Unacceptably large observation errors due to Surface Obs

[rtodling]

I have pointed this out quite some time ago, but perhaps the lack for a formal issue has put the problem in the back burn.

There is something incorrect or missing in the yaml files controlling the assimilation of Surface Marine and Surface Land Stations in the sense that the observation errors come out way too large:

CostJo : Nonlinear Jo(Surface Marine Stations) = 12142.6, nobs = 101149, Jo/n = 0.120047, err = 7.07671e+08
CostJo : Nonlinear Jo(Surface Land Stations) = 4957.76, nobs = 74935, Jo/n = 0.0661609, err = 9.68093e+09

the numbers in the last column should be order 1 or so.

[gmao-yzhu]

Wei, did you work on this type of obs? Are those bad data? Can you duplicate Ricardo's results?

[gmao-wgu]

This is most likely harmless. It indicates that some observations passed QC but were assigned very large observation errors due to the inflation procedure in surface pressure check.

However harmless this might be this will infect statics we end up doing in the various ops-output. Can we please look at this?