1: %\documentclass[traditabstract,referee]{aa}  

2: %\documentclass[traditabstract]{aa}  

3: \documentclass[oldversion]{aa}  

4: 

5: \pdfoutput=1

6: \usepackage{amsmath}

7: \usepackage{graphicx}

8: \usepackage{txfonts}

9: \usepackage{url}

10: \newcommand{\releasedate}{Feb 8, 2008}

11: \newcommand{\nspectra}{2470}

12: \newcommand{\ntargets}{1923}

13: \newcommand{\nassociations}{520}

14: \usepackage[authoryear]{natbib}

15: 

16: \usepackage{color}

17: \newcommand{\new}[1]{{#1}}

18: 

19: \begin{document}

20: 

21: \title{The Hubble Legacy Archive NICMOS Grism Data}

22: 

23: 

24: \titlerunning{HLA NICMOS Grism Data}

25: 

26: \author{Wolfram Freudling, 

27: Martin K\"ummel, 

28: Jonas Haase, 

29: Richard Hook, 

30: Harald Kuntschner, 

31: Marco Lombardi,

32: Alberto Micol, 

33: Felix Stoehr, 

34: \and

35: Jeremy Walsh}

36: \authorrunning{Freudling et al.}

37: 

38: \institute{Space Telescope -- European  Coordinating Facility}

39: \date{submitted May 29, 2008; accepted Sep 3, 2008}

40: 

41: \abstract{ {The Hubble Legacy Archive (HLA) aims to create calibrated science

42: data from the Hubble Space Telescope archive and make them accessible via

43: user-friendly and Virtual Observatory (VO) compatible interfaces.  It is a

44: collaboration between the Space Telescope Science Institute (STScI), the

45: Canadian Astronomy Data Centre (CADC) and the Space Telescope - European

46: Coordinating Facility (ST-ECF). Data produced by the Hubble Space Telescope

47: (HST) instruments with slitless spectroscopy modes are among the most difficult

48: to extract and exploit. } {As part of the HLA project, the ST-ECF aims to

49: provide calibrated spectra for objects observed with these HST slitless modes.

50: } {In this paper, we present the HLA NICMOS G141 grism spectra.  We describe in

51: detail the calibration, data reduction and spectrum extraction methods used to

52: produce the extracted spectra.  The quality of the extracted spectra and

53: associated direct images is  demonstrated through comparison with near-IR

54: imaging catalogues and existing near-IR spectroscopy.} { The output data

55: products and their associated metadata are publicly available

56: (\url{http://hla.stecf.org/}) through a web form, as well as  a VO-compatible

57: interface that enables flexible querying of the archive of the \nspectra\

58: NICMOS G141 spectra. \new{In total, spectra of \ntargets\  unique targets

59: are included.}} \keywords{NICMOS -- infrared spectroscopy -- Hubble Space

60: Telescope, Calibration} }

61: 

62: \maketitle

63: 

64: 

65: \section{Introduction}

66: 

67: Three of the current Hubble Space Telescope instruments include built-in

68: slitless spectroscopic modes: the Space Telescope Imaging Spectrograph

69: (STIS), the Near Infrared Camera and Multi-Object Spectrometer (NICMOS), and the

70: Advanced Camera for Surveys (ACS).  The main advantage of slitless spectroscopy

71: is that spectra can in principle be obtained from all objects within the field

72: of view of an instrument. The main disadvantages are that the size of the

73: object limits the achievable spectral resolution, that spectra might overlap,

74: and that the background is relatively high because no slit mask prevents the

75: full sky background from illuminating every pixel of the detector. In addition,

76: extracting one-dimensional spectra from such data is a complex process and is

77: often achieved using highly interactive procedures.  

78: 

79: A fraction of the slitless data in the HST archive has been collected to obtain

80: spectra of specific objects, but spectra of objects other than the primary

81: targets have in most cases not been extracted or analysed.  The goal of the

82: ST-ECF HLA project is to extract spectra of all objects that have been

83: observed with HST slitless spectroscopy modes and to serve these spectra

84: through an archive with associated descriptions of the spectra, such as how

85: much one spectrum is contaminated by those of nearby objects. 

86: 

87: Each of the HST slitless spectrograph modes provides a different capability to

88: HST.  STIS and the ACS prisms cover the UV part of the spectrum that is not

89: accessible from the ground. The NICMOS grisms cover IR wavelengths that can

90: partially be observed from the ground, but where the background is much lower

91: in space. The ACS optical channels covering the optical wavelength range also

92: benefit from a combination of lower background and higher spatial resolution

93: from space. \new{The STIS grating provides the highest  resolving power among

94: the HST instruments with slitless spectroscopy modes. }

95: 

96: The data analysis for each of the spectrographs presents a significant

97: challenge, and specialised, mostly interactive tools are available for

98: individual instruments such as NICMOS \citep{nicmoslook}. For ACS a set of

99: non-interactive software tools, called aXe, has been developed which, on the

100: basis of a catalogue of the objects on the associated direct image, extracts

101: one and two-dimensional spectra \citep{axe1}. This software package was, for

102: example, extensively used for extracting ACS Wide Field Camera slitless spectra

103: in the Hubble Ultra Deep Field (HUDF) \citep{axe2}.  For the HLA project, we

104: have produced a pipeline that is designed to extract spectra from large numbers

105: of datasets in an unsupervised manner and can be tailored to particular

106: instruments.

107: 

108: We chose to start the slitless spectroscopy HLA project with the NICMOS

109: G141 grism dataset. After a brief description of the available data in the

110: archive (Sect.~\ref{sec:data}), the basic reduction steps applied to the NICMOS

111: images are described in Sect.~\ref{sec:images}, followed by a detailed

112: discussion of the spectrum extraction procedure (Sec~\ref{sec:extract}). In

113: Sec~\ref{sec:cal}, we show a new set of NICMOS spectrum calibration data.

114: 

115: Our extraction pipeline, PHLAG, is described in (Sect.~\ref{sec:phlag}).  The

116: extensive set of metadata that allows the archive of spectra to be served

117: through Virtual Observatory (VO) interfaces is summarised in

118: Sect.~\ref{sec:metadata}.  The calibrated spectra were subject to quality

119: assessment in terms of their astrometric and spectrophotometric properties and

120: shown by internal and external comparisons to be well-calibrated (see

121: Sect.~\ref{sec:qc}).  

122: 

123: \begin{figure}[t]

124: \includegraphics[scale=0.48]{grismexample}

125: \caption{Example of one matching pair of a F160W image (left, image n4k6j1a4q)

126: and the corresponding G141 grism image (right, image n4k6j1zyq)  from HLA

127: dataset N4K6IZZCQ }\label{fig:grismexample}

128: \end{figure}

129: 

130: \section{NICMOS Grism Data}\label{sec:data}

131: 

132: NICMOS, a second generation HST instrument, was installed in 1997. One of the

133: three NICMOS cameras, NIC3 includes three grisms for slitless spectroscopy. A

134: grism combines a grating and a prism in such a manner that it  produces a

135: dispersed spectrum at the position of the undispersed image when the grism is

136: removed. At the nominal grism wavelength, the image through a grism will be, to

137: first order, identical to an image without a dispersive element.

138: 

139: In NICMOS, the grisms are mounted on a filter wheel that can switch between any

140: of the filters and grisms. Since there are no slit masks, spectra of all

141: objects in the fields are produced. Depending on the location of the object,

142: the zeroth, first or the second order spectra may be visible. The mode of

143: operation is to first take an image with a filter that matches the bandpass of

144: the grism followed by one or more exposures with the grism. To avoid image

145: defects, a small or large scale dither pattern is used in a well designed

146: observing run. Fig.~\ref{fig:grismexample} shows a typical pair of exposures of

147: a filter image with a matched grism image.

148: 

149: The most widely used of the NICMOS grisms is G141, which covers the wavelength

150: range from 1.10 to 1.95$\,\mu$m. This wavelength range is not easily accessible

151: from the ground, and the dark background in space makes G141 a sensitive mode

152: to obtain spectra of all objects within the 51 $\times$ 51 arcsec$^2$ field of

153: view of the camera. The dispersion  for the first order spectrum is about 8.0

154: nm per pixel, which results in spectra about 105 pixels long. 

155: 

156: In March 2002, the NICMOS Cryocooler System (NCS) was installed. With the NCS,

157: the NICMOS detectors operate at a higher temperature and this changed the dark

158: current as well as the quantum efficiency as a function of wavelength for each

159: pixel. Therefore, all calibrations have to be derived separately for data taken

160: with and without the NCS \citep{ncscal}.

161: 

162: In July 2007, the HST archive contained  about 9000 NICMOS G141 observations.

163: \new{These data include both pointed observations towards specific targets and

164: programmes which used NICMOS in parallel to other HST instruments.  More than

165: 50\% of the NICMOS data presented in this paper are from four different HST

166: programmes,  namely HST proposal IDs 9707 \citep{john1}, 9865 \citep{mal1},

167: 8082 \citep{john2} and 10226 \citep{mal2}.} The pointing positions are almost

168: randomly distributed on the sky and are shown in Fig.~\ref{fig:sky}.

169: 

170: \begin{figure}[t]

171: \includegraphics[scale=0.4]{sky}

172: \caption{Distribution of NICMOS G141 grism images on the sky.}\label{fig:sky}

173: \end{figure}

174: 

175: 

176: 

177: \section{Image Processing and Target Selection}\label{sec:images}

178: 

179: \subsection{Associations}

180: 

181: Each spectrum in this work is based on a collection of images, which includes

182: both grism images and undispersed images.  The images in each of these {\em

183: associations} overlap and were treated collectively. The undispersed images

184: were co-added for the creation of target lists, and the two-dimensional spectra

185: on the grism images were rectified and co-added before the extraction of

186: one-dimensional spectra. These steps are described in detail in this section.

187: 

188: The grouping of NICMOS datasets into associations was performed using the World

189: Coordinate System (WCS) in the image header to compute relative pointings. To

190: ensure accurate relative coordinates among the images in any given association,

191: only datasets obtained with the same set of guide stars were included.

192: 

193: The spectral dispersion of NICMOS grism images is along the x-axis of the

194: detector.  The orientation of the grism images on the sky determines which of

195: the spectra overlap. We therefore chose to co-add only grism images that differ

196: in orientation by less than one degree, whereas the orientation of undispersed

197: images was not restricted. \new{The result of this procedure is that, for some

198: targets, several spectra are extracted and not co-added. The reason for this

199: decision was that different orientations of the spectra result in different

200: contamination. Indeed, the very reason that many programmes have observed the

201: same field with different roll angles is that it is likely that at least one of

202: the spectra for each target is free of contamination. We leave it up to the

203: user to pick the best of the spectra for each target.   }

204: 

205: In total 962 NICMOS G141 associations were created from 9262 grism members (up

206: to June 2007).  The grouping efficiency is therefore more than 9.5 exposures

207: per association.

208: 

209: 

210: \subsection{Preprossessing of NICMOS Images}

211: 

212: To produce one-dimensional spectra of individual targets, we extracted the flux

213: and wavelength information from NICMOS grism images and associated undispersed

214: images that have been reduced with STScI's CALNICA reduction pipeline version

215: 4.1.1 \citep{nicmoshandbook}.  NICMOS supports non-destructive multiple

216: readouts, and virtually all NICMOS data are taken in this mode.  The STSDAS

217: program CALNICA \citep{calnica} produces calibrated images from these raw data,

218: the processing steps include bias correction, dark current subtraction, and

219: computation of the count rate for each pixel using all readouts. The NICMOS

220: images suffered from a number of peculiar anomalies, some of them are treated

221: by CALNICA, for others separate STSDAS tasks are available.  The default

222: parameters of CALNICA and some of the algorithms used in the auxilliary task

223: are optimised for use with undispersed images. In the rest of the section, we

224: describe the parameters and procedures we used for the NICMOS Grism HLA.

225: 

226: \subsubsection{Bars Correction}

227: 

228: Some NICMOS images suffer from pairs of bright and dark columns or rows,

229: so-called ``bars''.  \new{These bars are bias-related artefacts in the form of

230: noiseless offsets} of a few DNs along a pair of columns or rows, with the

231: pattern replicated exactly in all four image quadrants. CALNICA has a built-in

232: procedure to detect and remove such bars.  Unfortunately, the central two rows

233: of NICMOS spectra contain more than 80\% of the flux of point sources and the

234: spectra are therefore of similar width to the bars. As a result, in crowded

235: grism images when the spectra are well aligned with the rows of the detector,

236: CALNICA often mistakes the peak of the spectra for bars and tries to remove

237: them.  To avoid this problem, a value of 4$\sigma$ was used for the threshold

238: in CALNICA which controls the bar detection. This compares to a default value

239: of 3$\sigma$.  It was found that this selection removes the described problem

240: in virtually all cases.

241: 

242: 

243: \subsubsection{Bad Pixels}

244: 

245: Images reduced with CALNICA typically contain between 20 and 100 unflagged bad

246: pixels, which contain significantly higher flux than their neighbours. These

247: ``hot'' pixels can produce spurious emission lines in extracted spectra.  The

248: pixel spacing of the NICMOS NIC3 camera is 0.2$\,$arcsec which undersamples the

249: Point Spread Function (PSF).  Commonly used methods to distinguish between

250: point sources and hot or cold pixels, based on the sharpness of features on the

251: calibrated image, are therefore not very effective with NICMOS.  On the other

252: hand, the multiple readouts provide information that can be used to recognise

253: unreliable pixels. The following procedure was implemented to identify bad

254: pixels. For each pixel the accumulated count rate at each readout, which is one

255: of the outputs of CALNICA, was fitted by a line. A pixel was flagged as ``bad''

256: when either the slope of the line was different from zero by at least four

257: times the uncertainty of the slope, or the reduced $\chi^2$ of the fit was

258: greater than four.  Flagged pixels were ignored in the subsequent analysis.

259: 

260: \subsubsection{Cosmic Rays}

261: 

262: Because of the multiple readouts, the count rate of a pixel can be measured

263: even if that pixel is hit by a cosmic ray (CR). The extra charge deposited by

264: the CR between two readouts produces a sudden discontinuity in the count rate.

265: CALNICA includes a procedure to identify and remove such jumps from the data.

266: This procedure is effective for more than 90\% of all CR hits. \new{ CALNICA

267: version 4.1.1. sometimes fails and produces a bright pixel in the output

268: images, even though it correctly recognised that this pixel is affected by a

269: CR. A future release of CALNICA will fix this problem \citep{newcalnica} but is

270: currently not yet available. } Because of the undersampling of the NICMOS

271: images, distinguishing between stars and cosmic rays cannot be easily done on

272: the basis of the sharpness of a flux peak alone. We therefore adopted a

273: procedure to use both the charge built up and the sharpness to minimize the

274: number of false detections of CR hits.  Each pixel identified by CALNICA as

275: affected by CR was checked to determine whether this pixel had an unusually

276: high flux in the output images.  Specifically, such a pixel was flagged as bad

277: if its flux was more than 4$\sigma$ above the mean of its neighbouring pixels.

278: 

279: 

280: \subsubsection{Pedestal}

281: 

282: NICMOS images suffer from random DC offsets in the bias level, which may be

283: different in each of the four quadrants. The offsets are constant within each

284: quadrant. If not removed, these offsets imprint the structure of the flatfields

285: on the calibrated images during the flatfielding step. Since grism images are

286: not flatfielded as images, the pedestal is visible as different constant

287: offsets of the  quadrants. There are two STSDAS tasks, pedsky and pedsub,

288: available to remove the pedestal from the calibrated images.  Neither of them

289: works well with grism images. We therefore used the following approach to

290: improve the pedestal in the grism images. First, for each pixel of a quadrant

291: we normalised the count rates of the multiple readouts to the mean of the last

292: three reads. We then computed the median count rate for each readout of all the

293: pixels in a quadrant. We then computed the value of the DC offsets for each

294: quadrant that results in constant median count rate. These values were then

295: subtracted from all the pixels in the corresponding quadrant.

296: 

297: 

298: \subsection{Co-addition of Undispersed Images}\label{sec:coadd}

299: 

300: In order to create a deeper and cleaner direct image, to be used to prepare the

301: input object catalogue needed as input to the spectral extraction, the direct

302: images in each association were first combined.  The MultiDrizzle software

303: \citep{drizzle,Fruchter} was used to register each input image, based on its

304: header WCS, and to remove the small NICMOS geometric distortion using the

305: standard cubic polynomial coefficient set. Standard default settings were used.

306: The data quality arrays were also aligned by this step and registered output

307: weight images created. MultiDrizzle is not yet well optimised for NICMOS and

308: experiments revealed that a combination of inadequately precise alignment and

309: the strong undersampling of the NIC3 camera resulted in the cores of compact

310: objects being incorrectly flagged as cosmic rays during the ``driz\_cr''

311: MultiDrizzle step. This led to significant photometric errors. To avoid this

312: \new{ we used MultiDrizzle only up to the image combination step. } The

313: preparation of image bad pixel masks and the final image combination were

314: performed using the {\it imcombine} task in IRAF. 

315: 

316: 

317: \subsection{Target Catalogue}

318: \label{sec:undisp-image-catal}

319: 

320: The first step in the extraction of slitless spectra is to find objects  on the

321: undispersed image.  Object parameters relevant for the extraction are the

322: coordinates, the position angle, the object sizes, and the magnitude in a

323: reference spectral band.

324: 

325: \begin{table*}

326:   \caption{Relevant parameters used for the source extraction in the direct images.}

327:   \centering

328:   \begin{tabular}{lcl}

329:   \hline

330:     Parameter & Value & Description \\

331:     \hline

332:     \texttt{DETECT\_MINAREA} & $4$ & Minimum number of pixels above

333:     threshold \\

334:     \texttt{DETECT\_THRESH} & $3.0$ & Minimum threshold for detection

335:     \\

336:     \texttt{ANALYSIS\_THRESH} & $3.0$ & Minimum threshold for analysis

337:     \\

338:     \texttt{DEBLEND\_NTHRESH} & $8$ & Number of deblending sub-thresholds

339:     \\

340:     \texttt{DEBLEND\_MINCONT} & $0.03$ & Minimum contrast parameter for

341:     deblending \\

342:     \texttt{SEEING\_FWHM} & $0.26$ & Stellar FWHM in arcsec \\

343:     \texttt{BACK\_SIZE} & $64$ & Background mesh size \\

344:     \texttt{BACK\_FILTERSIZE} & $1$ & Background filter size \\

345:   \hline

346:   \end{tabular}

347:   \label{tab:sex}

348: \end{table*}

349: 

350: We used the SExtractor program \citep{ba} to generate the object catalogues.

351: The output exposure time image generated in the co-addition step was used as a

352: weight map, which significantly  reduces the number of spurious detections and

353: improves the accuracy of the photometry.

354: 

355: The relevant parameters used for the extraction are given in

356: Tab.~\ref{tab:sex}.  We used the ``windowed'' centroid (\texttt{XWIN},

357: \texttt{YWIN}) as the target coordinates, which is the  iteratively computed

358: first moment of  the object's surface brightness after convolution with a

359: Gaussian function with size matched to the object.  This leads to very accurate

360: centroid measurement, especially for point sources (the large majority of

361: sources in our sample).  In addition, we also measured ``un-windowed''

362: morphological major axes \texttt{A}, \texttt{B}, position angle \texttt{THETA},

363: and  magnitudes (\texttt{MAG\_AUTO}.  Finally, we kept track of the success of

364: these measurements by retaining the SExtractor \texttt{FLAGS} parameters (see

365: below).

366: 

367: The original SExtractor catalogue was then further filtered in order to remove

368: spurious detections:

369: 

370: \begin{itemize}

371: 

372: \item We removed all objects with $\mathtt{FLAGS} \ge 64$: that effectively

373: removes all objects for which a severe memory overflow occurred either during

374: deblending or during the extraction;

375: 

376: \item We removed faint objects for which we do not expect to be able to extract

377: a spectrum from the dispersed images ($\mathtt{MAG\_AUTO} \ge 23.5$), or

378: extremely elongated ones ($\mathtt{A} / \mathtt{B} > 4$), which are often

379: simply detector artifacts or cosmic rays;

380: 

381: \item We took objects with sizes smaller than one pixel ($\sqrt{\mathtt{A}

382: \mathrm{B}} < 1 \mbox{ pixel}$) as point-like sources, and set both their axes

383: to $1 \mbox{ pixel}$;

384: 

385: \item For extended objects, we computed the extension and orientation of the

386: virtual slits from the parameters \texttt{A}, \texttt{B}, and \texttt{THETA}

387: (see Appendix~A);

388: 

389: \item We discarded associations that suffered from significant crowding ($>140

390: \mbox{ sources arcmin}^{-2}$) \new{since such a high source density leaves too

391: little  sky background to allow the estimation of the local background levels.}

392: 

393: 

394: \item We included external 2MASS sources in the final catalogue, as described

395: in Sect.~\ref{sec:extern-2mass-sourc}.

396: 

397: \end{itemize}

398: 

399: 

400: \subsection{Astrometric calibration}

401: \label{sec:astr-calibr}

402: 

403: The astrometric coordinate system  of the raw NICMOS images is specified in the

404: WCS keyswords in the image headers.  Its accuracy  is ultimately  limited by

405: the accuracy of the catalogue used for pointing the telescope. For earlier

406: data,   the Guide Star Catalog~I \citep{1990AJ.....99.2019L} was used, and

407: since 2000 the Guide Star Catalog II \citep{gsc2} is used.  Both catalogues are

408: based on scans of the photographic Sky Survey plates at various epochs and

409: bandpasses.  The absolute accuracy reaches $\sim 0.3''$ over a large fraction

410: of the sky, but  errors can be as high as several arcsecs towards the edges of

411: the scanned plates.  A comparison of the astrometry in undispersed NICMOS

412: images with more accurate data showed such large offsets in some cases.

413: 

414: Accurate coordinates are not only important for the final astrometric fidelity

415: of our spectra, but also to identify potentially contaminating objects outside

416: of the NICMOS images (see Sect.~\ref{sec:contam}). We therefore decided to

417: carry out an independent astrometric calibration of all our undispersed images.

418: For that purpose, we first generated  preliminary versions of the  catalogues

419: described in Sect.~\ref{sec:undisp-image-catal} with coordinates based on the

420: WCS and used them to compute offsets to the WCS.

421: 

422: 

423: \begin{table*}

424: \caption[]{Reference Catalogues used for Astrometric Calibration.}\label{tab:refcats}

425: \centering

426: \begin{tabular}{lccc}

427: \hline

428: Catalogue & Acronym & Reference & Accuracy \\

429: \hline

430: USNO CCD Astrograph Catalog& UCAC2 & \cite{2004AJ....127.3043Z} & 0.06

431: arcsec\\

432: Two Micron All Sky Survey& 2MASS  &\cite{2006AJ....131.1163S} & 0.10 arcsec\\

433: Sloan Digital Sky Survey& SDSS-DR5  &\cite{2007ApJS..172..634A} & 0.15

434: arcsec \\

435: USNO-B1.0 catalog & USN0& \cite{2003AJ....125..984M} & 0.20 arcsec\\

436: Guide Star Catalog II & GSC2.3.2 & \cite{gsc2} & 0.30 arcsec\\

437: \hline

438: \end{tabular}

439: \end{table*} 

440: 

441: The offset correction applied to the WCS is the mean offset of the WCS

442: catalogue coordinates and a global astrometric reference catalogue obtained by

443: merging the five catalogues listed in Tab.~\ref{tab:refcats}.  The merged

444: astrometric catalogue includes sources from all reference catalogues and

445: assigns to each object  the coordinates of  the most accurate catalogue in

446: which it was detected.  The mean offset was then calculated using a weighted

447: average, with weights proportional to the inverse of the square intrinsic

448: astrometric errors associated with each source.

449: 

450: The most critical part of the whole procedure is the matching of the sources in

451: the astrometric catalogue with our object catalogue.  This step is challenging

452: because of the small field of view of the NIC3 camera and the relatively large

453: number of artefacts present in the NICMOS images.  The basic approach of our

454: adopted procedure is to first use the WCS coordinates and identify the nearest

455: neighbours in the merged reference catalogues, compute offsets to the WCS, and

456: iterate. The scatter in the offsets of different stars is then used as an

457: estimate of the astrometric accuracy.

458: 

459: Our method, however, differs in a number of ways from the simple scheme

460: outlined above.  Firstly, the initial astrometric solution assumed is not the

461: original HST astrometry, but rather the HST astrometry corrected for the median

462: offset computed among all the pre- or post-NCS observations. This median offset

463: is $\sim 1.1''$ for pre-NCS observations, and $\sim 0.3''$ for the post-NCS

464: observations (see Fig.~\ref{fig:astro_offsets}).

465: 

466: Secondly, we used the \textit{median\/} offset between the astrometric

467: catalogue and the NICMOS object catalogue in the first two iterations, and a

468: $\sigma$-clipped weighted average for subsequent iterations.  This ensures that

469: the initial, highly uncertain offset does not introduce any strong bias because

470: of wrongly matched pairs.

471: 

472: Thirdly, if during any iteration we found more than three matched pairs, we

473: uses a clustering analysis to remove spurious matches.  Specifically, we

474: identified clusters in the offset plane $(\mathrm{d}x, \mathrm{d}y)$.  We then

475: removed from each cluster those matches that increase the standard deviation of

476: the offsets by more than a factor of two.  This technique is more robust in

477: removing badly matched pairs than $\sigma$-clipping.

478: 

479: \begin{figure}[t]

480: \includegraphics[width=\hsize]{fig03}

481: \caption{Cumulative distribution of number of astrometric calibration stars

482: used in each field.}\label{fig:21}

483: \end{figure}

484: 

485: \begin{figure}[t]

486: \includegraphics[width=\hsize]{fig05}

487: \caption{Difference between objects coordinates computed from  the WCS in the

488: image header and coordinates listed in astrometric reference catalogues. The

489: crosses mark the mean of the distributions before and after the installation of

490: the NCS, and the circles make the 1$\sigma$ scatter around the mean. The

491: systematic error in the WCS coordinates is significant larger for the pre-NCS

492: data.  }\label{fig:astro_offsets} 

493: \end{figure}

494: 

495: \begin{figure}[t]

496: \includegraphics[width=\hsize]{fig04}

497: \caption{ Cumulative distribution of estimated astrometric accuracies of the

498: object coordinates before and after the astrometric correction.

499: }\label{fig:22}

500: \end{figure}

501: 

502: Finally, in the case of the 2MASS catalogue, we also compared H-band magnitudes

503: measured on the NICMOS images with those in the 2MASS catalogue and removed bad

504: matches.

505: 

506: At the end of the whole process the algorithm returned the following quantities:

507: 

508: \begin{itemize}

509: 

510: \item the estimated offset;

511: 

512: \item the theoretical statistical error on the offset, derived from a simple

513: error propagation of the intrinsic catalogue accuracies listed in

514: Tab.~\ref{tab:refcats};

515: 

516: \item in cases where at least two objects are matched, the sample variance of

517: the offsets of the matched objects was also calculated;

518: 

519: \item finally, the full list of matches was retained for inclusion in the

520: metadata.

521: 

522: \end{itemize}

523: 

524: In total, about 96\% of our released spectra have corrected coordinates, and

525: for 91\% of these we were able to match two or more astrometric sources The

526: distribution of the number of matches is shown in Fig.~\ref{fig:21} The

527: computed offsets are shown in Fig.~\ref{fig:astro_offsets}.  In

528: Fig.~\ref{fig:22} we compare the  distributions of the estimated astrometric

529: uncertainty (i.e.\ the square root of the sample variance of offsets) before

530: and after applying these offsets. The median  accuracy of the astrometry

531: improves from about 1.24 arcsec to about 0.27 arcsec.

532: 

533: 

534: \subsection{External 2MASS sources}

535: \label{sec:extern-2mass-sourc}

536: 

537: NICMOS grism images can include partial spectra of bright objects that are

538: located outside of the field of view (FOV) but close enough to the edge of the

539: image so that the grism still deflects part of the first order spectrum onto

540: the detector.  The zeroth order spectra of such objects can be fully included

541: in the grism image without a corresponding counterpart on the undispersed

542: image.  Such partial spectra might contaminate other spectra, and the zeroth

543: order spectra might be mistaken for bright emission lines.  We used the 2MASS

544: catalogue to identify bright sources close to each grism image that might

545: create such spurious spectra.  After the astrometric correction described in

546: Sect.~\ref{sec:astr-calibr}, the coordinates of the images are known well

547: enough to predict the location of such spectra accurately enough to take them

548: into account for contamination estimation.  By visual inspection of  the images

549: we determined that the maximum distance of sources from the image edges that

550: still produces a spectrum is 13 arcsec at the left image edge, and 3 arcsec at

551: the right image edge.

552: 

553: Although it is in principle possible to extract spectra of the external

554: sources, in the current release we did not attempt to do so, but only used the

555: external sources for the estimate of the contamination see

556: Sect.~\ref{sec:contam}.

557: 

558: \section{Extraction of NICMOS Spectra}\label{sec:extract}

559: 

560: \subsection{Methods}

561: 

562: One-dimensional integrated \new{first order} spectra of objects were extracted

563: from grism data with the help of the target catalogues. \new{Whilst second

564: order spectra are visible on NICMOS grism images for bright targets, in most

565: cases only small parts of these spectra are usable because they are not

566: completely contained within the NICMOS detector and usually suffer from

567: significant contamination.} We therefore chose to limit our project to the

568: first order spectra.

569: 

570: \new{The essential extraction steps} were to remove the local background close

571: to the spectrum, use the position of the object on the undispersed image to

572: obtain a wavelength scale, and then add up the flux values for each wavelength

573: bin. The wavelength dependent response of  each pixel was taken into account

574: during this extraction.

575: 

576: We used a modified version of the extraction software aXe \citep{axe1} version

577: 3.7 for that purpose.  aXe was designed to treat slitless spectroscopic data

578: and includes the transfer of source positions derived on undispersed images

579: onto the slitless images; the application of wavelength dependent  flat fields,

580: the co-addition and extraction of spectra and the  estimation of the

581: contamination. We now describe the modifications and additions that were made

582: to the basic aXe package, most of them address NICMOS specific properties of

583: the data. 

584: 

585: \subsection{Tracing of spectra} \label{sec:spectraces}

586: 

587: Spectra on G141 grism images are closely aligned with the rows of the detector.

588: The angle between the dispersion direction  and the detector rows depends on

589: the exact orientation of the grisms.  The grism is mounted on a rotating filter

590: wheel, which lead to slight differences in the positioning of the grisms each

591: time it is rotated into place. Typical values for the angle between the

592: dispersion direction and the x-axis of the detector ranged from 0 to 2$^\circ$.

593: We therefore determined this angle separately for each grism observation.

594: 

595: To ensure an identical extraction around the variable traces for the entire

596: data set, the spectral trace was measured on each grism image individually.

597: This was done by selecting the brightest three point-like objects on each image

598: and determining the trace solution for each object individually. The individual

599: solutions were then combined using the signal-to-noise ratio of the spectrum as

600: weight to determine the single trace solution for that grism image. 

601: 

602: 

603: \subsection{Background Subtraction}

604: 

605: The NICMOS grism images are  not flatfielded in the standard pipeline, because

606: the local quantum efficiency is wavelength dependent. Spatial variations in the

607: throughput were therefore taken into account during the extraction process for

608: each target spectrum, after the wavelength calibration. An imprint of the

609: flatfield convolved with the spectrum of the sky background in the grism

610: passband is therefore visible in calibrated grism images. This background has

611: to be {\em subtracted} before the flux in each wavelength bin is summed. In

612: addition to the imprint of the flatfield, a region of enhanced dark current was

613: visible in the lower right-hand part of all NICMOS images. This enhanced dark

614: current, which varies depending on the positioning of the field offset mirror

615: (FOM), is also an additive component that has to be subtracted.

616: 

617: G141 background images were constructed from grism images that contained no or

618: few visible spectra, using about twenty different images for each prepared

619: background.  The median of the scaled images was subtracted from each

620: individual image to isolate the contribution of the FOM to the background.  In

621: this manner, two different background images were prepared for different time

622: periods, one presenting the flatfield imprint on the background, and the other

623: the enhanced dark current.  Scaled versions of these backgrounds were

624: subtracted from each grism image.

625: 

626: The scaling of the subtracted background was a crucial parameter for the

627: flatness of the resulting image and therefore the quality of the extracted

628: spectra. Because of the uncertainties in the overall bias level and dark

629: current and the changing sky brightness, the measured background level was not

630: a good predictor of the structure seen in the background. The scaling for the

631: subtraction was therefore determined by finding the least-squares solution for

632: a line fit to  the pixel values in the background versus the ones in the grism

633: images. Only pixels that do not contain any spectrum were used in this

634: procedure.

635: 

636: For most data sets, this procedure succeeded in removing most variations in the

637: background. To further improve the quality of the extraction, a local

638: background was subtracted  for each extracted spectrum. This local background

639: is a linear fit to the region around each spectrum. 

640: 

641: 

642: \subsection{Virtual Slits}

643: 

644: The slit in a longslit spectrograph selects a part of the sky, and light from

645: that area is then dispersed and binned in wavelength.  In slitless

646: spectrographs, the light that reaches the final wavelength bins is only limited

647: by the extent of the targets.  The light distribution of target objects

648: therefore  defines the lines of constant wavelength on the dispersed image, and

649: the effective resolving power. The size, position and orientation of the lines

650: of constant wavelength are used to extract one-dimensional spectra from the

651: two-dimensional image. 

652: 

653: Target  objects can be of  an arbitrary shape that may depend on wavelength.

654: For the extraction of spectra, we made two approximations.  Firstly, we

655: neglected any wavelength dependence of the object shape and derived all shape

656: parameters from the undispersed image. Secondly, we treated all objects as

657: ellipses, i.e.  assumed that the isophotes were elliptical.  The spectral

658: resolution of the spectra extracted with these approximation was slightly lower

659: than an optimal extraction that uses knowledge of the exact shape of objects as

660: a function of wavelength. For these approximations, we computed the extraction

661: direction that optimizes the spectral resolution of the spectra. The details

662: are given in Appendix~A.

663: 

664: 

665: \subsection{Adjustment of Wavelength Scale}

666: 

667: As a result of the filter wheel non-repeatability, the spectra on any given

668: exposure can be shifted by up to a few pixels relative to the expected

669: wavelength zero point derived from the undispersed image.  The wavelength

670: zeropoint was therefore re-adjusted for each exposure by extracting the

671: brightest few ($\sim$4) spectra and cross-correlating the characteristic

672: spectral feature (i.e. the drop of sensitivity) at the red end of all spectra

673: with a fiducial template. This template was prepared from the sensitivity

674: curves shown in Fig.~\ref{fig:throughput}. In this way any wavelength

675: zero-point shifts introduced by the non-repeatability of the positioning of the

676: grism wheel were taken out with a precision of about $\pm$0.2 pixel or $\pm$16

677: \AA. Potential wavelength calibration changes as a function of position within

678: the field of view were not taken into account.

679: 

680: \begin{table*}[t]

681: \begin{center}

682: \caption{Tabulated values for the linearity correction $b$}

683: \label{tab:nonlin}

684: \begin{tabular}{ccccccccccccc}

685: \hline

686: 

687: $\lambda [\mu]$ m&.825&.875&.925&.975& 1.1&1.2&1.3&1.4&1.5&1.6&1.7&1.8\\

688: b &.069&.057&.052&.050&.049&.048&.041&.023&.013&.008&.004& 0\\

689: \hline

690: \end{tabular}

691: \end{center}

692: \end{table*}

693: 

694: \subsection{Pixel Response Function}\label{sec:prf}

695: 

696: The NICMOS detectors show significant sensitivity variations across individual

697: pixels \citep{sto99,xu03}. These variations, together with the large pixel size

698: and the small trace angles (see Sec~\ref{sec:spectraces}), result in a

699: modulation of the extracted spectrum with a wave-like pattern for point-like

700: objects. The amplitude of this pattern is about 4\%. To correct for this effect

701: the exact location of the trace must be known to a fraction of the pixel size. 

702: 

703: The subpixel accuracy for the trace location  within a spectrum cannot be

704: achieved with the fits described in Sec~\ref{sec:spectraces}. A sensitive

705: estimator of the trace location is the quantity $f = v_{1}/ v_{2}$ with  the

706: brightest pixel value $v_{1}$ and the second brightest pixel value $v_{2}$,

707: respectively. For each point-like object, the values $f$ along the spectrum

708: were determined. The position where this function peaks marks the wavelength

709: value at which the spectrum is located at the centre of a row.  For raw spectra

710: with sufficient signal-to-noise to determine the peaks were multiplied by a

711: correction function~$f_{\rm pr}$:

712: 

713: \begin{equation}

714: f_{\rm pr} = 1 + 0.04  \sin (2 \pi  y) ,

715: \end{equation}

716: where $y$ is the position of the spectrum on the detector determined with

717: subpixel accuracy as described above.

718: 

719: 

720: \subsection{Non-Linearity Correction}

721: 

722: The NICMOS detector responds to different infalling photon count rates in a

723: non-linear manner. Even if the total number of photons is identical, the

724: detector will report a different flux if the number of photons per unit time

725: differs.  This effect was described  in \cite{bohlinI}. Following

726: \cite{bohlinII}, we corrected point-like objects for the detected photon count

727: rates $c$ according to

728: 

729: \begin{equation}

730: c = c_{\rm obs} / [ 1 - 2 b(\lambda)  + b(\lambda)  \log_{10}(c_{\rm obs})], 

731: \end{equation}

732: where $c_{\mathrm obs}$ is the count rate derived from the grism image, and

733: the values $b(\lambda)$ are given in Table \ref{tab:nonlin}.

734: 

735: 

736: \subsection{Flux Calibration} \label{sec:flux}

737: 

738: Flux calibration of point sources was done by multiplying the count rate of

739: extracted spectra by the sensitivity curve (see Fig.~\ref{fig:throughput})

740: derived from flux standards.  Flux standards are stars and thus point-like

741: sources, and the spectral resolution is determined by the instrument setup. For

742: extended objects, however, the spectral resolution is degraded by the extent of

743: the object in the dispersion direction.  This smoothing of the spectra reduces

744: the amplitude of structure in the sensitivity curve and has to be taken into

745: account for the flux calibration of extended objects.

746: 

747: The flux calibration of extended sources was carried out by multiplying the

748: count rate by a smoothed version of the point source sensitivity function.

749: The width of the Gaussian smoothing kernel was

750: \begin{equation}

751: \sigma = f r\sqrt{s^{2} - p^{2}}  , 

752: \end{equation}

753: where $s$ is the FWHM of the object in the dispersion direction, $p$ is the

754: FWHM of the point spread function, and $r$ is the dispersion in $\AA$ per

755: pixel. The quantity $f$ is a correction factor that was found empirically by

756: experiment using extended source spectra; the adopted value was 0.65.

757: 

758: The procedure was tested both on  simulations of Gaussian objects using the aXe

759: simulation package aXeSIM \citep{axesim}, and on real extracted NICMOS spectra.

760: The tests showed that, in most cases, the procedure successfully removed the

761: imprinted structure resulting from the changes in sensitivity with wavelength

762: that is visible in spectra calibrated with the nominal point source sensitivity

763: curve. 

764: 

765: 

766: \subsection{Contamination}

767: \label{sec:contam}

768: 

769: One of the disadvantages of slitless spectroscopy is that spectra of science

770: targets may overlap with the spectra of random objects in the field that are

771: not masked by any slit mechanism. Only isolated objects are not affected by

772: this sort of contamination. The level of contamination depends on the

773: separation and relative flux level of the objects. Spectra that are

774: contaminated may still be useful if the expected flux from the contaminator is

775: significantly smaller than the flux in the spectrum of interest. Estimating the

776: contamination level for all spectra is essential to exploit them

777: scientifically.

778: 

779: \new{Contamination levels were estimated by aXe assuming flat contaminating

780: spectra \citep{axe1}. We used this feature to estimate the contamination both

781: from objects in the source catalogue and from bright 2MASS sources (see

782: Sect.~\ref{sec:extern-2mass-sourc}) that were close enough to the target object

783: so that their spectra might overlap.  The contamination level was estimated in

784: the following manner: The shape of spectra in the cross dispersion direction

785: was assumed to be Gaussian with a dispersion $\sigma$ equal to the object size

786: listed in the target catalogue.  The spectrum of each potential contaminator

787: was assumed to be flat in $f_{\lambda}$ with an integrated magnitude identical

788: to the one measured on the undispersed image.  With these assumptions,

789: two-dimensional spectra for all potentially contaminating objects were

790: produced. Finally, a spectrum at the position of the target was extracted in

791: exactly the same manner as the spectrum from the original grism image.  The

792: resulting spectrum was then used as an estimate of the contaminating flux.}

793: 

794: \new{

795: \subsection{Co-addition of Spectra}

796: 

797: The final step in the extraction of spectra is to sum up all flux contributions

798: for a given wavelength bin. The approach used by aXe is to rectify the two

799: dimensional spectra onto a predefined grid using the drizzle approach

800: \citep{drizzle}. In }\new{ our modified version of aXe, each spectrum of a

801: target is first drizzled to the same grid. We then use the IRAF task {\it

802: imcombine} with 4$\sigma$ rejection to combine the two-dimensional spectra.

803: One-dimensional spectra are finally produced by summing up all pixels that

804: correspond to the same wavelength. }

805: 

806: 

807: \section{The Pipeline for the Hubble Legacy Archive Grism data

808: (PHLAG)}\label{sec:phlag}

809: 

810: The whole extraction procedure from the retrieval of the input data to the

811: extraction of the calibrated spectra was carried out in a single Python script,

812: called {\bf P}ipeline for the {\bf H}ubble {\bf L}egacy {\bf A}rchive {\bf

813: G}rism data (PHLAG).  PHLAG works with the NICMOS calibrated images and  calls

814: the necessary external packages in a series of processing steps.  The steps

815: carried out are as follows:

816: 

817: \begin{enumerate}

818: 

819: \item {\bf Data preparation:} The data are prepared for the pipeline reduction.

820: The direct images are grouped according to the filter.  Pairs, consisting of

821: one direct image and one slitless image with small positional offsets, are

822: composed.

823: 

824: \item {\bf Image combination:} The direct images are rectified and co-added as

825: described in Sect.~\ref{sec:coadd}.

826: 

827: \item {\bf Object detection:} The object catalogues are extracted from the

828: undispersed image (Sect.~\ref{sec:undisp-image-catal}).

829: 

830: \item {\bf Spectral extraction:} The extraction of one  and two-dimensional

831: spectra is performed using the aXe software package with the modifications

832: described in Sect.~\ref{sec:extract}.

833: 

834: \item {\bf Metadata:} The spectra are prepared for ingestion into the database.

835: The metadata are collected or derived (see Sect.~\ref{sec:metadata}).

836: Associated products such as stamps and cutout-images are created. 

837: 

838: 

839: \end{enumerate}

840: 

841: 

842: \section{Calibration of NICMOS Spectra}\label{sec:cal}

843: 

844: \subsection{Flatfield Cube}

845: 

846: The quantum efficiency of each pixel on the detector is a function of

847: wavelength, and this function is different for each pixel. For slitless

848: spectroscopic images, the usual step of correcting the quantum efficiency

849: variations by flatfielding has to be replaced by a wavelength  dependent

850: correction. Such a correction requires a three-dimensional flatfield cube with

851: detector coordinates and wavelength as the three axis.  

852: 

853: The flatfield cube was derived from the NICMOS narrow band flat fields,

854: separately for the pre-NCS and post-NCS periods.  Tab. \ref{Table_flat} lists

855: the narrow band flats that were used.  For each pixel, we fitted a 5th order

856: polynomial  to the value of the pixel in each flatfield.  Fig.~\ref{Fig_flat}

857: illustrates the zeroth order (wavelength independent) and first order

858: (wavelength dependent) terms of the flat field cube. The mean value of the rms

859: on the polynomial fit for the post-NCS flat field cube is 0.15\%.

860: 

861: \begin{table*}[t]

862: \begin{center}

863: \caption[]{Flat field cube fitting for Pre- and Post-NCS flat fields.}

864: \vskip 0.1cm

865: \label{Table_flat}

866: \begin{tabular}{lll} 

867: \hline

868: Filter & Pre-NCS Flat & Post-NCS Flat \\

869: \hline

870: F108N & i191346kn\_flt.fits & 3\_F108N\_STEP16\_1On-AllOff\_sflt.fits \\

871: F113N & i191346mn\_flt.fits & 3\_F113N\_STEP16\_1On-AllOff\_sflt.fits \\

872: F164N & i191346pn\_flt.fits & 3\_F164N\_STEP16\_1On-AllOff\_sflt.fits \\

873: F166N & i191346qn\_flt.fits & 3\_F166N\_STEP16\_1On-AllOff\_sflt.fits \\

874: F187N & i191346sn\_flt.fits & 3\_F187N\_STEP16\_1On-AllOff\_sflt.fits \\

875: F190N & i191346tn\_flt.fits & 3\_F190N\_STEP16\_1On-AllOff\_sflt.fits \\

876: F196N & i1913470n\_flt.fits & 3\_F196N\_STEP16\_1On-AllOff\_sflt.fits \\

877: F200N & i1913471n\_flt.fits & 3\_F200N\_STEP16\_1On-AllOff\_sflt.fits \\

878: F212N & i1913472n\_flt.fits & 3\_F212N\_STEP16\_1On-AllOff\_sflt.fits \\

879: F215N & i1913473n\_flt.fits & 3\_F215N\_STEP16\_1On-AllOff\_sflt.fits \\

880: F240M & i1913475n\_flt.fits & 3\_F240M\_STEP2\_1On-AllOff\_sflt.fits  \\

881: \hline

882: \end{tabular}

883: \end{center}

884: \end{table*}

885:  

886: \begin{figure}

887: \includegraphics[bb=76 20 461 403,clip,width=0.47\hsize]{Fig_Flat_1}

888: \hfill

889: \includegraphics[bb=76 20 461 403,clip,width=0.47\hsize]{Fig_Flat_2}

890: \caption{The zeroth order (wavelength independent) plane of the 

891: post-NCS flat field cube (left) and the first order (wavelength

892: dependent) plane (right).  }

893: \label{Fig_flat}

894: \end{figure}

895: 

896: 

897: 

898: \subsection{Wavelength Calibration}

899: 

900: In-orbit wavelength calibration for G141 spectra was established by

901: observations of spectra of planetary nebulae (PNe). Two compact PNe were

902: observed -- Hubble 12 (Hb~12) and Vy2-2, although both are resolved by NIC3.

903: Fig.~\ref{Fig_Hb12} shows the one-dimensional extracted spectrum of Hb~12 with

904: the identified emission lines indicated. By matching the observed positions of

905: the lines, fitted by Gaussians, with known wavelengths, the G141 dispersion

906: solution was established. Vy2-2 was observed in 1997 as part of the early

907: calibration programme and Hb~12 in 2002 in the post-NCS era. For Hb~12 fifteen

908: spectra were analysed for first order dispersion solution and eight for the

909: second order spectrum; for Vy2-2 only three spectra were available to analyse,

910: and were only used to verify the solution derived from Hb~12. Within the errors

911: no difference was found between the wavelength solutions for both targets,

912: indicating no measurable change of the dispersion pre- and post-NCS. 

913: 

914: Up to fifteen emission lines were detectable in the first order spectra and

915: seven for the second order; the lines and their identifications are listed in

916: Tab.~\ref{Table_lines}. First and second order polynomial fits were made to the

917: variations of the pixel position with wavelength. The second order terms are

918: very small, typically $2 \times 10^{4}$ times smaller than the first order fit

919: coefficient; they were neglected and only a linear solution adopted. The

920: sampling of any spatial dependence of the wavelength solutions was poor and

921: single average values for the fit coefficients were adopted for the whole

922: detector. Tab.~\ref{Table_disps} lists the first and second spectral order

923: dispersion solutions in the form $ \lambda ({\rm \AA}) = A_0 + A_1  x_0$, where

924: $x_0$ is the pixel offset along the trace from the position of the direct

925: object.

926:  

927: \begin{table}

928: \caption[]{Nebular emission lines used for wavelength calibration.}

929: \label{Table_lines}

930: \centering 

931: \begin{tabular}{ll} 

932: \hline

933: ~~~$\lambda({\rm\AA})$ & Species \\

934: \hline

935: 10830.290 & HeI \\

936: 10938.095 & HI \\

937: 11164.403 & [Fe II] \\

938: 11305.854 & [SI] \\

939: 11969.059 & HeI \\

940: 12527.506 & HeI \\

941: 12818.08  & HI \\

942: 14072.70  & [Fe II] \\

943: 14706.272 & [Fe II] \\

944: 16109.313 & HI \\

945: 16407.192 & HI \\

946: 16806.520 & HI \\

947: 17362.108 & HI \\

948: 18174.121 & HI \\

949: 18751.01  & HI \\

950: \hline

951: \end{tabular}

952: \end{table}

953: 

954: \begin{table}

955: \caption[]{Dispersion solutions for G141 spectra.}

956: \vskip0.1cm

957: \label{Table_disps}

958: \centering 

959: \begin{tabular}{lrl} 

960: \hline

961: Spectrum & $A_0 ({\rm\AA})$ & $A_1 ({\rm\AA/pix})$ \\

962: order    &             &  \\

963: \hline

964:  1 & 13962.8 & 80.19 \\

965:  2 &  7141.7 & 40.54 \\

966: \hline

967: \end{tabular}

968: \end{table}

969: 

970:  

971: \begin{figure}

972: \includegraphics[width=\hsize]{wavecal}

973: \caption{NICMOS G141 spectrum of the planetary nebula Hb~12

974: showing the emission line identifications. The upper axis

975: shows the offset in pixels from the position of the direct image.

976: }

977: \label{Fig_Hb12}

978: \end{figure}

979: 

980: 

981: \subsection{Sensitivity Calibration}

982: \label{sec:fluxcal}

983: 

984: The sensitivity curve needed for the flux calibration of the spectra

985: (Sect.~\ref{sec:flux}) was derived from observations of the flux standards

986: GD153, G191B2B, and P330E.  Tab.~\ref{Table_fluxstd} lists the observations

987: used to derive the final sensitivity calibration.  Spectra were extracted from

988: those observations and co-added for each star separately for pre-NCS and

989: post-NCS observations. The sensitivity curve was then computed by dividing the

990: count rates of these spectra  by  tabulated standard fluxes for each star. The

991: standard fluxes were taken from the  HST CALSPEC library and are based on

992: Hubeny NLTE  models \citep{fluxcala}  in the case of G191B2B and GD153

993: (g191b2b\_mod\_004.fits,  gd153\_mod\_004.fits) and a combined STIS, NICMOS

994: spectrum in the case of P330E \citep[p330e\_stisnic\_001.fits,][]{fluxcalb}.

995: 

996: The average flux calibrations agreed to about 3\% between the three standard

997: stars and we did not detect significant variations as a function of time of

998: observation. However, variations as a function of  position within the field of

999: view indicated a scatter of about 8\%. The field of view coverage with  flux

1000: standard stars was not sufficient to allow a field-dependent  sensitivity

1001: calibration. Overall the absolute flux calibration for post-NCS observations

1002: was expected to be accurate to better than 10\%, while  for pre-NCS data it was

1003: difficult to assess the overall reliability due  to a lack of suitable standard

1004: star observations. The sensitivity files used in the pipeline reduction carry a

1005: random error of only 1-3\% that was propagated into the final spectra.  In

1006: addition, there was an uncertainty of about $\sim$10\% in the overall

1007: normalisation of the flux.  The final sensitivity curves derived for pre- and

1008: post-NCS periods  are shown in Fig.~\ref{fig:throughput}.

1009: 

1010: 

1011: \begin{table*}

1012: \caption[]{List of flux standard stars}

1013: \vskip0.1cm

1014: \label{Table_fluxstd}

1015: \centering

1016: \begin{tabular}{llrll}

1017: \hline

1018: Star    & Association & PROP ID & PROP PI & pre/post NCS \\

1019: \hline

1020: GD153   & N94A02E6Q & 10385 & Bohlin   & post NCS \\

1021: GD153   & N8U402MVQ &  9998 & Bohlin   & post NCS \\

1022: GD153   & N9U203NOQ & 11064 & Bohlin   & post NCS \\

1023: G191B2B & N9U201M3Q & 11064 & Bohlin   & post NCS \\

1024: G191B2B & N8U405TWQ &  9998 & Bohlin   & post NCS \\

1025: G191B2B & N94A03GDQ & 10385 & Bohlin   & post NCS \\

1026: P330E   & N8U406VZQ &  9998 & Bohlin   & post NCS \\

1027: P330E   & N8BR01ICQ &  8991 & Thompson & post NCS \\

1028: P330E   & N9U212P8Q & 11064 & Bohlin   & post NCS \\

1029: \hline

1030: G191B2B & N4IT01NUQ &  7696 & Calzetti & pre NCS  \\

1031: P330E   & N4VD01OYQ &  7959 & Calzetti & pre NCS  \\

1032: \hline

1033: \end{tabular}

1034: \end{table*}

1035: 

1036: \begin{figure}[h]

1037: \includegraphics[width=\hsize]{fig08}

1038: \caption{NICMOS sensitivity curves as a function of wavelength for the

1039:   G141 grism for pre-NCS and post-NCS data sets.}\label{fig:throughput}

1040: \end{figure}

1041: 

1042: 

1043: \section{Data Products and Distribution}

1044: 

1045: \subsection{HLA Portal}

1046: 

1047: 

1048: HLA data are distributed by both ST-ECF and STScI.  There are three main ways

1049: to search, browse, and access the NICMOS spectra:

1050: 

1051: \begin{enumerate}

1052: 

1053: \item{Archive Query Interface.} The HLA archive can be searched

1054: online (\url{http://archive.eso.org/wdb/wdb/hla/product_science/form}) and

1055: constraints on the target (e.g. the target name), the data properties (e.g.

1056: effective exposure time), the source properties (e.g. the magnitude) and the

1057: data quality (e.g. the signal-to-noise ratio) can be placed. The detailed

1058: result pages show the preview of the data as well as all available metadata.

1059: 

1060: \item{HLA archive at STScI.} The ST-ECF HLA data has been also integrated into

1061: the HLA interface of the Space Telescope Science

1062: Institute (\url{http://hla.stsci.edu}). A subset of the parameters of the

1063: ST-ECF HLA interface are shown and can be queried.

1064: 

1065: \item{Virtual Observatory.} We provide fully automated access to the HLA

1066: metadata and data via Virtual Observatory (VO) standards. A Simple Spectrum

1067: Access Protocol (SSAP) server has been established

1068: (\url{http://www.stecf.org/hla-vo}).  It serves VOTables in V1.1 format, which

1069: contain, in addition to the standard metadata, information about the footprints

1070: of the equivalent slits of the grism spectra. Our SSAP server has been tested

1071: with ESO's archive browser VirGO (\url{http://archive.eso.org/cms/virgo/}) as

1072: well as with SPLAT (\url{http://star-www.dur.ac.uk/~pdraper/splat/splat-vo/}, 

1073: also available at \url{http://starlink.jach.hawaii.edu/})

1074: and VOSpec (\url{http://esavo.esa.int/vospecapp}). 

1075: 

1076: \end{enumerate}

1077: 

1078: 

1079: \subsection{Distributed Files} \label{sec:metadata}

1080: 

1081: The spectra are distributed as sets of FITS files, which include two

1082: dimensional cutouts of each target from the rectified and calibrated  grism

1083: images, a cutout of the target from the undispersed filter image and  a

1084: one-dimensional extracted spectrum.

1085: 

1086: The two-dimensional grism stamp images and the direct image cutouts are

1087: multi-extension FITS files.  The one-dimensional spectrum follows the data

1088: formatting specified for FITS serialisation by the IVOA Spectral Data Model

1089: version 1.01 \citep{mcd}.  Each data point contains the wavelength in \AA, the

1090: count rate in electrons per second and the flux expressed in physical units

1091: along with associated errors and an estimate of the contaminating flux.

1092: Additional metadata include keywords to describe the contamination, the

1093: orientation of the dispersion direction on the sky and footprints.  

1094: 

1095: 

1096: 

1097: \begin{figure}

1098: \includegraphics[width=\hsize]{fig09}

1099: \caption{Distribution of the number of extracted spectra from HLA

1100:   datasets.}\label{fig:datasets}

1101: \end{figure}

1102: 

1103: 

1104: \begin{figure}

1105: \includegraphics[width=\hsize]{fig10}

1106: \caption{Magnitude distribution of the targets with spectra in the HLA release. The blue

1107: histogram are the extended sources. }\label{fig:maghist}

1108: \end{figure}

1109: 

1110: \begin{figure}

1111: \includegraphics[width=\hsize]{fig11}

1112: \caption{Measured signal-to-noise of the HLA spectra as a function of magnitude.}\label{fig:snmags}

1113: \end{figure}

1114: 

1115: \section{Results}\label{sec:qc}

1116: 

1117: \subsection{Data Release 1}

1118: 

1119: Data were released on \releasedate. This Data Release 1 (DR1) includes a total

1120: of \nspectra\ extracted from  \nassociations\ associations. 

1121: 

1122: The H-band magnitudes computed from the undispersed images of the targets range

1123: from 7 to 23.5  for point sources, and from 16 to 23.5 for extended sources as

1124: identified by the SExtractor program (see Fig.~\ref{fig:maghist}).  Only

1125: spectra with measured signal-to-noise ratios larger than three were included in

1126: the release, and  typical measured signal-to-noise ratios are about 50 for

1127: objects brighter than H$\approx$17 (see Fig.~\ref{fig:snmags}).

1128: 

1129: \subsection{Completeness}

1130: 

1131: Whether the spectrum of a particular target is extracted depends, among other

1132: things, on the detection of the target on the undispersed image, the  relative

1133: location and flux levels of other spectra on the grism image, the location of

1134: residual image defects and  the location of the spectrum  relative to the image

1135: edges.  The number of spectra extracted at each pointing therefore varies

1136: widely, and was typically about five.  The distribution of the number of

1137: spectra extracted from each association is shown in Fig.~\ref{fig:datasets}.

1138: 

1139: \new{To assess the completeness of the sample quantitatively, we have compared

1140: the DR1 catalogue with the survey by \citet{pat}. These authors searched

1141: two-dimensional NICMOS grism images for emission line objects and found a total

1142: of 33 candidates. Most of the emission lines are H$\alpha$ at redshifts between

1143: 0.75 and 1.9. Such a sample of galaxies is dominated by closely interacting

1144: systems and, because of contamination issues, our procedure selects against

1145: such objects. We found that a total of 6 of the \citeauthor{pat} candidates are

1146: included in DR1. We therefore consider 20\% to be a lower limit for the

1147: completeness of our catalogue.  }

1148: 

1149: The HLA includes any spectrum suitable for extraction, no attempt has been made

1150: to classify or match the targets with any catalogue. Some of the fields  have

1151: been observed several times with different roll angle. These spectra were not

1152: co-added and therefore several spectra might be available for one target.  In

1153: total, the number of unique targets in the release is \ntargets.

1154: 

1155: 

1156: \subsection{Photometric accuracy of the undispersed images}

1157: \label{sec:phot-accur-undisp}

1158: 

1159: The photometric accuracy of the undispersed direct images was assessed by

1160: comparing the magnitudes obtained in the F160W and F110W bands with the 2MASS H

1161: and J band magnitudes of matched objects.  When doing this comparison, we used

1162: aperture photometry with an aperture of 11 pixels, and applied a

1163: finite-aperture correction as provided by the NICMOS Data Handbook

1164: \citep{nicmoshandbook}; in addition, we converted the 2MASS magnitudes into the

1165: AB magnitude system.  

1166: 

1167: \begin{figure}[h]

1168:   \centering

1169:   \includegraphics[width=\hsize]{fig12}

1170:   \caption{\textbf{Top.} Direct image F160W magnitude, as a function

1171:     of the 2MASS $H$ magnitude.  \textbf{Bottom.}  Difference of the

1172:     two magnitudes as a function of the 2MASS magnitude .}

1173:   \label{fig:7}

1174: \end{figure}

1175: 

1176: The results obtained are shown in Fig.~\ref{fig:7}: a good agreement is

1177: obtained over the whole magnitude range, with negligible bias; the relatively

1178: large scatter observed, $\sim0.16 \mbox{ mag}$, is probably due to the

1179: photometric uncertainties present in the 2MASS catalogue for faint sources.

1180: Similar results were obtained for the F110W magnitudes.

1181: 

1182: 

1183: \subsection{Photometric accuracy of the spectra}

1184: 

1185: \begin{figure}[t]

1186:   \centering

1187:   \includegraphics[width=\hsize]{fig13}

1188:   \caption{Total throughput of the F160W filter for the NICMOS/NIC3 camera.}

1189:   \label{fig:4}

1190: \end{figure}

1191: 

1192: Quality control of the photometric calibration of the one-dimensional spectra

1193: was performed by comparing the integrated flux, as measured from the

1194: end-product spectra, with the magnitude derived from the direct images.  The

1195: following procedure was applied: 

1196: 

1197: \begin{itemize} 

1198: 

1199: \item all spectra  that had associated direct images in the F160W filter were

1200: selected; 

1201: 

1202: \item for the selected objects, only spectra which were considered complete up

1203: to the boundary of the total throughput of the F160W/NIC3 configuration (see

1204: Fig.~\ref{fig:4}), were selected.  In particular, all spectra that did not

1205: contain data within the wavelength range defined by $\mathrm{throughput} >

1206: 0.005$ were discarded; 

1207: 

1208: \item the flux of these objects was then integrated and compared with either

1209: the direct image magnitudes or, when available, with the 2MASS $H$ band

1210: magnitudes.  At the same time, the error on the magnitude due to flux

1211: contamination from other sources was also evaluated using data provided in the

1212: output products.  

1213: 

1214: \end{itemize}

1215: 

1216: \begin{figure}

1217:   \centering

1218:   \includegraphics[width=\hsize]{fig14}

1219:   \caption{\textbf{Top.} The ``spectroscopic'' magnitude, i.e.\ the

1220:     magnitude derived from the spectrum of the object, is shown as a

1221:     function of the direct image F160W magnitude.   \textbf{Bottom.}

1222:     The difference of the two magnitudes is plotted as a function of

1223:     the direct image magnitude.  \new{For both plots, the colour

1224:     of the points indicates the level of contamination, 

1225: coded according to the difference $\Delta m$ in the magnitude  of the target

1226: and that of the contaminator. The colour codes are as follows:

1227:  green:              $\Delta m > 8^{\sl mag}$,

1228:  yellow: $4^{\sl mag}<\Delta m < 8^{\sl mag}$,

1229:  orange: $2^{\sl mag}<\Delta m < 4^{\sl mag}$

1230: and  red: $0^{\sl mag}<\Delta m < 2^{\sl mag}$.

1231:  }

1232: }

1233:   \label{fig:5}

1234: \end{figure}

1235: 

1236: \begin{figure}[h]

1237:   \includegraphics[height=0.32\hsize]{fig15a}

1238:   \hfill

1239:   \includegraphics[height=0.32\hsize]{fig15b}

1240:   \hfill

1241:   \includegraphics[height=0.32\hsize]{fig15c}

1242: \newline

1243:   \includegraphics[width=0.31\hsize]{fig15aspec}

1244:   \hfill

1245:   \includegraphics[width=0.32\hsize]{fig15bspec}

1246:   \hfill

1247:   \includegraphics[width=0.31\hsize]{fig15cspec}

1248:   \caption{Examples of catastrophic discrepancies selected from the

1249:     outliers of Fig.~\ref{fig:5}.  Each panel shows

1250:     the direct image associated to the source for which an inaccurate

1251:     spectroscopic magnitude was obtained (marked in red); other

1252:     objects in the field used for the contamination calculation are

1253:     marked with blue ellipses. \new{The corresponding cutouts from

1254:     the grism image are shown below the undispersed images.}

1255:     \textbf{Left.} Source in a crowded

1256:     field, close to an undetected source to the

1257:     right. \textbf{Middle.} Source close to bright star on the

1258:     boundary. \textbf{Right.} Source contaminated by a bright star

1259:     outside the field (note the spike on the centre-left of the

1260:     image).}

1261:   \label{fig:6}

1262: \end{figure}

1263: 

1264: The results obtained for the comparison with the F160W direct images are shown

1265: in Fig.~\ref{fig:5}.  A very good linear response is obtained over a large

1266: range of magnitudes for both pre- and post-NCS data.  The effects of spectrum

1267: contamination are visible in the lower-right part of the graph, which is

1268: occupied mainly by \new{sources with  high levels of contamination, as directly

1269: evaluated from the spectra}.  Overall the agreement is good, with a median

1270: offset of 0.16 magnitudes for pre-NCS data, and 0.23 for the post-NCS case.

1271: The scatter is in both cases $\sim0.5$ magnitudes.

1272: 

1273: A few ``catastrophic'' cases of sources with apparent no-contamination but

1274: still large magnitude difference are however evident in this plot.  These cases

1275: were investigated further and simple explanations could be found for almost all

1276: of them.  In general, it was found that large discrepancies were associated

1277: with: (i) crowded fields, for which the contamination model might be

1278: inaccurate; (ii) objects close to bright sources not included in the analysis;

1279: (iii) objects close to extended, diffuse sources; or (iv) objects contaminated

1280: by sources outside the field of view of the direct image. \new{ Three examples

1281: of such cases are shown in Fig.~\ref{fig:6}. For each case, the undispersed

1282: image  with the marked problematic target is shown along with the corresponding

1283: spectrum cutout. In the left and centre panels, a bright undetected source

1284: produces significant  contamination in the extracted spectra.  In both cases,

1285: the contaminating source dominates the flux. In the right panel, the zeroth

1286: order of a very bright source outside the field produces a bright region  in

1287: the extracted spectrum.}

1288: 

1289: 

1290: \subsection{Noise}\label{sec:sn}

1291: 

1292: All extracted spectra include a predicted flux uncertainty  in each wavelength

1293: bin based on error propagation from the error array in the NICMOS images

1294: through to the final spectra.  Fig.~\ref{fig:signaltonoise} compares these

1295: predicted uncertainties with noise measurements on the final spectra using the

1296: DER\_SNR algorithm \citep{2007STECF..42....4S}. 

1297: 

1298: \begin{figure}[ht]

1299:   \centering

1300:   \includegraphics[width=\hsize]{fig16}

1301:   \caption{Noise measurements from the spectrum using the DER\_SNR

1302:     algorithm as a function the noise estimate obtained from the

1303:     NICMOS error array using error propagation.}

1304:   \label{fig:signaltonoise}

1305: \end{figure}

1306: 

1307: The SNR values are in good agreement with quite small scatter at the high flux

1308: end. At the faint end the scatter is larger and there is a slight but

1309: systematic trend for the DER\_SNR values to be higher than their predicted

1310: value. This might be due to a component, e.g. the noise in the background,

1311: which was not included in our noise model, or reflect some bias in the

1312: measurement of the noise. 

1313: 

1314: 

1315: \subsection{\new{Comparison of Spectra with NICMOSlook Extractions}}

1316: 

1317: \new{Most previous extractions of NICMOS grism spectra have used interactive

1318: tools, such as the NICMOSlook program \citep{nicmoslook}. NICMOSlook is a

1319: highly specialised tool that provides a large number of options and parameters

1320: that can be varied to optimise the extraction.  The main advantages of using an

1321: interactive program is that the background region can be adjusted for each

1322: individual spectrum and that the subpixel accuracy of the trace necessary for

1323: the pixel-response correction} \new{can be visually checked and fitting

1324: parameters can be modified.  In addition, NICMOSlook's optional deblending of

1325: overlapping spectra results in a more reliable estimate of contamination

1326: levels, and is accurate enough to recover heavily contaminated spectra.  The

1327: combination of optimising the background regions and deblending of overlapping

1328: spectra allows NICMOSlook to extract more spectra in crowded regions than is

1329: possible  with the automatic procedure used in this work. For example,

1330: \cite{pat} used NICMOSlook to extract the larger sample of H$\alpha$ emitters

1331: mentioned above.

1332: 

1333: For those spectra that are included in DR1, the main uncertainty is the

1334: background level. The impact is most visible in low signal-to-noise spectra.

1335: In Fig.~\ref{fig:nicmoslook}, we compare  the NICMOSlook extraction of three

1336: spectra  of the \citeauthor{pat} targets with the one from DR1, all of them are

1337: low signal-to-noise spectra.  The good agreement of the spectra suggests that

1338: both the relative and absolute flux calibration in DR1 is reliable.

1339: 

1340: To summarise, the accuracy of the completely unsupervised extractions of NICMOS

1341: spectra in HLA DR1 is close to what can be achieved with an interactive tool.

1342: In addition, we have taken into account contamination by sources outside the

1343: field of view of NICMOS, which is not possible using currently available

1344: interactive tools.  }

1345: 

1346: \begin{figure}

1347: \includegraphics[width=\hsize]{nicmoslook}

1348: \caption{\new{Comparison of HLA spectra  with extractions using 

1349: the NICMOSlook program. All three spectra are H$\alpha$ emitting galaxies

1350:  discovered by \cite{pat}. The blue curves are NICMOSlook extractions, whereas the

1351: red curves are the HLA extractions. The blue labels are the \citeauthor{pat}'s names.

1352: } }\label{fig:nicmoslook}

1353: \end{figure}

1354: 

1355: 

1356: \subsection{Comparison of Spectra with Published Data}

1357: 

1358: As a quality check, we compared some extracted spectra to spectra of the same

1359: objects taken from the ground, as well as previously published extractions of

1360: spectra from the same NICMOS data. Because of strong atmospheric lines, H-band

1361: IR spectroscopy is difficult to calibrate from the ground. In particular, the

1362: region around $\lambda=1.4-1.5\,\mu$ m is not accessible from the ground. 

1363: 

1364: In Figs.~\ref{fig:comp1} and~\ref{fig:1044}, we compare spectra of two high

1365: redshift QSOs to spectra from Gemini \citep{comp1} and the Italian Telescopio

1366: Nazionale Galileo  \citep[TNG,][]{maio}. In both cases, the shape of the line

1367: and continuum around the  CIII line at $\lambda\sim1.3\,\mu$m is well

1368: reproduced. The continuum at wavelengths longer than 1.5$\,\mu$m agrees to

1369: within about 5\% with the Gemini data, but differs by almost a factor of two

1370: from the TNG data. We attribute this discrepancy to the uncertain calibration

1371: of the data from the ground. 

1372: 

1373: Finally, in Fig.~\ref{fig:asr24}, we compare HLA spectra of the brown dwarf

1374: ASR~24 with the spectra extracted  from the same data set by \cite{asr24}.

1375: ASR~24 has been observed three times with NICMOS G141 with three different roll

1376: angles. The HLA therefore contains three separate spectra of this source,

1377: namely HNG\_J032911.32+311717.5\_N8VM06BEQ, HNG\_J032911.32+311717.5\_N8VM09G5Q

1378: and HNG\_J032911.32+311717.6\_N8VM16S3Q. \cite{asr24} did not derive a flux

1379: scale for their spectrum, this spectrum is therefore scaled to matched the HLA

1380: extraction. It can be seen that the absolute and relative flux calibration of

1381: the three HLA extractions agree to within 10\%.  We therefore conclude that the

1382: quality of the HLA NICMOS G141 spectra is comparable to the best extractions

1383: previously obtained from the data.

1384: 

1385: 

1386: \begin{figure}

1387: \includegraphics[width=\hsize]{fig17}

1388: \caption{Comparison of an HLA spectrum with a Gemini spectrum of the same

1389: source.  The blue curve is the heavily smoothed spectrum of SDSS

1390: J083643.85+005453.3 from \cite{comp1}, the red curve is

1391: HNG\_J083643.82+005453.4\_N6LE01ULQ.}\label{fig:comp1}

1392: \end{figure}

1393: 

1394: \begin{figure}

1395: \includegraphics[width=\hsize]{fig18}

1396: \caption{Comparison of an HLA spectrum with a TNG NICS spectrum of the same

1397: source.  The blue curve is the spectrum of SDSS 104433.04+012502.2 from

1398: \cite{maio}, the red curve is

1399: HNG\_J104433.08-012501.6\_N6LE03C7Q.}\label{fig:1044} 

1400: \end{figure}

1401: 

1402: \begin{figure}

1403: \includegraphics[width=\hsize]{fig19}

1404: \caption{Comparison of an HLA spectra of brown dwarf ASR~24 (red curves and

1405: error bars) with the spectrum published by \cite{asr24} (blue curve).  The

1406: latter has been scaled to match the mean flux level of the HLA

1407: data}\label{fig:asr24}

1408: \end{figure}

1409: 

1410: 

1411: \section{Summary and Conclusion}

1412: 

1413: The HLA NICMOS grism project provides a database of low resolution H-band IR

1414: spectra. Most of the spectra have never been previously extracted from the HST

1415: NICMOS grism data. The database is useful for work on cool stars and emission

1416: line galaxies at $z$ between 1.1 and 1.9, which are readily detected with the

1417: NICMOS grism. The calibration of the spectra is based on a new analysis of

1418: available calibration data and extraction of the spectra for isolated point

1419: source should be close to optimum.  Confused spectra and spectra of extended

1420: objects are identified in the data release and are also of high quality.  The

1421: absolute and relative flux calibration of the spectra is better than 10\%, and

1422: the wavelength calibration better than 5 nm.  The data release is accompanied

1423: by a wide range of auxiliary data and is available through several interfaces.

1424: We anticipate a revised data release based on a new version of the STScI

1425: CALNICA pipeline in early 2009.

1426: 

1427: \begin{acknowledgements}

1428: 

1429: This paper is based on observations made with the NASA/ESA Hubble Space

1430: Telescope, obtained from the data archive at the Space Telescope -- European

1431: Coordinating Facility.  We thank our HLA collaborators Brad Whitmore and  the

1432: STScI and CADC HLA teams.

1433: 

1434: \end{acknowledgements}

1435: 

1436: \clearpage

1437: 

1438: \appendix

1439: \newcommand{\diff}{\mathrm{d}}

1440: 

1441: 

1442: \section{Extraction of Slitless Spectra}

1443: \label{sec:slitl-meas}

1444: 

1445: 

1446: A crucial difference between longslit and slitless spectroscopy is the

1447: selection of the light to be dispersed. For slitless spetroscopy, the light

1448: that enters the spectrograph is only limited by the  object shape and the point

1449: spread function.  For the extraction, we neglected the fact that the shape, as

1450: well as the point spread function, can change with wavelength. For the

1451: following discussion, we also neglect the impact of the sensitivity curves on

1452: the spectrum. A more rigorous discussion of the parameters will be presented by

1453: \citet{theorypaper}. With the above  approximations, the specific intensity of

1454: the object $I(\vec x, \lambda)$ at the angular position $\vec x$ can be

1455: expressed as the product of two functions

1456: 

1457: \begin{equation} \label{eq:1} I(\vec x, \lambda) = I(\vec x) F(\lambda) \; ,

1458: \end{equation}

1459: where $I(\vec x)$, is the integrated intensity of the object as revealed by the

1460: direct image, and $F(\lambda)$ is the spectral energy distribution.

1461: $F(\lambda)$ is independent of the position $\vec x$.

1462: The image $\tilde I(\vec x)$ generated by the slitless spectrograph

1463: is then

1464: \begin{equation}

1465:   \label{eq:2}

1466:   \tilde I(\vec x) = \int I \bigl( \vec x - \vec r (\lambda -

1467:   \lambda_0), \lambda \bigr) \, \diff \lambda = \int I \bigl( \vec x -

1468:   \vec r (\lambda - \lambda_0) \bigr) F(\lambda) \, \diff \lambda \; ,

1469: \end{equation}

1470: where $\vec r$ is the dispersion vector and $\lambda_0$ a pivot

1471: wavelength.  The second equality shows that the result can be written

1472: as a simple convolution, carried out along the dispersion direction,

1473: of the original image with the object spectrum.

1474: 

1475: The goal of the spectral extraction is to add all flux values of pixels that

1476: represent the same wavelength. For slitless spectroscopy, the monochromatic

1477: light from a single wavelength covers a region  with the same shape as the

1478: target object on the detector. For simplicity, we assign all pixels on a

1479: straight line on the two-dimensional grism image to the same wavelength bin in

1480: the one-dimensional spectrum.  The size of the objects along the dispersion

1481: direction limits the spectra resolution of slitless spectra and has the same

1482: effect as the width of the slit in a longslit spectrograph.  In the following,

1483: we will refer to the box defined by the above straight line of constant

1484: wavelength and the size of the object along the dispersion direction  as the

1485: ``virtual slit''.

1486: 

1487: For complex objects, it is unavoidable that a straight line includes pixels

1488: with flux originating from different wavelengths. The goal of the extraction

1489: procedure is to choose the direction of the virtual slit so that variations of

1490: wavelength are minimised.  Our algorithm is designed to work exactly for the

1491: case of elliptical objects, i.e.  in that case each wavelength bin of the

1492: one-dimensional spectrum includes only pixels with flux from a single

1493: wavelength.

1494: 

1495: 

1496: 

1497: \begin{figure}

1498:   \centering

1499:   \includegraphics[scale=0.75]{angles}

1500:   \caption{Schematic drawing of a slitless spectrum measurement.}

1501:   \label{fig:angles}

1502: \end{figure}

1503: 

1504: 

1505: As shown above in Eq.~\eqref{eq:2}, the two-dimensional spectrum is a

1506: convolution of the object image with the one-dimensional spectrum.  Spectral

1507: features will be smoothed by the object intensity profile corresponding to each

1508: line of dispersion.  For example, for an object with a single emission line at

1509: $\lambda_\mathrm{e}$ so that $F(\lambda) \propto \delta(\lambda -

1510: \lambda_\mathrm{e})$,  the resultant image $\tilde I(\vec x)$ will be identical

1511: to $I(\vec x)$ except for a shift along $\vec r$. 

1512: 

1513: A natural choice for the orientation of the virtual slit is the direction of

1514: the line defined by the  ``centre of light'' of the intensity profile along the

1515: dispersion direction.  In this case,   the position $\vec x_\mathrm{e}$

1516: satisfies the equation

1517: 

1518: \begin{equation}

1519:   \label{eq:3}

1520:   \int \tilde I(\vec x_\mathrm{e} + \vec r \ell) \ell \, \diff \ell =

1521:   0 \; .

1522: \end{equation}

1523: 

1524: 

1525: For elliptical isophotes, the image of the object can be written as a function

1526: of the elliptical radius $\rho^2 \equiv \vec x^\mathrm{T} A \vec x$, 

1527: \begin{equation}

1528:   \label{eq:4}

1529:   I(\vec x) = f \bigl( \vec x^\mathrm{T} A \vec x\bigr) \; ,

1530: \end{equation}

1531: where $A$ is a $2 \times 2$ symmetric matrix, and where, for simplicity,

1532: we have chosen the origin of our coordinates at the centre of the

1533: elliptical object.  In the following, without loss of generality, we

1534: will also assume that the dispersion direction is along the horizontal

1535: axis $x_1$, so that $\vec r = (r, 0)$.

1536: For convenience, we convert to  a \textit{slanted\/}

1537: coordinate system $\vec x'$ defined by the linear

1538: transformation

1539: \begin{align}

1540:   \label{eq:5}

1541:   x_1 \mapsto x'_1 = {} & x_1 + k x_2 \; , &

1542:   x_2 \mapsto x'_2 = {} & x_2 \; . \\

1543:   \vec x \mapsto \vec x' = {} & T \vec x \; , & 

1544:   T = {} & \begin{pmatrix}

1545:     1 & k \\

1546:     0 & 1

1547:   \end{pmatrix} \; .

1548: \end{align}

1549: In this new coordinate system we then define the slanted direct image

1550: $I'(\vec x')$ as a simple remapping of $I(\vec x)$:

1551: \begin{equation}

1552:   \label{eq:6}

1553:   I'(\vec x') = I(\vec x) = I(T^{-1} \vec x') \; .

1554: \end{equation}

1555: Along a line parallel to the dispersion direction, the transformation

1556: $T$ is a simple translation: in other words, for fixed $x_2$, $x'_1$

1557: is a simple shift of $x_1$ by $k x_2$. 

1558: 

1559: To compute the orientation of the virtual slit, we choose the 

1560: value of $k$ in the  transformation $T$

1561: in such a manner that the image of the galaxy in the slanted coordinate

1562: system appears as an ellipse with one of the axes oriented along the

1563: dispersion direction.  This requirement is equivalent to choosing the

1564: matrix $T^{-\mathrm{T}} A T^{-1}$ to be  diagonal, which is guaranteed if

1565: $k = A_{12} / A_{11}$.  With this choice, we find

1566: \begin{equation}

1567:   \label{eq:7}

1568:   T^{-\mathrm{T}} A T^{-1} =

1569:   \begin{pmatrix}

1570:     A_{11} & 0 \\

1571:     0 & \det (A) / A_{11}

1572:   \end{pmatrix} \; .

1573: \end{equation}

1574: 

1575: In this new coordinate system  the virtual slits  are perpendicularly to the

1576: dispersion direction, because the object is now symmetric along the axis

1577: $x'_2$.  The corresponding lines in the original coordinate system can be

1578: obtained by transforming back the vertical lines, and as graphically

1579: illustrated in Fig.~\ref{fig:angles}, these lines can be obtained by joining

1580: the two tangent points of the ellipse describing the direct object with lines

1581: parallel to the dispersion direction.  

1582: 

1583: Note that the direction of the virtual slits is never along one of the axes of

1584: the elliptical isophotes except in the trivial case where the object is already

1585: oriented along the dispersion direction ($A_{12} = k = 0$).  The angle with

1586: respect to the vertical formed by the lines of constant wavelength is given by

1587: $\tan \beta = k = A_{12} / A_{11}$, and thus for highly elongated objects it

1588: can approach $\pm \pi/2$.

1589: 

1590: \def\aap{A\& A}

1591: \def\pasp{Proc.Astr.Soc.Pacific}

1592: \def\mnras{MNRAS}

1593: \def\apj{ApJ}

1594: \def\aj{AJ}

1595: \def\apjl{ApJL}

1596: \def\apjs{ApJS}

1597: \def\procspie{Proc. SPIE}

1598: 

1599: 

1600: \clearpage

1601: \begin{thebibliography}{}

1602: \bibliography{}

1603: 

1604: 

1605: \bibitem[Adelman-McCarthy et al.(2007)]{2007ApJS..172..634A} 

1606: Adelman-McCarthy, J.~K., et al.\ 2007, \apjs, 172, 634 

1607: 

1608: \bibitem[Bertin \& Arnouts(1996)]{ba} Bertin, E. \& Arnouts, S.\ 1996, \aap,

1609: 117, 393

1610: 

1611: \bibitem[Bohlin(2003)]{fluxcala}Bohlin, R. 2003, 2002 HST Calibration Workshop,

1612: ed. S. Arribas, A.  Koekemoer, \& B. Whitmore, (Baltimore:STScI),  115

1613: 

1614: \bibitem[Bohlin(2007)]{fluxcalb}Bohlin, R. C. 2007, in The Future of

1615: Photometric, Spectrophotometric,  and Polarimetric Standardization, ASP Conf.

1616: Series, Vol. 364, p. 315  ed. C. Sterken; astro-Ph 0608715

1617: 

1618: \bibitem[Bohlin, Riess \& de Jong(2006)]{bohlinI} Bohlin, R.C., Riess, A.  \&

1619: de Jong, R. 2006, Instrument Science Report, NICMOS, 2006-002

1620: 

1621: \bibitem[Bohlin, Lindler \& Riess(2005)]{bohlinII} Bohlin, R.C., Lindler, D.

1622: \& Riess, A. 2005, Instrument Science Report, NICMOS, 2005-002

1623: 

1624: \bibitem[Bushouse(1997)]{calnica} Bushouse, H. 1997, The 1997 HST Calibration

1625: Workshop with a New Generation of Instruments, 223, 223 

1626: 

1627: \bibitem[Dahlen et al.(2008))]{newcalnica} Dahlen, T., et al. 2008,

1628: Instrument Science Report, NICMOS, 2008-002

1629: 

1630: 

1631: \bibitem[Freudling(1997)]{nicmoslook} Freudling, W.\ 1997, Space Telescope

1632: European Coordinating Facility Newsletter, Volume 24, 7, 24, 7 

1633: 

1634: \bibitem[Fruchter \& Hook(2002)]{Fruchter} Fruchter, A. S., Hook, R. N. 2002,

1635: PASP, 114, 144

1636: 

1637: \bibitem[Greissl et al.(2007)]{asr24} Greissl, J., Meyer, 

1638: M.~R., Wilking, B.~A., Fanetti, T., Schneider, G., Greene, T.~P., \& Young, 

1639: E.\ 2007, \aj, 133, 1321 

1640: 

1641: \bibitem[Horne(1986)]{horne} Horne, K.\ 1986, \pasp, 98, 609 

1642: 

1643: \bibitem[Koekemoer et al.(2002)]{drizzle} 

1644: Koekemoer, A.M., Fruchter, A.S., Hook, R.N. \& Hack,W. 2002, 

1645:  Proceedings of the 2002 HST Calibration Workshop, 

1646: STScI, 337

1647: 

1648: \bibitem[Koekemoer et al.(2006)]{koe06} Koekemoer, A.M.,   Fruchter, A.S.,

1649: Hook, R.N., Hack W.,  \& Hanley, C. 2006, The 2005 Calibration Workshop, eds.

1650: A.\ Koekemoer, P.\ Goudfrooij, \& L.\ Dressel, L., 423 

1651: 

1652: \bibitem[K\"ummel et al.(2008)]{axe1} K\"ummel, M.,  

1653: Walsh, J.R., Pirzkal, N., Kuntschner, H., \& Pasquali, A., 2008, PASP, submitted

1654: 

1655: \bibitem[K\"{u}mmel, Kuntschner \& Walsh(2007)]{axesim}K\"{u}mmel, M.,

1656: Kuntschner, H., \& Walsh, J. R. 2007 , Space Telescope -- European Coordinating Facility Newsletter, 43, 8

1657: 

1658: \bibitem[Lasker et al.(1990)]{1990AJ.....99.2019L} Lasker, B.~M., Sturch, 

1659: C.~R., McLean, B.~J., Russell, J.~L., Jenkner, H., 

1660: \& Shara, M.~M.\ 1990, \aj, 99, 2019 

1661: 

1662: \bibitem[Lombardi et al.(2008)]{theorypaper} 

1663: Lombardi et al., 2008, in preparation

1664: 

1665: \bibitem[MacKenty(1997a)]{john2} MacKenty, J.\ 1997a, HST Proposal, 7907

1666: 

1667: \bibitem[MacKenty(1997b)]{john1} MacKenty, J.\ 1997b, HST Proposal, 8082

1668: 

1669: \bibitem[Malkan(2003)]{mal2} Malkan, M.\ 2003, HST Proposal, 9865

1670: 

1671: \bibitem[Malkan(2004)]{mal1} Malkan, M.\ 2004, HST Proposal, 10226

1672: 

1673: \bibitem[Maiolino et al.(2004)]{maio} Maiolino, R., Oliva, E., Ghinassi, F.,

1674: Pedani, M., Mannucci, F., Mujica, R., \& Juarez, Y.\ 2004, \aap, 420, 889 

1675: 

1676: \bibitem[McCarthy et al.(1999)]{pat} McCarthy, P.~J., et 

1677: al.\ 1999, \apj, 520, 548 

1678: 

1679: \bibitem[McDowell \& Tody(2007)]{mcd}  McDowell, J. \& Tody, D. (eds.) 2007,

1680: IVOA Spectral Data Model Version 1.01,

1681: http://www.ivoa.net/Documents/PR/DM/SpectrumDM-20070515.html

1682: 

1683: \bibitem[McLean et al.(2000)]{gsc2} McLean, B.~J., Greene, G.~R., Lattanzi,

1684: M.~G., \& Pirenne, B.\ 2000, Astronomical Data Analysis Software and Systems

1685: IX, 216, 145 

1686: 

1687: \bibitem[Mobasher \& Roye(2004)]{nicmoshandbook} Mobasher, B. \& Roye, E. eds.

1688: 2004, HST Data Handbook for NICMOS  V.6

1689: 

1690: \bibitem[Monet et al.(2003)]{2003AJ....125..984M} Monet, D.~G., et al.\ 

1691: 2003, \aj, 125, 984 

1692: 

1693: \bibitem[Pirzkal et al.(2004)]{axe2} Pirzkal, N., Xu, C., Malhotra, S., et al.

1694: 2004, \apjs, 154 501

1695: 

1696: \bibitem[Schultz et al.(2003)]{ncscal} Schultz, A.~B., et al.\ 2003, \procspie,

1697: 4850, 858 

1698: 

1699: \bibitem[Stern et al.(2003)]{comp1} Stern, D., Hall, P.~B., Barrientos, L.~F.,

1700: Bunker, A.~J., Elston, R., Ledlow, M.~J., Raines, S.~N., \& Willis, J.\ 2003,

1701: \apjl, 596, L39 

1702: 

1703: \bibitem[Storrs et al.(1999)]{sto99} Storrs, R., Hook, R., Stiavelli, M.,

1704: Hanley, C., \& Freudling, W., 1999,  Instrument Science Report, NICMOS, 99-005

1705: 

1706: \bibitem[Stoehr et al.(2007)]{2007STECF..42....4S} Stoehr, F., et al.\ 2007,

1707: Space Telescope European Coordinating Facility Newsletter, 42, 4

1708: 

1709: \bibitem[Thompson \& Freudling(2002)]{rodger} Thompson, R.~I., \& Freudling,

1710: W.\ 2002, HST Calibration Workshop: Hubble after the Installation of the ACS

1711: and the NICMOS Cooling System, 240

1712: 

1713: \bibitem[Skrutskie et al.(2006)]{2006AJ....131.1163S} Skrutskie, M.~F., et al.\

1714: 2006, \aj, 131, 1163 

1715: 

1716: \bibitem[Xu \& Mobasher(2003)]{xu03} Xu, C., \& Mobasher, B. 2003, Instrument

1717: Science Report, NICMOS, 2003-009

1718: 

1719: \bibitem[Zacharias et al.(2004)]{2004AJ....127.3043Z} Zacharias, N., Urban,

1720: S.~E., Zacharias, M.~I., Wycoff, G.~L., Hall, D.~M., Monet, D.~G., \& Rafferty,

1721: T.~J.\ 2004, \aj, 127, 3043 

1722: 

1723: \end{thebibliography}

1724: 

1725: 

1726: \end{document}

1727: 

1728: