Calibration Zeropoint Uncertainties

A. Relevant links

1. Nightly Photometric Zero Point Uncertainties from cal scans (V2 processing)
   Nightly photometric uncertainties were computed by taking the RMS of the residuals after fitting a function to the zero points vs. time. The functional fits are linear for J-band and constant (???) for H and K. In particular, see Figures 2, 3, and 4 on above web page which show histograms of the residuals RMS for J, H, and K.

2. Analysis of cal scans - higher order functional fits (V2 processing)
   Calibration zeropoint uncertainties were estimated by treating each calibration observation within a night as a test field. The test field was removed, the zero points for that night re-derived, and the difference between the new derived zero point and the observed zero point for the testfield was computed.

3. Analysis of cal scans - higher order functional fits (V3 processing)
   Same as above, but using V3 processing results.

4. Analysis of repeat observations (V2 processing)
   This analysis made use of scans from the Orion and Chamaeleon variability surveys, and duplicate scans from normal survey observations. The average magnitudes of the duplicate scans were defined as the "true" magnitudes, and the photometric offsets around these true magnitudes was computed on a tile-by-tile basis.

5. Here is a tabular summary of the average photometric uncertainties from the different analyses. The "CAL RMS" values were estimated by eye from Figures 2-4 on Roc's web page with linear fits for J-band. The remaining entries in the table using piecewise fits for J-band. Linear fits were used for H and K.

   Table 1

   Method	J	H	K
   Calibration fields (V2)	0.017	0.015	0.012
   Calibration fields (V3)	0.011	0.007	0.007
   Orion repeat fields	0.005	0.006	0.006
   Cham I repeat fields	0.008	0.005	0.007
   Duplicate tiles	0.008	0.006	0.007
   CAL RMS (north)	0.011	0.009	0.009
   CAL RMS (south)	0.013	0.011	0.009

6. Tom Chester's modeling of the observed zero-point scatter (V2 processing)
   This analysis suggests that the observed scatter of the nightly zero point RMS residuals can be explained by a random gaussian sampling of a constant zero point uncertainty. Tom suggests it would be better to adopt a single zero point uncertainty per band for the survey, rather than to adopt a different zero point uncertainty for each night.

B. Questions

1. Should we adopt a constant zero point uncertainty per band for the survey?
   Tom Chester's analysis strongly suggests the answer is yes, although the answer may not be so simple for J-band in the south (see Figure 2 from Roc's web page with linear J-band fits).

2. Does the zero point uncertainty vary with hemisphere?
   Figures 2-4 from Roc's old analysis (based upon linear fits for J-band calibration scans, not piecewise) indicate the J-band residuals could be at most 0.005 mag higher in the south than in the north.

3. What are the appropriate values to adopt for the zero point uncertainties?
   Roc's analysis of the calibration fields (v2) suggest larger values for the photometric uncertainties than my analysis of the repeat observations. As noted by Roc, this may be attributed to the fact that the calibration scans were observed in a wide variety of conditions, while the Orion and Cham I data were more controlled observations in that they were observed once a night (at similar airmass and time of day) over a 2-4 week time period.

   The normal 2MASS duplicate scans should provide a more representative sampling of the observing conditions. Even these rms values are about a factor of 2 smaller than the values derived from analysis of the calibration fields. But as noted by Roc, the calibration field analysis may overestimate the true uncertainty since the calibration is done effectively over 2 hours, rather than one hour. This will be most critical for J-band, where a piecewise fit is use to interpolate the zero points. In fact, looking at the RMS residuals from the old calibration procedure (Figures 2-4 on Roc's web page, the average RMS of the residuals is generally less than the zero point uncertainties derived by using the calibrations scans as "test fields".

   NOTE: Roc just completed processing the V3 calibration fields where he treated the calibration scans as test fields. The residuals in the V3 results are about a factor of 2 lower compared to V2 processing. Combined with the results for the duplicate fields, the results suggest that the zero point uncertainty is approximately 0.006 for H and K, and approximately 0.01 for J. It would be useful to redo the duplicate tile analysis with V3 processing.