The Great Observatories Origins Deep Survey (GOODS)
Spitzer Legacy Data Products
Interim Data Release (DR1+)

1.0 General information

2.0 Observations

3.0 Data reduction

4.0 Data products

Tables:

Other information:

Figures:

2.1 Fields and program IDs

The GOODS Spitzer Legacy program observations cover two fields on the sky. One of these fields (GOODS-N) coincides with the historical Hubble Deep Field North (Williams et al. 1996), while the other (GOODS-S) coincides with the Chandra Deep Field South (Giacconi et al. 2001). These fields have extensive observations at virtually every wavelength accessible from major space- and ground-based observatories, including deep, multicolor data from the Advanced Camera for Surveys (ACS) on the Hubble Space Telescope (Giavalisco et al. 2004).

Generically, the GOODS fields consist of sky regions that are approximately 10 x 16.5 arcmin on the sky. The orientations of these fields were originally chosen to match scheduling constraints for Spitzer observing (primarily for IRAC, where we observed the field at two epochs angles separated by 6 months and 180 degrees rotation). Coordinates, position angles, and other important parameters for the MIPS observations of GOODS-N are summarized in Table 1.

The GOODS observations are divided into two separate observing programs, one for each field: program 169 for GOODS-N, and program 194 for GOODS-S. The IRAC and MIPS observations of a given field share the same program identification number.

2.2 MIPS observing strategy and AORs

The design of the AORs for the GOODS MIPS observations is somewhat complex, and we will only give a high-level description here.

The observations for each field were broken into a series of Astronomical Observation Requests (AORs), each several hours long, that were designed to enable efficient scheduling. 24 AORs were used for the GOODS-N MIPS observations.

The MIPS observations were designed to overlay, as nearly as possible, the corresponding IRAC observations of the GOODS field, as described in the documentation for GOODS DR1. The overall pointing center for the GOODS-N field is given Table 1, along with the mean position angle (east of north) of the long axis of the field averaged over the duration of the MIPS observations.

The GOODS-N MIPS observations were executed over the course of several days during MIPS campaign number 8. This followed shortly after the first epoch of IRAC imaging for GOODS-N, which was carried out in IRAC campaign 8. The start and end times for the MIPS observations are summarized in Table 1; Table 2 gives a detailed list of the AOR labels, AORKEY identifying numbers, and start dates/times for each AOR. The telescope orientation rotates during this time, by approximately 1 degree per day, and hence so does the coverage of the GOODS field. The field rotates around the pointing center given in Table 1, and the range of rotation is also given in the table. Because MIPS campaign 8 followed the IRAC campaign 8 by about a week, the telescope position angle had rotated by approximately 7 degrees from that used for the corresponding IRAC observations, and therefore the fields do not perfectly align. However, when the effects of dithering are taken into account there is excellent overlap.

Table 1 - Parameters for GOODS-N MIPS observations

Parameter	GOODS-N
Spitzer program ID	169
RA (J2000)	12:36:54.87
Dec (J2000)	+62:14:19.2
Mean position angle	+41 deg
Range of PAs	4.2 deg
Start date/time	2004-05-27 22:08:46
End date/time	2004-06-01 06:29:48

Table 2 - Chronological AOR summary for GOODS-N MIPS observations

The observations were executed using the Large Source Photometry Astronomical Observing Template (AOT), which is described in detail in section 8.2.1.2.2 of the Spitzer Observer's Manual. The individual exposures used the 30 second frame time. In Large Source Photometry mode, the telescope is pointed at a position on the sky, and then the source is dithered along a series of 5 closely-spaced positions in the detector y-axis direction using small motions of the MIPS scan mirror. For the GOODS observations, the scan mirror cycle was repeated 5 times. This produces a set of 26 exposures dithered by small scan mirror offsets (5 positions x 5 repeats + 1 = 26; one extra frame is taken at the "home" position at the end of the Large Source Photometry sequence). The telescope is then offset by approximately 11.5 arcseconds in the cross-scan (x-axis) direction, and the scan mirror dither pattern is repeated. After this pattern is completed, the telescope is then offset by a large amount (> 5 arcmin) in the scan mirror direction, to image a completely disjoint "offset" field. The pattern of small scan mirror dithers with the small cross-scan offset is then repeated at this new pointing. In this way, two independent fields are imaged, each with a very tight set of small-angle dithers.

Within each AOR, we repeated this large source photometry pattern three times, with the major telescope pointing shifted to positions separated by of order an arcminute. These "major dither" pointings were themselves distributed over the pattern of a 12 point Reuleaux triangle, by means of cluster-mode offsets. Four AORs were needed to complete the pattern for any given MIPS field.

One may visualize the GOODS area as being divided into a 2 x 3 pattern of major MIPS "fields", each approximately 5 arcmin x 5 arcmin (modulo dithering). The MIPS scan mirror axis was aligned with the long axis of the GOODS field (i.e., <PA> = 41 for GOODS-N), so that the major "throws" of the pattern described above occurred in the direction where there are three fields. The AORs were divided into three basic types, each of which covered two of the fields along the long axis. If we label those fields as 1, 2, and 3, they were therefore imaged in combinations of 1+2, 2+3, and 1+3. These different types of AORs were given different labels (A, B and C) in the AORLABEL and the observing target name. The whole procedure was repeated for AORs in the parallel set of 3 fields.

Figure 1 illustrates the layout of the GOODS-N MIPS observations, with the dither positions overlaid on the GOODS MIPS image.

Figure 1: GOODS-N MIPS field layout with dither centers superimposed

3.1 SSC pipeline processing

The GOODS team started reductions of the data using products generated by the Spitzer Science Center (SSC) Basic Calibrated Data (BCD) pipeline. The GOODS-N MIPS data delivered to the GOODS team were processed with pipeline version S10.5.0, and the current data products are based on that version.

A very small number of frames were not correctly processed by the SSC pipeline, and thus no BCD products were available. We hope to recover those frames for future data products when the data are reprocessed by the SSC through the S11 pipeline.

3.2 Frame-level post-BCD processing

In post-processing of the individual BCD frames, we applied the following steps:

Unit conversion: BCD products were converted from units of MJy/sr back to DN by dividing by the conversion constant given in the FLUXCONV keyword in the BCD image headers, and multiplying by the exposure time given in the EXPTIME header keyword.

Scan mirror delta-flatfields: The 24 micron flat field is somewhat different for each scan mirror position. At the time the BCD data were processed by the SSC, there was not yet a fully reliable and automatic method for constructing scan mirror dependent flat fields, and the GOODS data needed additional processing to improve the flat fielding. We therefore constructed scan position-dependent delta-flats from our own science data themselves.

In Large Source Photometry mode, the scan mirror moves through 5 different positions, and delta flats were needed for each of them. Within each AOR, however, we only obtained 6 major, fully independent dither positions. (Each of these consisted 12 closely spaced small dithers due to the scan mirror motions and small telescope offsets of the Large Source Photometry AOT, but this small dither pattern was sufficiently compact that sources may overlap in the different positions, making sky flat construction more difficult.)

We therefore verified that the delta-flatfield pattern was stable from AOR to AOR, and that therefore data from all AORs could be combined in order to increase the number of independent large dither offsets and thus improve the quality of the delta-flat. In practice, the two dimensional structure of the images from the first two AORs showed a small gradient compared to the images in later AORs; the mean count levels of those images were also slightly different. These two AORs were taken right at the start of the MIPS campaign, and we attribute the differences to a transient "settling" of the detector into stable operations after the start of the campaign. The last two GOODS AORs were taken after Spitzer observed some bright A stars for another program, and latent images from these observations were faintly visible in median-combined stacks of the data from those AORs. We therefore excluded the first and last two AORs from the set used to construct the scan-dependent delta-flats. A small number of additional frames which showed electronic "jailbar" patterns (see below) were also excluded.

The remaining frames for each scan mirror position were scaled to a common count level and median combined to create the new delta flats. These were then divided into the BCD science images.

Second-order flat fields for AORs 1, 2, 23 and 24: Special second-order flat fields were constructed for the first and last two AORs to correct the problems noted above. These features appeared to be fixed in array coordinates and independent of scan mirror position, so data from all scan mirror positions (after delta-flatfield correction) were used to construct the second order flats. We note that at this time we cannot confirm whether these second order effects are really multiplicative (as we treated them) or additive, but their amplitude was so small that this is relatively unimportant, particularly considering the small number frames involved and the extensive dithering that was used.

Jailbar correction: In some images, a 4-column pattern of vertical bars could be seen, which corresponded to slightly different mean signal levels in the four interleaved readout amplifiers of the detector. This tended to occur for the first frame in an AOR, or in frames where there was a very strong cosmic ray event. All frames were inspected for such jailbars, and where they occurred, we applied an additive correction based on the median sky values for each of the four amplifiers to equalize their levels to a common median sky background for the whole image.

First and Second frame effect: As described in section 2.2 above, one basic sub-unit of the GOODS MIPS observing sequence is a series of 26 exposures taken as the scan mirror cycles 5 times through its pattern of 5 positions (plus one final exposure at the "home" position). This is always followed by a telescope offset. We noted that the first two frames of this 26 frame pattern always showed gradients with approximately 1% amplitude compared to the other frames. We interpret this as a small difference due to the different timing intervals between exposures dithered by the scan mirror offsets and exposures dithered by telescope offsets, perhaps akin to the IRAC ``first frame effect.'' We therefore constructed corrective images from the median of the frames affected in this way, and applied those multiplicatively as another second-order delta flat field correction. As above, we are not certain at this time whether these corrections should be multiplicative or additive, but given the very low amplitude and the small number of frames affected, this is not a serious concern.

Background subtraction: After the processing steps described above, a robust modal sky estimator was applied to each image, and the net sky level was subtracted (as a constant) from each image.

16 exposures showed small residual gradients in the scan direction. This seemed to occur in pairs of frames, but was otherwise apparently unpredictable. A linear function was fit to the gradient for these frames and subtracted from the images.

3.3 Astrometry, image registration, and image combination

The procedures for astrometric alignment of the images and combining them into a mosaic are closely interleaved, and therefore we describe them together here. The core routines for projecting pixels from the detector plane to an output image plane use the "drizzle" method of Fruchter & Hook (2002). The procedures used for outlier rejection are akin to the "multidrizzle" method used for HST/ACS GOODS data processing. Except where noted below (for intermediate steps in the outlier rejection process), image drizzling was done using the "point kernel", which ensures that each input detector pixel per image contributes only to a single output pixel in the drizzled mosaic. In this way, the noise values in adjacent pixels in the GOODS MIPS data products are uncorrelated with one another. This is different from most other drizzled data products with which users may be familiar, e.g., the Hubble Deep Field WFPC2 and NICMOS data sets, or the GOODS ACS mosaics. Those data products used finite drizzling kernels, resulting in significant noise correlation between adjacent pixels. The very large number of dithered exposures for the GOODS MIPS data sets allow us to take advantage of the point kernel, which ensures uncorrelated pixels (simplifying data analysis), and which maximizes the net angular resolution of the final drizzled images.

Mini-mosaics: First, it was necessary to derive an internally consistent astrometric solution for the MIPS images. Very few individual sources are readily detected in individual 30 second exposures, and therefore it was necessary for us to combine sets of images into "mini-mosaics" before detecting sources. Images from each set of 26 frames dithered by small scan mirror motions (see section 2.2) were combined using nominal offsets encoded in the world coordinate system solutions provided by the SSC. The geometric distortion solution provided in the BCD image headers was applied during this image combination step.

Because each output position of the mini-mosaics has redundancy of at least 5 exposures, and as many as 26, it was possible to reject most cosmic ray events at this step in the data reduction process. Images for each mini-mosaic were initially projected to a common tangent plane on the sky to form a data cube, using a drizzle scale=0.5 and pixfrac=0.7 (see Fruchter & Hook 2002 for an explanation of the drizzle parameters). At each sky pixel, a noise model was computed that accounts for the photon statistics from the sky plus source signals and the detector readout noise and dark current, as well as a local gradient term that boosts the effective local variance in regions (e.g., near the centers of bright sources) to account for the possibility of small registration uncertainties or PSF variations from frame to frame. These parameters were carefully tuned by experimentation to optimize the rejection of outlier pixels, e.g., cosmic rays and detector defects, while minimizing over-rejection at the positions of sources. (In practice, this required very little tuning for MIPS because most 24 micron sources are quite faint in individual 30s exposures, and have little impact on the cosmic ray rejection process.) The pixel values from each image were compared to the median for all images at that sky position. The differences between these values and the median were compared to the noise model, and outliers beyond a specified threshold were masked, so that they could be assigned zero weight in the final image combination.

After the outlier pixels have been identified, the images are then re-drizzled to form a final mini-mosaic image per AOR, with the outliers identified above assigned zero weight in the combination. This drizzling was done using the point kernel, so that noise values in adjacent pixels of the output mosaic are statistically independent of one another.

Astrometric alignment: The mini-mosaics have effective integration times of about 13 minutes, and many sources are readily detected in them. Source positions were measured in each mini-mosaic, and were cross-identified with their positions in other mini-mosaics within the same AOR, as well as with an external reference catalog based on positions of sources detected in the GOODS IRAC data covering the same portion of the sky. There is an excellent correspondence between 24 micron sources and counterparts in deep IRAC imaging (see Chary et al. 2004). The GOODS IRAC data are already tied to the very accurate astrometric reference frame from the GOODS HST/ACS image mosaics, and in this way the various Hubble and Spitzer data sets for GOODS have been astrometrically linked. A global astrometric solution for the MIPS mini-mosaics in each AOR was then derived, allowing for image translations, rotations, and an additional geometric distortion described by a cubic polynomial equation. (The mini-mosaics themselves were drizzled using the SSC distortion solution, so this additional polynomial distortion represents a (small) modification of the SSC solution.)

Although most cosmic rays and other pixel defects were removed during the step of mini-mosaic combination, and additional level of multidrizzle-type outlier rejection was performed by comparing the individual mini-mosaics to a median mosaic drizzled with scale = 1, pixfrac = 1. As before, pixels rejected as outliers were assigned zero weight in the final combination.

Background equalization: During the frame-level processing (section 2.2) we subtracted a background level from each 24 micron image. However, we found it useful to perform an additional step of background equalization before the final step of image combination. The images were stacked into a data cube aligned on the sky using the pixelization adopted for the final output images. The layers of this sky cube are then matched in pairs, following a sequence which maximizes image-to-image overlap within each pair. For each pair, the median of the image difference is subtracted from the second element of the pair. (Note that sources subtract away in this difference, because the images are aligned on the sky.) A median image is then formed by combining all of the images in the sky-equalized data cube. This median image is then projected ("blotted", in drizzle-speak) back to the input pixel coordinate frame for each individual input image. For each input image, a constant sky correction value is determined from the median of the difference between that image and the blotted median image. This constant is then subtracted, resulting in a highly uniform background level for all frames.

The final mosaics were drizzled onto an output grid of pixels with a uniform size of 1.200 arcsec/pixel, oriented on the sky according to the world coordinate standards defined for all GOODS data products (discussed in more detail below).

3.4 Exposure and weight map scaling

The output weight maps from the drizzling process used to construct the science mosaics have units of exposure time multiplied by the ratio of the drizzled pixel solid angle (exactly 1.44 square arcsec) to the original detector pixel solid angle (approximately 6.48 square arcsec). The exposure maps that are provided with this data release have been rescaled to remove this pixel solid angle ratio, so that they will represent the actual MIPS integration time at each point on the sky. Because of geometric distortion in the MIPS images, the original detector pixel solid angle varies slightly over the field of view, so a constant rescaling factor is not strictly correct. However, this difference is only a few percent at most over the field of view, and has been neglected here.

The MIPS images are essentially background limited, and therefore we would expect the RMS shot noise per pixel should be inversely proportional to the square root of the exposure time. We have conducted tests using split data sets to verify that this is indeed the case to a good degree of accuracy. We have therefore constructed noise maps (provided here as inverse variance images) by applying an empirical, constant multiplicative scaling factor to the exposure map images. To measure the intrinsic background shot noise, we first divided the list of input MIPS images into two independent halves which each uniformly sampled the overall dither pattern for the data set as a whole. We then drizzled each of these "half-datasets" together to produce independent images with virtually identical exposure times at all positions, and with all sources in the same pixel locations. By taking the difference of these images, all sources subtract away to a high degree of precision, leaving only the shot noise from the background plus sources themselves. The background shot noise dominates the total image noise except at the positions of the very brightest sources. The point kernel drizzling used here preserves the uncorrelated pixel statistics. Regions around sources were excluded in order to avoid the shot noise contribution from sources, as well as enhanced variance due to small image misalignments and PSF under-sampling issues. We measured the noise in the unmasked areas and computed the rescaling constant used to normalize the weight maps to match the measured (inverse) variance of the background.

It is important to note that the inverse variance maps represent only the shot noise component of the image noise at the sky background level of the images, i.e., the sky noise per pixel that would be present if there were no astronomical sources in the image. They do not include the Poisson shot noise from the sources themselves, nor any measure of photometric uncertainty due to image crowding or "confusion noise".

3.5 Flag maps

We have constructed flag maps that may be useful when making object catalogs and analyzing the GOODS MIPS images. These maps identify regions where there are and are not data in a given channel, and where the exposure time is low (i.e., around the edges).

The flag images are bit maps, i.e., integers that represent the sum of bit values, each of which indicates a different flag conditions. Table 3 describes the flag values, where the "bit number" starts at 0, and the "flag value" is the equivalent integer value for that bit setting. Bits not described in the table below are currently unused for flag settings.

Table 3 - GOODS MIPS flag map values

Bit number	Flag value	Condition
0	0	>50% of the modal exposure time
0	1	<50% of the modal exposure time
1	2	<20% of the modal exposure time
6	64	No data (zero retained exposure time)

These bit values will often appear in combination. For example, regions with < 20% of the modal exposure time (bit 1, flag value 2) also have < 50% of the modal exposure time (but 0, flag value 1). Therefore, those pixels will have flag values of 2 + 1 = 3. Regions with no data will have flag values 64 + 2 + 1 = 67. For reference, the "modal exposure time" used for setting the thresholds used in the flag maps is 37095 sec, or 10.304 hours.

Note that regions with Flag = 2 (i.e. < 20% of the typical exposure time) still have integration times up to 2 hours - hardly shallow by Spitzer/MIPS standards! However, those regions will have a fairly steep gradient in their exposure time and local noise amplitude.

4.1 File names

File names for these GOODS data products include the following components, separated by underscores ("_"):

As an example, the GOODS-N MIPS channel 1 superdeep epoch 1 science image (version 0.36) is named "n_mips_1_s1_v0.36_sci.fits".

Note the GOODS IRAC program includes both "superdeep" and "ultradeep" images, and that each IRAC data set was obtained in two separate observing epochs. The MIPS observations correspond only to the "superdeep" IRAC data (there is no separate "ultradeep" MIPS program), and were taken in only a single epoch (after the first IRAC epoch). Therefore, the "s1" designation (component 4 of the filename as outlined above) is formally superfluous, but has been retained for symmetry with the naming convention used for the GOODS IRAC data products.

• Complete list of FITS images in the GOODS interim data product release.

4.2 World coordinate system

All GOODS imaging data products are generated using a common scheme for world coordinates and pixel projection, which we briefly describe here. The images are projected on a tangent plane, with a the tangent point (CRVAL1,2) selected to be near the center of each field (GOODS-N, GOODS-S). They are aligned with north up (+y) and east left (-x). The pixel scales for GOODS imaging data products from different telescopes and instruments are always chosen to be integer multiples of one another. For the MIPS GOODS images, this scale is 1".200/pixel, which is approximately (but not exactly) half the native MIPS pixel scale. (Other scales for GOODS public-release data sets include 0".60/pixel for the IRAC data, 0".15/pixel for the ESO/VLT ISAAC CDF-S data, and 0".03/pixel for the HST/ACS Treasury Program images.) The pixel position (CRPIX1,2) that corresponds to the tangent point (CRVAL1,2) is always set to be a half-integer value. In this way, GOODS imaging data products from different telescopes and instruments can always be mapped to one another by simple integer rebinning, if desired.

Since the release of the GOODS HST/ACS v1.0 data products on 29 August 2003, we have found that the absolute astrometry for the GOODS-N field is slightly offset in declination from the reference frame defined by VLA 20 cm source positions from Richards (2000). This difference is approximately

dec(VLA) - dec(GOODS-N) = -0.38 arcsec

with negligible offset in right ascension. We have not yet adjusted the coordinate reference frame of the GOODS-N MIPS or IRAC data products to account for this, in order to keep the Spitzer and Hubble GOODS data products on the same world coordinate system and pixel grid. We intend to adjust the coordinate reference systems for both the Spitzer and Hubble GOODS data products in future releases.

4.3 Science images

The pixel intensities for GOODS MIPS data products are given in units of DN per second, derived from the original SSC BCD products (which have units of MJy/sr) using the FLUXCONV BCD header keyword (see section 3.2). The MIPS Data Handbook, section 3.7, discusses the flux conversion factors for MIPS as derived by the SSC. GOODS images have been drizzled to remove geometric distortion and thus have a constant pixel solid angle over the field of view. From the SSC-derived FLUXCONV factors and the original pixel solid angle at image center (6.483 square arcsec), we derive the conversion from instrumental to flux density units. We summarize this information in Table 4, providing the flux densities in micro-Janskys and the AB magnitudes that correspond to a count rate of 1 DN/sec. This information is also recorded in the image headers in the keywords FLUXCONV and MAGZERO (see section 4.7). For reference, we also list the detector gain (electrons/DN) in each channel; note that the effective gain for the GOODS mosaics (which are normalized to DN/sec) varies with the exposure time as a function of position.

Note that no adjustment for source color has been applied to the SSC-derived flux calibration (see the MIPS Data Handbook, section 3.7.4, for a discussion).

Table 4 - MIPS flux and magnitude conversion factors

Channel	Wavelength	FLUXCONV uJy/(DN/s)	MAGZERO AB for 1 DN/s	Detector gain (e/DN)
1	24 microns	6.691	21.836	5

4.4 Exposure maps

The exposure maps represent the MIPS integration time in seconds at each position on the sky in the co-added image mosaics, after rejection and masking of outlier pixels (e.g., cosmic rays, pixel defects, etc.). Fine-scale granularity from pixel to pixel in the exposure maps is a consequence of the process of drizzling the images onto a subsampled pixel grid using the point kernel, as described in section 3.4.

4.5 Weight (inverse variance) maps

The weight maps represent the inverse square of the RMS pixel-to-pixel noise (in DN/s) at the background level of the images. The construction of these maps is described in section 3.4. This represent the shot noise component due to the sky background and instrument noise only, and does not include Poisson noise from sources, nor any measure of photometric uncertainty due to source crowding or confusion.

4.6 Flag maps

The flag maps use bit values to flag regions of the images with and without data in a given channel, and with reduced exposure time. The flag map values are given in Table 3.

4.7 FITS headers

The FITS headers of the GOODS data product images incorporate various useful and relevant information about the images.