[Isocops] Fwd: Full covariance and mapmaking

[Nathan Schwadron]

Begin forwarded message:

> From: "DeMajistre, Bob" <Bob.DeMajistre at jhuapl.edu>
> Date: December 22, 2010 6:01:55 PM EST
> To: Nathan Schwadron <nschwadron at guero.sr.unh.edu>, Ken Fairchild <fair-play at comcast.net>, Roland Vanderspek <rvdspek at bu.edu>, "Maher A. Dayeh" <maher.aldayeh at swri.org>, Paul Janzen <paul.janzen at umontana.edu>
> Cc: Mark Tapley <mtapley at swri.org>
> Subject: Full covariance and mapmaking
> 
> All,
> 
> I just committed the changes necessary to generate maps using the full correct covariance (rather than the approximate method that Geoff implemented).  You should probably wait for the nightly to run (cleanly) before trying this stuff out – it depends on stuff in $IBEX_OPS, which the nightly will update (but I have no idea how to update directly myself). If the nightly explodes tonight you can cast your heaps of derision on me. 
> 
> There should be no change to the way everything works with your existing cat files. In order to turn the changes on, you’ll have to set new variables in the cat file.  The new variables are as follows:
> 
> #### Full covariance flags OFF IS A NULL STRING
> ef_fullcov=1            #have hilo_reflux produce covariance file
> ef_cov2var=1          #copy fullcov stuff into reflux variances
> ft_fullcov=1             #turn on full covariance in flux translate
> fe_fullcov=1            #turn on full covariance in hilo_monoe
> fe_cov2var=1          #copy fullcov stuff into hilo_monoe variances
> 
> If you set all of these, then the fully propagated error will replace the old estimates in the final mapping files.
> 
> Now is a good time to stop reading if you don’t care about the details.
> 
> The C-G corrected maps are built by scripts that use the following executables (starting from me_bin output)
> hilo_reflux  - corrects the fluxes for finite energy width of the ESA channels
> flux_translate – takes orbit by orbit products and places them on a map
> hilo_monoe – puts everything on a common energy grid. 
> 
> Note that the FIRST STEP (using hilo_reflux) introduces dependence between energy channels, and therefore a full covariance begins here. 
> 
> The flux translate step operates on one channel (energy) at a time, there is no energy channel crosstalk. HOWEVER, the full covariances produced by hilo_reflux must be propagated because the energy channels in the map ARE NOT INDEPENDENT.
> 
> The flux_monoe  step does the CG correction and therefore further couples the energy channels.
> 
> To do this problem, I’ve introduced a new file type for carrying around the covariance matrices around. A new covariance file is generated at each step along the way – there is also the option to replace the current variance estimates with the fully propagated variances.
> 
> If you set ef_fullcov=1, the covariance matrix will be calculated and stored when hilo_reflux is run. If you also set ef_cov2var=1, the variances (i.e., the diagonal of the covariance matrix) are placed in the fvar1a files. Otherwise the fvar1au files will have Geoff’s propagation in them. You don’t have to set  ef_cov2var=1 to continue to the next step of full covariance propagation (though you do have to set ef_fullcov=1 to proceed with the propagation). Note that these are still orbit by orbit files.
> 
> Setting ft_fullcov=1 will continue the propagation through flux_translate. Note that flux_translate is run one channel at a time. flux_translate reads in the covariance file produced by hilo_reflux, and then propagates the covariance for the channel for which it is being run. The first time through, flux_translate produces a NEW covariance file with only one channel populated. Subsequent runs update this file. So by the end of the script – which runs on all 6 hi channels, the new covariance file is filled. Note that since there is no crosstalk in flux_translate itself, the error propagation goes on a usual. If you set ef_cov2var=1 (i.e., you told hilo_reflux to replace the estimates with the full covariance), the variance results from flux translate will have the full propagation as well.
> 
> Setting fe_fullcov=1  instructs hilo_monoe to open the full covariance produced by flux_translate and propagate the covariance. Thus a new output covariance fill will be built. Again, setting fe_cov2var=1 will move the variances from the full propagation file into the appropriate fvar output file.
> 
> I’ve also built some IDL readers for the covariance files (in the APL area) that I can tell you how to use. 
> 
> Happy hunting,
> Bob D.

Nathan Schwadron
n.schwadron at unh.edu

University of New Hampshire
Morse Hall - Room 353
8 College Road
Durham NH 03824
USA

(603) 862-3451



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.sr.unh.edu/mailman/private/isocops/attachments/20101223/a04fcaf9/attachment.html