all with a resolution of R = 1000. Each pointing consisted of a time series of about 30 minutes in duration, from individual shots with 20 s exposure time. A nearby star of similar brightness (KIC 8462763, V = 11.86 mag) was included in the same slit and used as reference in the photometric analysis. The target's spectra were divided into five wavelength ranges, and the fluxes at each of these wavelengths are averages of the 30 minutes time series. A detailed description and analysis of the GTC data are presented in Deeg et al. (2018).

A.2.1. Low-resolution

A.2.2. High-resolution