ECMWFTraining course 2005 slide 1
Observations Preprocessing
Carla Cardinali
ECMWFTraining course 2005 slide 2
Observations preprocessing
MakeCMA
Screening
Minimization
MatchUp
FeedBack
Forecast
ECMA
CCMA
ECMA
BUFR
ODB
BUFR
ECMWFTraining course 2005 slide 3
Observations preprocessing
ECMA/ODB
CCMA/ODB
Output BUFRs
ECMWFTraining course 2005 slide 4
Observations preprocessing
What is ODB?
Developed software at ECMWF to manage large amounts of satellite data.
ODB/SQL language is a subset of ANSI/SQL query language. Compiler translates
ODB/SQL in C-codes files
How do use it?
Access to ODB database is through Fortran90 modules functions. A ECMA data
base contains table with column entries lat, lon, date, time observation type
which is used to locate observation at a given moment
Obs. Ident ( sat. id, lat, lon, st.alt., date, time)
Obs. V. (wind, temp.,…per pressure) and (radiances per ch., inst. type)
Various flag: active, blacklisted….
Departures (obs-background, obs-analysis)
ECMWFTraining course 2005 slide 5
Observations preprocessing
Incoming Observation
The observations arrive at ECMWF through GTS and they are stored in a
decoded format in Report Data Base after some rudimentary quality control e.g.
observation format and position, climatological and hydrostatic limit and
temporal consistency. An observation file suitable for assimilation is created
6-hour Preprocessing Observation Array
Format conversions, change of some observed variables (relative humidity
from dry and wet bulb temperature) and assignment of observation error
statistics
ECMWFTraining course 2005 slide 6
Observations Screening
Select the “best” observation
At the first trajectory run the model counterparts for all the observations are
calculated through the nonlinear observation operators. For the observation
screening, the background errors are interpolated to the locations of the
observed variables. The "extended" observation array (ECMA) contains
observations complemented by the background departures and quality control
information for most of the observations.
This array is stored for later feedback.
After the screening a "compressed"
array is passed to the minimization
(CCMA).
ECMWFTraining course 2005 slide 7
Independent Observations Screening
The screening logic is to make first those decisions that are not
depending on any other ones
Preliminary checks: completeness of the report
Blacklist
Data Selection: which observation types, variables, vertical ranges will be used
in the assimilation.
Monthly selection: discarding stations that have been reporting excessively
noisy or biased compare to the background field.
Impact studies.
ECMWFTraining course 2005 slide 8
Independent Screening Decision: Conventional Observations
Background Quality Control
The BgQC is applied to all the variables that are intended to be used in the
assimilation
BgQC of wind observations is done simultaneously for both wind components. For wind direction the error limits of 60, 90 and 120 apply for flags 1,2 and 3
2 2
3
2
12
o b b
Hx y
uo
ub
ECMWFTraining course 2005 slide 9
Independent Screening Decision: Conventional Observations
ECMWFTraining course 2005 slide 10
When is not working
ECMWFTraining course 2005 slide 11
Independent Screening Decision: Satellite Observation
Bias Correction
Bias correction coefficients are recomputed from the past 2/4 weeks of
departure statistics. The feedback files are used for monitoring the performance
of the observing and assimilation system: e.g. ATOVS and SSMI radiances,
SCATT wind.
If the removed bias is a model forecast bias, the subsequent assimilation will enforce it. Usually, only half bias is removed.
Gross check Measured and background brightness temperatures are present for all required channels. Bias correction coefficients, satellite id, and scan position are all valid before proceeding. For all channels a cloud detection is performed
290K
230K
ECMWFTraining course 2005 slide 12
Dependent Screening Decision: Satellite Observation
HIRS
3
2
1
land +oceanChannels below rejected
SSMI
ocean: LWP=f(ch3/4) If LWP > cloud+rain
ECMWFTraining course 2005 slide 13
Dependent Screening Decision: Satellite Observation
ocean: LWP=f(ch1,ch2,ch3)If LWP > cloud+rainland: if (ch1-ch3) > rain because of large emissivity over land no cloud contamination detection
1 3
Tb
no cloud
cloud
rain
AMSU-A
12
3
4
5
6
ECMWFTraining course 2005 slide 14
Independent Screening Decision: Satellite Observation
Background Quality control: ATOVS, SSMI, METEOSAT
Background temperature, specific humidity and
ozone profiles are checked to make sure they are close to or within the range
for which the radiative transfer model is valid. Temperature is within the range
150-350 K, specific humidity is positive and not supersaturated and the ozone
is within climate extremes.
Radiance at the 2 extreme edge positions of the swath are not used in 4D-Var
O Rad RadT
b O HBH2
ECMWFTraining course 2005 slide 15
Independent Screening Decision: Satellite Observation
ECMWFTraining course 2005 slide 16
Independent Screening Decision: Satellite Observation
ECMWFTraining course 2005 slide 17
Dependent Screening Decision: Conventional Observation
Before performing the dependent screening decisions, the flag information
gathered so far is converted into a report status, namely active, passive,
rejected or blacklisted i.e the RDB datum flag.
Vertical consistency of multilevel reports
Duplicated levels are removed. If some consecutive layers are of suspicious
quality they are rejected and for geopotential obs. also the layer above
Removal of duplicated report
Search globally for duplicated (AIREP) data. Usually, there are aircraft reports
that have the same date/time, roughly the same location and, at least partially,
the same data. Usually, the station ID is slightly different.
ECMWFTraining course 2005 slide 18
Dependent Screening Decision: Conventional Observation
Redundancy check
Removes redundant SYNOP/PAOB :
co-located reports with the same station ID are searched for and only the
closest active report to the analysis time is retained. For simultaneous
reports, the one with more active data is retained. The same is done for
reports that have equal time difference to the analysis time. Co-located
SYNOP (mass obs) is redundant to TEMP (geop.) that is within 50 hPa
Removes redundant SHIP/DRIBU:
moving platforms within a circle of 1 ° are considered as potentially
redundant: reports closest to the centre of the screening time with most
active date are retained.
ECMWFTraining course 2005 slide 19
Dependent Screening Decision: SYNOP
The effect of the observation screening on SYNOP surface pressure observations. Column height gives the number of observations available, while the shaded part displays those actually used in the assimilation.
4D-Var Screening for 4D-Var
3D-Var Screening for 3D/4D-Var
ECMWFTraining course 2005 slide 20
Dependent Screening Decision: Conventional Observation
Level selection for TEMP/PILOT redundancies:
for one time window rejection in layers around std-levels according the
following priorities
maximum datum flag
time difference to analysis time
distance to the std-level
significant level (turning points of the sounding)
temp over pilot
ECMWFTraining course 2005 slide 21
Dependent Screening Decision: Thinning
AIREP thinning
horizontal thinning
from the same platform, a minimum distance between the nearby reports is
enforced: Box 125 62 km
vertical thinning
is performed around the model vertical levels: one aircraft measurement
per model level.
ECMWFTraining course 2005 slide 22
Dependent Screening Decision: Thinning
VERT increased number of AIREPfrom 15 to 60 vertical levels
+
+HOR also increased number of AIREP in horizontal
ECMWFTraining course 2005 slide 23
Screening Decision: Thinning
TEMP
GTSObs
thin
ECMWFTraining course 2005 slide 24
Dependent Screening Decision: Thinning
ATOVS thinning: a repeated scan is performed to get the
observation resolution of 70 km (it depends on sensors/channels)
a sea sounding is preferred to a land one
clear sounding to a cloudy one
closeness of observation time to the centre of the screening
time window
A second thinning takes place that selects one observation every
140km
ECMWFTraining course 2005 slide 25
Dependent Screening Decision: Thinning
The usage of ATOVS reports in the assimilation on the North Eastern Atlantic. Filled rings mark reports contain one or more channels used in the assimilation, whereas the empty rings denote rejected reports. Most of the rejections are due to the horizontal thinning and much less due to the quality reasons.Note that both edges of the swath are rejected.
ECMWFTraining course 2005 slide 26
Dependent Screening Decision: Thinning
SCAT:ERS-2
The process is defined with respect to the particular measurement geometry of
the instrument. The backscatter data are acquired within individual cells related
to a 450 km wide grid with a mesh of 25 km in the across and along track
directions. 19 measurement nodes are thus defined across the scatterometer´s
swath, numbered from 1 to 19 as the incidence angle increases, while 19 rows
are also considered in the along track direction to gather the data in squares of
19 by 19 points. The thinning is then achieved by keeping only every fourth
point within these squares. The data are thus used at a resolution of 100 km
instead of the original 25 km sampling distance.
Sea-ice contamination < 273K
High wind rejection test > 25m/s
Normalized distance to the cone or wind residual. During the assimilation this quantity is computed and data rejected if a large value is found.
ECMWFTraining course 2005 slide 27
Dependent Screening Decision: Thinning
Triplet obs
.. ..
.. ..
..
.. ..……....
.. 3 .. .. .. 7
check the distancebetween observations
1-step
2-step
ECMWFTraining course 2005 slide 28
QuikSCAT
50 km
Courtesy of Hans HersbachLars Isaksen Mark Leidner
Quadruplet obs
ECMWFTraining course 2005 slide 29
Dependent Screening Decision: Thinning
162
1
()(())OBSMOD
iii
MLE
uu
High resolution trajectory to choose only one
ECMWFTraining course 2005 slide 30
Dependent Screening Decision: Thinning
30°N30°N
40°N 40°N
50°N50°N
180°
180° 170°W
170°W
Obs$rv$d winds abov$ 1. m/sAmbig. clos$st to ana, from 2001 0518 0412 to 2001 0518 0733QSCAT 50km ($6dr): data cov$rag$
30°N30°N
40°N 40°N
50°N50°N
180°
180° 170°W
170°W
Rej: only 1 beam measur.active
Minima are not well defined
Var-QC
White areas: insuf. info for MLE inversion e.g rain contamination
ECMWFTraining course 2005 slide 31
Screening Statistics
281515286821701848026058052492792140
Reports Active Passive
SynopAirepSatobDribuTempPilotSatemPaobScat
Obs. Type
2421829969405001708602730558631380
Rejected
3933228991765183094375193416760
Blacklisted
00100699007154867760
ECMWFTraining course 2005 slide 32
After Screening
Compression of CMA-file
After screening only 15% of all observations are active
10-20% TOVS left
40% Conventional left
The observation are resorted among the processors for a more
optimal load balancing of the parallel computer
ECMWFTraining course 2005 slide 33
After Screening
Parallel computing environment
The observation screening should result in exactly the same selection of
observations when different number of processors are used. Independent
decision can be made at different processor fully in parallel. But global view of
observation array is needed when a dependent decision has to be taken which
implies that some communication between the processors is required. The
observation array is too large to be copied in each individual processor then
only a minimum necessary information is globally communicated
Global Array
Time
IDLocation 1
2