Issues with weights in multiscale ensemble analysis

[jderber-NOAA]

While making some optimization changes to the multiscale ensemble analysis, an error in the weights was noticed. For the global mean (0,ig) weights, all of the values were set to the default of 1.0. Apparently the correct code was lost in the process of incorporating code into the GSI. The values should be 1.0 for ig=1 and 0.0 for all other values of ig.

The values were updated for the global mean in the init_mult_spc_wgts routine (apply_scaledepwgts.f90). Some other minor changes to this routine were done including setting proper variable type for constants, removing unnecessary arrays, using logical rather than 1 and 0. Also noted was a line that said spc_multwgt=max(spc_multwgt,0.0_r_kind). This max was moved inside the i and ig loops and a min included to ensure that the weight is not greater than one. Finally, a message is printed out if rtem2 < 0. and spc_multwgt(i,2) is set to zero.

The code is being put into my multiscalebug branch.

Tests on the values of spc_multwgt are being performed. Identical results are produced for the spc_multwgt values except for the global mean (0,ig>1).

[jderber-NOAA]

All regression tests passed (I don't think any of the tests use the routine changed). Should only change global mean wgt for ig > 1.