All Observation rejected in I4DVAR_Impact

[mumzy]

I have assimilated AVISO_sla and surface current in the new ROMS 3.7version, some obs were rejected and soma were accepted in the assimilation step as follow:
Number of State Observations Processed: ObsTime = 4773.0000, 2013-01-25 00:00:00.00

Variable IstrObs IendObs Count Rejected

zeta 7266 14530 7265 5767
u 14531 14533 3 0
v 14534 14536 3 1

Total 7271 5768
Obs Tally 14536 11549

Wrote TLM increments at observation locations, datum = 0007266 - 0014536

The obs number accepted in the model would change with the number of STD threshold when the BackGround Quality Control in terms of obs provenance in i4dvar.in, but all obs were rejected in I4DVAR_impact step:

Number of State Observations Processed: ObsTime = 4773.0000, 2013-01-25 00:00:00.00

Variable IstrObs IendObs Count Rejected

zeta 7266 14530 7265 7265
u 14531 14533 3 3
v 14534 14536 3 3

Total 7271 7271
Obs Tally 14536 14536

Wrote 4DVAR observation sensitivity, datum = 0007266 - 0014536

The BGQC is same with the assimilation and this situation doesn't change with the STD threshold.
Anyone knows? thanks so much!