Automatic analysis of Swift-XRT data
aa r X i v : . [ a s t r o - ph ] N ov Astronomical Data Analysis Software and Systems XVII
P5.3
ASP Conference Series, Vol. XXX, 2008J. Lewis, R. Argyle, P. Bunclarck, D. Evans, and E. Gonzales-Solares, eds.
Automatic analysis of Swift-XRT data
P. A. Evans, L. G. Tyler, A. P. Beardmore, J. P. Osborne
Department of Physics and Astronomy, University of Leicester,Leicester, LE1 7RH, UK
Abstract.
The
Swift spacecraft detects and autonomously observes ∼ ∼
96% of which are detected by the X-raytelescope (XRT). GRBs are accompanied by optical transients and the field ofground-based follow-up of GRBs has expanded significantly over the last fewyears, with rapid response instruments capable of responding to
Swift triggerson timescales of minutes. To make the most efficient use of limited telescopetime, follow-up astronomers need accurate positions of GRBs as soon as possi-ble after the trigger. Additionally, information such as the X-ray light curve,is of interest when considering observing strategy. The
Swift team at LeicesterUniversity have developed techniques to improve the accuracy of the GRB po-sitions available from the XRT, and to produce science-grade X-ray light curvesof GRBs. These techniques are fully automated, and are executed as soon asdata are available.
1. Introduction
The
Swift satellite triggers on ∼
100 Gamma Ray Bursts (GRBs) per year, andthe X-ray telescope (XRT, Burrows et al. 2005) provides localisations accurateto ∼ >
90% of these. However, GRBs, and their optical counterparts,fade rapidly and ground-based observers have limited telescope time available touse for observations. It is thus desirable to produce precise, accurate positionsrapidly. Furthermore, a key indication of the scientific interest of a GRB comesfrom the
Swift -XRT light curve. These are non-trivial to produce correctly, butit is desirable to generate them accurately and rapidly.We describe how these two challenges are being met by the
Swift team atLeicester, providing automatic data analysis which gives results of publication-grade quality. While our work is specific to
Swift , rapid response time-domainastrophysics is a fast-growing field, and this type of software will be increasinglyuseful to astronomers.
2. Positions
To locate sources on the XRT detector we first perform a cell-detect search. Wethen centroid accurately on these sources using the point spread function (PSF)-fitting technique described by Cash (1978). Since the XRT’s PSF is known, ourfit has just two interesting free parameters, the x and y position of the object.Further, the presence of hot pixels has a very minor effect on the fit and willbe correctly accounted for in the uncertainty. Following a micrometroid impact1 Evans et al.
RA: 320.88357 DEC: −53.02672 ERR: 1.9" (90%)GRB060614 D E C RA s s h m s s s " −53 o ’ " " " " Figure 1.
Left : An image of GRB 070429A, generated from SPER data.The red circle shows the automatically determined GRB position. Despitethe bad columns, the centroid is accurate.
Right : The UVOT-enhanced XRT position of GRB 060614. The image isthe UVOT V -band image. The magenta circle shows the final position. Thegreen circles are the individual positions which contributed to this, and theyellow circle was rejected as an outlier. in mid-2005, there area several bad columns on the XRT detector, which arepermanently masked out. The location of these is known, and the fitted PSF isadjusted to account for this, giving accurate positions even when the object liesright over the bad columns (Fig. 1). When
Swift detects a GRB, the XRT takes up to three short ( < The above positions are subject to the uncertainty in the spacecraft boresight,derived from the on-board star tracker, which is approximately 3.5”. We havedeveloped a technique to remove much of this uncertainty.In addition to the XRT,
Swift contains an ultra-violet and optical telescope(UVOT, Roming et al. 2005) which takes data simultaneously with the XRT.We have deduced the transformation from a position on the XRT detector to utomatic data analysis V filter is in use), thus for any sourcedetected in XRT we can determine where it would appear on the UVOT. Wethen use the standard Swift software to translate this into an initial sky position,match serendipitous sources in the UVOT field of view with the USNO-B1 cat-alogue to determine an aspect correction, and apply this correction to find thetrue sky position. This removes the spacecraft boresight from the loop entirely,the position accuracy being limited by the accuracy of the XRT to UVOT trans-formation and the accuracy of the aspect solution.
Swift usually takes multipleobservations of GRB, so we have have multiple datasets on which the abovetechnique can be applied. We can then take the weighted mean of the positionsthus produced which reduces the uncertainty arising from the aspect solution.Any outliers are detected and removed, and the weighted mean is recalculated.Fig. 1 shows an example GRB with the weighted mean position, the individualpositions which contributed to this, and an outlier. The 90% error radii of thesefinal ‘UVOT-enhanced XRT positions’ are < .
9” 50% the time, a factor of 2reduction compared to the normal, unenhanced positions.This process is fully automated, and runs as soon as
Swift
3. Light curves
The standard approaches to light curve creation assume essentially uniform eventdata and produce uniform bin sizes, however for GRBs, whose essential be-haviour is to fade, this is not appropriate. Further,
Swift -XRT data are compli-cated by the dead columns on the CCD, the XRT’s innovative mode-switchingtechnology which allows the XRT to determine its operating mode based onsource brightness (which changes), and pile-up – where multiple photons arriveon the same XRT pixel during one readout cycle, and are thus recorded as onlya single photon. It is vital that these effects are properly accounted for: opticalastronomers, who must decide whether to invest their limited telescope timeon a given GRB, must have confidence in light curve features such as gradi-ent changes (‘breaks’) or flares, which may guide their decision on whether toobserve a given burst.To this end, we have developed a fully automated script which runs everytime new
Swift data of GRBs arrive at the UKSSDC, approximately every ninetyminutes. This software dynamically varies the source extraction region and thedata-bin size based on the source count-rate (corrected for losses due to pile-up) It builds separate light curves for the different operating modes, and thencombines them. Where data from multiple modes overlap, it decides whetherboth data points are valid, or one is spurious (Fig. 2). If the instrument istoggling rapidly between modes, the data from both modes should be correct.However, if the instrument spends a considerable amount of time in one mode,but a datapoint from another mode spans this interval, the latter point will
Evans et al.
100 1000200 500 20000.010.1110100 C oun t R a t e ( . − k e V ) ( s − ) Time since BAT trigger (s) 0.20.5 C oun t R a t e ( . − k e V ) ( s − ) F r a c . E x p Time since BAT trigger (s)
Figure 2. Examples of readout-mode switching/
Left : GRB 060929, show-ing a giant flare. Where data from the two modes overlap the red data-pointaroiund 500 s is spurious, and will be rejected by our code (it is left in here forillustrative purposes).
Right : GRB 050315. Here, the readout mode toggledrapidly, and our software correctly decided that both datasets are both valid,will keep them.
4. Closing Remarks
The automated science-grade analysis of data is both achievable, and essentialin some fields. With the increasing popularity of time-domain astrophysics, andthe advent of rapid response hardware, the demand for such software will onlyincrease.