Test and Implement Default Parameters for Jointcal Outlier Rejection

added an option to stop outlier rejection early. We need to investigate what value is appropriate for the new astrometryRelativeOutlierTolerance parameter and to set that value as the default for processing HSC data.

Results
The default jointcal configuration performs chi2 minimization and outlier rejection steps in which iterations of outlier rejection are continued until no more outlier measurements are found. When jointcal is run on hundreds of visits, this means that runtime can stretch to multiple weeks. The astrometryOutlierRelativeTolerance parameter adds the option to stop iterations of outlier rejection at the point when rejecting additional outliers has a minimal effect on the chi2. Iterations stop when the fractional change in chi2 is below the astrometryOutlierRelativeTolerance parameter.

Initial testing on sets of 10, 100, and 378 visits in Tract 9813 indicated that setting the parameter to 0.002 provided a good compromise between speeding up the runtime while having a minimal effect on the final model. For very large datasets, runtime should be cut to a couple of days in lieu of a couple of weeks. Plots showing these results are attached to . Running with this value was subsequently tested on RC2.

Running the validateDRP and pipe_analysis scripts on both these results (identified in plots as “optionON”) and the results of running jointcal in the default configuration (“optionOFF”) shows that stopping outlier rejection iterations earlier has a minimal effect on the final astrometric model. For the System Science Requirements metrics AM1 and AM2, turning on the optional parameter changes the metric values by less than 0.1 marcsec in general, or about 2%, with a few cases where the metrics are up to 0.2 marcsec higher. The AF1 and AF2 metrics are similarly only slightly changed, and in many cases are better with the optional parameter turned on. A summary of the changes in the AMx and AFx results is shown in the attached plot valDRP_synopsis.png, and the full diagnostic plots are shown in the comments below.

Similarly, comparison of the diagnostics output by visitAnalysis.py shows that the results from the jointcal runs with the parameter turned on are equivalent, or in some cases better, than the results from the default configuration. This is particularly true in the cases, many in HSC-Y, where the mean and standard deviation are high in the default configuration. The attached plot visitAna_synopsis.png shows the change in mean and standard deviation for a sample of twenty visits. The {{plot-vNNNN-matches_

{ra/dec/raCosDec}-sky-stars.png}} plots and {{plot-vNNNNN-matches_{ra/dec/raCosDec}

-psfMagHist.png}} for a selection of visits are in the comments below.

The full outputs of matchedVisitMetrics.py, visitAnalysis.py, and colorAnalysis.py for optionON and optionOFF are available at lsst.ncsa.illinois.edu/~csaunder/DM-30534/. The output of compareVisitAnalysis.py is also available in the optionOFF visitAnalysis directories.

The conclusion of these tests is that running jointcal with the parameter astrometryOutlierRelativeTolerance turned on and set to a value of 0.002 for HSC will have a negligible impact on the final output of the pipeline.