NMC method

1
NMC method

• As proxy for background error, use 48-24hr
  forecasts verifying at same time
• Used originally with SSI because background error
  represented in spectral space
• Not clear at the time how to derive from
  innovation statistics as done with OI

2
NMC method

• Has worked surprisingly well in SSI
• Not clear that this is best approach for
  estimating parameters in GSI
• For SSI, complete spectrum of correlation
  estimated, along with latitudinally dependent
  variances, but physical space correlations
  homogeneous
• For GSI, only correlation length and variance
  estimated, but both can be functions of position

3
NMC method

• Statistical balance between psi and psfc, T, chi
• - in SSI, coupling scale dependent
• - in GSI, no scale dependence

4
NMC method

• Can flow dependent correlations help?
• Should we try to use innovation information?
  How?

5
NMC method

• Problem with using statistics derived from one
  model in a different model
• Vertical interpolation to new vertical coordinate
  does not work for statistical balance
  parametersproblem worst near surface
• This was overcome by projecting stream function
  errors onto a smooth basis in the vertical before
  computing the statistical balance parameters, so
  that the coupling parameters were a smooth
  analytic function that could be evaluated at any
  height. This is useful when statistics for a
  particular model are not available.