Spitzer 70 and 160-micron Observations of the COSMOS Field
ABSTRACT We present Spitzer 70 and 160 micron observations of the COSMOS Spitzer survey (S-COSMOS). The data processing techniques are discussed for the publicly released products consisting of images and source catalogs. We present accurate 70 and 160 micron source counts of the COSMOS field and find reasonable agreement with measurements in other fields and with model predictions. The previously reported counts for GOODS-North and the extragalactic First Look Survey are updated with the latest calibration, and counts are measured based on the large area SWIRE survey to constrain the bright source counts. We measure an extragalactic confusion noise level of sigma_c = 9.4+/-3.3 mJy (q=5) for the MIPS 160-micron band based on the deep S-COSMOS data and report an updated confusion noise level of sigma_c = 0.35+/-0.15 mJy (q=5) for the MIPS 70-micron band. Comment: Accepted AJ, 15 Aug. 2009. Data available at http://spider.ipac.caltech.edu/staff/frayer/mycosmos/ until released by IRSA
Accepted AJ, 2009 August 15
Spitzer 70 and 160 µm Observations of the COSMOS Field
D. T. Frayer1, D. B. Sanders2, J. A. Surace3, H. Aussel4, M. Salvato5, E. Le Floc’h2, M. T.
Huynh1, N. Z. Scoville5, A. Afonso-Luis6, B. Bhattacharya1, P. Capak3, D. Fadda1, H. Fu5, G.
Helou1, O. Ilbert2, J. S. Kartaltepe2, A. M. Koekemoer7, N. Lee2, E. Murphy3, M. T. Sargent8,
E. Schinnerer8, K. Sheth3, P. L. Shopbell5, D. L. Shupe1, L. Yan3
We present Spitzer 70 and 160 µm observations of the COSMOS Spitzer survey
(S-COSMOS). The data processing techniques are discussed for the publicly released
products consisting of images and source catalogs. We present accurate 70 and 160 µm
source counts of the COSMOS field and find reasonable agreement with measurements
in other fields and with model predictions. The previously reported counts for GOODS-
North and the extragalactic First Look Survey are updated with the latest calibration,
and counts are measured based on the large area SWIRE survey to constrain the bright
source counts. We measure an extragalactic confusion noise level of σc= 9.4 ± 3.3mJy
(q = 5) for the MIPS 160 µm band based on the deep S-COSMOS data and report an
updated confusion noise level of σc= 0.35±0.15mJy (q = 5) for the MIPS 70 µm band.
Subject headings: galaxies: evolution — galaxies: starburst — infrared: galaxies
The Cosmic Evolution Survey (COSMOS) is a deep multi-wavelength wide-area (2 deg2) pro-
gram for studying the evolution of galaxies and active galactic nuclei (AGN) (Scoville et al. 2007).
1Infrared Processing and Analysis Center, California Institute of Technology 100-22, Pasadena, CA 91125, USA
2Institute for Astronomy, 2680 Woodlawn Drive, University of Hawaii, Honolulu, HI 96822, USA
3Spitzer Science Center, California Institute of Technology 220–06, Pasadena, CA 91125, USA
4CNRS, AIM-Unit´ e Mixte de Recherche CEA-CNRS-Universit´ e Paris VII-UMR 7158, F-91191 Gif-sur-Yvette,
5Astronomy Department, California Institute of Technology 105–24, Pasadena, CA 91125, USA
6Instituto Astrofiscia de Canarias, Via Lactea, 38200 La Laguna, S/C de Tenerife, Spain
7Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218, USA
8Max-Planck-Institut f¨ ur Astronomie, K¨ onigstuhl 17, D-69117 Heidelberg, Germany
arXiv:0902.3273v2 [astro-ph.CO] 25 Aug 2009
– 2 –
The COSMOS Spitzer (S-COSMOS) survey is comprised of the Infrared Array Camera (IRAC, 3.6–
8µm) and Multiband Imaging Photometer for Spitzer (MIPS, 24, 70, and 160 µm) data (Sanders et
al. 2007). This paper presents the far-infrared (FIR) 70 and 160 µm MIPS observations of the field.
Although the mid-infrared (MIR) 24 µm array is more sensitive to the detection of distant galaxies
than the MIPS-Germanium (MIPS-Ge) 70 and 160 µm detectors, the 24 µm data are biased toward
warm AGN and are affected by broad mid-infrared PAH emission and silicate absorption features
redshifted into the band. The strong MIR spectral features along with the large variations of the
FIR/MIR continuum ratios (e.g., Dale et al. 2005) yield highly uncertain bolometric corrections.
The long-wavelength 70 and 160 µm observations directly measure the FIR peak of the spectral-
energy distributions (SEDs) for redshifts z ? 1.5 and are key for constraining the total infrared
luminosities and star-formation rates of galaxies within the COSMOS field. The MIPS 70 and 160
µm data provide an important piece of the puzzle in the quest of understanding galaxy evolution.
The goal of the data paper is to document the data products to facilitate the ongoing research
of the COSMOS field. The data products and a description of the observations and data reduction
are provided as part of the large public repository of multi-wavelength data for the COSMOS
field (Hubble Space Telescope Advanced Camera for Surveys (ACS), Koekemoer et al. 2007; radio,
Schinnerer et al. 2007; X-ray, Hasinger et al. 2007; and optical and near-infrared, Capak et al.
2007). We present two scientific results here: (1) the 70 and 160 µm source counts, and (2) the
measurement of the confusion noise for the MIPS 160 µm band.
The MIPS S-COSMOS observations were carried out in 4 campaigns from 2006 January
through 2008 January (Table 1).The project (Spitzer programs 20070 and 30143) represents
over 450hr of MIPS observations. All observations were taken using the scan mapping mode dur-
ing nominal “Cold” MIPS campaigns (telescope temperatures low enough to yield good quality
160 µm data). The initial observations taken in 2006 Cycle-2 are described in Sanders et al. (2007).
The Cycle-2 observations were comprised of a shallow wide-area (1.75deg×1.97deg) survey to
quantify the level of cirrus within the field and a small (0.5deg×0.33deg) deep “test” field. After
the successful completion of the Cycle-2 program, deep observations over the entire COSMOS field
were carried out in Cycle-3 (depth of about 2800s, 1350s, and 270s in the MIPS 24, 70, and 160 µm
The Cycle-3 astronomical observational requests (AORs) were optimized specifically for the
MIPS-Ge bands, without compromising the 24 µm data. In contrast, some early MIPS programs of
other groups were designed for the 24 µm band at the expense of the MIPS-Ge bands. The forward
and return scan legs were offset by 148??which provides sufficient overlap for the 70 µm array.
Cross-scan dither offsets of ±0, 42, 83??and in-scan dither offsets of ±0, 18??were used between
multiple maps to account for the unusable parts of the MIPS-Ge arrays and the unobserved central
row of the 160 µm footprint. The values of the cross-scan dithers were also chosen to avoid the
– 3 –
overlap of 24 µm readouts between consecutive maps. We carried out both forward and reverse
scan maps to help characterize the long-term transients of the MIPS-Ge detectors.
The majority of the data were taken in slow scan mode. At the slow scan speed (2.6??s−1),
each AOR consisted of 4 scan legs of 1.5deg. Each AOR mapped 1.5deg×592??, and 10 AORs were
used to map the field once. In total, 13 slow maps (29.5hr per map) were carried out in Cycle-3,
along with one map at the medium scan rate (12.5hr) to complete the awarded time. Five AORs
were lost due to satellite downlink issues in the second epoch of Cycle-3 and were re-observed in
early 2008 (Cycle-3c, Table 1).
The scan direction of MIPS is determined by the date of observation, and the observations
were carried out on the days that minimized the zodiacal light. Since the field is near the ecliptic
plane, the zodiacal background contributes significantly to the total noise budget for the MIPS
24 and 70 µm bands. The zodiacal light is not significant at 160 µm. The galactic cirrus level is
low in the direction of the COSMOS field (Sanders et al. 2007) and is not the dominant source of
confusion noise within the MIPS-Ge bands. Figures 1 and 2 show the final MIPS 70 and 160 µm
images, combining all epochs (Table 1). Since the 70 µm and 160 µm arrays are on the opposite
side of the MIPS field of view, the overscan regions yield slightly non-symmetric coverage. For both
Cycle-3a and Cycle-3b, the entire ACS field was observed, and the MIPS over-scan regions provide
coverage for the IRAC data outside of the ACS field.
3. Data Reduction
The raw MIPS-Germanium 70 and 160 µm (MIPS-Ge) data were downloaded from the Spitzer
Science Center (SSC) archive and were reduced using the Germanium Reprocessing Tools (GeRT,
version 20060415) and additional specialized scripts developed for processing the MIPS-Ge survey
data. The GeRT uses an offline version of the SSC pipeline to produce the basic calibrated data
products (BCDs). The basic MIPS-Ge processing steps are discussed by Gordon et al. (2005)
and within the GeRT documentation. The processing for S-COSMOS made use of lessons learned
from the processing of the extragalactic First Look Survey (xFLS, Frayer et al. 2006a) and the
deep GOODS-North observations (Frayer et al. 2006b). Additional processing enhancements were
derived here using the S-COSMOS data.
The main processing steps were carried out in the following order: (1) calculation of the
data ramp slope, (2) stimulator-flash interpolation, (3) improved stimulator-flash response solution,
(4) calibration of the slope image to yield the BCD product, (5) enhanced filtering of the BCD
product, (6) data co-addition, (7) identification of bright sources, (8) masking bright sources and
re-calculation of the filtering corrections, (9) final data co-addition, and (10) final source extraction.
For comparison, the on-line SSC pipeline only performs steps (1), (2), (4), basic filtering, and step
(6), and the GeRT includes software for steps (1), (2), (4), and (8). The 10 processing steps are
summarized in the following sub-sections.
– 4 –
3.1. Basic Processing
The optimal processing for step (1) depends on the background, the length of the data ramp
(MIPS-Ge data are recorded with non-destructive reads sampled at 0.131 s), and the rate of cosmic
rays at the time of the observations. The SSC on-line pipeline is tuned for the short data ramps (3s
and 4s). We tuned the pipeline modules (cosmic ray detection and removal and slope estimation)
of step (1) to minimize the noise level for the longer 10s ramps of the S-COSMOS data. The tuning
solutions for 70 µm are similar to those derived for the GOODS-North photometry data which have
the same ramp length (Frayer et al. 2006b). MIPS-Ge uses stimulator-flash observations to track the
response of each detector as a function of time. After the calculation of the initial stimulator-flash
solution, which is basically a linear interpolation between stimulator-flash measurements (step 2),
we removed outlier values and re-derived a smoothed stimulator-flash solution (step 3). For 70 µm,
the stimulator-flash response function was smoothed by about 2 minutes (slightly longer than the
stimulator-flash cycle) to provide the lowest noise. For 160 µm, the solution was smoothed by
about 8 minutes to yield the best results.
In step 4, the BCD data are calibrated as BCD(t) = FC[U(t)/SR(t) − DARK]/IC, where
U(t) is the uncalibrated slope image, SR(t) is the stimulator-flash response solution derived in step
(3), and DARK is the dark calibration file. The IC calibration file is the illumination correction
which corrects for the flat-field response and the non-uniformity of the stimulator flash (Gordon
et al. 2005). The flux conversion factor (FC) converts the instrument units into physical surface
brightness units of MJysr−1. For self-consistency, we adopt the same DARK and IC files used
for the official calibration of the MIPS instrument (Engelbracht et al. 2007; Gordon et al. 2007;
Stansberry et al. 2007). The calibration FC values of 702 MJysr−1per MIPS-70-unit (Gordon et al.
2007) and 41.7 MJysr−1per MIPS-160-unit (Stansberry et al. 2007) are adopted. The calibration
of MIPS is based on stellar SEDs (Sν∝ ν2). We have placed the data on a constant ν Sνscale by
dividing the data by the color correction factors of 0.918 and 0.959 for the 70 and 160 µm bands,
respectively (Stansberry et al. 2007). These color corrections are appropriate (accurate to better
than 2%) for a wide range of galaxy and AGN SEDs (Sν∝ ν−α, α = 0–3) across the filter bandpass.
3.2. Data Filtering
Optimization of the processing steps (1-4) provided sensitivity improvements of about 20%
in comparison to the default parameters, while the filtering (step 5) can provide more than a
factor of two improvement in point-source sensitivity. Filtering is key in the removal of systematic
instrumental effects which impede the ability to integrate down with deep observations. The two
main artifacts impacting MIPS-Ge data are the latents due to the stimulator flashes and variations
of the slow response (> 2min) as a function of time. The slow response is removed at 70 and
160 µm by subtracting a running median per detector as a function of time, i.e., a high-pass filter.
The latent artifacts due to the stimulator flashes are not fully removed by a simple high-pass filter,
– 5 –
since these variations occur on shorter time-scales. At 70 µm the stimulator-flash artifacts are
correlated by column. Since the scan direction is nearly along the columns of the array, these
artifacts contribute to the streaking within the maps if not corrected. We remove the column
residuals by subtracting the median of the values along each column for every BCD at 70 µm. The
combination of the column median filter and a high-pass median time filter per detector removes
the instrumental artifacts at 70 µm.
There is no equivalent column filter at 160 µm to remove the high-frequency (short-time
scale) latent images introduced by the stimulator-flashes. Fortunately at 160 µm, these artifacts
are repeatable and can be determined by stacking the data as a function of DCENUM (Data-
Collection-Event Number) within the stimulator-flash cycle. Since the scan-mirror position also
varies with DCENUM within the stimulator-flash cycle, we stacked the data per AOR for each
scan-mirror position and took the median value of the stack to derive the correction as a function
of detector and scan-mirror position. These corrections were subtracted from the BCDs to remove
artifacts due to the stimulator-flashes. The combination of this stacking correction and a high-pass
median time filter removes the instrumental artifacts at 160 µm.
With the S-COSMOS data, we tuned the filtering techniques (step 5) to minimize the noise
in the output maps. For both 70 and 160 µm, the short-time scale stimulator-flash artifacts were
removed first, followed by a high-pass median time filter to remove the longer time-scale transients.
We adopted a time filter width of 12 frames (2 minutes) to yield the best results.
initial filtering, the brightest sources have negative “side-lobes” in the maps since bright sources
bias the calculation of the median for neighboring pixels. To remove these filtering artifacts, the
filtering was done in two passes. The data from the first filtering pass (step 5) were co-added
(step 6), and sources were extracted (step 7) to find the location of the bright sources. The source
positions within the original BCDs were masked and new filtering corrections were calculated in
a second pass, ignoring the pixels containing sources. This two-pass filtering technique minimizes
the artifacts while preserving the point-source calibration. After the second filtering pass (step 8),
the data were co-added to produce the final maps (step 9, Sec. 3.3) and sources were extracted to
produce the catalogs (step 10, Sec. 3.4).
The SSC mosaicking software (MOPEX, Version 16.3.7, Makovoz & Marleau 2005) was used
to combine the data and make the final images. A fast plane-to-plane coordinate transformation
method was used to project the data onto the sky (Makovoz 2004) with the default MOPEX
interpolation scheme. We carried out the imaging steps following the techniques discussed for the
xFLS MIPS-Ge data (Frayer et al. 2006a). An important improvement available after the processing
of the xFLS data is a more robust outlier rejection technique within MOPEX. The updated method
rejects data around the median of a data stack for each sky pixel instead of rejecting data with
respect to the average of the data stack. This enables more aggressive outlier rejection without
– 6 –
compromising the calibration of point sources. The best sensitivity was obtained using rejection
thresholds of ±2.5σ. The new outlier rejection method improved the sensitivity in the maps by
The pipeline uncertainties of the BCDs and integration times were not used as weights in the
co-addition of the data; all of the data (155,411 BCDs in total) flagged as good were given equal
weight. Bad data flagged on a detector basis during the pipeline processing or identified as outliers
by MOPEX were not included. The vast majority of the data were taken with 10s integrations
and no correction is needed for the 4s BCDs (all BCDs are calibrated correctly in MJys−1). The
pipeline uncertainties (calculated from the formal error propagation of the pipeline steps) do not
fully represent the actual noise characteristics of the data, and underestimate the real noise slightly
in low background regions. The main utility of the pipeline uncertainties for these data is to provide
a lower limit to the input noise for the MOPEX outlier rejection algorithm.
3.4. Source Detection and Extraction
Sources were detected and extracted from the final images using the Astronomical Point-Source
Extraction (APEX) tools within the MOPEX software package and using additional specialized
scripts. For optimal source detection and extraction, accurate background subtraction and noise
estimates are needed. The filtering process (steps 5&8) yields a small systematic negative offset of
the background level in the image. This offset was estimated by taking the median of the image
within the central regions after masking sources detected at levels greater than 3σ. After the
removal of the global offset level (−0.05 MJysr−1at 70 µm and −0.07 MJysr−1at 160 µm), there
are still local background fluctuations across the image depending on the local density of sources
(both from detected sources and un-detected/confused sources associated with infrared galaxies
seen at 24 µm). The local background level was derived by taking a median within a box around
each pixel (after masking sources detected at levels greater than 3σ). For source detection, we used
a small box with a linear size of 5 Full-Width Half-Maximum (FWHM) widths to remove the local
background, and for source extraction (fitting) we used a larger background box with a size of 9–10
FWHM widths to conserve the calibration.
Several types of noise images can be produced by the MOPEX and APEX software, but none
are optimal for these data. The “std” noise image produced by MOPEX represents the empirical
scatter from the repeated observations per sky pixel divided by the square-root of the number of
good observations. For deep observations the “std” file underestimates the true noise since it does
not account for the pixel-to-pixel correlated noise or the confusion noise. APEX computes the
spatial pixel-to-pixel noise (“noise” file) by calculating the noise within a local box surrounding
each pixel after rejecting positive outliers. The output “noise” image has variations that depend on
the outlier parameter and the number of sources within the local box. To avoid local biases due to
sources, we derived a “noise” image after the extraction of sources, adopting a box size with a linear
scale of 9-10 FWHM widths (the same size used for the local background subtraction for source
– 7 –
fitting). To preserve both the small and large scale spatial variations of the uncertainty across the
science image, both the “std” (representing small-scale variations) and “noise” (representing large-
scale variations) files were used. The quality of “std” and “noise” images was first improved by
smoothing the images by 1 and 3 FWHM widths, respectively, and then combined in quadrature.
Equal weights were given to the “std” and “noise” images at 70 µm. Since the confusion noise
is important at 160 µm, the weights at 160 µm were based on the relative contributions of the
instrument and confusion noise (Sec. 5.2). The std image was given a weight corresponding to the
instrument noise, and the noise image was given a weight corresponding to the confusion noise to
produce the combined uncertainty file at 160 µm. The median level of the combined uncertainty
file was scaled to match the total average noise level derived from fitting the 1σ width of a Gaussian
to the data histogram of the image after source extraction. Source extraction and noise estimates
were repeated until the results converged.
After proper background subtraction and deriving an accurate uncertainty image, sources
were detected using the APEX peak algorithm. Peaks with a signal-to-noise ratio (SNR) of greater
than 3 were fitted using the Point-source Response Function (PRF, which includes the effects of the
detector size and the adopted sub-sampling of the detectors) image. At the spatial resolution of the
MIPS-Ge bands, the PRF is stable. At 70 µm we adopt the PRF (FWHM = 18.6??) made previously
from the xFLS data (Frayer et al. 2006a). At 160 µm we made a new PRF using the COSMOS,
xFLS, and the data from all of the fields from the Spitzer Wide-area Infrared Extragalactic Legacy
Survey (SWIRE, Lonsdale et al. 2004). None of the individual fields has a large number of isolated
160 µm sources with high SNR to make a high-quality PRF. At 160 µm, the brightest sources (? 2
Jy) cannot be used due to the non-linearity of the detectors (Stansberry et al. 2007). In total, we
used 149 isolated sources from the Spitzer 160 µm surveys which have SNR ? 30 and S160 < 1Jy1
to produce an empirical PRF (FWHM = 39??).
The detection and extraction of sources was done in multiple passes. Initial source lists were
visually analyzed and cleaned to remove sources with low coverage and potential spurious sources
in the Airy rings around bright sources. The detection table was updated to remove bad sources,
and source fitting was redone. In the few cases where the detection routine failed in the proper de-
blending of two or more sources, the detection table was modified, and source fitting measurements
were redone using better initial source positions.
The absolute pointing reconstruction of the Spitzer telescope is typically better than 1??. The
pointing uncertainties are much less than the large FWHM width of the data (18.6??at 70 µm and
1Throughout this paper the flux densities of the MIPS bands are defined as S160 = Sν(155.9µm), S70 =
Sν(71.4µm), and S24 = Sν(23.7µm).
– 8 –
39??at 160 µm). We verified the pointing solutions by comparing the positions of the 70 µm sources
with their VLA radio counterparts (Schinnerer et al. 2007). For the approximately 400 sources
detected at SNR> 5 at both 1.4 GHz and 70 µm, we find an average positional difference (Spitzer
- VLA) of ∆RA = −0.08±0.08??and ∆Dec = 0.17±0.08??. These small offsets are well within the
scatter measured for individual sources of 1–2??, so no positional corrections were made to the data.
MIPS is calibrated assuming the point-spread functions (PSF, which does not include the
effect of the detector size) calculated using the Spitzer TinyTim (S-TinyTim) models (Krist 2002).
The modeled PSFs have been shown to match the observations for all bands (Engelbracht et al.
2007; Gordon et al. 2007; Stansberry et al. 2007). An important aspect of the MIPS PSFs is that
significant flux arises outside of the first Airy ring on large spatial scales. We corrected the derived
flux densities for emission outside of the adopted PRF images using the S-TinyTim models. For the
adopted PRFs (87??×87??at 70 µm and 190??×190??at 160 µm), we derive an aperture correction of
1.15 at both wavelengths based on modeled PSFs assuming a constant ν SνSED. Empirically, it is
difficult to measure the level of emission outside of the first Airy ring for MIPS-Ge data. However,
the calibration factors assume the modeled PSFs out to very large spatial scales, and we must
apply this correction for consistency. We verified the calibration consistency of our techniques
using archived calibration observations taken over the lifetime of the mission. For comparison,
observations of the calibration star HD180711 (S70 = 447.4mJy, Gordon et al. 2007) were used
at 70 µm, while observations of the ULIRG IRAS 03538–6432 were used at 160 µm (Stansberry
et al. 2007; Klaas et al. 2001, (S160 = 1.04Jy)). Using an aperture correction for a modeled PSF
with a stellar SED and similar reduction techniques carried out for S-COSMOS, we measure a flux
density of 450 ± 18mJy for HD180711. This is consistent with expectations (flux density ratio of
1.01 ± 0.04). At 160 µm we measure a flux density ratio of 0.97 ± 0.05 compared to expectations
for IRAS 03538–6432 using similar reduction techniques carried out for S-COSMOS. These results
suggest that the calibration of the S-COSMOS data is consistent with the MIPS calibration papers;
the absolute calibration uncertainty of MIPS is 5% at 70 µm (Gordon et al. 2007) and 12% at 160
µm (Stansberry et al. 2007).
3.7. Corrections for Completeness and Eddington Bias
The completeness levels were estimated by simulations and calculations based on the SNR
threshold. For the simulations, sources were injected at random locations into the images and
extracted using the same techniques adopted for the production of the catalogs. For high SNR
≥ 5.0, the SNR threshold itself is the dominant effect in the determination of the completeness
level. At lower SNR, other effects such as the details associated with source detection and fitting
become significant. The completeness levels vary significantly across the image as a function of
– 9 –
coverage and flux density. Based on the SNR ≥ 5.0 threshold, the average effective completeness
level across the image can be calculated as a function of flux density as the fractional area within
the image for which S + I ≥ 5.0U, where S is the flux density, I is the science image, and U is
the uncertainty image. Figures 3&4 show completeness calculations as a function of flux density
and coverage. The simulations match the expected curves very well at 70 µm for both the deep
and typical regions within the image. For the deep region at 160 µm, the simulated completeness
values are relatively noisy and are lower than expected for 60–90mJy, potentially due to the effects
of confusion and/or the small number of independent beams within the simulated area.
completeness calculations at 160 µm for the nominal coverage of greater than 20 are similar to the
simulations and the calculations for the typical range of coverage values (25–33). For the derivation
of the S-COSMOS source counts (Sec. 5.1), we adopt the completeness curves for coverages greater
than the nominal values of 100 and 20 at 70 and 160 µm, respectively (solid lines, Fig. 3&4).
In addition to the completeness corrections, the counts are affected by the Eddington bias (flux
boosting). At faint flux densities the observed flux densities are slightly higher on average than the
true flux densities since sources on positive noise features are preferentially selected. This effect of
”flux boosting” is not as important for SNR ≥ 5 as for lower SNR, but would still yield a small
systematic bias in the measured counts if not corrected. The same simulations used to help quantify
the completeness corrections were used to check the importance of the Eddington bias. The ratio
of the observed to input flux densities (Sobs/Strue) of each bin were measured for both the deep
region and the wide area with ”full-blown” simulations that injected sources at random positions
within the image (one at a time) and exacted the output flux densities using the same methods
adopted to produce the source catalogs. Given the variation of coverage in the data, it is difficult
to obtain sufficient statistics as a function of flux density and coverage across the image using
these full-blown simulations (as was the case for the completeness estimates). Instead, we carried
out simpler calculations that are consistent with the full-blown simulations, but are significantly
more accurate. Input flux densities (Strue) with the same power-law distribution as the observed
source counts were randomly added to the noise distribution given by the uncertainty image, and
the output observed flux densities were derived. The effective Eddington bias of Sobs/Strue was
calculated over the entire image for each flux density bin for the adopted SNR ≥ 5 cut and used
to correct the observed flux densities for the derivation of the counts (Sec. 5.1). This method fully
accounts for the variation of coverage and noise across the image.
The S-COSMOS 70 and 160 µm products are available on-line at the NASA Infrared Science
Archive (IRSA) at the Infrared Processing and Analysis Center (IPAC). The products described
here are version 3 of the data. Version 1 and version 2 were early quick-look products based on
simplified reductions of sub-sets of the observations. Version 3 represents the combination of all
data from the S-COSMOS MIPS program and is the first version to be calibrated properly and
– 10 –
processed with the best known data reduction techniques.
The science images have been background subtracted with the removal of a global offset, but
the local background fluctuations have not been removed (Sec. 3.4). The images are in surface-
brightness units of MJysr−1assuming the latest calibration and have been color-corrected to match
SEDs with a constant ν Sνscale (Sec. 3.1). The uncertainty images have also been color-corrected
and are in units of MJysr−1. The uncertainty images (1σ) represent both the small-scale and
large-scale spatial noise properties associated with the science image (Sec. 3.4).
The coverage maps provide the effective number of observations (after data editing) per point
on the sky for the science images. At 70 µm for the nominal coverage of greater than 100, the
median effective exposure time is 1350s while the ultra-deep region with a coverage of greater than
250 has an exposure time of 2800s. At 160 µm for the nominal coverage of greater than 20, the
median effective exposure time is 273s, while the ultra-deep region with a coverage of greater than
50 has an effective exposure time of 567s.
Table 2 lists the image properties and sensitivities. The sensitivities represent the total noise,
including confusion. The surface-brightness noise for the adopted pixel scale was derived from
fitting the 1σ width of a Gaussian to the data histogram of the image after source extraction. To
derive the effective point-source noise, including the effects of correlated noise between the pixels,
we carried out aperture measurements at random locations within the residual image after the
extraction of sources. We derive conversion factors between the point-source noise and surface-
brightness noise of 13.3 ± 1.3mJy(MJysr−1)−1and 74 ± 7mJy(MJysr−1)−1for the 70 µm and
160 µm images respectively. The average point-source noise (1σ) is 1.7mJy and 13mJy at 70 and
160 µm, respectively. These point-source values include the aperture correction of 1.15 for emission
outside of the measured PRF (Sec. 3.6).
Point-source catalogs were made for a SNR ≥ 5.0. In total, 1512 sources at 70 µm (Table 3)
and 499 sources at 160 µm (Table 4) are cataloged. The catalogs are single-band source lists and
are independent from each other and the MIPS 24 µm data. Although the catalogs are not biased
by data at other wavelengths, we did use the 24 µm and radio data to confirm the reliability of
the catalogs. Within the central area cataloged at 24 µm, only seven 70 µm sources do not have
a 24 µm counterpart (S24 > 60µJy) within the approximate 70 µm beam radius of 9??(which
corresponds to a relatively large chance positional coincident of about 50% for a 24 µm source). Of
these seven, four are associated with a radio source, two do not have a radio counterpart, and one
is outside the radio coverage. For the three cases without a current 24 µm or radio counterpart, the
– 11 –
70 µm position is located between a blend of two or three 24 µm sources. These blended sources
could represent valid detections at 70 µm. All 160 µm sources have possible 24 µm counterparts,
which is not unexpected given the low spatial resolution of the 160 µm data (where on average
there are about three 24 µm sources per 160 µm beam). Although we find no obvious spurious
detections within the catalogs, users should be cautious and check the images when comparing
catalogs at different wavelengths due to potential source blending. Tables 3&4 show the format for
an example portion of the S-COSMOS 70 µm and 160 µm catalogs published in the on-line edition
of the Journal.
PRF-fitting flux densities and aperture measurements were made using the APEX software.
The aperture and PRF measurements are in statistical agreement. Since PRF measurements are
significantly more accurate for faint sources (e.g., Frayer et al. 2006a), the PRF flux densities are
used for the vast majority of sources (flag of “p”). Aperture measurements are adopted for sources
not reasonably well fitted by the PRF (extended sources or unresolved blends) and are given a flag
of “a” within the catalog. Resolved blends fitted well by multiple PRFs are given a flag of “p”.
For the one blend of two resolved sources which is not fitted well by two point sources at 70 µm,
the total aperture flux is divided among the two components based on their relative peaks (flag
of “ap”). For consistency with the adopted calibration of MIPS, the measurements include the
aperture correction of 1.15 for emission outside of the measured PRF (Sec. 3.6). No corrections
for flux non-linearity have been made (Stansberry et al. 2007). Only one source (SCOSMOS160
J100027.0+032226, S160 ∼ 11Jy) is bright enough to be significantly impacted by flux non-linearity
(the only source at 160 µm with an aperture measurement, flag of “a”); its flux density should be
treated with caution.
The errors on the fitted flux densities derived by APEX are not used since they are underes-
timated by about a factor of 3 for these data. We adopt errors based on the SNR measurements
which represent the fitted peak flux density divided by the uncertainty image at the location of
the source. The errors on the measured flux densities (S) represent the random errors given by
the SNR ratio combined with the systematic calibration uncertainty (?cal). The flux density error
σ(S) = (1/SNR + ?cal)S, where ?calis 0.05 and 0.12 for the MIPS 70 µm (Gordon et al. 2007)
and MIPS 160 µm (Stansberry et al. 2007) bands respectively. The few sources (9) requiring large
aperture measurements have additional measurement errors of 10–20%.
We find that the positional fitting errors calculated by APEX are 2.0 times larger on average
than the expected radial positional errors of ≈ 0.6(FWHM/SNR) given by Condon (1997) for
all SNRs. We adopt the APEX errors, but treat them as 2σ uncertainties. For these data the
relationship of Condon (1997) is appropriate for typical SNRs, but underestimates the errors for
the highest SNRs. We adopt a floor on the uncertainty of ?pos= 0.5 pixel (? 0.1 FWHM, Table 2),
representing the difficulty in deriving positions to better than a fractional pixel regardless of the
SNR. The cataloged radial positional errors (2σ uncertainties) are given by (σ2
where σxand σyare the fitting errors in the x-y plane derived by APEX.
– 12 –
5.Results and Discussion
The source counts were derived from SNR≥ 5.0 source lists corrected for completeness over the
region within the images with high coverage (> 100 at 70 µm and > 20 at 160 µm), corresponding
to an effective survey area for the 70 and 160 µm fields of 2.471 and 2.438 sq-deg, respectively.
The public catalogs presented in Sec. 4.2 include sources outside of this nominal area with lower
coverage. The effective completeness value for each flux density bin was computed by integrating
over the completeness curves (Fig. 3&4) as a function of flux density across each bin with weights
based on the measured slope of the source counts.
Figures 5&6 show the Euclidean-normalized differential source counts (dN/dS×S2.5) at 70 µm
and 160 µm, and the results are tabulated in Tables 6&7, respectively. The counts are calculated
for independent flux density bins. The error bars represent the Poisson errors associated with the
number of sources and the uncertainties on the completeness corrections. The grey-region shows
the range of values implied from the most statistically accurate counts within each flux density
bin, including the additional uncertainty due to the systematics associated with the calibration of
MIPS (5% at 70 µm and 12% at 160 µm).
At 70 µm the S-COSMOS counts are measured down to a level of 10mJy which is near the peak
of the Euclidean-normalized differential source counts (Frayer et al. 2006b). The faint (∼ 10mJy)
S-COSMOS counts at 70 µm (Fig. 5) agree with those measured for GOODS-N (Frayer et al.
2006b) and the xFLS (Frayer et al. 2006a). The bright counts for S-COSMOS also agree with
the measurements from the SWIRE survey. At intermediate flux densities of around 20–30mJy,
the counts for S-COSMOS are slightly lower than those found for the xFLS and model predictions
of Lagache et al. (2004). All the counts have been placed on the same scale by matching the
calibration, color corrections, and the PRF adopted for S-COSMOS. The correction factors for the
other data sets are given in Table 5. The previous results for the xFLS, SWIRE, and GOODS-N
did not include the aperture correction for the flux density outside of the measured PRF. The
counts for SWIRE are based on the public catalogs (2005 November, Data Release 3 [DR3] ) which
cover 49 sq-deg. The SWIRE counts at 70 or 160 µm have not previously been published and are
presented here to constrain the counts at the brightest flux densities. Only the high SNR (? 10)
SWIRE sources with completeness levels near one are presented in Figures 5&6.
At 160 µm the S-COSMOS counts are measured down to a level of 60mJy. Measurements of
deeper counts are limited by confusion (Sec. 5.2). The measured counts at 160 µm agree with the
counts measured previously in the xFLS (Frayer et al. 2006a) and the counts derived here based
on the SWIRE survey. As done for 70 µm, all counts are placed on the current calibration scale
(Table 5). These correction factors also include a decrease in the flux densities due to the new
PRF derived here (Sec. 3.4). The previous PRF images used by the xFLS and SWIRE surveys are
affected by the flux non-linearity for bright sources at 160 µm.
– 13 –
The faintest (S160 < 80mJy) and brightest (S160 > 500mJy) source counts are consistent
with the Lagache et al. (2004) model, but the observed counts for all three surveys at intermediate
flux densities are about a factor of 1.5 times lower than the model implies. The observed 160
µm counts show a steep increase in the differential counts for decreasing flux densities (dN/dS ∝
S−3.5±0.2) for S160 < 150mJy. The slope for the faint (10–20mJy) S-COSMOS 70 µm source
counts (dN/dS ∝ S−3.1±0.2) is roughly consistent with the 160 µm slope and is slightly steeper
than the model predictions. The differences between the model and the source counts at 70 µm
is not as clear as those seen at 160 µm. The results for 160 µm may suggest the importance of
galaxies having more cold dust than assumed in the models. Observations with the future Herschel
telescope should provide better constraints on the FIR SEDs of distant galaxies.
5.2. Confusion Level of the MIPS 70 and 160 µm Bands
The S-COSMOS 70 µm data are dominated by instrument uncertainties and are not deep
enough to constrain the confusion noise. Frayer et al. (2006b) measured the MIPS 70 µm confusion
noise based on the much deeper 70 µm observations of GOODS-North. Based on Frayer et al.
(2006b) and the calibration scaling factors (Table 5), the updated extragalactic confusion noise
level for the MIPS 70 µm band is σc= 0.35 ± 0.15mJy (q = 5), including the updated systematic
error on the flux calibration.
In contrast to the 70 µm data, the S-COSMOS 160 µm data contain a significant noise compo-
nent due to confusion. We estimate the confusion noise at 160 µm following the empirical technique
performed at 70 µm (Frayer et al. 2006b). A direct empirical measurement of the confusion noise
for the MIPS 160 µm band has not been published previously. Dole et al. (2004) report a confusion
level at 160 µm based on the techniques of Dole et al. (2003) and the models of Lagache et al. (2004).
They find that the predicted confusion noise from their models is in reasonable agreement with the
observations, assuming that the instrumental noise (σI) follows the theoretical MIPS model and
integrates down as σI∝ t−0.5. However, the early processing of MIPS-Ge data was not optimal,
and an accurate empirical measurement of the total instrumental noise (including photon noise,
detector noise, and noise associated with the data processing) is required to measure the confusion
The instrument noise was estimated empirically by subtracting pairs of data subsets with the
same integration time and covering the exact same region on the sky (which removes sources and
any remaining cirrus structure after filtering). We fit the instrument noise measurements as a
function of integration time for combinations of deep pairs of data sets and obtain σI∝ t−0.49±0.03.
This result is consistent with idealized data (σ ∝ t−0.5) and highlights the success of the reduction
methods in removing systematic artifacts. The extrapolation of instrument noise for half of the
data to the full data set yields σI= 0.1134±0.0033MJysr−1, where the uncertainty represents the
rms measurement error combined in quadrature with the error associated with the extrapolation.
– 14 –
Following the terminology of Dole et al. (2003), the total noise (σT) represents the noise after
the extraction of sources above a limiting flux density (Slim), and the photometric confusion noise
(σc) represents fluctuations due to sources with flux densities below Slim. The confusion noise is
given by σc= (σ2
noise after the extraction of sources. We iterate between source extraction at different limiting flux
densities and confusion noise measurements until we converge to a solution with q ≡ Slim/σc= 5.
For the q = 5 solution, we derive σT= 0.1772±0.0089MJysr−1and σc= 0.1362±0.0119MJysr−1,
for a limiting source flux density of S160 = 50mJy.
I)0.5, which is appropriate for an approximate Gaussian distribution of the
Noise measurements in surface brightness units (MJyrsr−1) depend on the pixel scale, and
all measurements here are based on the 8??pixel scale of the 160 µm image. The uncertainties
on the noise measurements given in MJysr−1also do not include the 12% systematic calibration
uncertainty. Including the systematic calibration uncertainty and the measured conversion between
surface-brightness noise and point-source noise (Sec. 4.1), the point-source confusion level of the
MIPS 160 µm band within the COSMOS field is σc= 10.0±3.1mJy. The systematic uncertainties
contribute 71% to the total error budget, while the random errors contribute 29% to the total error
The measured confusion noise is not entirely due to galaxies. Unlike the case for the MIPS 70
band µm (e.g., Dole et al. 2003; Frayer et al. 2006b), Galactic cirrus is not negligible at 160 µm.
We estimate the level of confusion due to Galactic cirrus for the MIPS 160 µm band using the
background estimates from the Spitzer tool for planning observations (SPOT) which is based on
the interstellar medium (ISM) maps from the Diffuse Infrared Background Experiment (Schlegel et
al. 1998). At the effective wavelength of 155.9 µm of the MIPS 160 µm band, the ISM background
in the direction of the COSMOS field is about 2MJysr−1. Using the calculations of Jeong et al.
(2005), this background level corresponds to an ISM confusion noise of σc(ISM) ? 3.4mJy. Hence,
the confusion noise due to unresolved galaxies is σc(gal) = (σ2
The derived confusion limit agrees fairly well with predictions. Dole et al. (2004) report a
Source Density Criterion (SDC) limit of 40mJy (σc(qSDC= 3.8) = 10.6mJy) based on the model
predictions of Lagache et al. (2004). For a direct comparison with the empirical measurement, the
Dole & Lagache et al. predictions suggest σc(q = 5) = 12.5mJy. We find an extragalactic confusion
level for the MIPS 160 µm band of σc(q = 5) = 9.4 ± 3.3mJy, which is just slightly lower than the
6. Concluding Remarks
We present the 70 and 160 µm observations of the COSMOS field and describe the products.
This is the first extragalactic survey available to the public at 70 and 160 µm that has been placed
on the calibration scale derived from the recent MIPS calibration papers. We provide updated
correction factors for the previously released catalogs of the xFLS (Frayer et al. 2006a) and SWIRE
– 15 –
(Lonsdale et al. 2004) programs. Counts are presented based on the S-COSMOS and previous
surveys and are found to be in reasonable agreement with the model of Lagache et al. (2004).
However, the faint 160 µm source counts are significantly steeper than model predictions. We
measure an empirical extragalactic confusion noise level of σc = 9.4 ± 3.3mJy (q = 5) for the
MIPS 160 µm band. In comparison, the expected confusion noise at 160 µm for Herschel is about
1.5–2mJy (Herschel Observation Planning Tool, HSPOT, version 3.4). Future observations with
the Herschel telescope should constrain the counts and far-infrared properties better than can be
done currently due to confusion for the MIPS 160 µm band.
We thank our colleagues associated with the Spitzer mission who have made these observations
possible. This work is based on observations made with the Spitzer Space Telescope, which is
operated by the Jet Propulsion Laboratory, California Institute of Technology under a contract with
NASA. Support for this work was provided by NASA through an award issued by JPL/Caltech.
Capak, P., et al. 2007, ApJS, 172, 99
Condon, J. J. 1997, PASP, 109, 166
Dale, D. A., et al. 2005, ApJ, 633, 857
Dole, H., et al. 2004, ApJS, 154, 93
Dole, H., Lagache, G., & Puget, J.-L. 2003, ApJ, 585, 617
Engelbracht, C. W., et al. 2007, PASP, 119, 994
Frayer, D. T., et al. 2006a, AJ, 131, 250
Frayer, D. T., et al. 2006b, ApJ, 647, L9
Gordon, K. D., et al. 2007, PASP, 119, 1019
Gordon, K. D., et al. 2005, PASP, 117, 503
Hasinger, G., et al. 2007, ApJS, 172, 29
Jeong, W.-S., Mok Lee, H., Pak, S., Nakagawa, T., Minn Kwon, S., Pearson, C. P., & White, G. J.
2005, MNRAS, 357, 535
Klaas, U., et al. 2001, A&A, 379, 823
Koekemoer, A. M., et al. 2007, ApJS, 172, 196