1. The impact of satellite observations over land and sea-ice surfaces within the ECMWF system
Cristina Lupu, Niels Bormann, Reima Eresmaa
The WMO 6th Workshop on the Impact of various observing systems in NWP
Shanghai, China, May 10-13, 2016

2. Outline Surface-sensitive MW sounder data IR sounder & imager data
Use over land and sea-ice at ECMWF
OSEs study impact results
FSOI evaluation
IR sounder & imager data
Impact results and challenges
Conclusions

3. Use of surface-sensitive MW sounder data over land/sea-ice
Use of surface-sensitive radiances requires reliable estimates of surface emissivity and skin temperature.
Over sea: Accurate fast emissivity model (FASTEM), Sea-surface temperature (SST)
Over land + sea-ice:
Use surface emissivity retrieved from window channel observations and FG (“dynamic emissivities”, Karbou et al. 2006), complemented by an atlas where required.
Chosen window channel should have similar frequency as sounding channels and good surface sensitivity.
Skin temperature taken from land-surface model.

4. OSE study: Channels considered and their use over land/sea-ice
Instrument
Satellites
Snow-free land
Snow-covered land
Sea-ice
AMSU-A
(clear-sky)
Metop-A, Metop-B, NOAA-15,
NOAA-18, NOAA-19
Channels 5-7
Channels 6-7
Channel 5 over N.Hem.
ATMS
S-NPP
Channels 6-8; 18-22
Channels 6-8 -
MHS
(all-sky)
Metop-A,
Metop-B,
NOAA-18,
NOAA-19
Channels 3-5
Channel 3
Channels 3-4
SSMI/S
F-17
Channels 9-11
Channels 10-11
Temperature-sounding in red; Humidity-sounding in blue;
Channel-dependent orography screening also applied.

5. OSEs for surface-sensitive MW sounder data over land/sea-ice
Experiments over 8 months:
2 June – 30 Sept 2014; 2 Dec 2014 – 31 March 2015
Base: No surface-sensitive MW sounder data over land and sea-ice
Base + MW sea-ice: Add surface-sensitive MW sounders over sea-ice
(i.e., 5 x AMSU-A, MHS, SSMI/S)
Base + MW sea-ice + MW WV land: Add MW humidity sounders over land
(i.e., ATMS, 4 x MHS, SSMI/S)
Base + MW sea-ice + MW land: Add surface-sensitive MW temperature sounders over land (i.e., data usage as in operations)

6. Impact over sea-ice and land

7. Impact of surface-sensitive MW sounders over sea-ice and land
Base + MW sea-ice + MW land
Base + MW sea-ice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis
Bad
Good

8. Impact of surface-sensitive MW sounders over sea-ice and land
Normalised change in Stdev error of vector wind,
Base + MW sea-ice + MW land vs Base,
Strong impact over extra-tropics/higher latitudes.
Modest impact in tropics.
Good
Bad
-0.04
-0.02
0.00
0.02
0.04

9. Short-range forecast impact against other observations
Stdev(o-b) normalised by Base, global statistics, 8 months
Good
Bad
Good
Bad
TEMP-T
TEMP-q
CONV-VW
IASI
H2O O3
Window/
lower T

10. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Base + MW sea-ice + MW land
Base + MW sea-ice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis
June – Sept
2014
Dec 2014 –
March 2015

11. Seasonal dependence of impact of surface-sensitive MW data over land and sea-ice
Base + MW sea-ice + MW land
Base + MW sea-ice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis.
June – Sept
2014
Larger sea-ice extent
Dec 2014 –
March 2015

12. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Base + MW sea-ice + MW land
Base + MW sea-ice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis.
June – Sept
2014
Less sea-ice
Dec 2014 –
March 2015

13. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Base + MW sea-ice + MW land
Base + MW sea-ice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis.
June – Sept
2014
Less impact in winter:
Difficulties due to snow?
Sampling?
Dec 2014 –
March 2015

14. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Normalised change in Stdev error of
Z 500 hPa
Base + MW sea-ice + MW land
vs Base
T+48 h
0.15
0.10
0.05
June – Sept
2014
0.00
-0.05
Dec 2014 –
March 2015
-0.10
-0.15

15. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Data coverage, MHS channel 4 (Metop-B)
Base
Base + MW sea-ice + MW land
Aug
2014
Difference in sea-ice extent
Not used over snow in winter
Feb
2015

16. Impact from temperature and humidity-sounding channels

17. Impact of surface-sensitive MW humidity and temperature sounders over land
Base + MW sea-ice + MW land
Base + MW sea-ice + MW WV land
vs Base + MW sea-ice, normalised
difference in RMSE for Z 500 hPa.
Bad
Good

18. Short-range forecast impact against other observations
Stdev(o-b) normalised by Base, global statistics, 8 months
Good
Bad
Good
Bad
TEMP-T
TEMP-q
CONV-VW
IASI
H2O O3
Window/
lower T

19. Seasonal dependence of impact of MW humidity and temperature sounders over land
Base + MW sea-ice + MW land
Base + MW sea-ice + MW WV land
vs Base + MW sea-ice
Dec 2014 – March 2015
June – Sept 2014
Bad
Good

20. FSOI: surface-sensitive MW sounders over land and sea-ice
1.53% MW land
1.59% MW seaice
2.27% MW land
0.87% MW seaice

21. Forecast sensitivity to observations impact (FSOI) - Globe
Instr.
FSOI
%,Land
%, Sea-ice
AMSU-A
2.39 %
1.22 %
ATMS
1.45 %
0.37% T sounding
1.07% q sounding -
MHS
1.32 %
1.07 %
SSMI/S
0.22 %
0.17 %
Total
5.38 %
2.76% T sounding
2.61% q sounding
2.46 %

22. Conclusions
Surface-sensitive MW sounder data over land and sea-ice provide significant positive forecast impact in the operational ECMWF system.
Very significant positive impact from data over sea-ice over S. Hemis.
Comparable impact from temperature and humidity-sounding channels over land.
Results suggest significant seasonal dependence of the forecast impact, most likely due to changes in surface characteristics.
N.Hemisphere impact much weaker over winter.
Most likely due to limitations in surface emissivity or skin temperature estimation over snow, leading to a more restricted/sub-optimal use.
S.Hemisphere impact appears to be linked to sea-ice extent.
FSOI evaluation results are well in agreement with the OSEs impact study results.

23. Operational use of IR sounder data over land and sea-ice
IR sounder radiances are currently more exploited over sea than over land:
Over sea: ISEM sea-surface emissivity model;
Over land: Limited number of channels insensitive to surface emission (emissivity is assigned default value independent of the wavenumber or SZA); a background error of 5K is assumed for the Tskin.
Over sea-ice: Limited to channels in the long-wave CO2 absorption band;
Instrument
Satellites
Sea
Land
Sea-ice
IASI
Metop-A/B
188 channels
No data
162 channels
AIRS
Aqua
136 channels
48 stratospheric channels
88 channels
CrIS
S-NPP
77 channels
30 stratospheric channels
52 channels

24. Impact of operational IR radiances assimilated over land and sea-ice
Experiments over 4 months: 2 June – 30 July 2014; 2 Dec 2014 – 23 January 2015 Base + IR seaice + land Base + IR seaice
vs Base, normalised difference in RMSE for
Z 500 hPa, verification against own analysis
Bad
Good

25. Trials to extend the use of advanced IR sounders data over land
The set-up over land is just the same as over sea, but not over high orography
Instrument
Satellites
Sea
Land
Sea-ice
IASI
Metop-A/B
188 channels
162 channels
AIRS
Aqua
136 channels
88 channels
CrIS
S-NPP
77 channels
52 channels
Mixed impact on forecast scores (e.g., RMSE Z 500 hPa) and on the background fit to other observations (e.g., MW observations)
Bad
Good

26. Way forward
The future use of infrared radiances over land is likely to rely on fundamentally different cloud detection and improved handling of skin temperature for satellite data assimilation.

27. Impact of emissivity modelling on BT simulations
emis=atlas
emis=0.98
BT differences between IASI
observations and simulations
RTTOV IR UWiremis land emissivity atlas (Borbas et al., 2007)
Window ch cm-1
Assuming constant emissivity, the simulated IASI spectra in the window regions significantly deviate from measurements. Need good a-priori information of emissivity over land !

28. Surface skin temperature at ECMWF - issues
Tskin is not independently observed;
Highly reactive in space and time;
Polar orbiting satellites have a very biased diurnal sampling of the skin temperature (2 passes per day)
Complex to maintain, is changing during minimisation
It is an independent variable, separate to the atmospheric temperature and skin temperature at other locations
Atmospheric information aliases into Tskin and is “lost”
Does not influence the model forecast of skin temperature;
Strategy:
Evaluate the possibility of moving away from sink variable to a full field control variable.
Better characterize the background errors for Tskin.

29. Strategy for cloud detection over land
Large uncertainties in surface emission make the current cloud detection for IR radiances unfit for use over land
Cloud detection land using observed TB only
So far experimenting with IASI only
Initial attempts rely on four super-channels, each super-channel ~10-12 channels
(1) Channels in wavenumber range 706—715 cm-1
(2) Channels in wavenumber range 722—740 cm-1
(3) Channels around wavenumber 875 cm-1
(4) Channels around wavenumber cm-1
Two checks for presence of cloud in the field-of-view
Sounding-channel check
using super-channels (1) and (2)
Check passed if data falls between these lines
Window-channel check
using super-channels (3) and (4)
Check passed if data falls above these lines
Clear
Cloudy
Clear
Cloudy

30. Locations of clear data on a window channel
Sea
Snow or ice surface
Desert
Vegetation
Tune the scheme for data over vegetation, desert and snow
False alarms: Canada and tropical rainforests?
FG-departure clear data (shown over vegetation)
Non-gaussianity, cloud-contaminated data getting through?

31. Conclusions
Surface-sensitive MW sounder data over land and sea-ice provide significant positive forecast impact in the operational ECMWF system.
Very significant positive impact from data over sea-ice over S. Hemis.
Comparable impact from temperature and humidity-sounding channels over land.
Results suggest significant seasonal dependence of the forecast impact, most likely due to changes in surface characteristics.
N.Hemisphere impact much weaker over winter.
Most likely due to limitations in surface emissivity or skin temperature estimation over snow, leading to a more restricted/sub-optimal use.
S.Hemisphere impact appears to be linked to sea-ice extent.
IR sounder data over land and sea-ice – some progress has been made, but significant challenges persist. Tskin and the cloud detection are the main limiting factors in using more IR satellite sounding data over land.

32. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Data coverage, AMSU-A channel 6 (Metop-B)
Base
Base + MW sea-ice + MW land
Aug
2014
Difference in sea-ice extent
Reduced use over snow in winter
Feb
2015

33. Seasonal dependence of impact of surface-sensitive MW data over sea-ice and land
Data coverage, AMSU-A channel 6 (Metop-B)
Base
Base + MW sea-ice + MW land
Aug
2014
Some data rejections over Sahara, probably due to skin temperature issues
Feb
2015