WIRC Data Reductions

T. Jarrett (IPAC/Caltech)
Feb 17, 2004

The Palomar 200" Wide Field Infrared Camera (WIRC) is a wonderful addition to the stable of instruments aboard the Big Eye. With its wide 9 arcmin field of view, sensitive and cosmetically clean array, it's like using a CCD in the NIR. Nonetheless, WIRC also presents a number of reduction challenges that are discussed in this document. Some links to official WIRC stuff: instrument specifications, Filter Transmission Profiles, Software User Manual and the HAWAII-2 Detector Characterization. Here is a nice compilation plot of the WIRC filters: pdf of wirc filters.

Contents

Cookbook Reductions

So you are ready to WIRC. Click here to get it on.

Standard Reductions: lesson learned

For the previous camera that was used at Palomar, PFIRCAM, the standard reduction was to construct the flat-field (or pixel-to-pixel gain response function) from either "dome", "twilight" or "dark sky" imaging. For sky subtraction, construct a "dark sky" image from a set of images that are mostly free of large scale structure (e.g., galaxies). The reduction procedure is then

This method produced the most photometrically consistent reduced images. But we have found that it does not work very well with WIRC. Specifically, the flat-field image seems to be problematic. Flats constructed from dome imaging or from "twilight" imaging do not work at all. After further testing at the telescope, my colleagues and I (Phil Appleton and Bill Reach) discovered that the light illumination level mattered crucially to the flat-field. If the target observations had a different illumination level than the flatfield images (either constructed from dome light or twilight), then the flat did not work at all.

Instead, the only way to construct a flat that worked was to use the "dark sky" imaging from the target observations (and hence, the light illumination was by design the same in both). Hence,

Subsequently we discovered that even this method induces unusual systematics in the reduced image that are filter dependent (more on this later). Moreover, we discovered (the hard way) that WIRC may be spectacularly linear over most of its dynamic range, but it is not linear at the very faint end (where our narrow-band images inhabit this phase space). So, we now linearize our WIRC data before carrying out any reductions. The linear results are shown below.

After talking with Roc Cutri about the characteristics of HgCdTe arrays and how 2MASS handled flatfielding, the solution became clear -- use the pixel response created from linearity measurements. The virtue of using the slope in the linearity is that it is independent of the filter. The reduction is now:

This method has proved to be the most robust and clean way to produce consistent photometry. Typically 5-7% RMS in comparison to 2MASS photometry. There is still one major unresolved problem -- reduced WIRC images exhibit a consistent and significant flux bias pattern across the focal plane. The nature of this problem may be related to the field correction optics (since it resembles "vignetting"). In any case, the solution is to compute this bias and correct for it it. The reduction is now:

The Jarrett WIRC reduction package will do this as an option. See the cookbook instructions for reducing WIRC images.

The detailed linearity measurements and the flux bias is described next.

Linearity Measurements

A set of linearity data was acquired by Margrethe Wold, Oct 15, 2003, using the P200 and WIRC. With this data we are able to derive the curvature in exposure vs. flux, enabling rectification (linearization) of the science data.

Method: Described in Jarrett et al. (1994), for each pixel a quadratic is fit to derive the curvature:

Places results in 4 fits files: a0.fit, a1.fit, a2.fit and a3.fit. The coefficient images can then be used to rectify (linearize) science data. Note that before the measurements are made, the dark current image (appropriate to exposure time) is subtracted from each raw image.

Results:

a_0 coefficient
(lower boundary)
a_1 coefficient
(slope ==
pixel response function)
a_2 coefficient
(curvature ==
non-linearity)
a_3 coefficient
(saturation bound)

Other linearity tests: see a test using stars to measure the calibration as a function of integrated flux.

Flux Bias (*new results: May 23, 2006)

We noticed early on that our NGC6946 images suggested a flux bias, between 5 and 10%, from the center of focal plane out to the edges. We first measured this bias in the diffuse light of N6946. We then verified that it is also endemic to point sources. How this bias arises is unknown, and how it defeats the flat-field pixel response correction is unknown. In any event, it is easily measured in the reduced images. The procedure is to first calibrate the reduced image using 2MASS stars; e.g,


The plot shows the 2MASS Ks-band mag versus the WIRC Ks-filter mag. The RMS is 0.041 mag. All looks well for this image, but built-in to this RMS is scatter due to a focal-plane dependent flux bias.

The next step is to take all of the calibrated measurements (2MASS vs. WIRC == delta mag) and place them in a large grid based on the WIRC focal plane. We then compute the median delta mag for each grid. We expect to find the average delta mag to be zero from grid to grid. The actual result for the Ks-band filter is shown here:


Notes: this image shows the latest results using FeII, H2 and Ks observations (May 16, 2006) of the M3 globular cluster; the orientation is standard North up, East to the left)

Flux bias pattern measured for the Ks-band filter. The central bias is +0.06 mag (meaning, the WIRC mags are systematically too faint), and the edge bias is approximately -0.04 mag (meaning, the WIRC mags are systematically too bright).

After applying this correction to the reduced Ks-band images, the resultant photometry is significantly improved; e.g.,


The plot shows the 2MASS Ks-band mag versus the WIRC Ks-filter mag with the flux bias removed. The RMS is 0.029 mag (compared to the previous, 0.041 mag, shown above). The remaining 3% scatter in the photometry is due to the WIRC S/N measurements (which were for the most part between 50 and 100) and the RMS in the 2MASS absolute photometry (expected to be 2 to 3%). We conclude that our corrected WIRC Ks-band photometry is as good as 2MASS will allow it to be.

We have investigated how this flux bias varies from filter to filter. Our best measurements come from a set of observations of the globular clusters M3 and M15, and from fields in the Milky Way where the source density is high. We also have measurements from fields near N6946, M33 and from various fields. All in all, we have the following statistics and results:
(note: the orientation is standard North up, East to the left)

The flux bias pattern is approximately the same for all filters. This result is consistent with the flux bias origin associated with the camera optics (i.e., outside of the focal plane).

Based on these results, I will assume that the flux bias is same for all filters, and the best correction map is given by the last one in the table -- FeII + H2 and Ks.

To see the long and sordid history behind this flux bias investigation, click here. You will also find other mysterious phenomenon we encountered, including the amazing Crescent Moon feature.

The following table comes from results derived from 2003 observations, now superceded by the new M3 measurements (see above).

filter name number of star
measurements 		correction grid elements pattern image
(2048X2048 pixels)
Ks-band 16,776 32 X 32
H2 (1-0) 9499 32 X 32
Kcont 2097 16 X 16
FeII 6367 32 X 32
Hcont 1735 16 X 16
J-band 1390 16 X 16
J + Hcont + FeII +
Kcont + H2 + Ks 		37622 64 X 64