horizontally varying background error variance

[travissluka]

Currently, the background error variance is effectively 1C. A horizontally varying background error variance needs to be used by providing a LinearVariableChange

• I'm not sure if the saber::StdDevVariableChange can be used for this, probably not, it might only be used for the standard deviation when SABER is used to calculate the full covariance from a sample ensemble.
• otherwise, we need our own variable change class that reads in the background error standard deviation and implements the multiply and multiplyAD methods. Conceptually it shouldn't be very difficult. See https://github.com/JCSDA-internal/soca/blob/develop/src/soca/Transforms/BkgErr/BkgErr.h for an example with SOCA (though we of course we will but using all C++ instead)

For actual standard deviation values to use: we can just create a file with some made up values for now, and then later work on using scientifically meaningful values (perhaps use the data files from the current OISST?)

[loganchen39]

@travissluka Are we supposed to solve this issue right now? Can we make some tests now like you and Jim mentioned at our last meeting?

[travissluka]

@loganchen39 I think we can wait on this. For the first iteration of tests we can use a constant global correlation length and background error variance as discussed. I'll do a PR shortly that fixes the reason it is no longer compiling

[loganchen39]

@travissluka I'm back from the SODA project, now I want to work on this project. For starting the tests, my initial thoughts are:

1. Create a UMD-SST/src/mains/test.cc ,
2. Create a UMD-SST/test/testinput/test.yml , and set up parameters.
3. Prepare initial input SST data which is partially filled, I've put an example at feature/test branch, test/Data/20090701120000-ESACCI-L3C_GHRSST-SSTskin-AVHRR19_G-CDR2.1_night-v02.0-fv01.0.nc
   I know this file has 56m, it is only for temporary use.
4. The output should be a totally filled SST record?

Is my understanding correct? Do you have any comments?