Identifying high-redshift GRBs with RATIR

Gamma-ray bursts (GRBs) (Gehrels et al., 2009; Mészáros, 2006) are bright objects that emit across the entire electromagnetic spectrum and therefore can be seen to high-redshift (Lamb & Reichart, 2000). This allows observers to identify the locations of high-redshift, faint, host galaxy population otherwise missed by the current and future magnitude limited surveys (Tanvir et al., 2012). Such discoveries will allow us to unveil the properties of early star formation (up to $z\sim 10$) and the epoch of reionization.

To date a number of high-redshift GRBs (all in the redshift interval $6<z<10$) have been identified including GRB 050904 (Totani et al., 2006), GRB 080913 (Greiner et al., 2009a), GRB 090423 (Tanvir et al., 2009) and GRB 090429B (Cucchiara et al., 2011). However, it is only in a more recent example, GRB 130606A at $z=5.913$ (Chornock et al., 2013), that high signal to noise ratio spectrum has been obtained of a high-redshift GRB afterglow. Using a method similar to that employed for quasars (Fan et al., 2006; Becker et al., 2001), the detection of Ly$\beta$ and Ly$\gamma$ transmission in this spectrum allowed Chornock et al. (2013) to identify that the intergalactic medium (IGM) is mostly ionized at this redshift, and that the epoch of reionization must have occurred earlier in the history of the Universe.

Castro-Tirado et al. (2013) also present broadband photometric and spectroscopic observations of GRB 130606A, finding a lower HI column density of $N_{\rm HI}=7.1\times 10^{19}$ cm^-2. Totani et al. (2013) consider a possible intervening damped Lyman-$\alpha$ (DLA) system at $z_{\rm DLA}=5.806$ as a contributor to the absorption via HI gas. Considerations of the required silicon abundance in such a DLA system lead Totani et al. (2013) to instead attribute the residuals in their host-only model to diffuse IGM along the line of sight at $5.83<z_{\rm IGM}<5.91$.

To obtain high signal-to-noise ratio spectra of potential high-redshift GRB, afterglows must first be rapidly identified. The primary method used to do so is the measurement of photometric redshifts from optical “dropout” candidates. As radiation travels through the Universe, clouds of neutral hydrogen attenuate the transmitted radiation (Madau, 1995). This effect occurs along the line of sight between source and observer, and so this attenuation encroaches into increasingly redder parts of the observed optical and near infrared (NIR) spectrum for sources (and therefore atomic hydrogen clouds) at higher redshifts.

In practice, few facilities can provide the necessary photometry to identify high-redshift candidate as most automatic, robotic telescopes use optical filters and as such are not sensitive to high-redshift GRBs. In some instances, where large aperture ground-based facilities suffer observational constraints, spectral information may not be obtained prior to the GRB afterglow fading below instrumental sensitivity limits, such as for GRB 090429B (Cucchiara et al., 2011), providing further motivation to obtain photometric redshift, $z_{\rm phot}$, estimates for high-redshift dropout candidates. Whilst $z_{\rm phot}=9.38_{-0.32}^{+0.14}$ (90% likelihood range) was later obtained, additional rapid-response from GRB-dedicated NIR multiple filter instruments would have assisted in providing a measure of $z_{\rm phot}$ and allowing large aperture telescopes to begin observations at the earliest possible epoch after the initial trigger.

The variable $z_{\rm phot}$ is estimated by fitting templates to the optical and NIR spectral energy distributions (SEDs; see for example Krühler et al. 2011 and Curran et al. 2008) observed for the afterglow of the GRB, which has an intrinsic synchrotron spectrum (Granot & Sari, 2002; Sari et al., 1998). Through this template fitting method the redshift of a GRB, the host galaxy dust extinction and the optical/NIR spectral index of the GRB afterglow can be estimated. Fitting a host extinction law allows a study of the prevalence of dust in early galaxies along GRB sight lines (Zafar et al., 2011), whilst the local spectral index can be compared to that obtained in other energy ranges to determine key parameters of the underlying synchrotron spectrum (Perley et al., 2013).

To determine $z_{\rm phot}$, Krühler et al. (2011) make extensive use of data from the Gamma-Ray Burst Optical/Near-Infrared Detector (GROND; Greiner et al. 2008). GROND simultaneously images in seven filters: g’, r’, i’, z’, J, H and K_S, providing a broad optical and NIR SED. Such an SED was used to determine the $z_{\rm phot}=4.35\pm 0.15$ for GRB 080916C (Greiner et al., 2009b) and another indicated helped identify that GRB 080913 was a high redshift event ($z=6.695\pm 0.025$) and warranted further observations from large aperture facilities (Greiner et al., 2009a).

Another instrument capable of providing the photometric observations needed to determine $z_{\rm phot}$ is the Reionization And Transients Infra-Red (RATIR; Butler et al. 2012) camera, which is mounted on the 1.5-meter Harold L. Johnson telescope of the Observatorio Astronómico Nacional on Sierra San Pedro Mártir in Baja California, Mexico. This facility, which became fully operational in December 2012, conducts autonomous observations (Watson et al., 2012; Klein et al., 2012) of GRB triggers from the Swift satellite (Gehrels et al., 2004). Since RATIR obtains simultaneous photometry in r, i, Z, Y, J and H, it is an excellent instrument for estimating $z_{\rm phot}$ for high-redshift GRBs and as such allows us to optimize spectroscopic observations with highly oversubscribed large telescopes.

RATIR and GROND operate in similar manners, both with best responses of order a few minutes after the initial Swift trigger time (Butler et al., 2013c; Updike et al., 2010). The filters employed by both instruments are similar, but not identical. GROND has a broader spectral coverage, extending to both higher and lower wavelengths. The lower wavelength coverage aids in the identification of $2.3<z_{\rm phot}<4$, although RATIR is specifically designed to target objects in the high-redshift Universe. While GROND also has a K_S-band filter, extending coverage to higher redshifts, RATIR contains a Y-band filter that GROND lacks. With such functionality, RATIR is better able to provide more precise estimates of $z_{\rm phot}$ when a GRB occurs $7<z<8$. RATIR is also 100% time dedicated to GRB follow-up. Perhaps the most important distinction between the two instruments is in their locations. RATIR is situated at latitude $\phi=+31^{\circ}02^{\prime}43^{\prime\prime}$, while GROND is mounted on the 2.2-meter MPI/ESO telescope at La Silla Observatory, Chile, with a latitude $\phi=-29^{\circ}15^{\prime}15^{\prime\prime}$, meaning the both form a complimentary pair of instruments routinely accessing different parts of the sky.

This work describes the development of our template fitting routine. In §2 we describe the models employed to produce the templates fitted to the RATIR optical and NIR data. In §3 we describe the Swift and RATIR observations of GRB 130606A, before in §4 presenting the results from analysis of the early epoch RATIR data. We then extensively test the capabilities of our algorithm with simulated RATIR data in §5. Finally, we present our conclusions in §6

To measure $z_{\rm phot}$ we adopt a template fitting methodology similar to that of Krühler et al. (2011), who use combined data sets from GROND and both the Ultraviolet/Optical Telescope (UVOT; Roming et al. 2005) and X-ray Telescope (XRT; Burrows et al. 2005) on board the Swift satellite. Another similar methodology is that of Curran et al. (2008), although the algorithm we implement is significantly less computationally expensive. To this end we use models of the underlying physical spectrum of the source and the mechanisms by which this spectrum is altered before reaching an observer at Earth. These processes are dust absorption from the host galaxy, attenuation from the Intergalactic Medium (IGM) and the response of the RATIR filters to incoming photons.

Unlike Krühler et al. (2011) our algorithm does not consider UVOT photometry, as only approximately 26% of Swift GRBs having UVOT detections, compared to the $\sim$60% of bursts detected by ground-based facilities (Roming et al., 2009). Following Schady et al. (2012), we also favor the extinction laws of Pei (1992), instead of the more general “Drude” model (Liang & Li, 2009) used by Krühler et al. (2011). This was done in an attempt to limit the number of parameters being fitted as the “Drude” model has four parameters compared to the one ($A_{V}$) used in the standard Pei (1992) extinction laws.

The Burst Alert Telescope (BAT; Barthelmy et al. 2005) on board Swift triggered on GRB 130606A (Swift trigger 557589) at 21:04:39 UT on 2013 June 6^th (Ukwatta et al., 2013). BAT identified prompt structure at the trigger time and later, brighter $\gamma$-ray emission at approximately 150 seconds after the initial trigger time.

Swift/XRT detected an uncatalogued fading X-ray source, providing a more accurate position for ground-based follow-up with narrow field instruments.

The enhanced-XRT position for GRB 130606A was found to be ${\rm RA(J2000)}=16^{{\rm h}}37^{{\rm m}}35\fs 12$, ${\rm Dec(J2000)}=+29\arcdeg 47\arcmin 46\farcs 4$ with an uncertainty of $1\farcs 5$ (Osborne et al., 2013). This uncertainty is a radius which signifies 90% confidence. This position was calculated using the Swift/UVOT to astrometrically correct the positions of sources in the XRT field of view.

Chornock et al. (2013) observed the afterglow of GRB 130606A with both the Blue Channel spectrograph (Schmidt et al., 1989) on the Multiple Mirror Telescope (MMT) and the Gemini Multi-Object Spectrograph (GMOS; Hook et al. 2004) on the Gemini North telescope. These observations had midpoints of 7.68 and 13.1 hours after the initial BAT trigger, for the Blue Channel spectrograph and Gemini-N/GMOS respectively. From this rapid follow-up with large aperture facilities, Chornock et al. (2013) obtained high signal to noise ratio spectra allowing them to measure $z=5.913$.

RATIR first observed the field of GRB 130606A between 7.38 and 14.19 hours after the BAT trigger. Earlier observations were precluded by the GRB occurring during daylight hours at the Observatorio Astronómico Nacional. RATIR obtained 4.42 hours of exposure in the r- and i-bands and 1.85 hours in the Z-, Y-, J- and H-bands (Butler et al., 2013b). The observations for all six photometric bands from this first night are shown in Figure 1.

A second epoch of observations were taken the following night between 31.14 and 37.86 hours after the Swift trigger (Butler et al., 2013a). With 4.96 hours of exposure time in the r- and i-bands both yielding upper limits, showing that the source had faded in the i-band. The burst remained bright enough in each of the NIR filters to allow detections in all four bands with a total exposure time of 2.07 hours. In all four filters the burst had faded by more than two magnitudes, as shown in Table 1.

The RATIR camera consists of two optical and two infrared cameras (Butler et al., 2012), allowing for simultaneous image capture in four bands ($riZJ$ or $riYH$). Split filters immediately above the infra-red detectors allow for near-simultaneous images in six bands by dithering the source across the infra-red detector focal plane. We employ a dither pattern that allows for sequential exposure in $ZJ$ followed by $YH$. Dithering also allows for subtraction of the sky background and the detector dark signal at the source position. We capture 80s exposure frames in $ri$ and 67s exposure frames in $ZYJH$ due to additional overhead. We apply the same image reduction pipeline – with twilight flat division and bias subtraction routines written in python and using astrometry.net (Lang et al., 2010) for image alignment and swarp (Bertin, 2010) for image co-addition – to data taken from each camera.

Photometry is calculated by running sextractor (Bertin & Arnouts, 1996) over individual science frames and mosaics over a range of apertures between 2 and 30 pixels diameter. A weighted average of the flux in this set of apertures for all stars in a given field is then used to construct an annular point-spread-function (PSF). This PSF is then fitted to the annular flux values for each source to optimize signal-to-noise for point source photometry. We perform a field-photometric calibration for the $riZ$ bands using the Sloan Digital Sky Survey Date Release 9 (SDSS DR9;Ahn et al. 2012). The RATIR $riZ$ filters have been shown to be equivalent to the SDSS filters ($riz$) at the $\lesssim$ 3% level (Butler et al., in prep). We calibrate the $YJH$ bands relative to the Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006), employing the United Kingdom Infrared Telescope (UKIRT) Wide Field Camera (WFCAM; Hodgkin et al. 2009; Casali et al. 2007) to estimate $Y$ from $JH$ for field stars. Photometric residuals for these bands are also approximately 3%.

The spectroscopic observations for GRB 130606A allow us to assess the accuracy of our template fitting routine using this burst as a test case. The standard operation of our algorithm requires input RATIR photometry and an estimation of $\beta_{\rm X}$. The former is taken from the initial RATIR GRB Coordinates Network (GCN) circular, whilst the latter is obtained from the UK Swift Science Data Centre (UKSSDC) automated analysis data products²²2http://www.swift.ac.uk/xrt_spectra/ (Evans et al., 2009).

The algorithm then uses a broad 1-dimensional grid in redshift ($0\leqslant z_{\rm grid}\leqslant 12$) to find $z_{\rm phot}$. This value provides a rapid and robust estimate of $z_{\rm phot}$ that aids in determining whether follow-up with large aperture facilities is warranted.

We then use 2-dimensional grids to refine our estimate of $z_{\rm phot}$. Two such grids are used for each extinction law. The first fits $A_{V}$ and model normalization, $N_{0}$, at fixed points in $z_{\rm grid}$ $\beta_{\rm opt}$ parameter space over the range $0\leqslant z_{\rm grid}\leqslant 12$. By determining the region of this parameter space with the smallest total $\chi^{2}_{\rm eff}$, we then produce a second grid which focuses on the best fit solution with a higher resolution in both $z_{\rm grid}$ and $\beta_{\rm opt}$.

In practice, the results from the finer, 2-dimensional grid provide a more precise estimate of $z_{\rm phot}$. However, when coordinating follow-up from large aperture facilities, those produced from the 1-dimensional grid $z_{\rm grid}$ provide sufficiently robust estimates such that further observations can be requested at the earliest possible epoch after the initial Swift/BAT trigger.

For GRB 130606A we used RATIR photometry obtained between 7.38 and 7.79 hours after the Swift/BAT trigger time (Butler et al., 2013b). As the RATIR image reduction pipeline runs on data as it is available from the instrument, it was already possible at this epoch to identify GRB 130606A as an r-band dropout candidate. We also obtained $\beta_{\rm X}=0.86_{-0.13}^{+0.14}$ from the UKSSDC.

The best fit for each of the four extinction laws, the three standard templates from Pei (1992) and a fit where we assumed no host extinction, are detailed in Table 2. The corresponding templates are plotted with the RATIR photometry in Figure 2. It is important to note that we only include dust extinction from the host galaxy. For GRB 130606A, there is a possibility of an intervening system, with its own unknown dust content (Totani et al., 2013). If present, $z_{\rm DLA}=5.8$, meaning that we would be unable to distinguish between the dust in the host and any potential dust in the DLA. Even should the dust content of the DLA be high, our value of $z_{\rm phot}$ remains robust due to the small difference in redshift better then host galaxy and DLA.

Table 2 suggests that either the MW or no-dust solution best represents the data. Visual inspection of Figure 2 suggests the LMC and SMC models are of at least comparable quality, however, the magnitude compared to that measured by RATIR is an average across the wavelength coverage of each filter. For the LMC and SMC templates, the average values across the $J$- and $H$-bands are more discrepant than those of the MW or no dust models. When using a MW type extinction law, a large quantity of dust is favored, with $A_{V}=1.48$. Consequently, $z_{\rm phot}$ for the MW template is reduced, as a large quantity of dust contributes to the strong suppression of intrinsic GRB flux in the $r$- and $i$-bands.

Figure 3, which contains the probability maps over all allowed ranges of $z_{\rm grid}$, shows the best fit solutions are located in a narrow redshift range at $z_{\rm phot}\sim 6$ for all four extinction laws. The finer resolution probability maps in Figure 4 allow more detailed structure to be discerned. In the no dust model there are two regions of the parameter space where the prior weighted probability is maximized. The details of the local maximum of each are in Table 3. The best found solution has $\beta_{\rm opt}\sim\beta_{\rm X}$ and corresponds to $z_{\rm phot}=6.01$. This is in good agreement with the cruder estimate made using the 1-dimensional grid in redshift, which is $z_{\rm phot}=5.97_{-0.08}^{+0.16}$ for an SED with no dust extinction (see Table 2).

The LMC and SMC probability maps in Figure 4 contain similar structure. The first solution is at $\beta_{\rm opt}\sim\beta_{\rm X}$, as found when $A_{V}=0$. With the inclusion of a non-zero $A_{V}$ this solution is more extended in the $z_{\rm phot}$ axis. This is because at marginally lower redshifts a higher $A_{V}$ allows absorption from the IGM to contribute to the flux deficit seen at shorter wavelengths. This elongation in the $z_{\rm phot}$ axis is more pronounced in the lower $\beta_{\rm opt}$ solution seen for both the LMC and SMC extinction laws. Some values of $\beta_{\rm opt}$ do not exactly coincide with the expected values resulting from a cooling break. These result from an enhancement in the prior probability distribution between $\beta_{\rm opt}=\beta_{\rm X}-0.5$ and $\beta_{\rm opt}=0.5$ where the slow and fast cooling break solutions overlap.

As with the $A_{V}=0$ extinction law, the $\beta_{\rm opt}\sim\beta_{\rm X}$ is the better of the two alternative solutions, and indeed a fitted value of $A_{V}=0$ is retrieved for both the LMC and SMC extinction laws.

The top right panel of Figure 4 shows the same parameter space using the MW extinction law. In this instance there are four local maxima in the prior weighted probability map. The template SED for each is shown in Figure 5. The two higher $z_{\rm phot}$ solutions are equivalent to those found when $A_{V}=0$. There are two additional solutions, however, at lower redshift. It is the shape of the MW extinction law that allows for these two solutions, as the 2175 Å bump can contribute to the suppression of low wavelength emission. Even with the contribution from this feature, the required $A_{V}$ is high, with the $\beta_{\rm opt}\sim\beta_{\rm X}$ solution requiring $A_{V}=0.73$ and the lower $\beta_{\rm opt}$ solution requiring $A_{V}=1.46$.

An additional point to note is the selection of the best fit from the four possible MW solutions. Table 3 shows that $\chi^{2}$ is lowest for solution 2 ($z_{\rm phot}=5.69$, $A_{V}=0.73$, $\beta_{\rm opt}=0.86$), however, when weighting by the $\beta_{\rm opt}$ prior probability distribution, the effective $\chi^{2}$, $\chi^{2}_{\rm eff}$ as defined in Equation 7, is lower for solution 1 ($z_{\rm phot}=5.62$, $A_{V}=1.46$, $\beta_{\rm opt}=0.42$). When comparing the multiple solutions for each extinction law to the obtained spectroscopic redshift of Chornock et al. (2013), we find a solution with $\beta_{\rm opt}\sim\beta_{\rm X}$ is most consistent with the reported value of $z_{\rm spec}=5.913$ in all cases.

As an independent test to determine which of the dust extinction laws best represents the data, we used the generic three component extinction law measured in Massa & Fitzpatrick (1986) and Fitzpatrick & Massa (1990, 1988, 1986). Applying the prior probability distribution detailed in Reichart (2001) yielded a fit where $z_{\rm phot}=6.01_{-0.40}^{+0.13}$, $\beta_{\rm opt}=0.88$, $R_{V}=3.21$ and $A_{V}=0.00$. Low dust extinction disfavors solutions 1 and 2 found using a MW type extinction law (see Table 3), which both had the most discrepant values of $z_{\rm phot}$ when compared to $z_{\rm spec}$ as reported in Chornock et al. (2013).

We constructed simulated data to be processed with our algorithm in order to test the regions of the parameter space in which robust values of $z_{\rm phot}$ could be recovered from RATIR photometry. Specifically, we wanted to establish the range of GRB redshifts for which we could estimate $z_{\rm phot}$, demonstrate whether the algorithm could correctly recognize the input extinction law and consider the effects of a high amount of dust extinction.

We have presented a template fitting algorithm used to determine photometric redshifts from RATIR data when a dropout candidate is present. This algorithm represents the intrinsic GRB spectrum with a physically motivated synchrotron model, includes dust extinction from the GRB host galaxy and absorption from intervening clouds of neutral hydrogen along the line of sight. Each fitted SED therefore provides estimates of $z_{\rm phot}$, $\beta_{\rm opt}$ and $A_{V}$.

Applying this algorithm to the first optical dropout candidate observed by RATIR, GRB 130606A, we successfully recover a value of $5.6<z_{\rm phot}<6.0$ (this is the range of best fit solutions), with the exact solution being dependent on the extinction law applied. This agrees well with the spectroscopic redshift obtained by Chornock et al. (2013) of $z=5.913$. Furthermore, we demonstrate that using the prior probability distribution from Reichart (2001) for $\beta_{\rm opt}$ can help discern between the multiple potential solutions, as demonstrated particularly for the MW type extinction law.

We also present the typical output of the algorithm, including SEDs and probability maps of the parameter space for each extinction model. Analysis of the finely gridded probability maps, which focus on the most probably region of $z_{\rm phot}$ $\beta_{\rm opt}$ parameter space, shows that for an LMC or SMC type extinction law the favored model is of a host containing negligible quantities of dust. The degeneracy between several model solutions with a MW extinction law is partially lifted by using the Reichart (2001) prior probability distribution, which indicates a low value for $A_{V}$ best fits the data. With this prior probability ruling out high $A_{V}$ solutions, the best fit MW solution becomes one with negligible host dust content. With this solution, the three extinction laws all favor a near identical fit, which is consistent with a zero $A_{V}$ solution in both the obtained parameter values and $\chi^{2}$ fit statistic.

To ensure our algorithm is robust, we then create simulated RATIR photometry, typical of the quality expected from the first night of observations after the Swift/BAT trigger. We firstly show that when there is no dust extinction in the host galaxy, we can successfully recover $z_{\rm phot}$ using the template fitting, in the ranges $4\lesssim z_{\rm sim}\lesssim 7.5$ and $8.75\lesssim z_{\rm sim}\lesssim 9.5$, for a variety of $\beta_{\rm opt}$ values. When $7.5\lesssim z_{\rm sim}\lesssim 8.75$ we remain able to identify that $z_{\rm phot}$ is within this range, and is therefore a target of high interest to larger facilities.

Introducing a moderate quantity of dust extinction to our simulations allows us to draw some conclusions about the dust present in the host. MW type dust is the most easily identifiable, due to the prominent feature at 2175 Å. For weaker amounts of dust extinction it is difficult to differentiate between LMC and SMC extinction laws. However, at $A_{V}\approx 0.5$ the two can be distinguished from one another.

An interesting result from our tests using a MW extinction law and $A_{V}=0.5$ is that the strong 2175 Å feature in the extinction law allows for a resolved value of $z_{\rm phot}$ even when neutral hydrogen absorption occurs slightly below the RATIR wavelength coverage. With a large presence of MW type dust in a host galaxy our algorithm can determine $z_{\rm phot}$ to an improved lower limit of $z_{\rm phot}\sim 3$.

We also considered simulated data containing strong host dust extinction. When simulating RATIR photometry with an SMC dust template, our algorithm was unable to resolve a solution for most values of $z_{\rm sim}$. This is due to the large amount of suppression of low wavelength emission from the intrinsic GRB spectrum by the dust population within the host galaxy. For both the MW and LMC extinction laws precise values of $z_{\rm phot}$ were obtained for $z_{\rm sim}\lesssim 7.5$. After this point the accuracy of $z_{\rm phot}$ reduces, although we remained able to robustly state that the GRB was of high redshift. The associated 3$\sigma$ errors were higher with the MW extinction law due to the greater prominence of the 2175 Å feature in this type of extinction law. This allows slightly lower redshift solutions to be found where the 2175 Å bump contributes to the suppression of lower wavelength emission.

We also note that low mass stars in the Milky Way provide a population of interlopers that may occur near a Swift/XRT GRB position. To quantify the probability that such a source is present in RATIR observations we considered thick and thin disk components of the Milky Way with exponential vertical density profiles (Jurić et al., 2008). We determined the local density of M-stars within a radius of 20 pc using the Hipparcos Main Catalog (Perryman et al., 1997). This volume was chosen to simultaneously maximize completeness and the total number of sources. We cross referenced these sources with the 2MASS All-Sky Catalog of Point Sources (Cutri et al., 2003), using the cross-match service provided by CDS, Strasbourg, to obtain their J band magnitude. We found the chance probability that an M-star with J$<21$ will be present in a 300 square arcsecond region centered around the Swift/XRT position is less than 2.3 %. This corresponds to one (or fewer) chance M-star in observations of approximately 43 different fields of view. Such events can be classified as non-GRB by their lack of temporal variability and the blackbody shape of the SED at low wavelengths.

Since December 2012, RATIR has observed 56 GRB fields of view, with one instance of an M-star interloper. RATIR observations of the field of GRB 131127A found a red source near the Swift/XRT position, suggesting a high-redshift candidate ideal for follow-up (Butler et al., 2013d). This source was observed at later epochs, revealing no significant evidence for fading (Butler et al., 2013e; Im et al., 2013), leading to the conclusion that this source was not a GRB.

The presented methodology is aimed specifically at identifying potential high-redshift GRBs, and providing a preliminary estimate of redshift. With $z_{\rm phot}$ in hand, the justification of triggering larger spectroscopic facilities to measure a highly resolved spectroscopic redshift is significantly higher. With the automated capabilities of RATIR it is therefore possible to request observations from such large facilities at early epochs after the initial Swift trigger time, thereby obtaining high signal-to-noise ratio spectra required to answer fundamental questions about the high-redshift Universe.