1. MJO and monsoon Simulations in the ECMWF VarEPS-Monthly System
Frédéric Vitart , Franco Molteni
and Laura Ferranti
European Centre for Medium-Range Weather Forecasts

2. Outline The ECMWF VarEPS-monthly forecasting system 2. MJO Prediction
Monsoon prediction
4. Conclusion

3. Ensemble Forecasting systems at ECMWF
Weather and Climate
Dynamical Forecasts
Medium-Range
Forecasts
Day 1-15
Monthly
Forecast
Day 10-32
Seasonal
Forecasts
Month 2-7
Product
Ocean model
Atmospheric model
Wave model
Atmospheric model
Wave model
The monthly forecasting system fills the gap between two currently operational forecasting systems at ECMWF: medium-range weather forecasting and seasonal forecasting. Medium-range weather forecasting produces weather forecasts out to 10 days, whereas seasonal forecasting produces forecasts out to 6 months. The two systems have different physical bases. Medium-range weather forecasting is essentially an atmospheric initial value problem. Since the time scale is too short for variations in the ocean significantly to affect the atmospheric circulation, the ECMWF medium-range weather forecasting system is based on atmospheric-only integrations. SSTs are simply persisted. Seasonal forecasting (2-6 months forecasts), on the other hand, is justified by the long predictability of the oceanic circulation (of the order of several months) and by the fact that the variability in tropical SSTs has a significant global impact on the atmospheric circulation. Since the oceanic circulation is a major source of predictability in the seasonal scale, the ECMWF seasonal forecasting system is based on coupled ocean-atmosphere integrations.
Tool

4. Current system (VAREPS/Monthly):
Monthly Forecasting
Old MOFC system:
Coupled forecast at TL159
Initial condition
Day 32
Current system (VAREPS/Monthly):
EPS Integration at T399
Initial condition
Coupled forecast at TL255
Heat flux, Wind stress, P-E
Slide 14: New monthly forecasting system. Previously the monthly forecasting system consisted of coupled ocean-atmosphere integrations with an atmospheric horizontal resolution at TL159. Now, the monthly forecasting system and the VarEPS have been merged.
This graphic displays the new configuration: atmosphere-only at TL399 forced by persisted SST anomalies till day 10 twice a day. After day 10, the atmospheric model at a TL255 resolution is coupled to an ocean model till day 32.
The coupled model consists of the ECMWF atmospheric model (the same cycle as the deterministic forecast), coupled to an ocean general circulation model, which is a version of the Hamburg Ocean Primitive Equation model (HOPE), developed at the Max Plank Institute for Meteorology, Hamburg. The ocean model has lower resolution in the extratropics but higher resolution in the equatorial region, in order to resolve ocean baroclinic waves and processes, which are tightly trapped at the equator. The ocean model has 29 levels in the vertical. The atmosphere and ocean communicate with each other through a coupling interface, called OASIS, developed at CERFACS, France. The atmospheric fluxes of momentum, heat and fresh water are passed to the ocean every 3 hours and, in exchange, the ocean sea surface temperature (SST) is passed to the atmosphere.
Day 10
Day 32
Ocean only integration

5. The ECMWF VarEPS-monthly forecasts
A 51-member ensemble is integrated for 32 days every week
Atmospheric component: IFS with the latest operational cycle and with a T399L62 resolution till day 10 and T255L62 after day 10.
Persisted SST anomalies till day 10 and ocean-atmosphere coupling from day 10 to day 32.
Oceanic component: HOPE (from Max Plank Institute) with a zonal resolution of 1.4 degrees and 29 vertical levels
Coupling: OASIS (CERFACS). Coupling every 3 hours
Slide 15: Description of the VarEPS-monthly forecasting system.
Each week, the coupled model is integrated forward to make a 32 day forecast with 51 different initial conditions, in order to create a 51-member ensemble.

6. The ECMWF monthly forecasting system
Atmospheric initial conditions: ERA40 and ECMWF operational analysis
Oceanic initial conditions: Last ocean analysis + accelerated analysis.
Perturbations:
Atmosphere: Singular vectors + stochastic physics
Ocean: wind stress perturbations during data assimilation.
Slides 16 and 17
In order to initiate monthly forecasts, initial conditions for both the ocean and atmosphere are required. Atmospheric and land surface initial conditions are obtained from the ECMWF operational atmospheric analysis/reanalysis system. Oceanic initial conditions originate from the oceanic data assimilation system used to produce the initial conditions of the seasonal forecasting system 2. However, this oceanic data assimilation system lags about 12 days behind real-time. The lag is partially due to the fact that the SST, obtained by interpolating in time the weekly OIv2 SSTs produced by NCEP, can be up to 12 days behind real-time. A first option would be to wait for the oceanic initial condition to be created by the data assimilation system to start the forecast, as in seasonal forecasting. This would mean losing 12 days of forecast and is not therefore suitable for monthly forecasting. A second option would be to persist the SST anomalies of the latest ocean analysis. However, we have some information about the wind stress and heat fluxes during the last 12 days of the ECMWF atmospheric analysis; this information can be used to help determine the present ocean initial state. Therefore, the option that has been chosen for monthly forecasting consists in integrating the ocean model from the last ocean analysis forced by analyzed wind stress, heat fluxes and P-E. During this 'ocean forecast', the sea surface temperature is relaxed towards persisted SST, with a damping rate of 100 W/m2/K.

7. The ECMWF monthly forecasting system
Background statistics:
5-member ensemble integrated at the same day and same month as the real-time time forecast over the past 18 years.
This represents a 90-member ensemble
It runs once every week, 2 weeks in advance (5-week window)
Real-time forecasts
01/01/08
01/01/07
01/01/06 .
Re-forecasts
Slide 19:
Because of model errors, a drift occurs in the coupled system. In order to evaluate this model drift, the coupled model is integrated with 5 different initial conditions (5-member ensemble) at the same day and month as the real time forecast, but over the past 18 years, creating a 90-member climate ensemble.
01/01/90

8. The ECMWF monthly forecasting system
Anomalies (temperature, precipitation..) -
Slide 20:
Anomaly maps are similar to seasonal forecasting charts, but with weekly means instead of monthly means. Over each point of the map, atmospheric variables such as 2-metre temperature, total precipitation, mean sea-level pressure or surface temperature, have been averaged over a weekly period (week 1: day 5 to 11, week 2: day 12 to 18, week 3: day 19 to 25, and week 4: day 26 to 32) and also over the 51 members of the real-time forecast and the 60 members of the back statistics. The plots display the difference between the ensemble mean of the real-time forecast and the ensemble mean of the back-statistics. The product therefore displays the shift of the forecast ensemble mean from the estimated "climatological" mean (created from ensemble runs over the past 18 years).
In addition, a Wilcoxon-Mann-Whitney test (WMW-test, see for instance Wonacott and Wonacott 1977) has been applied to estimate whether the ensemble distribution of the real-time forecast is significantly different from the ensemble distribution of the back-statistics. Regions where the WMW-test displays a significance less than 90% are blank. Regions where the WMW-test displays a significance exceeding 95% are delimited by a solid contour (blue or red depending on whether the anomaly is positive or negative respectively).
The blanking of "non-significant" shifts does not mean that there is no signal in the blanked regions, but only that, with the particular sampling we have, we cannot be sure that there is a signal. For this reason, there are likely to be many areas where a signal is real but remains undetected.

9. The ECMWF monthly forecasting system
Probabilities (temperature, precipitation..) -
Slide 21: Probability and tercile maps are also produced. An example of probability map for the period day is displayed on this slide.

10. ROC score: 2-meter temperature in the upper tercile
Skill of the ECMWF Monthly Forecasting System
ROC score: 2-meter temperature in the upper tercile
191 cases
Day 5-11
Day 12-18
Day 19-25
Day 26-32
Slide 30: Map of ROC scores of the probability that 2-meter temperature averaged over the period day is in the upper tercile. Only the scores over land points are shown. The terciles have been defined from the model climatology. The verification period is Oct 2004-May Red areas indicate areas where the ROC score exceeds 0.5 (better than climatology). This plot shows that the coupled model performs better than climatology for the period days
For the period days 19-26, the skill is much lower than for days 12-18, as expected. The red is largely dominating overall, suggesting that the model generally performs better than climatology at this time scale. Europe seems to be a difficult region, with very low skill at this time range. Tropical regions display the strongest skill after 30 days, suggesting that the coupled model at this time range starts to behave more like seasonal forecasting.

11. ROC score: Precipitation in the upper tercile
Skill of the ECMWF Monthly Forecasting System
ROC score: Precipitation in the upper tercile
191 cases
Day 5-11
Day 12-18
Day 19-25
Day 26-32
Slide 30: Map of ROC scores of the probability that 2-meter temperature averaged over the period day is in the upper tercile. Only the scores over land points are shown. The terciles have been defined from the model climatology. The verification period is Oct 2004-May Red areas indicate areas where the ROC score exceeds 0.5 (better than climatology). This plot shows that the coupled model performs better than climatology for the period days
For the period days 19-26, the skill is much lower than for days 12-18, as expected. The red is largely dominating overall, suggesting that the model generally performs better than climatology at this time scale. Europe seems to be a difficult region, with very low skill at this time range. Tropical regions display the strongest skill after 30 days, suggesting that the coupled model at this time range starts to behave more like seasonal forecasting.

12. 3. Performance over the Northern Extratropics
ROC Area of the probability that 2-meter temperature in the upper tercile
Northern Extratropics
Day 5-11
Day 12-18
Day 19-32
EUROPE
Day 5-11
Day 12-18
Day 19-32

13. Outline The ECMWF VarEPS-monthly forecasting system 2. MJO Prediction
3. Monsoon prediction
4. Conclusion

14. Prediction of the Madden Julian Oscillation (MJO)
Monthly Forecast starting on 7 December 2007

15. Prediction of the Madden Julian Oscillation (MJO)
Monthly Forecast starting on 7 December 2007

16. MJO metric
Forecast anomalies of OLR, U850, & U200 were obtained by subtracting the hindcast climatology of the respective fields from the forecast.
The forecast anomalies were then averaged over latitudes (-15, 15).
The forecast anomalies of the 3 fields were divided by the respective global standard deviations calculated from the NCEP reanalyses.
The resulting anomalies were projected onto the first two EOF patterns, the latter being computed by Matt Wheeler from NCEP reanalyses.
The two projection coefficient time series were further divided by the standard deviations of first two PC (also from NCEP reanalyses), to obtain the RMM12.
To remove the interannual variability, a 120-day running mean of RMM12 is subtracted.

17. Operational MJO prediction

18. MJO Prediction
EXPERIMENT: A 5-member ensemble has been run every day from 15 December 1992 to 31st January 1993.
The MJO diagnostic is based on the same method as in Wheeler and Hendon (2004). We use combined EOFs of:
Velocity potential anomalies at 200 hPa
Outgoing Long-Wave Radiations anomalies
Zonal wind at 850 hPa anomalies
Anomalies are relative to the past 12 years climate. The fields have been averaged between 10N and 10S and normalized, before computing the combined EOFs.
Slide 32: In order to assess the skill of the monthly forecasting system to predict an MJO event, the following experiment has been performed: a 5-member ensemble monthly forecast has been integrated for each day between 15 December 1992 and 31 January 1993 (there was a very strong MJO event during that period). An analysis using combined EOFs of velocity potential at 200 hPa, U850 and OLR has been applied to the ensemble integrations (as in Wheeler and Hendon, 2004).

19. MJO Prediction
The combined EOFs have been computed on ERA40 daily data from to 2004.
Slide 33: MJO forecasts. This plot shows the 2 first combined EOFs of velocity potential at 200 hPa, 850 U-velocity and OLR averaged between 10N and 10S.

20. VP200 - Forecast range: day 15
ERA40
28R3
29R1
31R1
32R2
32R3
33R1
29/12
05/01
12/01
days
20/01
28/01
04/04
12/02
10/04
04/05
09/06
06/07
11/07
06/08

21. Amplitude of the MJO PC1 PC2
28R3
29R1
30R2
31r1
32r2
32r3

22. MJO Propagation
PC1
PC2
28R3
29R1
30R2
31r1
32r2
32r3

23. WINTER 2007/2008: CY32R3

24. Winter 2007/2008: CY32R3

25. Forecast Biases and Convection Precipitation for DJF against GPCP for different cycles: from 15 year 5 months integrations for
CY31R1 Sep 2006
CY32R2 June 2007
CY32R3 Nov 2007

26. MJO Prediction. CY30R2 PC1 PC2 Persisted SSTs Coupled VAREPS Ocean ML
Slide 39: Anomaly correlation between observations and forecasts at different time lags. The left panel represents the results obtained with PC1 and the right panel the results obtained with PC2. This slide suggests that the monthly forecast displays skill in predicting the evolution of the MJO up to 14 days. The scores are higher than those obtained when running the atmospheric model forced by persisted SSTs (as in EPS) rather than using a coupled ocean-atmosphere model. The scores are lower than those obtained by integrating the atmospheric model coupled from day 0 and even lower than those obtained with an atmospheric model coupled to an oceanic mixed-layer model. This suggests that SSTs play a very important role in the propagation of the MJO and that the vertical resolution of SSTs is important. On the other hand, an increase of the horizontal resolution of the atmospheric model does not lead to better MJO forecasts.
Persisted SSTs
Coupled
VAREPS
Ocean ML

27. MJO Prediction. CY32R3 PC1 PC2 Pers. Atm. Initial conditions Ocean ML
Slide 39: Anomaly correlation between observations and forecasts at different time lags. The left panel represents the results obtained with PC1 and the right panel the results obtained with PC2. This slide suggests that the monthly forecast displays skill in predicting the evolution of the MJO up to 14 days. The scores are higher than those obtained when running the atmospheric model forced by persisted SSTs (as in EPS) rather than using a coupled ocean-atmosphere model. The scores are lower than those obtained by integrating the atmospheric model coupled from day 0 and even lower than those obtained with an atmospheric model coupled to an oceanic mixed-layer model. This suggests that SSTs play a very important role in the propagation of the MJO and that the vertical resolution of SSTs is important. On the other hand, an increase of the horizontal resolution of the atmospheric model does not lead to better MJO forecasts.
Pers. Atm. Initial conditions
Ocean ML
Coupled

28. Interim Reanalysis- Forecast scores
PC1
PC2 ER E4 EI

29. Conclusion
The model has some skill to predict the evolution of the MJO up to about 14 days. The amplitude of the MJO has improved since 2004, but the model has now difficulties to propagate the MJO across the Maritime continent.
Using a high vertical resolution mixed layer model improves the MJO scores.
MJO scores are improved when using the latest ECMWF reanalysis.

30. Outline The ECMWF VarEPS-monthly forecasting system 2. MJO Prediction
3. Monsoon prediction
4. Conclusion

31. Operational Prediction

32. Example of Verification
08/05-15/05/06
15/05-22/05/06
22/05-29/05/06
29/05-05/06/06
ANA
Day
5-11
Day
12-18

33. Indian monsoon 2006
Forecast Day 12-18
[20-5N, 60-90E]

34. Indian Monsoon 2007
, , , , , , , , , , , , , , , ,

35. 2008 Monsoon
Rainfall for the first week is for the period 1st June 2008 to 4th June 2008 Sunday, June 15, 2008 (New Delhi)
The monsoon has finally arrived in the capital. This is the earliest monsoon to arrive in Delhi in 108 years as it comes a record 14 days ahead of its normal date of June 29. Available weather records in Delhi date back to 1901 and not once has monsoon hit the city before June 16. Delhi has been receiving a healthy rainfall for the past few days. Monsoons have already been declared in many parts of north and northwest India including Uttarakhand, Himachal Pradesh and Jammu and Kashmir.
28-4/5
9-15/6
14-20/4
19-25/5
26-1/6
23-30/6

36. Evaluation of the skill the monthly forecasting system to predict Indian rainfall
Experiment’s setting:
46 day forecasts (60 km resolution from day 0 to 10, 80 km from day 10 to day 45)
15 ensemble members
Starting dates: 15 May
Model Cycle 31R2 (operational cycle from 12/06 to 06/07)

37. Experiments

38. June Rainfall

39. June Rainfall
VarEPS
(80 km)
MOFC
(120 km)
Observations

40. June Precipitation over India

41. June Precipitation over India
VEPS
PERS
MOFC ML
Correlation
0.57
(+/- 0.05)
0.59
(+/- 0.06)
0.43
(+/- 0.07)
0.62
(+/- 0.04)
RMS error
0.92
1.6
0.98
0.84

42. June pentads 1-5 June 6-10 June Forecast range: Day 21-25

44. June SST prediction
Difference of skill to predict SSTs(linear correlation with analysis) between mixed-layer model and coupled model.

45. Rainfall Climatology
JUNE
JULY
AUGUST
SEPT.
VEPS
OBS

46. July Precipitation over India (VEPS-CY32R3)

47. Precipitation over Indian: July 2002
VarEPS
(ensemble mean)
Observations

48. August Precipitation over India

49. September Precipitation over India

50. Skill to Predict Pentads
June
July
August
September

51. Conclusion
The VarEPS-Monthly system is skilful to predict Indian rainfall in June and September. The skill is very low is August.
Increasing horizontal resolution or using an ocean mixed layer model lead to better prediction of June Indian rainfall.

52. Recent or planned changes to the ECMWF physical parametrizations
Improved treatment of ice sedimentation, auto- conversion to snow in cloud scheme and super-saturation with respect to ice (CY31R1)
RRTM-SW + McICA
MODIS albedo + revised cloud optical properties (CY32R2)
New formulation of convective entrainment
Variable relaxation timescale for closure
Reduction in background free atmosphere vertical diffusion
(CY32R3)

53. Aircraft data New scheme Default
Simple ECMWF scheme: comparison to Mozaic aircraft data (from Gierens et al.)
Aircraft data
New scheme
Ignore red stars (different scheme)
Default

54. Impact on relative humidity (RH) climatology
31r1 – 30r1 annual mean difference
Largest changes in the tropical upper troposphere

55. Convection changes to operational massflux scheme
New formulation of convective entrainment:
Previously linked to moisture convergence
– Now more dependent on the relative dryness of the environment
New formulation of relaxation timescale used in massflux closure:
Previously only varied with horizontal resolution – Now a variable that is dependent on the convective turnover timescale i.e. variable in both space and time also
Impact of these changes is large including a major increase in tropical variability

56. Composite vertical profile for west pac, JJA
Minimum cloud top heights distributions (CloudSat)
precipitating clouds
Of note:
Trimodality (quadra-modal) heights
precipitating clouds are deeper than non precipitating clouds
non- precip
clouds

57. Average tropical cloud profiles
In the Tropics now a tri-modal cloud distribution becomes apparent, with a strong increase in mid-level clouds. CloudSat data will be used to verify these results.