After careful checks, the estimated result is still below the measurement. For our geometry, the result from SK2009 is similar to that from Harry2002.

 So here are the results

• The calculation for coating Brownian noise from SK2009 (finite size mirror) is similar to Harry2002 result (half infinite mirror).
• I double checked my code by changing my parameters to those used in SK2009 paper and got the same result. Apparently, their spotsize is very big compare to ours (1/3 of the mirror radius vs 1/40 of the mirror radius).
• I revisited my comsol model again to check Brownian noise from substrate and spacer when the mirror is curve (more realistic model). The difference is small. (add fig).
  

The optical bonding area in this model is similar to the real cavity, compared to what is done before where the bonding area is everywhere on the mirror beside the bore hole. So it is quite certain that it is the noise from the substrate/spacer.

Since the measurement has 1/f slope, it is very likely to be Brownian thermal noise (Thermoelastic/ TO will have different slopes). It might be that the 1998 mirrors have high loss. We will see that with shorter cavity measurement.