Transcript cdc analysis has smaller standard deviation

INTERCOMPARISON – HIRLAM vs. ARPA-SIM
CARPE DIEM AREA 1
Per Kållberg Magnus Lindskog
what I want to show here
the HIRLAM system and experiments
comparison of arpa and hirlam analyses
verification of forecasts against own analyses
verification of forecasts against observations
precipitation forecasts
conclusions
HIRLAM (HIgh Resolution Limited Area Model)
•0.1º to 0.4° rotated lat/long grid
•Hydrostatic, hybrid coordinates
•Spectral (double Fourier with extension zone)
•Or gridpoints on the C-grid
•Eulerian or semi-Lagrangean time-stepping
•Lateral boundary relaxation - usually ECMWF LBC
•ISBA soil model
•TKE (Turbulent Kinetic Energy) turbulence closure
•Kain-Fritsch convection or ’STRACO’ condensation
HIRVDA (HIRlam Variational Data Assimilation)
•3D-Var or 4D-Var
•Multivariate statistical balance: vorticity - divergence - mass - moisture
•Scale and latitude dependent geostrophy
•Boundary layer friction
•’NMC’-method background error statistics
•Ensemble assimilations to replace ’NMC’-method
•Moisture effects with a revised moisture control variable
•Initialization: normal modes or a weak digital filter
•Observation operators include:
•Conventional (TEMP PILOT AIREP DRIBU SYNOP SHIP SATOB)
•Raw radiances (TOVS, ATOVS)
•Integrated humidity from GPS
•Radial winds from Doppler radars
hirlam experiments – nov.3 to nov.8
cdc
HIRLAM 6.1.
0.1°/ 0.1° 40 levels
3D-Var data assimilation
digital filter initialization (DFI)
ECMWF operational analyses on the
boundaries
ECMWF operational conventional
observations
’straco’ condensation & ’cbr’ turbulence
cdd
no data assimilation at all, just a +144
forecast from 3 november
cde
as cdc but with revised horizontal structure
functions
(slightly smaller scales)
comparisons between
the arpa and the hirlam
data assimilations
analysis
differences
arp – cdc
850 hPa
geopotential
6 Nov. 00Z
we have used
different
orographies
this affects the
post-processing to
pressure levels
and
mean sea level
2 metre temperature analysis differences
00UTC (left) and 12UTC (right)
arpa (left) and hirlam (right) 10 metre wind analyses
mean sea level pressure analysis and SYNOP observations
5 November 1999 12ÙTC
arp
cdc
mean sea level pressure analysis and SYNOP observations
6 November 1999 00ÙTC
arp
cdc
comparisons between
the arpa and the hirlam
forecasts
(verified against ’own’ analyses)
the analyses and the +24h forecast errors at the analysis time
mean sea level pressure on November 7th 1999
cdc
cdc
12Z
00Z
arp
arp
verification against observations
sea level pressure (arp & cdc) fit to SYNOP/SHIP
screen level temperature (arp & cdc) fit to SYNOP/SHIP
screen level dewpoint (top) and total clouds (bottom)
(arp & cdc) fit to SYNOP/SHIP
10 metre windspeed (arp & cdc) fit to SYNOP/SHIP (top)
and SHIP only (bottom)
850hPa geopotential (arp & cdc) fit to TEMP
850hPa windspeed (arp & cdc) fit to TEMP/PILOT
conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
more conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
•
P_msl 24h forecasts have comparable qualities
more conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
•
P_msl 24h forecasts have comparable qualities
•
10-metre windspeeds.
– cdc biased high, both in anlyses and – worse – in forecasts
– arp strong diurnal variations in the analysis biases – good at night, to weak
at day
more conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
•
P_msl 24h forecasts have comparable qualities
•
10-metre windspeeds.
– cdc biased high, both in anlyses and – worse – in forecasts
– arp strong diurnal variations in the analysis biases – good at night, to weak
at day
•
screen level temperature - analyses
– cdc biased warm at daytime, cool at night
– arp biased cool at daytime, warm at night
more conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
•
P_msl: 24h forecasts have comparable qualities
•
10-metre windspeeds
– cdc biased high, both in anlyses and – worse – in forecasts
– arp strong diurnal variations in the analysis biases – good at night, to weak
at day
•
screen level temperature - analyses
– cdc biased warm at daytime, cool at night
– arp biased cool at daytime, warm at night
•
screen level temperature – forecasts
– cdc has a cooling drift (well known in SMHI operations)
– arp quite biasfree forecasts
more conclusions from the comparisons with observations
•
P_msl: cdc analysis has smaller standard deviation
– (arp postprocessing is noisier)
•
P_msl: 24h forecasts have comparable qualities
•
10-metre windspeeds
– cdc biased high, both in anlyses and – worse – in forecasts
– arp strong diurnal variations in the analysis biases – good at night, to weak
at day
•
screen level temperature - analyses
– cdc biased warm at daytime, cool at night
– arp biased cool at daytime, warm at night
•
screen level temperature – forecasts
– cdc has a cooling drift (well known in SMHI operations)
– arp quite biasfree forecasts
•
total clouds: arp has more clouds than cdc. cdc has a diurnal cycle
more conclusions from the comparisons with observations
•
850hPa geopotential: analyses and forecast essentially similar fits
•
850hPa windspeed: cdc somewhat smaller bias and standard
deviation
accumulated precipitation
• cdc
6 Nov. 06Z + 24h
• cde
6 Nov. 06Z + 24h
• arp
6 Nov. 00Z + 24h
• arp
6 Nov. 12Z + 24h
• Rubel
6 Nov. 06Z - 7 Nov. 06Z
• cdd
6 Nov. 06Z – 7 Nov. 06Z
24-hour accumulated precipitation 6 Nov 06Z to 7 Nov 06Z
exp:cdc
24-hour accumulated precipitation 6 Nov 00Z to 7 Nov 00Z
exp:arp
24-hour accumulated precipitation 6 Nov 12Z to 7 Nov 12Z
exp:arp
24-hour accumulated precipitation 6 Nov 06Z to 7 Nov 06Z
(Rubel & Rudolf, Wien )
24-hour accumulated precipitation 6 Nov 06Z to 7 Nov 06Z
exp:cdd
the somewhat tighter structure functions used in
the hirlam cde experiment
experiment yields somewhat more intense precipitation
than the cdc control experiment
24-hour accumulated precipitation 6 Nov 06Z to 7 Nov 06Z
exp:cdc
24-hour accumulated precipitation 6 Nov 06Z to 7 Nov 06Z
exp:cde
general conclusions from the comparisons
•
pressure and mean sea level differences due to different orographies
and different post-processing algorithms
– arp noisier, especially Pmsl and geopotential at 850
•
too large scale of the hirlam background errors (0.4°/ 0.4° grid)
– new, smaller scale background errors yield slightly more intense
precipitation
•
•
•
analysis increments on model levels problematic in steep orography
dfi initialization not ideally tuned for this resolution and such a small
area
long integration (cdd) without D.A. still skillful, but D.A. improves the
quality
•
cdc Pmsl forecasts have generally smaller errors against own analysis
•
precipitation forecasts qualitatively good,
– arp has some very intense spots, cdc is somewhat smoother
•
and not bad quantitatively either
what we still want to do
•one more hirlam assimilation with a revised turbulent momentum flux
•run some forecasts from each other’s analyses
Grazie mille per la vostra attenzione!