1. Impact of AMDAR/RS Modelling at the SAWS
Warren Tennant

2. Weather Forecast Modelling at the SAWS
UK Met Office Unified Modelling system running operationally at the SAWS since September 2006
Installed under an operational licence with Met Office that includes:
using the model for operational forecasting responsibilities
Includes limited commercial use of model output
Real-time feed of model input data (initial conditions and observations) from Met Office
Software support and upgrades
Scientist exchange and visits

3. SAWS National Responsibility
SAWS is responsible to the public of South Africa to provide:
national forecasts of basic weather conditions up to three days ahead
advisories of high-impact weather conditions up to 48-hours ahead and longer where possible, and
warnings of imminent severe weather, to
safeguard lives and property and mitigate the impact of weather on South Africans

4. SAWS SADC Responsibility
SAWS serves as a WMO Regional Specialised Meteorological Centre (RSMC) for the Southern Africa Development Community (SADC)
Plays a backup role in the event of national disasters, e.g. Tropical Cyclone in Mozambique
Provides NWP guidance products to SADC as part of the WMO/CBS Severe Weather Forecasting Demonstration Project (SWFDP)
Process to role-out UM SA12 forecasts (images and data) to SADC countries has been started

5. SA12 Regional Model (based on Met Office NAE model)
Reconfigure global input for 48hr fcst at 00Z
3DVAR at 06 & 12Z from above => 48hr fcst at 12Z
Continuous 3DVAR 6-hourly cycle => 48hr fcst at 00Z

6. Observations in southern Africa
SAWS has a moderate upper-air network
8 GPS stations
2 Island stations
QC and availability good
Rest of southern Africa is a rawindsonde void

7. Global AMDAR Availability 00Z

8. Global AMDAR Availability 06Z

9. Global AMDAR Availability 12Z

10. Global AMDAR Availability 18Z

11. SAA-AMDAR Coverage

12. OSE Experiment Design Continuous 3dVAR 6-hourly cycle
Control: All observations
Experiment: No AMDAR or Rawindsondes
Independent: Interpolated global model 4dVAR initial conditions (no LAM DA)
Winter case: 1 May to 8 Jul 2007
Summer case: 24 Oct 2006 to 18 Jan 2007
Verification:~2000 rainfall stations and ~10 rawindsonde stations in South Africa

18. Wind-speed verified against Rawindsondes
Obvious impact on analyses
Significant impact at 24 hours
Little impact at 48 hours (slight degradation at 250hPa)
Forecast without DA better scores!

19. Spatial ACC of wind-speed :: MJJ
Impact of AMDAR/RS on wind speed is positive throughout
True even if using global 4dVAR analysis as verification standard
Run with no-DA sometimes better than DA run – especially after 48 hours

20. Temperature temporal correlation to Radiosondes
Mostly positive impact
Some cases where no-DA works better
possibly because 4dVAR initial conditions from global model better
Inconsistency at LBCs from LAM DA initial conditions

21. Spatial ACC of temperature forecasts
Similar to wind speed results
Positive impact throughout
Less dependency of impact on forecast lead-time

22. Rainfall forecast verification :: BIAS
Bias expressed as a percentage of the observed rainfall
Summer Case:
No strong impact on forecast day 1
On day 2 more positive impact, except very light rain

23. Rainfall forecast verification :: BIAS
Winter Case:
Bias similar on forecast day 1
Slight negative impact for light rain amounts on day 1
On day 2 not a positive impact as with summer case

24. Rainfall forecast verification :: RMSE
Summer Case:
No strong impact on forecast day 1
On day 2 positive impact for all thresholds
DA runs better than no-DA run except for heavy rain

31. Summary and Conclusions
AMDAR/RS have a definite positive impact on regional model 3dVAR forecasts in southern Africa
Impact decreases with forecast lead-time :: signal probably influenced by LBCs
Best impact found with wind in mid-upper troposphere in tropics
Data assimilation plays an important role in rainfall initialisation
Impact on rainfall best seen in day 2 forecasts and with heavy rainfall in winter