Experiment

I (dstn) built a normal r-band index, then tried to solve SDSS u, g, r, i, and z fields. I used the same set of fields imaged in the five bands, so the only difference between the input fields should be the relative brightness and visibility of objects. I ran a total of about 35,000 fields in each band.


Results

Round Field Objs Code Tol NAgree u g r i z
1 0-50 0.005 2 82.65 % 93.33 % 94.58 % 93.41 % 91.82 %
2 0-75 0.005 2 92.77 % 98.58 % 98.95 % 98.66 % 98.29 %
3 0-50 0.01 2 94.91 % 99.09 % 99.37 % 99.21 % 98.93 %
4 0-50 0.01 1 96.90 % 99.42 % 99.58 % 99.48 % 99.29 %

Analysis

As you might expect, the 'r' band performs best, and performance drops off the further we get from 'r'. Bands 'g' and 'i' have similar performance, while 'u' performs significantly worse than 'z'.

http://trac.astrometry.net/browser/trunk/doc/summary1.png?format=raw

However, as we put in more effort in subsequent rounds of solving, the 'u' band improves significantly and shows no sign of plateauing before solving nearly all the fields.

http://trac.astrometry.net/browser/trunk/doc/summary2.png?format=raw

http://trac.astrometry.net/browser/trunk/doc/summary3.png?format=raw


Conclusion

We get acceptable performance across the ugriz spectrum, but we'll have to be careful outside this range, especially on the low side.


The index is sdss-24. The data is in raid3/RUNS/103.