THE RCT 1.3 m ROBOTIC TELESCOPE: BROADBAND COLOR TRANSFORMATION AND EXTINCTION CALIBRATION

The KPNO 50 inch was developed in the early 1960s by the Space Sciences Division at Kitt Peak National Observatory (KPNO) as an engineering research platform for the development of remote control protocols for future orbital telescopes. It was to be operated on Kitt Peak remotely from Tucson, AZ via commercial telephone lines, a concept in-line with today's robotic telescopes. The telescope was installed in its present location in 1965 and operated as a remote observatory test bed (Maran 1967b) until 1969, when the facility was converted into a full-time on-site visitor facility. The telescope was then fitted with an aluminum-coated ceramic-vitreous primary mirror cell, replacing a nickel-coated cast aluminum cell (Maran 1967a) that warped with changes in ambient temperature. For the next 25 years, the KPNO 50 inch served as an optical and near-infrared facility for visiting astronomers, as well as a test bed for National Optical Astronomy Observatory (NOAO) near-infrared instruments, such as the Infrared Imager, the Simultaneous Quad Infrared Imaging Device, the Cryogenic Spectrograph, and the Cryogenic Optical Bench. However, budget constraints eventually forced decommissioning of the 50 inch as part of KPNO portfolio in 1996. In 2000, the RCT Consortium successfully bid to operate the telescope (dubbed the Robotically Controlled Telescope) and completed its full refurbishment as a robotic instrument in 2009. Tests of the engineering and optical quality continued through 2012, over which time science programs such as this calibration program also commenced. It is now fully operational and open for regular use by the RCT Consortium partners.

The refurbishment and automation of the telescope was driven by a group of researchers, most of whom were from primarily undergraduate institutions, with complementary science goals that would benefit from an automated 1 m direct imaging facility. Broadly, the envisioned projects required precise photometric monitoring (σ_mag ≲ 0.1) of relatively bright targets (≲ 20 mag), in visits of a few minutes often executed periodically over long timespans (weeks to years). The details of each project set different requirements for photometric precision, often weighed against acquisition constraints and practical limits on the overall versatility of the automatic scheduling. Here we focus on a few projects that demonstrate some of these challenges and thus exemplify the science requirements for the RCT.

Measuring transiting extrasoloar planets. An important goal in probing stellar system evolution is to at least detect large planets about relatively small stars, for example, Neptune-sized planets about main sequence F or G stars. Through the transiting method, the change in apparent brightness, or transit depth (Δm), is at best dependent on the relative cross sections of each component by

$$\begin{equation} \Delta m = -2.5\,\log \left (1-\frac{R_p^2}{R_\star ^2}\right ), \end{equation} \tag{ 1 }$$

where R_p and R_⋆ are the radii of the planet and star, respectively. A Neptune-like planet transiting an F-star would result in a Δm of, at most, ∼0.001 mag. This objective places strong constraints on photometric precision for the telescope. Single-exposure estimates based on the RCT exposure time calculator (RCT-ETC) suggest σ_mag ≃ 0.001 should be achievable for bright (roughly V ≲ 14 mag) targets in 2–3 minute exposures (McGruder et al. 2003). The RCT-ETC (http://rct.wku.edu/kittpeak/etc.html) is based on the IRAF task ccdtime and provides approximate exposure times, signal-to-noise ratio (S/N), and magnitude limits based on optimized estimates of the site and telescope conditions. We discuss comparisons to real data in Section 3.2.

Ephemerides of minor planets. Refining the ephemerides of minor objects in the solar system requires precise astrometry, which is often complicated by the non-sidereal motion of the object itself. The goal is to obtain sufficient S/N in exposures short enough to keep star field trailing significantly below the FWHM of the point spread function (PSF). For many comets and asteroids, this limits exposure times to only a few tens of seconds, which should be sufficient for σ_mag ⩽ 0.1 on objects as faint as V ∼ 17 mag.

Monitoring active nuclei. Blazar variability monitoring, particularly on short timescales, can provide important clues to the nature of the supermassive black holes that they harbor. Well-sampled light curves of the variation in optical continuum provide probes the size of emission regions that, when combined with complementary data from gamma-ray to radio wavelengths, provide strong constraints for model conditions of these active galactic nuclei. These variations, however, are unpredictable and irregular occurrences lasting several minutes or several years, thus requiring a level of periodic monitoring that often challenges classical scheduling at ground-based telescopes. The objective for the RCT is keep to an observing cadence for some programs and use other periodic programs as fillers to keep the overall schedule efficient. The automated scheduling of the RCT, as discussed further in Section 2.2, is efficient enough to wedge small programs between optimal windows of larger ones, repurpose windows of failed observations, and utilize fractions of nights in marginal or suboptimal conditions that might otherwise be lost. Combined with options for repeated requests, the scheduling allows a regular monitoring of blazars for well-sampled long-term light curves, for complementary studies with, for example, RXTE, VERITAS, and Fermi (Carini et al. 2004).

Light curves of supernovae and gamma-ray bursts. Both gamma-ray bursts and supernovae are now routinely discovered at a rate of several hundred per year, respectively, now enabling a more comprehensive understanding of the nature of these extremely energetic explosions. There is still much to ascertain on the most and the least energetic ranges of both supernovae and gamma-ray burst optical transients, especially at early times or within hours of explosion, and most critically for gamma-ray burst of the shortest duration. Doing so for a sufficiently large sample will likely require coordinated action from rapid-response observatories, ideally with queues that can be appended to or reconstructed automatically (within minutes) to catch high-priority events (McGruder et al. 2004). As discussed further in Section 2.2, the RCT scheduler is designed to take the relative priority of observations into account when building its schedule, which is rebuilt frequently to accommodate new targets throughout night. The coordination of targets of opportunity will eventually come from large-scale "all-sky" surveys, such as PanSTARRS, SNfactory, and Palomar Transient Factory (and eventually LSST), which will act as "event brokers" for the community via VOEvents, GCN, ATEL, CBET, or other electronic announcements (see compendium from "Hot-wiring the Transient Universe," http://hotwireduniverse.org).

Narrowband imaging of resolved nebulae and comets. The RCT is equipped with two large filter wheels, allowing—in addition to standard broadband filters—immediate access to one of two sets of narrowband interference filters. The narrowband filters are centered on diagnostic nebular emission lines from H⁺, He⁺⁺, S⁺, N⁺, and O⁺⁺ in one set and molecular diagnostic lines (with complementary continuum) useful for comets and minor planets in a second set. These narrowband filters extend the versatility of the RCT, allowing two-dimensional (2D), spacially resolved maps of extinction (Hα/Hβ), electron temperatures, and densities; ionization maps for nearby nebulae; and extensive coverage of comets pre- and post-perihelion to monitor outburst and fragmentation, determine volatile production rates, and trace the evolution of the comae (Walter et al. 2004).

The RCT camera operates with a 2048 × 2048 SITe SI-424A backside-illuminated CCD with 24 μm pixels. The CCD is housed in an IRLabs dewar,¹¹ which is cooled to 172 K via a closed-cycle, PT-14 gas Polycold refrigeration system from Brooks Automation, Inc.¹² The camera sits at the f/13.5 Cassegrain focus to produce a plate scale of 1189 mm⁻¹ (0285 pixel⁻¹) and an unvignetted 96 × 96 field of view.

The filter-wheel assembly sits above the dewar window, housing a pair of nine-slot filter wheels, within each are eight circular mounts for 4.5 inch diameter filters and one unthreaded open mount, to provide a total of 16 possible filter options (and one double-open position) continuously available. Table 2 provides a complete list of the filters available, along with effective wavelengths, bandwidths (or FWHM), and maximal throughputs.

The RCT broadband set consists of Johnson/Bessell UBVRI filters from Andover Corporation.¹³ Figure 1 shows the nominal and effective filter throughputs, corrected for the CCD quantum efficiency, nominal atmospheric transmission at zenith from a smoothed MODTRAN5 model (Berk et al. 2006), dewar window transmission, and primary mirror reflectance, as measured in 2009. These are ideal measures, as factors associated with degradation and frequency of maintenance reduce the overall system response, which is discussed further in Section 3.2.

Zoom In Zoom Out Reset image size

While the RCT 1.3 m can be operated on-site or remotely, its main function is in a full-robotic imaging mode. The autonomous scheduling algorithm is an essential component of this observatory and has been designed to manage numerous requests with various observing constraints. It is built from the INSGEN list generator and process spawner originally developed for the Berkeley Automatic Imaging Telescope (Richmond et al. 1993), which makes scheduling decisions through a least-cost insertion algorithm.

The method of the INSGEN scheduling algorithm (or scheduler), as fully described in Richmond et al. (1993), is as follows. Users provide "requests" that detail observation information such as the target(s), the number of exposures, the exposure times, and filters, and the observing constraints such as the range of airmass, minimum angle of Sun/Moon avoidance, or specific start or end times. Users also assign a priority to their requests, which in principle can be managed by partner fractions. The scheduler calculates "windows of opportunity" for each target in each request on the basis of the target's rise and set times, the block length of the request, readout and slew times, sunrise and sunset, and other constraints.

A "best time" is defined within each window, which is generally centered at the meridian transit unless otherwise constrained. The scheduler sorts the request list by priority and reserves blocks at the best times according to priority. As the scheduler proceeds through the prioritized list, it searches outward from a given request's best time for free blocks within the target's window of opportunity. If no free window can be identified, the scheduler then attempts to accommodate the block by pushing higher priority requests away from their optimal spots without pushing them out of their observable windows or reordering the sorted list. If this cannot be accomplished, the request with the lower priority is skipped until the following scheduling instance, in which it discards the original list and starts again.

Scheduling instances are implemented periodically throughout the daytime hours and each time an observation has completed. With each instance, the previous observation is rejected from the list (if applicable), and the allowable windows are reconstrained by either sunset or the current time, whichever is latest. This allows for real-time ingesting of new observation requests at any time, even while the telescope is observing, and significant flexibility to keep a highly efficient schedule in case previous requests have a shorter execution window than anticipated (usually slew times shorter than the estimated 2 minute intervals) or fail because of a lack of available guide stars. Scheduling instances take little time to execute, usually less than 10 s for the hundreds of active requests currently in the queue.

The INSGEN scheduler operates the RCT with highly efficient observing plans, over periods of continuous observing, or without instrument outages or poor weather. As a measure of the telescope's efficiency, we show in Figure 2 the fraction of each night the imager exposed on fields, which was generally between 40% and 60%. This efficiency figure includes the CCD readout time for each exposure, as this is set by the binning mode or similar details for a specific observation. But the fraction omits the slew time to each field, filter setup, or guide star acquisition (when applicable) as these are limitations more inherent to the facility. The remainder of the time was lost to weather and instrument servicing, with less than 10% spent on field acquisition and other overheads.

Zoom In Zoom Out Reset image size

The observation requests are collected and logged in an on-site database, populated by users via a web interface available to partners. There are several convenience options at the interface, including target resolution via NED, SIMBAD, JPL Horizons, or the expanding RCT database. Users can monitor the schedule via the same web interface, seeing both tabular and graphical representations of the scheduled, skipped, and executed requests.

The RCT executes a calibration sequence in a separate request, taken nightly or as priority allows. Bias images were taken in a dark dome (after twilight) as a precaution, although the instrument shows no evidence of light leaks and the dark current is estimated to be less than 1 × 10⁻⁵ e⁻ s⁻¹. Flat-field correction images were derived from images of a lamp-illuminated screen on the interior of the dome, which are done as a matter of course in at least each of the broadband filters. All science data in this analysis were bias and dome flat-field corrected using the IRAF imred.ccdred package, with respective nightly correction images. Comparisons of a few dome-flat and bias images yielded a nominal read noise of 15.1 e⁻ and gain of 2.6 e⁻ DN⁻¹.

Raw RCT images have only an approximate World Coordinate System map to their pixels. To accurately correlate sources with known positions of standards, we pattern matched detections (via SExtrator; Bertin & Arnouts 2010) with the USNO B1 catalog (Monet et al. 2003), using a triangle-matching algorithm that solves for linear source-to-source transformations, including shifts, rotations, and changes in scale, in which the plate scale is a free parameter. This automated fitting required greater than approximately five detections per image, evenly distributed, for accurate matching (to rms <02). For most observations in the redder passbands of these standard fields, where there is greater sensitivity and a larger number of red sources, this criteria was easily met. But for bluer passbands where there were fewer detected sources, the failure rate in direct pattern matching was as much as two times more frequent than for the redder passbands, resulting in about half as many correlated measures.

We may have had more success in pattern matching if we reduced the free parameters in the matching algorithm, fixing the plate scale to set value for example, although it was important to independently verify the plate scale of the image plane. We might also have made use of the back-to-back execution of the observing sequence to create "detection frames" from combinations of images, in the same filter or in all-filter combinations. Although that would have required some slight image-to-image forced alignment, through cross-correlation or human source identification, as small (∼05) drifts between exposures occur in the sequence when tracking was momentarily reset. However, the final number of matched targets were more than sufficient to precisely determine the photometric transformations, even after the rejection pass, and so it was not deemed necessary to put effort into recovering these acceptable losses.

Instrumental aperture photometry was performed in a 6.5 pixel (371) radius on rectified images, scaled by their exposure times to units of DN s⁻¹. The aperture was chosen by determining the growth curve of several hundred stars in selected images of each passband, as illustrated in Figure 4, and selecting an aperture consistently near the confusion limit (6–7 pixel radii, on average). This aperture was large enough to accommodate changes in the PSF due to seeing and other instrumental effects. Aperture corrections were estimated from ratios of the encircled energy of a 2D Gaussian with symmetric FWHM equal to the mean measured FWHM of the image, at the confusion aperture relative to an infinite aperture. In all passbands, these corrections were found to be less than 0.01 mag. However, it is important to note, as can be seen in Figure 4, that the shape of the PSF was typically more complex than the 2D Gaussian fit. An extreme case due to instrumental failure is also shown in Section 3.3.

Zoom In Zoom Out Reset image size

A first-pass calibrated magnitude (m) was determined in each filter from the instrumental magnitude using

$$\begin{equation} m = {-}2.5\log _{10}({\rm DN\,\,s}^{-1})+Z+A_0X+ A_1({\rm color})+A_2X({\rm color}), \end{equation} \tag{ 2 }$$

where Z is the zero point (magnitude at 1 DN s⁻¹), A₀ is the airmass term coefficient, A₁ is the color term coefficient (using colors later indexed in Table 3), and A₂ is the airmass × color term coefficient with the same color indices. The calibration coefficients in Equation (2) (Z, A₀, and A₁, respectively) were fit via the weighted Levenberg–Marquardt nonlinear least squares algorithm provided by the IRAF apphot package. The cross extinction–color corrections (or A₂ terms) are often difficult to constrain when simultaneously fit with the color transformations and first order extinctions, and leaving these terms free resulted in degeneracies that were inconsistent with the data. It was expected, however, that these terms would be negligible, with |A₂| < 0.02. Therefore, we fixed A₂ ≡ 0.0 for all filters throughout this analysis. Figure 5 shows the difference in instrumental and cataloged magnitudes as functions of each of the fit parameters. Equilibrium binned data (11 bins of approximately 250 measures in each bin) and the slopes of the preliminary photometric calibration coefficients are shown for comparison. Note that the fits are to the data and not to the binned points.

Zoom In Zoom Out Reset image size

Figure 6 shows the residuals of the fits, biased to the initial zero point magnitudes for each filter, with additional offsets for clarity. The RCT is equipped with a modest "cloud monitor," a thermoelectric infrared sky temperature sensor that compares sky temperature with ground temperatures to derive, in principle, an approximate fraction of the sky covered in clouds. The residuals in Figure 6 are shaded in grayscale by these relative cloud values. As non-photometric data are inherently fainter, their measures stream downward on this plot. It is apparent from the figure that the cloud values do little to discern non-photometric (or otherwise faint) data. At present, it is unclear if the cloud monitor is too imprecise, or inaccurate because of failure, but it would seem that it is not yet a useful indicator of photometric quality and thus insufficient for isolating photometric data for this analysis.

Zoom In Zoom Out Reset image size

The distribution of residuals in the fits do, however, provide a guide for isolating photometric data. The histograms inset to the right of Figure 6 show the distribution of the residuals relative to the zero points derived in this first pass analysis. As one would expect, these distributions grossly show modes near the bright end of each calibrated set, with skewed smaller populations trailing toward the faint end from gradually less photometric observations. However, there also appears to be bimodality in the distributions, with fainter secondary loci located in the tail of the distributions, most visibly in the B-band set. Such distributions could result from data obtained in systematically similar (and unfavorable) weather conditions or from a systematic error in the functionality of the telescope. We discuss some likely possibilities, with focus on the latter, in Section 3.3. For the sake of furthering this analysis, we assume these two populations are distinct and can be reasonably separated into photometric and non-photometric (or otherwise obscured) sets. We fit bimodal Gaussian probability distribution functions (PDFs) to the histograms, solving for the mean and standard deviation of each mode. These distributions are overplotted on the histograms of Figure 6.

We then performed a second-pass analysis, following the same apphot procedure to determine calibration coefficients outlined at the beginning of this section, on only the data constrained within 1σ of the bright modes in the histograms of Figure 6. The results of this final pass are shown in Figure 7, and the final calibration coefficients are shown in Table 3.

Zoom In Zoom Out Reset image size

The final derived color transformation coefficients are in excellent agreement with available data from comparable telescopes with optical imagers and similar broadband filters at KPNO, specifically the WIYN 0.9 m and the KPNO 2.1 m (Massey et al. 2002). The extinction coefficients are reasonably similar, although our analysis has been carried out to larger airmasses than presented by Massey et al. (2002). As discussed previously, no attempt was made to fit the airmass × color terms for possible atmospheric reddening, as it was assumed the effect was negligible. The lack in notable trend in the data of Figures 5 and 7 generally confirm this assumption.

The photometric calibration coefficients were determined with total errors in the mean of less than 0.3%. The photometric errors (or errors in the calibration, presented as 1σ errors in Table 3) are also very reasonable, although there may be systematic contributions to these residuals attributable to faults in the facility, which is a likelihood further discussed in Section 3.3. As an independent test, these photometric corrections were applied to the early RCT direct imaging data of SN 2011fe, obtained at an average airmass of 1.9. The results shown in Figure 8 are in excellent agreement with data from various observatories of this nearby Type Ia supernova.

Zoom In Zoom Out Reset image size

As another test of the calibration quality, we compared the calibrated magnitudes of all 22, 059 data points with their measured photometric errors, in a S/N comparison shown in Figure 9. The figure shows an orderly agreement between the calibrated magnitudes and the measured magnitude errors that would otherwise be absent (resulting in increased scatter along the abscissa) if the calibration were inaccurate. There are two issues, however, to note. First, the trend in increasing noise occurs at much shallower magnitudes (by factors of 2–4) than expected from estimates in S/N given by the RCT-ETC, given the appropriate exposure times, seeing conditions, and average sky brightnesses. Some deviation from these ideal estimates are expected, as the overall system response changes over time. It is not likely this is an issue with the aluminization of the primary, as measures of the reflectivity of the primary mirror have changed little since 2009 and are appropriately accounted for in the RCT-ETC models. The atmospheric corrections used in the RCT-ETC are also well matched to the MODTRAN5 models for the site. It is most likely this is a degradation in the broadband filter A/R coatings, reducing the response since their initial measures over a decade ago.

Zoom In Zoom Out Reset image size

The second issue to note is the spur of high-noise data seen at relatively bright magnitudes in each passband, indicated in gray in Figure 9. These data points are correlated in time, uniquely associated with a range of dates between 2010 December 26 and 2011 December 31. They do not consistently correspond to any range in photometric quality, either in cloud value or residual in calibration, as is seen in Figure 6. Maintenance logs over the same period indicate a herring-bone noise pattern in the biases that was later attributed to a failing (and eventually failed) power supply card in the camera controller, which was subsequently replaced. Measures of the read noise in that time frame were consistently an order of magnitude larger than at other more nominal times. These noisier data do not have a substantial effect on the determination photometric calibration coefficients. They do, however, add to the overall measurement error budget.

Throughout this analysis there were concerns over possible errors or failures of the remote observatory and the degradation of the facility over the long refurbishment, as budgets limited servicing and regular maintenance schedules. This calibration program was implemented, in part, to monitor the working of the observatory, and the results indeed show larger intrinsic scatter than one might expect.

The largest source of systematic error is likely attributable to the RCT dome. The RCT has the same dome that was installed in 1965 for the Remote Controlled Telescope, which is just wide enough to accommodate the full entrance pupil of the telescope. Slight errors in dome azimuth encoder positions partially occult the opening angle and dim target sources. With few redundant monitors, a problem with the encoders went largely unnoticed until observations in late 2011 of a very bright target produced secondary, out-of-focus reflection images off the dewar window, which were clearly knife-edged. The problem was later identified as an intermittent failure of the PMAC card, specifically on the channel that controlled and monitored the dome encoder. The card has since been replaced. While it is reasonable that dome occultation could have been responsible for the fainter loci of residuals seen in Figure 6, it remains unclear how many images were previously affected. It is possible, however, that the amount of occultation could have been consistent at repeated azimuthal positions.

There are other systematic concerns but at a much lower level. Two instrumental issues led to significant variations in PSF. The focus encoders have seasonal drifts that require manual assessment and correction, and at times this calibration was late. These few, slightly out of focus images were generally not a concern to this large aperture photometry. In addition, the mirror support cells were at times under-inflated because of a problem with the air compressor and the bladder system that also contributed to an unusual stellar shape. A comparison of the image quality under these conditions to those under more nominal conditions is shown in Figure 10. Clearly, PSF photometry would not be possible with these complex issues in optical quality, which is why large aperture photometry was used in this analysis. Tests on the growth curves, exemplified in Figure 4, show the selected apertures were large enough to enclose approximately 99% of the source light, even with these variations in PSF. The standard star targets were not in crowded fields, and source blending was not an issue. The variations in PSF due to seeing at the site was also not a factor. The issue with the mirror support system has been addressed, but the focus still requires seasonal human intervention.

Zoom In Zoom Out Reset image size

Lastly, there are concerns over the loss of reflectivity (or additional scattering) due to dust settling on the telescope optics and the oxidation of the primary and secondary cells. Measures of reflectivity and scattering following a cleaning in 2009, and again in 2012, show as much as a 10% loss of signal over the same period, affecting bluer wavelengths more than red. Empirically, there does not appear to be an appreciable trend (to >0.1 mag) over the period from late 2011 to the end of 2012, when the majority of data for this analysis was obtained. There is a clear decline, however, from 2012 to 2013, and no data from that period are used in our final calibration analysis.