1. Arctic observing system for regional NWP
Harald Schyberg (met.no), Frank Thomas Tveter (met.no) Roger Randriamampianina (met.no), Trygve Aspelien (met.no)

2. Outline Introduction Observing system and impact studies
Ongoing work on improving satellite data usage in high-latitude
Results are from projects
EUCOS space-terrestrial network studies
Norwegian IPY-THORPEX project
DAMOCLES (EU FP7 project)

3. Operational NWP LAM modeling – some characteristics
Should add information to products from global modeling centres (ECMWF, …)
Higher resolution
More frequent data assimilation, shorter cutoff times (some satellite data types arrive late)  Forecasts from a given analysis time available before those from global model

4. Numerical Weather Prediciton in the Arctic
NWP quality in general determined by:
Quality of model formulation
Physics of processes in the Arctic
Quality of boundary forcing
In particular lower boundary: Ice/snow/ocean: heat and moisture flux, momentum flux, radiative fluxes
For LAMs: Lateral boundaries
Quality of initial state estimate
As determined in data assimilation – issues of observation coverage/usage in the Arctic

5. WMO workshop on impact of obs systems, 2008:
Synopsis from many studies using “observing system experiments”
Figure shows impact of observing systems in terms of gain in forecast range for “medium range” forecasts

6. Regional experiments – do we get consistent results with this picture?
Two approaches in anaysing data denial impact studies:
Statistical, long term Case studies
No weighting to favor high-impact weather
Statistical significance?

7. EUCOS impact studies
Two questions in these “Space-terrestrial studies”:
In the presence of the full conventional observing system, what is the impact of adding the various satellites
In the presence of the full satellite observing system, what it the impact of the various components of the conventional observing system
Met.no (and several other global and regional centres) participated in the second part funded by EUCOS

8. EUCOS impact trials at met.no
Use of HIRLAM 3D-Var Winter period and summer period 2005

9. Results- All scenarios (more details later)
Control Scenario
(all available in-situ
observations)
Baseline
scenario
Add E-ASAPs
Add AIREPs

10. Results from met.no HIRLAM OSEs in EUCOS terrestrial network study
Conventional observations have large positive impact in our system in the prescence of satellite data
Probably higher impact than in ECMWF system (we use less satellite obervation types)
TEMPs still dominating factor for analysis quality, wind more than temperature
Aircraft obs complement radiosondes (positive impact of adding aircraft in the presence of sondes)
Significant positive impact from E-ASAP network

11. The Arctic observing system for regional NWP
An observation gap in the conventional observing network:
Lack of radiosondes
Lack of surface observations, but some buoys, more during IPY
The routinely available Arctic Ocean observing system consists basically of buoys and satellite data
AMSU, AIRS/IASI radiances and MODIS winds
Most of the satellite data are free atmosphere, surface/lower troposphere info lacking
The total observing system in the Arctic Ocean:
Lacks surface/near surface observations
Relies on efficient use of satellite data in assimilation schemes
Use of satellite sounding data is less straightforward over sea ice and land than over ocean
 Met.no focus on including IASI and on use of surface affected AMSU measurements

12. Challenges for Arctic data assimilation: Radiosonde observation coverageff

13. Challenges for Arctic data assimilation: aricraft obs coverage

14. Challenges for Arctic data assimilation: SYNOP surface obs coverage

15. Challenges for Arctic data assimilation: buoy surface obs coverage

16. Coverage from some satellite observation types

17. The Norwegian IPY-THORPEX project: Assimilation of extra campaign data and IASI observations (R.Randriamampianina, T. Aspelien)
Observing system experiments during the Norwegian THORPEX campaign (several polar lows):
After implementing and optimizing IASI assimilation: What is the impact of assimilating IASI data (Norwegian version of HARMONIE NWP system)
What is the impact of assimilating campaign data: Extra radiosonde launches from Norwegian and Russian Arctic stations, Norwegian coast guard vessels, dropsondes, … (HARMONIE and LAMEPS)

18. HARMONIE and its assimilation system
(Hirlam Aladin Regional/Meso-scale Operational NWP In Europe)
Model domain: rotated Lambert pr.
Dx=dy= 11 km, 60 vertical levels
up to 0.2 hPa
3D-Var assimilation system
Use of conventional and satellite data
Obs operator for radiance data: RTTOV- 8.7

19. Exploring the use of IASI during the Norwegian IPY-THORPEX campaign period
IASI has 8461 channels, of which we extract 366 for potential use.
Four experiments have been performed using 41 active channels
Period: –
(Warming period 5 days)

20. Impact as function of forecast range over the period
Left effect of IASI with campaign data, right without campaign data
Verification against ECMWF analyses

21. A forecast sensitivity measure to various observations during the campaign period:

22. Polar low case March 2008
+24 hrs forecasts of pressure and precip
IASI, Caobs
No IASI, Caobs
No IASI, no Caobs
IASI, no Caobs

24. 3 March 2008: Probabilistic variables (Wind, + 51 hrs)
wind > 15 m/s
Green=
With campaign
Red=Without
Campaign data

25. Probabilistic variables (Precipitation + 51 hrs)
Green:
Regular + Campaign Obs
rr > 1 mm/3hr
Red:
Regular Obs rr > 1 mm/3hr

26. Developments towards using surface affected AMSU-A data over sea ice

27. AMSU-A channels over sea ice
We simulate observations from NWP fields using radiative transfer model RTTOV-8 (“B”) and compares against the real observations (“O”)
When using fixed emissivity and NWP surface temperatures, typical O-B rms magnitudes over sea ice are:
Ch 3 ~5K
Ch 4 ~3K
Ch 5 ~2K
Ch 6-9 ~ 0.5K
Previously ch has been used over sea ice.
Can we improve the use of ch 6-7 and add lower peaking channels?
How to define emissivity and surface temperature?

28. 3 issues in using surface-affected microwave sounding (AMSU-A) over sea ice
The sea ice emissivity
Accounting for penetration depth of the radiation and vertical temperature gradient within the sea ice
Getting a realistic (first guess of) surface temperature to RTTOV

29. Emissivity: Help using data from Ocean and Sea ice SAF
High-latitude centre: Hosted by Met.no and DMI
Daily sea ice products: Concentration, edge, type etc on 10 km grid
Will also include info on sea ice microwave emissivity in future
See and saf.met.no

30. Emissivity (1)
Use daily OSISAF concentration chart to find near-100% ice covered area
In this area multi- year sea ice from OSISAF was used as predictor for sounding ch emissivity

31. Emissivity (2)
An alternative approach is to estimate the emissivity from a window AMSU-A channel using simplified radiative transfer theory leading to:  = (Tb obs – Tb sim(=0)) / (Tb sim (=1)-Tb sim(=0))
One could then use the assumption that emissivity varies slowly with frequency and also apply it for other channels
Statistical analysis showed that using the MY ice map from SSM/I gave a better fit between modeled and observed data

32. Handling of surface emitting temperature (figures from R.Tonboe)
Variations in penetration depth, increasing temperature with ice depth:
The colder, the more misrepresentative is the surface temperature for the emitting layer

33. Example: O-B statistics ch 4 using RTTOV-8 over sea ice
Right panel: Departures as a function of ground temperature for first-year sea ice
Left panel: The same for multi-year sea ice
The colder, the more misrepresentative is the surface temperature for the emitting layer
a) 0-10% multiyear b) 70-80% multiyear

34. Implementation in variational data assimilation
J(x) = ½ (x-xb)T B-1 (x-xb) + ½ (y-H(x))T O-1 (y-H(x))
Multiyear ice fraction based surface emissivity dependence included in the observation operator H(x)
Variational bias correction:
A linear correction extension added to H(X) for handling dependence of emitting temperature on surface temperature
Control variable x extended with slope coefficient for this dependence
An optimal dependence on surface temperature is determined intrisically, constrained by other information available in the analysis

35. Setup of NWP assimilation experiment over sea ice, using HIRLAM 4D-Var
Reference experiment: no variational bias correction, constant sea ice emissivity, emitting temperature equal to surface temp, channels 6 and 7
Experiment 1: variational bias correction, channels 6 and 7
Experiment 2: variational bias correction, channels 5,6 and 7

36. Impact study of changing the surface handling using channels 6 and 7 in HIRLAM 3D-Var
Small effects on”standard” surface verification averaged over the whole European network
Some positive effect seen on profile verification against radiosondes
Problem of lack of verifying obs near sea ice areas
Red: Fixed sea ice emissivity and surface temperature from HIRLAM
Blue: Emissivity estimation based on OSISAF products and surface temperature in the VarBC control variable

37. Problem of modeling surface temperatures over sea ice:
Several present NWP system utilizes digital ice concentration maps from passive microwave (SSM/I, SSMIS, AMSR) for instance OSISAF
Difficult for low-resolution satellite data like SSM/I to detect small areas of open water within the sea ice
But the effect of such areas on surface fluxes very important
Present passive microwave concentration algorithms show small, but unrealistic, concentration fluctuations over closed sea ice
This also affects present assimilation of AMSU data

38. Typical example: Fraction of ice used in operational HIRLAM NWP model (April 2010)
Ice concentrations above 95% set to 100%,
but still some unrealistic values remaining between 90 and 95%

39. Example: Corresponding 2m temperature
Temperatures typically in the range from -10 to - 30

40. Example: Corresponding difference T (2m) – T (0m)
Erroneous small water areas creates large differences (and erroneously large heat fluxes)
Also a problem for use of satellite sounding data
No straightforward solution to this

41. Summary
Gaps in the Arctic observation network: Particularly in lower troposphere
A potential for improvement with conventional observations even in precense of satellite. Profile info most valuable.
A good coverage of satellite data of the Arctic will be available for the foreseeable future: It is probably cost-effective to exploit the still unrealized potential there
Particular issues on surface description for satellite observation usage over sea ice

42. Thank you …

43. Backup slides follow

44. Challenges for NWP in the Arctic application
Turbulence modeling
Stable boundary layer
Unstable boundary layer
Surface fluxes over sea ice
Heat
Radiative
Momentum
Routine observing system is basically satellite
Utilization of satellite data over sea ice: surface properties, cloud detection, moisture/cloud assimilation
Arctic cloud microphysics parametrization
Leads, meltponds, albedo...
Roughness

45. Some main paths from observations to products for numerical atmospheric modeling
Focus of this talk
Arctic atmospheric observations
Process studies
Data assimilation
Improved NWP model parametrizations
Reanalysis
Analysis
Modeling of future climate
Monitoring of past and present climate
Short/medium range operational NWP

46. Scatter in O-B can be further reduced by introducing surface temperature dependence
Physical basis described by Mathew et al, 2008
Leads to empirical expression:
Temitting = aT2m + b

47. Can this be dealt with through linear correction (“bias correction”)?
Linear dependence of observed brightness temperature on Ts
But:
slope is of Tb vs Ts slightly dependent on ice characteristics such as type
need to simultaneously include multi- year ice fraction dependence
Have chosen to implement this using variational bias correction

48. Some targets for atmospheric process studies
Turbulence modeling
Stable boundary layer
Unstable boundary layer
Surface fluxes over sea ice Leads, meltponds, albedo...
Heat
Radiative
Momentum (Roughness)
Arctic cloud microphysics parametrization
Should improve numerical modeling capability of the Arctic atmosphere  NWP skills, reanalysis quality, climate modeling capability

49. EUCOS Impact Scenarios at met.no with HIRLAM
Winter period: 14 December 2004 – 20 January 2005
Summer period: 1 August – 31 August 2005
6 hours cycling, forecasts up to +48hrs from 00Z cycle stored
Baseline scenario:
All available satellite observations:
AMSU-A over ocean (EUMETSAT retransmission)
QuikScat winds (100 km product)
MSG cloud drift winds (AMV/SATOBs)
SYNOPS from GSN station list
Radiosondes from GUAN climate station network.
Bouys (no ship data)

50. Role of the in situ observation network
Hard to see an increase in the Arctic conventional network for day-to-day NWP
Observation campaigns expected to be important
Process studies which feed back to model physics
Validation studies – identify NWP model problem areas
Validation/calibration of satellite observing system

51. Some more conclusions
Many recent studies shows clear positive impact of adding sounding data and improving use of satellite data: A potential for improving forecasting and reanalysis capability
A main contributor to the Arctic observing system for data assimilation will be satellite data
Surface observations missing in the Arctic
Improvements in accounting for the sea ice contribution to satellite measurements
Improvements in describing boundary layer processes over sea ice
Future progress in Arctic atmospheric modeling will take place on the interface between expertise on sea ice, data assimilation and atmospheric physical processes