DISCLAIMER: The zero points described in this document apply to point sources. They are also applicable to photometric calibration of the fluxes of distant galaxies with reasonable accuracy. However, be warned that the information in this document does not apply to extended sources e.g. local, large galaxies. It has been discovered that the glue holding the IRAC CCDs in place scatters about 40% of incident light, and thus extended sources have a calibration factor about 40% higher. Look elsewhere for details about calibrating extended sources.
NOTE added Dec 04 Be aware that IRAC suffers from a (color term) effect, resulting in the calibration being a function of an object's position on the CCD (and, of course, of the spectral slope of the source). This means that the true flux for objects lying near the edge of the chip may vary significantly (up to 10%) from the fiducial value measured for calibration stars lying near the center of the CCD. The IRAC Instrument Team is still working to quantify this problem. There is some more information here
Skip to the conclusions
The IRAC data is calibrated based on a plan which was developed before flight. The data is calibrated using 11 primary calibrators stars (PCs) and (eventually) about 30 secondary calibrator stars (SCs). The primary calibrators are located in the continuous viewing zone (CVZ) and are observed twice per campaign. The secondary calibrators are located in the ecliptic plane are are observed once during each period of autonomous observation (PAO), or about once per twelve hours. The PC's (SC's) provide measures of longterm (shorterm) baseline stability. Since the SC's are observed close to downlinks, any candidate must fall in a window diametrically opposed to the earth by approx 180 degrees in the ecliptic plane. This window drifts by a full 360 degrees per year so any one candidate is visible only for a couple of campaigns at most (Decause the telescope gradually lags behind the earth the window visible on any given date will not coincide exactly with the window visible on the same date the previous or following year).
The primary calibrator stars were chosen by Tom Megeath (IRAC IST based at CfA) based on ground-based observations of candidate stars. The calibration stars are always A dwarves or K giant stars because these are most easily modelled (featureless spectra). Due to an increase in the slew ragte approved for Spitzer after nominal operation campaign 4, the CVZ shrank. This resulted in the loss of two original PCs for several months of each year. Three new PCs were added to the program at that time.
Depending on the magnitude of the star, the exposure times were either 0.2 (1.2), 1.2 (10.4) or 10.4 (26.8) seconds for channel 1 and 2 (channel 3 and 4). This is to ensure decent signal-to-noise in each case (but precludes deducing anything about the linearity response of the CCD directly from the PCs).
The primary calibrators (increasing magnitude) are:
The secondary calibration stars were chosen by me based on a candidate pool provided by Tom Megeath. I eyeballed Digitized Sky Survey plates and selected the stars with fewest nearest neighbors or other contamination. Sometimes fewer than two candidates were available for each campaign, or the chosen star turned out to be variable or otherwise unsuitable. The "bad" stars will be culled from the pool of potential candidates in future years.
The secondary calibrators were as follows. The campaign is shown in parentheses. (One star from each of campaigns 2 and 3, and both from campaign 4 did not get processed)
I wrote a suite of perl scripts to automate the analysis since IRAC makes about 1000 observations of primary and secondary calibration stars per campaign. For either case (PCs or SCs), the calibration star BCD's are dark-subtracted and flatfielded. A local background is subtracted and then I run an object detection algorithm and create catalogs. I select the star of interest from each catalog based on its RA and DEC. Each time a calibration star is observed, it is observed five times, with the image falling in the center, and four corners of the CCD. I calculate the mode of star flux for each of the five observations within apertures of radii 2, 4, 6, 10 and 20 pixels. For each case I then reject any observation with flux falling outwith 10% of the mode value. This ensures, in particular, that observations containing cosmic ray hits are successfully rejected and not included in the analysis. There are more cosmic rays in channel 3 and 4 data, due partly to the different make-up of those chips, but mainly due to the longer exposure times required for those channels. Mini stamp cutouts of 64 x 64 pixel regions around each observation are created and can be eyeballed to ensure there are no problems. The mean (and standard deviation) of the retained observations are then calculated. The measured mean flux is then compared to the predicted flux from models generated by Martin Cohen. (Because IRAC is far more sensitive than previous cameras it is not possible to calibrate to any existing observations of infrared standard stars) Note the results quoted below compare the flux measured within a 10 pixel radius aperture i.e. 12.2 arcsec. Knowing the zero point for each channel, one can then calculate the DN per second to Jansky conversion factor for each observation of each star.
For the gory details i.e. cookbook and scripts look here.
The zero point (luminosity of a 0th mag star) in each channel is :
| ch1 | ch2 | ch3 | ch4 | |
|---|---|---|---|---|
| Jansky | 277.5 | 179.5 | 116.6 | 63.1 |
The table shows the conversion factor averaged over first eight nominal operations campaigns (about 150 independent observations). The primary calibrators (PCs) only are used in this analysis. The quoted uncertainty in parenthesis is the root mean square of all the observations.
fluxconv is also a very useful quantity. Calculated from the first eight campaigns, this is the value which should be applied by the BCD pipeline (in converting from raw image units of DN/s to BCD image units of MJy/sr). A constant factor of 34.98 relates DN/s per microJy and fluxconv. Clearly, FLUXCONV, the value calculated by me from the first few campaigns, and currently being applied by the BCD pipeline to all campaigns, seems pretty accurate.
| ch1 | ch2 | ch3 | ch4 | |
|---|---|---|---|---|
| DN/s microJy | 0.256 (0.009) | 0.208 (0.005) | 0.048 (0.001 ) | 0.142 (0.004) |
| fluxconv (MJy/sr per DN/s) | 0.112 (0.004) | 0.138 (0.003) | 0.593 (0.018) | 0.201 (0.006) |
| FLUXCONV (MJy/sr per DN/s) | 0.1125 | 0.1375 | 0.5913 | 0.2008 |
NOTE : The calibration analysis was performed using an aperture of radius 10 pixels (12.2" ). This table provides (empirically derived) aperture corrections factors (uncertainties in parentheses) which should be applied to correct to the flux measured using a fiducial 10 pixel radius.
| Aperture radius in pixels (arcsec) | ch1 | ch2 | ch3 | ch4 |
|---|---|---|---|---|
| 2 ( 2.44") | 1.20 (0.05) | 1.21 (0.05) | 1.35 (0.06) | 1.58 (0.04) |
| 4 ( 4.88") | 1.07 (0.04) | 1.09 (0.04) | 1.08 (0.04) | 1.09 (0.03) |
| 6 ( 7.32") | 1.04 (0.04) | 1.04 (0.04) | 1.03 (0.04) | 1.05 (0.03) |
| 8 ( 9.76") | 1.01 (0.04) | 1.01 (0.04) | 1.02 (0.04) | 1.02 (0.03) |
| 1.23 (1.5") | 1.55 (0.08) | 1.62 (0.08) | 1.89 (0.08) | 2.03 (0.09) |
| 1.64 (2.0") | 1.29 (0.06) | 1.35 (0.06) | 1.57 (0.05) | 1.72 (0.04) |
| 2.05 (2.5") | 1.18 (0.05) | 1.20 (0.05) | 1.32 (0.05) | 1.56 (0.04) |
| 4.10 (5.0") | 1.07 (0.04) | 1.08 (0.04) | 1.08 (0.04) | 1.09 (0.03) |
The units of BCD data are MJy/sr. This table provides the zero points for converting from MJy/sr into Vega magnitudes, for images with "raw" IRAC pixelsize of 1.22". In almost all cases, an aperture radius much smaller than 12.2" should be used when measuring fluxes from BCD data (because of contamination from neighbor objects). Use the aperture correction table above to figure out the appropriate correction to apply to the zero point.
The table also provides the Vega to AB magnitude conversion.
| ch1 | ch2 | ch3 | ch4 | |
|---|---|---|---|---|
| Vega zero point | 17.25 | 16.78 | 16.31 | 15.68 |
| m(AB) = m(Vega) + | 2.79 | 3.26 | 3.73 | 4.40 |
What is the repeatability for a given star (i.e. relative calibration)?
About 1.8% (ch1), 0.8% (ch2), 1.4 % (ch3), 0.5%(ch4), using e.g. NPM1p680422
| Star (DN/microJys) | ch1 | ch2 | ch3 | ch4 |
|---|---|---|---|---|
| NPM1p670536 | 0.249 (0.006) | 0.203 (0.002) | 0.048 (0.001) | 0.142 (0.001) |
| HD165459 | 0.259 (0.006) | 0.208 (0.002) | 0.048 (0.001) | 0.143 (0.001) |
| NPM1p680422 | 0.249 (0.004) | 0.199 (0.002) | 0.047 (0.001) | 0.140 (0.001) |
| NPM1p640581 | 0.265 (0.007) | 0.213 (0.003) | 0.049 (0.000) | 0.148 (0.001) |
| NPM1p660578 | 0.258 (0.004) | 0.215 (0.002) | 0.050 (0.001) | 0.148 (0.001) |
| KF09T1 | 0.254 (0.002) | 0.204 (0.001) | 0.048 (0.001) | 0.142 (0.001) |
| NPM1p600581 | 0.271 (0.002) | 0.214 (0.003) | 0.049 (0.001) | 0.145 (0.002) |
| KF06T1 | 0.241 (0.006) | 0.209 (0.003) | 0.047 (0.002) | 0.136 (0.002) |
| KF08T3 | 0.258 (0.006) | 0.207 (0.003) | 0.049 (0.002) | 0.143 (0.002) |
| KF06T2 | 0.262 (0.001) | 0.215 (0.001) | 0.050 (0.001) | 0.139 (0.004) |
| s1808347 | 0.259 (0.007) | 0.208 (0.003) | 0.048 (0.002) | 0.139 (0.004) |
What is the scatter between different stars (i.e. absolute calibration)? Is this due to the models?
See "Plots" section. Based on the first six campaigns, the scatter is about 2 or 3%. [3.5% (ch1), 2.4% (ch2), 2.1 % (ch3), 2.8%(ch4)]
Yes, in part. Martin quotes 2 or 3% uncertainties on the estimated values of the fluxes for each star based on his models.
Are there any trends on the campaign level (measured from the CVZ calibrators) or on 12-hour timescales (from ecliptic calibrators)?
If there are trends they are subtle and not discernible at present.
Here is a breakdown for the PCs by campaign.
| Campaign # (DN/microJys) | ch1 | ch2 | ch3 | ch4 |
|---|---|---|---|---|
| All | 0.255 (0.009) | 0.208 (0.005) | 0.048 (0.001) | 0.142 (0.004) |
| 1 | 0.249 (0.008) | 0.212 (0.005) | 0.049 (0.001) | 0.143 (0.005) |
| 2 | 0.246 (0.008) | 0.206 (0.004) | 0.048 (0.002) | 0.143 (0.004) |
| 3 | 0.259 (0.008) | 0.206 (0.004) | 0.048 (0.002) | 0.141 (0.005) |
| 4 | 0.258 (0.008) | 0.207 (0.004) | 0.048 (0.002) | 0.142 (0.005) |
| 5 | 0.259 (0.008) | 0.209 (0.005) | 0.048 (0.001) | 0.142 (0.003) |
| 6 | 0.260 (0.006) | 0.207 (0.005) | 0.048 (0.001) | 0.141 (0.003) |
| 7 | 0.261 (0.008) | 0.208 (0.005) | 0.048 (0.002) | 0.142 (0.003) |
| 8 | 0.262 (0.007) | 0.208 (0.006) | 0.048 (0.001) | 0.142 (0.003) |
Here, I analysed all the PCs observed at the beginning of the first six campaigns, and all the PCs observed at the end of the first six campaigns separately. No trend appears.
| Campaign (DN/microJys) | ch1 | ch2 | ch3 | ch4 |
|---|---|---|---|---|
| All | 0.255 (0.009) | 0.208 (0.005) | 0.048 (0.001) | 0.142 (0.004) |
| Beginning | 0.255 (0.010) | 0.208 (0.005) | 0.049 (0.002) | 0.142 (0.004) |
| End | 0.255 (0.009) | 0.207 (0.005) | 0.048 (0.002) | 0.142 (0.004) |
Email me : gillian@ipac.caltech.edu
31st July 2004