CMSAF water vapour in comparison to radiosondes and CHAMP
Ralf Lindau
Water vapour from ATOVS
Global daily fields of Layered Precipitable Water LPW are calculated by kriging method on 90 km resolution.
Advantage of kriging : An error map is available for each field.
Example: 1.10.2004
850 – 700 hPa
Radiosondes
173 GUAN
(GCOS Upper-Air Network) stations well distributed over the globe.
It‘s the resolution, stupid
For any comparison of two data sets resolution is crucial.
If resolutions are not equal, the often actually seriously presented conclusion would be:
The low resolved observations underestimate high values and overestimate low values.
ATOVS fields: Daily , 90 km
Radiosondes: 4 times a day, point measurements
Thus, average the radiosondes over 1 day.
But still Radiosonde data include additionally the spatial variance of daily
means within 90 x 90 km
2Variability of Water Vapour
Famous paper:
Lindau, R. and E. Ruprecht, 2000:
SSM/I-derived total vapour content o ver the Baltic Sea
compared to independent data, Met.Zeitschrift, 9, No.2, 117-123.
1 day include 8 mm2 of variance 90 by 90 km2
include 3 mm2 of variance The spatial variance of temporal means
must be smaller than that of individuals (3 mm2).
October 2004
+ 3.82 mm + 17 %
High correlation: r = 0.95 But wet bias of 3.82 mm2
Happy with correlation ?
Correlation reduction due errors is:
1 - r = e2 / (2 + e2 )
= 7.5 mm2 / 225 mm2
= 0.03
More than half of the scatter is explained by random errors in ATOVS and RS.
But there is still the bias....
The error of ATOVS is explicitely calculated within kriging.
The mean error variance for October 2004 and for those 173 gridboxes, where RS data is available is:
e
AT= 6.19 mm
2The error variance of daily means from RS is calculated by:
Internal variance / observ number:
e
RS= 7.65 mm
2Different layers
+ 2.35 mm
+ 21 % + 0.18 mm
+ 5 %
1000 – 850 hPa 700 – 500 hPa
Time series of the bias
The bias is not confined to October 2004.
It persists through
10 month
Discriminating GUAM data
Ocean / Land Height above sea
or is it the ice surface in the Antarctica?
Ocean, Height, Ice ?
+ 2.85 mm + 16 %
+ 1.68 mm + 9 %
- 0.18 mm - 7 %
Different First-guesses
No
NCEP
Different first-guess (taken as basis for the retrieval) result in strong differences.
Example: 1.9.2006, 700-500 hPa
ATOVS vs Radiosondes
No
GME
NCEP
- 0.82 mm - 3 %
+ 3.67 mm + 14 %
+ 7.08 mm + 27 %
The corresponding bias vanishes for „No first guess“
and is doubled, if NCEP is used.
... and what about Champ ?
Example for 1 month of Champ data.
4239 obs/month 1 obs/10 min
Irregularly spreaded over the globe
ATOVS vs Champ
850 – 700 hPa700 – 500 hPa - 3.20 mm
- 15 %
- 0.15 mm - 5 %
+ 0.02 mm + 2 %