Changes between Version 146 and Version 147 of ppStack_testing_201111

Timestamp:

Jun 28, 2012, 3:09:23 PM (14 years ago)

Author:

Mark Huber

Comment:

--

ppStack_testing_201111

@@ -747 +747 @@
 == Target PSF for Convolved Stacks ==
 
-Want to explore not necessarily setting the convolved "matched/equilized/homogenized" deep-stack target PSF to what generally appears to be driven by the worst input. Have seen cases when the worse inputs could be image rejected and in that case not a good idea to use anyways. What level of deconvolution is acceptable?
-
- * for large N, a rerun of PSF evelope/target PSF would be of little time cost or just reject outright as little input to final stack? 
- * for small N may want to reject anyways? Adding poorest image versus dropping 20-30% of possible inputs?
+Want to explore not necessarily setting the convolved "matched/equilized/homogenized" deep-stack target PSF to what generally appears to be driven by the worst input. Have seen cases when the poor inputs could be image rejected and in that case not a good idea to use anyways, and the target PSF re-chosen and stack remade. How can poor imputs be pre-filtered better? Is a most common input based target PSF better? What level of deconvolution is acceptable? 
+
+ * for large input image N, a rerun of PSF evelope/target PSF would be of little time cost but rerun of all the image convolutions is terribly expensive. Will want better pre-rejection at least, however, addition of a deconvolved image may be acceptable.
+ * for small input image N, may want to reject anyways? Adding poorest image versus dropping 20-30% of possible inputs?
  * instead of a rerun of target PSF with additional image rejection, could choose mean/median PSF or an optimal and use poor deconvolution to reject worse images?
  * rejecting inputs for convolved stacks is the same for unconvolved stacks (for the pixel rejection method), so excessive rejection not good for either.
  * expect the rejecting and target to be different for MD (similar good inputs, easy to clip worse seeing tail without significant loss of depth) versus LAP (similar to wildly variable quality inputs, maybe few far outliers or maybe even spread of quality levels).
 
-What are the convolved stacks used for?
+Purpose of convolved stacks?
  * staticsky photometry -- convolved stacks have a similar PSF across the field for photometry and morphology measurements. However, also then goes through another round of (de)convolution/smoothing to match the PSF of all input filters.
- * other?
+ * matched PSF over skycell, semi-matched PSF over larger areas? How necessary is that currently?
+ * specific science reasons for either or?
+
+Test cases:
+ 0. Greatly reduce the input image FWHM limits and accept possibly fewer inputs into the stacks. For stacks that fail without enough inputs, have a second pass with more relaxed limits. Pass 1 could also require a larger number of minimum inputs (larger than the normal 2-6 inputs currently) with a very restrictive "FWHM" cut (good). Pass 2 more relaxed allowing fewer minimum inputs and/or relaxed cuts (good as going to get).
+  * affects the unconvolved and well as the convolved however
+ 1. Enhance the already exising simple model target PSF code to interally use the already calculated input FWHM mean (and stdev) to set a simple target PSF.
+   * is a convolution to a simple model PSF workable
+   * how much deconvolution is acceptable in the inputs to the stack
+ 2. Either modify the PSF envelope code to trend towards the more likely target PSF or sub-select the inputs for the the target PSF to be set to.
+   * open question if the PSF envelope code is behaving as intendent and needs to be tested further
+   * then similar to case 1 deconvolution acceptable over the field
 
 Test runs (same as defined above [wiki:ppStack_testing_201111#TestSets]): 
- * all previous improvements, like SYS.ERR=0, included comes after this stage for the convolution mainly so not important for picking target, but is important for final photometry/property comparisons.
- * since comparing between different runs, will want to use same SEED values for ppStack and psphot at least initially
- * be sure to note PSF model choices for the two samples, will want to try match initially but also look at when not well matched.
+ * all previous improvements included, like SYS.ERR=0, comes after this stage for the convolution mainly so not important for picking target, but is important for final photometry/property comparisons.
+ * since comparing between different runs, will want to use same SEED values for ppStack and psphot at least initially. Will also want exactly same input warps copies (i.e., from simtest run and not regenerate for each to avoid any differences even if using same seed there as well).
+ * be sure to note PSF model choices for the two samples, will want to try match initially, but also look at when not well matched later.
  * what is the baseline to use for relative comparison? Unconvolved stacks are made at same time, but can depend on the comparison. Simulated set has known values for its creation.
- * will want to include spatial variations, simtest sample should be uniform but MD04 will not be and may have "strange" inputs.
+ * will want to include spatial variations, simtest sample should be uniform but MD04 will not be and may have "strange" inputs to note and make use of.
  * not sure how to included information rejected images by some IQ or deconvolution limits.
  * Simtest sample:
-  * may need to make a sample distribution of simulated PSFs, currently sample is flat. Would need to revise the making of the stack.mdc list of inputs, or will just be worse/upper limit of possible stack degrading for including deconvolved images.
+  * may want to make a sample distribution of simulated PSF FWHMs, currently sample is flat. Would need to revise the making of the stack.mdc list of inputs, or will just be worse/upper limit of possible stack degrading for including deconvolved images.
   * start with sample N=20, not clear what N=100 would help to do here.
   * PSF model is _GAUSS for simulated images, start with matched target model and mpdel in psphot. Will want to look at different matched target models.