Jan 09, 2016
• EnKF analysis reduced the innovations but smaller than GSI
Start from 0900 with adaptive inflation but with max inflation up to 1.2 (green)
Experiments described in the report with adaptive inflation but with max inflation up to 1.2 (red)
As second experiments but with unlimited adaptive inflation (blue)
As third but add perturbation to the forecast fields before running EnKF(pink)
GSI analysis (black)
ps
Pw: GPSIPW GPS-INTEGRATED PRECIPITABLE WATER
Three hours forecast Innovations
Potiention temperature
Water vapor mixing ratio
Three hours forecast Innovations
X component wind
Y component wind
Three hours forecast Innovations
GSIEnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
Surface u
24 hours analysis increment of GSI and EnKF
GSI
24 hours analysis increment of GSI and EnKF
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
Top level t
GSIEnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
Surface temperature
24 hours analysis increment of GSI and EnKF
GSIEnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
Top level u
24 hours analysis increment of GSI and EnKF
24 hours Inflation of T, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
24 hours prior spred of T, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
24 hours post spred of T, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
24 hours Inflation of U, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
24 hours prior spread of U, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
24 hours post spread of U, surface
EnKF adaptive, max inflation=1.2
EnKF adaptive, unlimited EnKF adaptive unlimit+pert init
Know issues
• The increment of EnKF analysis after 24 fours is smaller in the CONUS domain. It is not reasonable since the most of observations are within the CONUS Domain.
• The spread is higher outside the CONUS domain and smaller within the CONUS domain which indicates the convergence of EnKF is too faster in the observation area.
Next move• Double check the localization. And the U and V
coordinate.• Excluded the observation type ‘pw’, because the
innovation of ‘pw’ is larger than GSI even at the first 24 hours analysis
• Add properly perturbation to the three hours forecast fields before EnKF analysis. I have tested to use the random cv methods but crashed after 3 days cycle, I’m thinking a way add perturbation in the observation area, since the spread is smaller in the CONUS domain.
inflation' 2 ' 2
' 2 ' 2
analpertwt *( analpertwt , analpertwt 1r1=
analpertwt *( , analpertwt 1
logsigma ln( / )
r = abs(logsigma/lnsigcovinfcutoff)
R=r1+taper(r)
) / (
*covinflate
anal
) 1
) / ( )
pertwt 0.8;
b a
b a
pres prssfc
x x
x x
covinflate 0.1; lnsigcovinfcutoff 6
taper(r)
logsigma ln( / ) -2.995732274
(logsigma / lnsigcovinfcutoff ) 0.4992
5
88712
0
10
3
00
prestop prssfc
r ab
prestop
prssfc
s