TYPE: new feature

KEYWORDS: WRFDA, ep, ep_format

SOURCE: Jamie Bresch (NCAR)

DESCRIPTION OF CHANGES:

ep_format=1:  (default) original format, double precision, each ep file is for one variable and one member, as the output from the current gen_be_ep2.
ep_format=11: same as ep_format=1 except data are in single precision
ep_format=2:  single precision, each ep file is for one variable and all members
ep_format=3:  single precision, each ep file is for one variable and all members but on decomposed patch domain

No impact on existing ep reading results.
New WRFDA code should work without any changes from the user when ep files are generated using the existing very inefficient gen_be_ep2 program.
New utility for generating ep in new formats (ep_format=1, 11, 2, 3 through namelist option) is proposed in PR #912.

LIST OF MODIFIED FILES:

modified:   Registry/registry.var
modified:   var/build/depend.txt
modified:   var/da/da_main/da_solve.inc
modified:   var/da/da_main/da_wrfvar_top.f90
modified:   var/da/da_setup_structures/da_setup_flow_predictors.inc
new file:   var/da/da_setup_structures/da_setup_flow_predictors_ep_format2.inc
new file:   var/da/da_setup_structures/da_setup_flow_predictors_ep_format3.inc
modified:   var/da/da_setup_structures/da_setup_structures.f90

TESTS CONDUCTED:

A case with 80-member 414x324x50 running with nproc=216.
The ep reading time:

| EP Format     | Time (s)  |   
| -------------:|----------:|
|     1         |   52.0    |
|     11        |   45.0    |
|     2         |   24.0    |
|     3         |    0.3    |

ep_format=1   ~ 52 sec
ep_format=11  ~ 45 sec
ep_format=2   ~ 24 sec
ep_format=3   ~ 0.3 sec 
Another major benefit to ep_format=3 is the reduced memory requirement.
New ep_format=11,2,3 produce identical analysis results.

RELEASE NOTE: New DA ep_format option to read in ensemble perturbation (ep) generated by existing gen_be_ep2 and new gen_be_v3 utilities.

…ting tool (#1088)

TYPE: enhancement

KEYWORDS: WRFDA, gen_be_v3, memory, be.dat for cloud_cv_options=2

SOURCE: Jamie Bresch (NCAR)

DESCRIPTION OF CHANGES:
1. For memory issue:
For applications that are for large number of cases and large domain sizes,
the gen_be_v3 program needs to run on a large-memory node.
To deal with the memory issue, add a namelist option pert1_read_opt
to choose how the program accesses pert1 data (output of do_pert_calc=true) internally.
pert1_read_opt=1: (default, original behavior) read and store all cases in memory at once.
pert1_read_opt=2: read from pert1 file when need it.
This option writes out additional vertical-mode-projected fields when do_eof_transform=.true.
Users can set pert1_read_opt=2 if memory insufficiency occurs.
With pert1_read_opt=2, the program runs slower than pert1_read_opt=1 but the memory usage is reduced

2. For reformatting individual be_[varname].dat into be.dat that WRFDA reads:
PR #912 already includes var/gen_be_v3/util/combine_be_cv7.f90 that handles basic variables.
This PR adds another var/gen_be_v3/util/combine_be_cv7_ccv2.f90 that reformat GEN_BE_V3 output into the same format as GENBE_2.0 for cloud_cv_options=2 applications.

LIST OF MODIFIED FILES:
M       var/gen_be_v3/README.gen_be_v3
M       var/gen_be_v3/gen_be_v3.F90
A       var/gen_be_v3/util/combine_be_cv7_ccv2.f90

TESTS CONDUCTED:
The results are identical with pert1_read_opt=1 and pert1_read_opt=2.