The Far-Infrared Deep Extragalactic Legacy Survey (FIDEL)
Spitzer Legacy Data Products
First Spitzer Data Release (DR1)

1.0 General information

2.0 Observations

2.1 Fields and program IDs
2.2 ECDFS epoch 1 MIPS observations and AORs


3.0 Data reduction

3.1 MIPS 70 micron data reduction

3.1.1 BCD pipeline processing and filtering
3.1.2 Image combination, calibration, and astrometry

3.2 MIPS 24 micron data reduction

3.2.1 BCD pipeline processing
3.2.2 Frame-level post-BCD processing
3.2.3 Astrometry, image registration, and image combination


4.0 Data products

4.1 File names
4.2 World coordinates
4.3 Science images and flux units
4.4 Exposure maps
4.5 Standard deviation images
4.6 FITS headers


Tables:

Table 1: Parameters for ECDFS epoch 1 MIPS observations
Table 2: Chronological AOR summary for ECDFS epoch 1 MIPS observations


Other information:

List of FITS images in the FIDEL interim data product release
Example of FIDEL MIPS data product FITS header

Figures:

Figure 1: ECDFS epoch 1 MIPS field layout

1.0 General Information

This document describes the first release of data products from the Far-Infrared Deep Extragalactic Legacy Survey (FIDEL), a Spitzer Space Telescope Legacy Science program (PI: Mark Dickinson, NOAO).

This data release (DR1) consists of "best-effort" reductions of the first epoch of 24 and 70 micron data for the Extended Chandra Deep Field South (ECDFS) with the Multiband Imaging Photometry for Spitzer (MIPS, Rieke et al. 2004). These products are being delivered to the Spitzer Science Center (SSC) for public release in June 2007. The second epoch of ECDFS observations will be presented in the second FIDEL data release (DR2), in September 2007.

The imaging data products are described in detail below, and consist of a science image mosaic plus associated exposure and noise maps.

2.0 Observations

Here, we provide a brief description of the FIDEL MIPS observations.

2.1 Program information and survey fields:

FIDEL is Spitzer program ID number 30948, and was originally submitted with the title "A Deep-Wide Far-Infrared Survey of Cosmological Star Formation and AGN Activity."

The program is obtaining data in three fields on the sky. The bulk of the data is being taken in two fields, the Extended Chandra Deep Field South (ECDFS) and the Extended Groth Strip (EGS). A smaller amount of additional data is being obtained in the GOODS-North area, in order to augment partial 70 micron coverage of that field from a GO-1 program (PID 3325) by Frayer et al. (2006b, ApJ, 647, L9).

The bulk of the ECDFS MIPS observations have being carried out in two epochs, separated by approximately six months (September 2006 and March 2007). This allows the Spitzer telescope orientation to rotate by 180 degrees, ensuring reasonably symmetric coverage in the MIPS 24 and 70 micron bandpasses. Data release DR1 consists of the first epoch observations for the ECDFS.

Spitzer MIPS observations are now organized into "warm" and "cold" campaigns, in order to maximize the cryogenic lifetime of the telescope. "Cold" campaigns result in greater cryogen usage, and provide a lower background to allow 160 micron observations to be taken. The ECDFS MIPS observations in epoch 1 were taken during a "warm" MIPS campaign, and hence only 24 and 70 micron data were obtained. Additional MIPS scanning observations of the ECDFS were taken during cold campaigns in January and March 2007, and will provide 160 micron coverage for FIDEL.

Generically, we consider the ECDFS to cover a square region approximately 30'x30' on the sky, oriented along the J2000 celestial axes. The target coordinates (roughly, the center of the ECDFS itself) and observing dates for the first-epoch MIPS observations are summarized in Table 1.

2.2 ECDFS epoch 1 MIPS observations and AORs

The design of the AORs for the FIDEL MIPS observations is somewhat complex, and we will only give a high-level description here. Please refer to the Spitzer Observer's Manual for detailed explanations of the various observing modes and terminology. This first release consists of data taken during the first observing epoch for the ECDFS.

The observations for were broken into a series of Astronomical Observation Requests (AORs), each several hours long, that were designed to enable efficient scheduling. For the first epoch of ECDFS observations, we had 27 AORs, falling into two basic types. The majority (24/27) of the AORs were taken with the 70 micron array as the prime instrument (henceforth, "70-prime"), with the observing strategy optimized for that wavelength. These AORs have TARGET=ECDFS. Three other AORs were taken with the 24 micron array as the prime instrument ("24-prime"). These AORs have TARGET=ECDFS-mips24, and were used to fill gaps in the 24 micron areal coverage, for reasons described in more detail below. A detailed chronology of the AORs is given in Table 2.

Table 1 - Parameters for ECDFS epoch 1 MIPS observations



Parameter	ECDFS
Spitzer program ID	30948
RA (J2000)	03:32:30.37
Dec (J2000)	-27:48:19.3
Start date/time	2006-09-01 18:18:45
End date/time	2006-09-05 16:11:25

Table 2 - Chronological AOR summary for ECDFS epoch 1 MIPS observations



All ECDFS epoch 1 MIPS observations were using the photometry-mode Astronomical Observing Templates (AOTs), which are described in more detail in section 8.2.1.2 of the Spitzer Observer's Manual. In this mode, the field of view is dithered using a combination of telescope and scan mirror offsets around a particular pointing position. The 70-prime AORs used the 70 micron compact source (small field) mode, while the 24-prime AORs used the 24 micron large source (large field) mode.

The 70-prime AORs observed pointing positions that covered roughly the bottom two-thirds of the 30'x30' ECDFS. The parallel 24 micron data taken at the same time covered roughly the upper two thirds of the field. The 70 micron pointings were carefully placed to avoid overlap with previous, deep 70 micron observations taken in a GO-2 program (PID 20147, PI: D. Frayer) which observed roughly the central 10'x10' of the GOODS-S field. This leads to the "gap" in the U-shaped 70 micron coverage seen in Figure 1. The first epoch 70 micron observations have their maximum exposure time along the bottom third of the ECDFS, and roughly half that exposure time in the left and right "legs" in the middle third. For the second epoch observations (taken in March 2007), this pattern is rotated by 180 degrees, covering the top third of the field and adding more exposure time in the middle portion. The central hole on GOODS-S will eventually be filled with the (currently proprietary) data from the Frayer et al. GO program 20147, and a complete, combined image (also incorporating additional scanning mode data from FIDEL and from previous GTO programs) will be released as a "version 1.0" data product circa June 2008.

The MIPS observations were optimized to achieve roughly uniform exposure time at 70 microns when the epoch 1 + 2 observations are combined. This, however, leads to non-uniformity in the depth of the parallel 24 micron observations. By avoiding the central field targeted by the Frayer 70 micron GO-2 program, we also introduce some gaps in the 24 micron coverage. These, as well as some "protruding corners", were filled by the additional 24-prime AORs, which were taken in epoch 1 (only), and are thus included in the data products that we release here.

Figure 1 illustrates the layout of the ECDFS epoch 1 MIPS observations, showing the images and the relative exposure time maps at 24 and 70 microns. In this DR1 data set, at 70 microns, the deeper regions of the image have net exposure time approximately 4800s, while the shallower regions have about 2400s. At 24 microns, the exposure time varies more. The deeper region has typical net exposure time approximately 12600s. Shallower regions have exposure times approx. 6300s, while typical exposure times int regions covered by the 24-prime "filler" AORs range from 2000s to 3400s. The exposure times for the ultimate FIDEL data products will be longer in all areas when the second epoch data are included, as well as scanning mode data (which cover most of the ECDFS area), both from both FIDEL and from previous GTO observing programs.


Figure 1: ECDFS epoch 1 MIPS field layout
The 24 and 70 micron images are shown at left, and their corresponding exposure maps are shown at right.
The red box is the same in all panels, and has dimensions 30'x30'.

Here, we briefly describe the reduction of the FIDEL MIPS data and the construction of the data products.

3.1 MIPS 70 micron data reduction
3.1.1 BCD pipeline processing and filtering

The S14.4 on-line filtered BCD (fbcd) products were downloaded from the Spitzer archive to make a quick-look image and an initial source list. The raw data were then reduced from scratch using the Ge Reprocessing Tools (GeRT, version 060415 [2006 April 15]). The GeRT BCD processing was done with the best set of optimized pipeline parameters derived previously from deep 70 micron photometry data of GOODS-North (see Frayer et al. 2006b).

Filtering is crucial for deep 70 micron photometry data. The calibration stimulator (stim) flashes are used every 6 data collection events (DCEs), and latents due to these stim flashes accumulate over time. These latents correlate with column. To remove stim flash latents, we used a median column filter. After column-filtering, the residual drifts of the detectors as a function of time were removed by the application of a median high-pass time filter per detector (with a filter width of 16 DCEs). This combination of filters has been shown to give the best sensitivity for deep photometry data. The positions of bright sources in the BCDs were flagged so that the filtering corrections were not biased by the presence of sources. This has been shown to maintain point-source photometry while removing the "side-lobe" artifacts seen in the on-line filtered products (see Frayer et al. 2006a, AJ, 131, 250). The median filtering techniques yield small offsets from zero in the average level of the filtered-BCDs. These offsets correlate with the DCE position within the stim cycle and were removed by subtracting the median level from each fbcd. With offline re-processing of the data, we have improved the sensitivity of the products by more than 20% over the on-line filtered products.

3.1.2 Image combination and astrometry

The data were coadded onto a sky grid with a scale of 4.0 arcsec/pixel using the MOPEX mosaicing and source extraction software (version 030106). The astrometry for the 70 micron data uses the default pointing solutions from telescope telemetry without further correction. Typical uncertainties in the Spitzer pointing solution from AOR to AOR are of order 1 to 1.5 arcsec, which is small compared to the 70 micron beam size and pixel scales. We did refine the 24 micron pointing solution (see section 3.2.3), and may apply these refinements to the 70 micron data as well in future re-reductions.

3.2 MIPS 24 micron data reduction
3.2.1 BCD pipeline processing

The epoch 1 ECDFS 24 micron MIPS data reductions described here begin with the products generated by the SSC Basic Calibrated Data (BCD) pipeline version S14.4.0.

3.2.2 Frame-level post-BCD processing

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

Create masks for latent images: Bright sources observed with MIPS can produce latent images that persist into subsequent exposures. This often happens when Spitzer observes relatively bright stellar targets before or in between FIDEL AORs. We took the median of all 24 micron exposures per AOR and compared them to one another in the time order of their observing sequence. Time-varying latent images, typically at the 0.5% level, were apparent in some AORs. We produced masks to set affected pixels to zero weight when combining the data into image mosaics.

Scan mirror delta-flatfields: A median of all data (combining all frames from all AORs in each group, i.e., 70-prime or 24-prime) show that the BCD frames have residual flatfielding artifacts at the 1% level. These artifacts include some variability with scan mirror position, suggesting that the pipeline spot-map corrections are not quite perfect. Additionally, the default scan mirror position (CSM=2007.) shows large scale features (in the 70-prime data only) which are not evident in either the other four scan mirror position medians or in any of the 24-prime data.

We created delta flats for each of the five scan mirror positions and divided these into the data. This was done separately for the 24-prime and 70-prime images. The frames affected by the latent artifacts are masked before creating the delta flats, and all low-number data collection events (DCEs) visibly affected by the bias boost (see below) are excluded from the delta-flat medianing. The corrections are applied to the unmasked frames, including the low-number DCEs.

Remove cross-scan background gradient from low-number DCEs: Bias boosts occur at the start of every sequence of data collection events (DCEs) indicated by a unique EXPID number. We created median images from all observations grouped by DCE number (DCENUM) to look for any residual structure correlated with time since the bias boost. The first frame in each sequence (DCENUM=0000) shows large-scale residual features after division by the scan mirror delta-flatfields. While most of these features have disappeared by DCENUM=0001, the mean background level of the low-number DCEs (with DCENUM approximately less than 0010) remains depressed relative to the high-number DCEs (with DCENUM > 0015), with a 0.2-0.3% gradient in the cross-scan direction across the chip. These effects are more obvious in the shorter, 10-second 70-prime frames, and tend to fade as the DCE number increases.

For the DCENUM=0000 frames, we used the median image as a tertiary flat-field correction. In the 70-prime data, for 0001 <= DCENUM <= 0013, we fit a low-order function along the row axis of the median images, and divided that shape into the individual frames with DCENUM=0001 to 0013. For the 24-prime data, the individual DCEs have longer exposure times (30s instead of 10s), and these DCE-dependent effects seem to be greatly reduced. There, we applied corrections only to the DCE=0000 and 0001 frames.

Remove AOR-to-AOR flatfield differences: In some data sets, after the above steps were performed, large scale, low-level patterns were still seen, varying from AOR to AOR. These were not prominent in the ECDFS epoch 1 data considered here, but were seen in other data sets. We interpreted these as larger-scale latent image patterns. In order to remove these, we constructed new median images of the dithered data within each AOR, and then subtracted these from each individual image within that AOR. For uniformity, we applied this step to all 24 micron data taken in 70-prime mode, regardless of whether a clear residual pattern was seen. This correction was only applied to the 70-um prime photometry-mode data; the relative small number of independent dither points per AOR in the 24-prime observations left us with insufficient spatial sampling to construct reliable median images per AOR without effects of source contamination which might subsequently affect photometry.

Jailbar patterns: Low level jailbar patterns (vertical striping associated with small bias offsets in the four interleaved readout amplifiers) could be seen in the DCE=0000 images. However, these were effectively removed by the multiplicative corrections described above. (The jailbar correction should probably be additive, not multiplicative, but the amplitude is so small (0.1-0.2%) that this should have very little impact.) No other jailbar patterns were visible in any other data sets examined, including with very strong cosmic ray hits (some jailbars in such data had been observed during the GOODS data reduction). Therefore, in the end, no jailbar corrections were applied to the the ECDFS epoch 1 24 micron data.

Sky subtraction: Cosmic rays and bad pixels were masked in all frames during the MOPEX combination step (see below). After a first pass run of MOPEX, these pixel defect masks were saved, and then used to discard pixels when calculating a median sky background level (iteratively sigma-clipped) for each individual science image. These median values were subtracted from the data prior to (re)mosaicing.

3.2.3 Astrometry, image registration, and image combination

Image combination for the FIDEL MIPS data was carried out using the MOPEX software (version 030106) provided by the Spitzer Science Center.

First, we created preliminary 24 micron mosaics for each AOR. MOPEX aligns the images on the assumption that the world coordinate information in the BCD image headers, which are generated based on telescope telemetry, are correct. Cosmic rays and other image defects were rejected as outliers in the stack of repeated observations at a given sky position. MOPEX generates masks of these outliers, which were checked to ensure reasonable behavior.

For each AOR-mosaic, we generated a catalog of 24 micron sources and matched this to the positions of ECDFS IRAC sources from a preliminary catalog (v1.0) based on the data from SIMPLE, a GO-2 Spitzer Legacy program (PID 20708; PI: P. van Dokkum). External checks indicate that the SIMPLE astrometry (which is tied to that from the MUSYC project) agrees with that for the GOODS Spitzer and HST data within approx. 0.1 arcsec on average, although there may be local residuals that are somewhat larger.

We computed an iteratively sigma-clipped mean astrometric offset between the MIPS 24 micron positions and the IRAC positions. We also visually inspected vector fields of the astrometric residuals looking for other systematic trends, and found none of significance. The astrometric offsets were applied to the world coordinate information (CRVAL1,2) in the headers of each image within that AOR. The process was repeated for each 24 micron AOR.

The astrometrically corrected images were then re-mosaiced, combining data from all AORs. The 70-prime and 24-prime observations were combined separately with MOPEX because they have different exposure times per image. However, the same "fiducial image frame" was adopted for the output mosaics to ensure that they would both be projected onto the same pixel grid. MOPEX produces "coverage maps" showing the number of input images that contribute to a given pixel position in the combined mosaic. We rescaled these to approximate exposure maps by multiplying by the exposure time per DCE (10 and 30 seconds, respectively, for the 70-prime and 24-prime data sets), and then combined the 70-prime and 24-prime sub-mosaics, weighting by their respective exposure maps.

The final image mosaics were produced on a tangent plane projection, with a scale of 1.200 arcsec/pixel and with the pixel axes aligned with the J2000 celestial coordinate axes.

4.1 File names

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

1) "fidel" prefix
2) Field (here, "ecdfs")
2) Bandpass (here, "24" or "70")
3) Data set and epoch (here, "p1", for photometry-mode, epoch 1)
4) Data version number
5) Image type ("sci" = science image; "exp" = exposure map; "std" = standard deviation image)


As an example, the FIDEL ECDFS 70 micron epoch 1 (photometry mode) version 0.31 science image is named "fidel_ecdfs_70_p1_v0.31_sci.fits". The version numbers are based on FIDEL internal nomenclature, and have no particular significance. For this release (DR1), the version numbers for the 70 and 24 micron images are v0.31 and v0.30, respectively.

• List of FITS images in FIDEL data release 1.

4.2 World coordinate system

The 70 and 24 micron ECDFS images use a tangent plane sky projection, with scales of 4.0 and 1.20 arcsec/pixel, respectively, and pixel axes aligned with the J2000 celestial axes. The WCS tangent points (CRVAL1,2) of the 70 micron and 24 micron images are not the same in the current reductions. Therefore, a transformation (beyond simple linear translation and resampling) would be needed in order to achieve accurate pixel alignment from 24 to 70 microns. However, the world coordinates of the two images should agree within the practical measurement uncertainties for source positions.

4.3 Science images and flux units

The pixel intensities for FIDEL ECDFS epoch 1 science images (*_sci.fits) are in units of MJy/sr, as per the SSC BCD data products. Standard calibration conversion factors (FLUXCONV) from instrumental units to MJy/sr were used: 702 MJy/sr per MIPS70-unit, with an absolute uncertainty of 7%, and 0.0447 MJy/sr per 24 micron DN/s, with an absolute uncertainty of 4% (Gordon et al. 2007, submitted). Pixels in areas where there are no data are set to zero.

Multiplying the data units (MJy/sr) by the the pixel solid angles for the reprojected MIPS images, the effective "zeropoints" to convert data units (N70, N24) in the FIDEL data products (e.g., from a source, summed over some aperture) to flux densities (f70, f24) in millijanskies (mJy) are:

f70 (mJy) = 0.37607 * N70
f24 (mJy) = 0.03385 * N24

No color-corrections have been applied to the flux values. The MIPS flux conversion factors are calibrated for 10000K stellar spectral energy distributions. Galaxies, which can have significantly different spectral energy distributions, may require a multiplicative color correction term. This is discussed in the MIPS Data Handbook section 3.7.4 (version 3.2.1), where color correction factors are tabulated. At 70 microns, typical color correction factors for galaxies are of order 1.07 to 1.09, depending on the shape of the SED. The correction factors are generally smaller at 24 microns.

[We note that the units for the FIDEL 24 micron science image data products are different than those adopted for the GOODS 24 micron data reductions, e.g. for the GOODS-S field, which is embedded within the ECDFS. The GOODS team converted the MIPS BCD data units back to DN/sec.]

Due to the interpolation used by MOPEX when subsampling the MIPS pixels to produce the mosaiced images, data values in adjacent pixels of the science images are strongly correlated with one another. [Again, this is unlike the case for the GOODS MIPS data products, where point kernel drizzling resulted in uncorrelated sky pixels. We will investigate such drizzling for future FIDEL versions of the FIDEL data products.] This correlation leads to an apparent suppression of the background noise, effectively a local smoothing. Therefore (as in many astronomical data sets), one should use caution when interpreting the measured RMS noise of the sky background. The standard deviation images (see section 4.5) provide an estimate of the local image noise, but one that should also be considered cautiously, as discussed below.

4.4 Exposure maps

The exposure maps (*_exp.fits) represent the approximate total MIPS integration time in seconds at each position on the sky in the co-added image mosaics. The MOPEX mosaicing software produces "coverage" maps, which represent the number of individual exposures which contribute to each pixel in the output image mosaic, and we have multiplied these by the approximate mean exposure time per data collection event (DCE) to produce the exposure maps. The precise exposure time per pixel is likely to be slightly different (probably smaller), due to rejection of outlying values such as cosmic rays, array defects, etc.

4.5 Standard deviation images

We provide standard deviation maps (*_std.fits), which are an output product from MOPEX. At each pixel position in the output mosaic, these maps are computed from the standard deviation of the individual pixel values from images that contribute to that mosaic position, divided by the square root of the number of exposures taken at that position. The units of these images are MJy/sr, as for the science images. Regions of the images where there are no data are set to 0 in the standard deviation images.

We caution that these standard deviation maps should be regarded as indicative only, and regarded with some skepticism! They assume a certain noise model, namely, that the shot noise for N exposures decreases as 1/sqrt(N), and that no other systematics contribute to the image-to-image standard deviation that is measured. Although careful studies of noise properties in MIPS data have shown that indeed the noise does reduce roughly as 1/sqrt(t) (see, e.g., Frayer et al. 2006b), this scaling is not precise and in particular has not been validated for these particular FIDEL data. The actual image noise that affects photometry may also have contributions from source blending and confusion, which is not taken into account in these standard deviation maps. Finally, sources can appear as "rings" of enhanced RMS in the maps, primarily due to small variations in image alignment or due to pixel undersampling, such that regions of strong image gradients (i.e., near the FWHM point of a point source) lead to enhanced RMS.

4.6 FITS headers

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

• Example of FIDEL MIPS data product FITS header