Dear all, here is a short and belated report on calibration of SDSS data obtained in non-photometric conditions, and its implications for LSST. A more detailed report (a journal paper) will be available in a month or so. Cheers, Zeljko ------------------------------------------------------------------- = Summary = LSST will be able to observe in partially cloudy (non-photometric) nights because such data can be calibrated with a sufficiently dense network of calibration stars. Such a dense network will be self-calibrated by LSST very soon after the first light. Given such a network, what is the largest cloud extinction for which LSST can maintain calibration accuracy of 1%? To answer this question, one needs to know the spatial cloud structure (structure function) and the density of calibration stars on the sky. Both of these can be addressed with SDSS data. The extrapolation of SDSS data to LSST parameters indicates that the upper limit on cloud extinction which still permits 1% accurate calibration is between 1 and 3 magnitudes. = Analysis = A few plots are available as http://www.astro.washington.edu/ivezic/sdss/calib/report-lsst.ps Many SDSS-II SNe runs are obtained in grossly non-photometric conditions (up to 6 magnitudes of cloud extinction). As such, they are ideal for studying the properties of cloud extinction and how clouds impact the accuracy of photometric calibration. Despite the fast variations of large cloud extinction, these data can be calibrated with the aid of a very dense standard star catalog. This catalog includes the majority of stars brighter than its faint cutoff (r~20.5 for photometric accuracy of 1% per star, see fig. 1), and was constructed by averaging repeated SDSS scans with ~10 observations per star (the photometric accuracy of individual observations is ~0.02 mag). An example of photometric calibration of such a cloudy run is shown in fig. 10. During the first 150 fields (corresponding to 1.5 hours of time) the cloud extinction varies between 0 and 6 mag, with variations as fast as 2 mag/second. An example of a more moderate, though still grossly non-photometric run, is shown in fig. 12. As the bottom two panels show, although the cloud extinction can be up to 0.6 mag, the zeropoint accuracy is well below 0.01 mag. This accuracy is achieved by calibrating data in small 5 arcmin2 patches that have a sufficient number of stars for constraining zeropoints (from ~10-20 in the u and z bands to ~40-50 in the gri bands), and yet are small enough that the cloud extinction variation across the patch is typically negligible. The patches are rectangular with the aspect ratio of 1:20. The need for such an aspect ratio is a result of drift scanning (TDI) technique and the fact that clouds move on the sky with typical angular speeds of 4-15 deg/min. The correlation between zeropoint error and cloud extinction for a very cloud run is shown in fig. 17 (each symbol is one calibration patch). For 95% of patches, the achieved zeropoint accuracy is better than 0.05*X, where X is the cloud extinction. This level of accuracy is determined by the size of calibration patch. For example, a smaller patch would suffer less from the spatial variation of cloud extinction, but it wouldn't have enough stars to beat down the noise of their individual photometric measurements (~0.02 mag). A similar approach to the calibration of LSST data (assuming that a standard star catalog is available, e.g. from photometric nights) would benefit from several effects: 1) The calibration patches can be squares; for the same area, this results in a 5 times smaller angular scale, compared to 1:20 rectangles used to calibrate SDSS TDI data. On these angular scale, the cloud structure function is roughly linear (a quantitative plot not provided, but fig. 20 serves as an qualitative illustration) and thus the zeropoint error would be of the order 1% through 1 mag thick clouds 2) LSST will be deeper by about 3 mag. With a conservative assumption that log(N) \propto 0.3*m (0.6 for Euclidean counts), the surface density of stars will be about 10 times larger for LSST than for SDSS. This larger density enables 10 times smaller patches, or about 3 times smaller angular scale for calibration, resulting in another factor of 3 improvement of accuracy. 3) Fitting a smooth function for cloud opacity over several calibration patches would result in further improvements. 1) and 2) predict that LSST data could be calibrated with a 1% accuracy even through 3 mag thick clouds. Given the extrapolations, I adopt the range 1-3 mag as the upper limit on the acceptable cloud opacity. ------------------------------------------------------------------- _______________________________________________ LSST-calibration mailing list LSST-calibration@lsstmail.org http://www.lsstmail.org/mailman/listinfo/lsst-calibration