A Large Ground-based Observing Campaign of the Disintegrating Planet K2-22b

Near-infrared light curves of K2-22 were obtained using the IRIS2 camera on the 3.9 m AAT (Tinney et al. 2004), located at Siding Spring Observatory (SSO) in Australia. IRIS2 is a 1K × 1K camera utilizing a HAWAII-1 HgCdTe detector, read out over four quadrants in the double-read mode, achieving a field of view of 77 × 77, at a pixel scale of 04486 pixel⁻¹.

Observations were obtained on UT 2017 March 15 (transit epoch = 2669) and UT 2017 March 16 (transit epoch = 2671) with IRIS2 in the Ks band, at 30 s exposure time. These observations were scheduled to accompany simultaneous optical photometry from the LCO 1 m telescope at SSO. The IRIS2 observing procedure and photometric reductions largely follow that described in Zhou et al. (2014). Dark exposures were subtracted from the data frames, and flat-field division was performed with the aid of on-sky offset frames taken before and after the time-series observations. To ensure the target star remained on the same pixel throughout the time-series observations, we made use of an off-axis guider, as well as manual adjustments to the telescope pointing via real-time plate solutions from the science frames. Plate solving and final aperture photometric extraction were performed with the fitsh package (Pál 2012). Relative photometry was performed with a set of four reference stars in the field.

We used KeplerCam on the 1.2 m telescope at the FLWO on Mt. Hopkins, Arizona to observe 5 transit windows of K2-22b. KeplerCam has a single 4 K × 4 K Fairchild CCD486 with a 0366 pixel⁻¹ and a field of view of 231 × 231. The observations were made on UT 2017 March 8 (transit epoch = 2650), UT 2017 March 9 (transit epoch = 2652), UT 2017 April 16 (transit epoch = 2752), UT 2017 April 21 (transit epoch = 2765) and UT 2017 May 22 (transit epoch = 2846). All observations were obtained in an SDSS i filter with a 120 s exposure time. Images were reduced using standard idl routines as outlined in Carter et al. (2011). AstroImageJ (Collins et al. 2017) was used to perform aperture photometry on the processed images. We used an aperture radius of 8 pixels, corresponding to a 293 radius.

We observed three transits of K2-22 on three consecutive nights from UT 2017 March 15 to UT 2017 March 17 (transit epochs = 2669, 2672, 2674) using the 0.7 m Thai Robotic Telescope—Gao Mei Gu Observatory (TRT-GAO), in Lijiang, China, with an Andor iKon-L 936 2 K × 2 K CCD camera with a scale of 0613 pixel⁻¹. The observations were performed through a Cousins-I filter with 60 s exposures. The total science images on each night are 250 (UT 2017 March 15), 140 (UT 2017 March 16) and 150 (UT 2017 March 17). The calibration was carried out using the iraf tasks along with astrometric calibration using Astrometry.net (Lang et al. 2010). The aperture photometry was carried out using sextractor (Bertin & Arnouts 1996) with an adaptive scaled aperture based on the seeing in an individual image. The final apertures used were between 4 and 5 pixels equivalent to 245–307 (Bertin & Arnouts 1996).

We were awarded Director's Discretionary Time to use the OSIRIS instrument (Cepa et al. 2000) mounted on the 10.4 m GTC to observe 1 transit of K2-22b (PI: E. Pallé). The data were obtained on UT 2017 May 17 (transit epoch = 2835) using OSIRIS long-slit spectroscopy mode with the R1000R grism (spectral coverage of 510–1000 nm), 2 × 2 binning mode (0254 pixel⁻¹), readout speed of 200 kHz, gain of 0.95 e⁻/ADU, and readout noise of 4.5 e⁻. A custom built long slit with a width of 12'' was used to obtain spectra of K2-22 and one reference star. The exposure time was set to 250 s and the total observing time was 3.3 hr (airmass varied from 1.12 to 1.97). A total of 45 science images was acquired during the run. Images were reduced using standard procedures (bias and flat calibration), and the spectral extraction of K2-22 and the reference star was done using an aperture of 40 binned pixels (102) in width. For more details on the GTC data reduction procedure, we refer the reader to Section 7 of Sanchis-Ojeda et al. (2015).

We collected 24 light curves of K2-22 between UT 2017 March 1 and UT 2017 May 29 in either the Sloan g or Sloan i band from multiple 1 m telescopes in the LCO network (Brown et al. 2013). The corresponding transit epochs and site of each observation are given in Table 2, where CTIO refers to Cerro Tololo Inter-American Observatory in Chile, McDonald refers to McDonald Observatory in Texas, SAAO refers to South African Astronomical Observatory in Sutherland, South Africa, and SSO refers to Siding Spring Observatory near Coonabarabran, New South Wales, Australia. The 1 m LCO telescopes each have a 4 K × 4 K Sinistro detector with a 26' × 26' field of view and a pixel scale of 039 pixel⁻¹. Calibrated data were downloaded from the LCO archive and then analyzed to extract the photometry following a similar procedure as described in the following section.

We collected two light curves of K2-22 on UT 2017 March 13 (transit epoch = 2663) and UT 2017 April 11 (transit epoch = 2741) using the E2V 4 K × 4 K CCD at the 40 inch Swope telescope at Las Campanas Observatory. Observations were acquired in both cases with the SDSS i filter but with different exposure times. Due to the closeness of the target to the Moon on the March 13 observations, the exposure times were set at 70 s. This allowed us to achieve photometry at the ∼2% level only, due to the local sky brightness. The observations spanned almost 5 hr, covering a significant portion of the orbit of K2-22b. On April 11, the Moon did not interfere with our observations and thus the exposure times were set to 110 s, which allowed us to reach photometry at the 0.5% level per point over 6.5 hr, covering a significant fraction of the orbit of K2-22b. The data were reduced using a standard photometric pipeline which makes use mainly of tools from the Astropy package (Astropy Collaboration et al. 2013), and performs bias, dark and flat-field corrections, along with astrometric identification of stars in the field using Astrometry.net (Lang et al. 2010) and subsequent extraction of photometry for all the stars in the field. Differential photometry for K2-22 was produced by using an ensemble of 10 comparison stars over a 10 pixel radius, which implies a radius of 435 as the pixel scale is 0435 pixel⁻¹.

At TNO, in Chiang Mai, Thailand, we conducted photometric observations of K2-22b on the 0.5 m Thai Robotic Telescope—Thai National Observatory (TRT-TNO) and the 2.4 m Thai National Telescope (TNT) on UT 2017 March 15 (transit epoch = 2669) and UT 2017 April 13 (transit epoch = 2745), respectively.

A transit observation was conducted with the TRT-TNO through a Cousins-R filter using an Andor iKon-L 936 2 K × 2 K CCD camera attached to the 0.5 m Schmidt–Cassegrain Telescope. The field-of-view of each image is 234 × 234. The observation was conducted using 60 s exposure time and 250 science images were obtained. The calibration and photometry were performed using the same routine as TRT-GAO observations.

For the TNT observations, the target was observed with ULTRASPEC (Dhillon et al. 2014) in the i-band. ULTRASPEC is a high-speed camera using a frame-transfer electron-multiplying CCD (EMCCD). The target was observed in full-frame mode, with a field-of-view of 77 × 77 and a pixel scale of 045 pixel⁻¹. The exposure time was 42.78 s, with a dead time of 15 ms between each frame. More than 400 science images were obtained, but only the first 221 images are used in our analysis as latter images were affected by clouds. Plate matching and aperture photometric extraction were performed using the fitsh package (Pál 2012). Relative photometry utilized a largely similar set of reference stars as the LCO observations (barring differences due to the different field of view). Light curves were extracted over six different apertures and background annulus radii to optimize the point-to-point photometric scatter.

We observed five complete transit windows of K2-22b on UT 2017 March 08 (transit epoch = 2650), UT 2017 March 09 (transit epoch = 2652), UT 2017 March 15 (transit epoch =2668), UT 2017 May 22 (transit epoch = 2846), and UT 2017 May 25 (transit epoch = 2854) using the ULMT located at the Mt. Lemmon summit of Steward Observatory, AZ. The observations employed a 0.6 m f/8 RC Optical Systems Ritchey–Chrétien telescope and an SBIG STX-16803 CCD with a 4K × 4K array of 9 μm pixels, yielding a 266 × 266 field of view and 039 pixel⁻¹ image scale. All observations were conducted using 200 s exposure times and no filter. Conditions were clear except for some thin clouds during the observations on UT 2017 March 09 and UT 2017 March 15, which caused higher scatter in the light curves.

We used AstroImageJ (Collins et al. 2017) to perform bias, dark, and flat-field corrections, and to perform fixed radius circular aperture photometry on the final processed images. We used aperture radii of 10 pixels (39) and a comparison ensemble of four or more stars to produce differential target star light curves.

We observed one complete transit window of K2-22b on UT 2016 December 13 (transit epoch = 2427) using the WIYN²⁵ High-Resolution Infrared Camera (WHIRC) installed on the 3.5 m WIYN telescope at Kitt Peak National Observatory in Arizona. WHIRC is a Raytheon Virgo HgCdTe detector with a 2048 × 2048 array, a pixel scale of 01 pixel⁻¹, and a 34 × 33 field of view. The Ks-band observations began about one hour before the expected transit and ended approximately one hour after the end of the expected transit, when morning twilight began. Some clouds were present during the observations.

The first step in processing the WHIRC data involved using the wprep iraf script available on the WHIRC instrument website²⁶ to trim the images and perform a linearity correction. We then corrected the images against darks, masked bad pixels, removed the pupil ghost using an image generated by dividing Ks and J flats, and performed flat-fielding. We used AstroImageJ (Collins et al. 2017) to perform circular aperture photometry on the final processed images, using an aperture radius of 13 pixels (13) and one comparison star to produce a light curve.

To better understand the distribution of transit depths in our data and our detection limits, we performed a transit light curve fit using the Mandel & Agol (2002) model to all 31 data sets that covered most of the expected transit window (as per the ephemeris from Sanchis-Ojeda et al. 2015). The following constraints were imposed on the fits: the transit duration²⁷ was required to be <2 hr, the planet-star radius ratio (R_p/R_*) to be <1, and the mid-transit time should be within 1 hr of the predicted transit time. To account for long-term trends in the baseline of a light curve, we also allowed for a quadratic trend to be fitted simultaneous to the transit fit. The resulting values measured for R_p/R_* are given in Table 2. If no uncertainty in R_p/R_* is provided in the table, then the value for R_p/R_* is the 1σ upper limit. If no value for R_p/R_* is given at all, then that particular light curve did not cover a complete transit window of K2-22b. Figure 1 shows the detrended data and best-fitting models to the 31 light curves that fit our criteria.

As identified in Sanchis-Ojeda et al. (2015), K2-22 has a faint nearby companion located ∼2'' away. The angular proximity of this neighboring star means that in most of our photometry, the flux from the target and the neighbor are blended. To mitigate the effects of the neighboring star, we follow Ciardi et al. (2015) to calculate the true planet radius (and therefore the true planet-star radius ratio): ${R}_{p}(\mathrm{true})/{R}_{p}(\mathrm{observed})=\sqrt{{F}_{\mathrm{total}}/{F}_{1}}$, where F₁ is the flux of K2-22 as it is the star that is being transited and F_total is the combined flux of the primary and neighbor star. We used magnitudes provided in Table 3 of Sanchis-Ojeda et al. (2015) for K2-22 and its neighbor to compute F₁ and F_total for the relevant bandpasses where we measured either a detection or upper limit of the transit of K2-22b. We provide the corrected planet-star radius ratios in Table 2.

Given the proximity of the neighbor, there is some probability that it is the source of the transit signal rather than K2-22. However, Sanchis-Ojeda et al. (2015) argue why K2-22 is the most likely transit host rather than the neighbor. Furthermore, the WIYN observations presented here resolved K2-22 from the neighbor and provided a tentative detection of the K2-22b transit. More importantly, the neighbor did not display an obvious deep transit during the WIYN observations, which is additional evidence it is not the source of the transit signal.

To determine how our measured transit depths compare to previous observations, we returned to the K2 Campaign 1 discovery data collected between 2014 May and August and followed the procedure outlined in Section 3 to measure the depths of transits in that long-cadence data (where each measurement has an integration time of 30 minutes). Depths were measured for transits that had at least two points in-transit, and the results are shown in Figure 2 as a function of epoch. The detected K2 transits have depths ranging from 0.21 to 0.57% (the 1σ distribution) with a median depth of 0.37%.

Zoom In Zoom Out Reset image size

In Figure 3, we show the measured transit depth (or upper limit on the transit depth) as a function of epoch from our ground-based follow-up campaign, along with the Probability Distribution Function (PDF) of the transit depths measured from K2 data. Due to the overall shallowness of the transit, we detected the transit in just ∼one-third of the 31 complete transit windows observed. We still find that our measured depths are consistent with what K2 detected, within the uncertainties.

Zoom In Zoom Out Reset image size

Our most robust detection was obtained from the GTC, with a measured transit depth of ∼0.3% (Figure 1). To test whether we should have recovered a transit of this depth in our other individual data sets, we injected the GTC signal into the other light curves we collected. The resulting Box-fitting Least Squares (BLS; Kovács et al. 2002) spectrum is shown in Figure 4. When performing the BLS search on our data, we searched for transits with q_min = 0.02, q_max = 0.15, and orbital periods between 0 and 10 days, where q is the transit duration/orbital period. Figure 4 illustrates that if the transit of K2-22b had a consistent depth of ∼0.3%, we would have recovered it in the other individual observations. This suggests that the transit depth of K2-22b was changing over the course of our campaign. Figure 3 further supports this: for example, an initial detection of the K2-22b transit was made in 2016 December followed by a cluster of non-detections (upper limits) in early 2017 March (epoch 2638), and then beginning in 2017 mid-March (epoch 2669) we had several positive detections of the transit. Several of the light curves where we had a non-detection had sufficiently high photometric precision that we should have recovered a ∼0.3% transit. Therefore, we conclude that over the course of our observing campaign, the transit depth was indeed changing and was at times deeper and at other times shallower than the highest-precision event obtained by the GTC on 2017 May 17 (Figure 1).

Zoom In Zoom Out Reset image size

Comparing the high-precision transits observed with the GTC in 2015 and in 2017 (Figure 5) also reveals that the 2017 transit is shallower than previously measured in 2015, however, the 2017 observations are still consistent with depths measured from K2 overall. This instead suggests that the 2015 transits may have been observed at a time of relatively higher activity (e.g., enhanced ultraviolet and/or X-ray radiation) that led to increased mass-loss and deeper transits than typically seen by K2.

Zoom In Zoom Out Reset image size

One additional note of interest is that the GTC transit observed in 2017 appears to have occurred exactly when predicted and to be more symmetric than the previously observed transits from the GTC. The other light curves presented in Figure 1 have poorer individual photometric precision than the GTC, so we are unable to perform a robust investigation of any changes in the transit shape or ephemeris over each epoch. While some appear to have longer durations than expected and/or occur outside the predicted transit window, due to the relatively poor photometric precision in those cases we do not claim that those deviations in transit duration or time are significant.

Because we have 24 light curves from LCO in total (17 in Sloan i band and 7 in Sloan g band), we explored whether stacking the phased light curves would provide a clear detection of the K2-22b transit. Figure 6 shows the stacked light curves from LCO in both bands, where the photometric error bars are defined as the standard deviation of the binned light curve outside of the expected transit window. Interestingly, while we do not see an obvious transit in the i band data, the g band data reveal a potential ∼0.3% transit occurring slightly earlier than predicted. We note the g band data were collected between UT 2017 March 16–18 between transit epochs 2671 and 2676, while the i band data were collected over several months between UT 2017 March 1 and May 29. Since the g band data were collected over a relatively short timescale, this suggests the g band detection is robust, although these data have notably poorer photometric precision than the i band data. The g band depth is also consistent with what we measured from the high-precision GTC light curve collected two months later on UT 2017 May 17. However, given the significant scatter in the g band data, we cannot definitively conclude that an early transit of K2-22b was detected in the stacked light curves.

Zoom In Zoom Out Reset image size

In addition, there are a few individual measurements that may be anomalous and therefore affecting our interpretation of the data. The first detection in 2016 December (with WIYN; see Figure 1) is highly dependent on detrending. If we assume the transit is longer than expected, then the curvature seen in the light curve could be indicative of a transit as we present here. On the other hand, if we assume the transit has the expected duration of ∼48 minutes, then detrending the data results in a null detection of a transit and reduces the apparent slope in the transit depth over time. The uncertainty in the transit duration and depth for the WIYN data likely stems from the fact that we only had one suitable comparison star in the field of view. LCO observations at epoch 2865 (Figure 1) also show a significant trend at the beginning of the light curve, but as the data acquired at the subsequent epoch do not show any anomalous trend, we believe this is not astrophysical and instead some type of instrumental artifact. The best-fitting model is correspondingly a poor fit given this anomalous light curve feature.

The few epochs where we have multi-wavelength coverage over a complete transit window (i.e., epochs 2652, 2669, 2672, 2846 in Figure 1) do not reveal any significant color dependence. While the R and I band data at epoch 2669 are suggestive of a depth difference, we do not have the photometric precision to make a robust claim.

Figure 6 shows the stacked light curves from LCO in Sloan i and g bands, and there we see that the g band data are suggestive of a ∼0.3% transit. On the other hand, we see no clear evidence for a transit in the i band data. That we find evidence of a chromatic transit that is deeper at bluer wavelengths is consistent with what has been observed for, e.g., KIC 12557548b (Bochinski et al. 2015) and KIC 8462852 (Boyajian et al. 2018; Deeg et al. 2018) (see Section 5 for further discussion of these other systems). The caveat is that we are comparing stacked light curves collected over several days to months, and the g band data have poorer photometric precision than the i band data. We therefore cannot make any definitive claims about the chromaticity based on the LCO data and instead conclude that the LCO data are suggestive of a deeper transit at blue wavelengths.

We also analyzed the spectroscopic data collected from the GTC, where in Figure 7 we present the GTC data split into blue and red wavelengths and also in three different color bands. During this one transit, no significant color dependence in the depth is measured. We note that of the three transits observed previously with the GTC, only one displayed a color-dependent depth (with a deeper transit seen at blue wavelengths than red), and that was the deepest transit measured on UT 2015 February 4 as shown in Figure 5 and Sanchis-Ojeda et al. (2015). Ridden-Harper et al. (2018) recently demonstrated that the transit depth of disintegrating rocky planets is wavelength-independent for optically thick tails. The single wavelength-dependent transit observed for K2-22b to date could therefore be a sign of an optically thin tail at that particular epoch.

Zoom In Zoom Out Reset image size