Farnham PBechtold

Benefits and Limits of NWP through convective parameterization“Repetitio est mater studiorum”Verification with observations is the mother of meteorology : 

Benefits and Limits of NWP through convective parameterization “Repetitio est mater studiorum” Verification with observations is the mother of meteorology Farnham Castle 6-8 January 2004

Outline: 

Outline Give you an overview of what current massflux schemes can do and where the limitations are: Midlatitudes and Tropics A big list of things to check: global Scores, verification with radiosondes, tropical and midlatitude cyclones, precipitation verification, MJO, diurnal cycle etc. Numerics and others A little bit meteorology/forecasting of the 4/5 May 2003 Tornado Outbreak, 9 September 2002 French flood, 1-3 December 2003 French flood Conclusions and short term activities

Operational Forecasts and Analysis at ECMWF: 

Operational Forecasts and Analysis at ECMWF 4DVAR-12h window Analysis, and T511 (40 km, 60 vertical levels) 10-day deterministic forecast – twice a day. In about two years it is planned to run at T799 (25 km) and 91 vertical levels 10-day Ensemble forecast at T255 (80 km) using 50 ensembles Monthly coupled ensemble forecast (once a week) Experimental 6 months coupled seasonal forecast system

Tropical and global budgets (1): 

Tropical and global budgets (1)

Long term verification Scores (1):Tropical winds: 

Long term verification Scores (1): Tropical winds

Long term verification Scores (2): NH+SH Geopotential: 

Long term verification Scores (2): NH+SH Geopotential

Long term verification Scores (3):Europe precipitation: 

Long term verification Scores (3): Europe precipitation

“Grid-point storm” problem (1): 

“Grid-point storm” problem (1) If convective heating/mixing (stabilization) is not adequately represented in the model, the model might get saturated under moist and/or strong forcing conditions – it then develops an explicit turnover to get rid of the instability. However, these resolved-scale updrafts are not at the right scale in models with grids larger than say 5-10 km (actual convective updraft radius are generally smaller than 1-2 km). These unphysical strong ascents in the model produce excessive “stratiform” rain, too deep lower tropospheric pressure systems and strong divergence at upper levels, destroying the actual Jet structure – these model errors then propagate and grow quickly, affecting heavily the forecast skill of the model.

Mai is always the most difficult monthin Nh ……: 

Mai is always the most difficult month in Nh …… Problems we had last year were related to “grid-point” storms over US (and also in Tropics) - strong upper-level divergence. The problem was related to insufficient trigger during night …. The simple solution was to check for all parcels in the lowest 700 hPa, and to increase the precipitation efficiency and entrainment in the convection. But the consequence was to verify all the climate, tropical biases, diurnal cycle etc. ……… as a change in the partition between convective and stratiform precipitation changes the moistening/drying of the troposphere and also changes the deepening/damping of synoptic perturbations and tropical waves.

“Grid-point storm” problem (2)Instability - CAPE: 

“Grid-point storm” problem (2) Instability - CAPE This problem is particularly important over regions with high convective instability (CAPE), i.e. over North America (Great Plaines) during Northern Hemispheric Spring, South America (Southern hemispheric spring), but also over the tropical Pacific Ocean (Indonesia region). As an example here is the monthly mean distribution of CAPE for May 2002 at 00 and 18 UTC – Nota: typical values for Europe for this period are just about half

“Grid-point storm” problem (3)48h forecasted convective and stratiform rainfall with different versions of convection scheme/trigger: 

“Grid-point storm” problem (3) 48h forecasted convective and stratiform rainfall with different versions of convection scheme/trigger Note the large amount of stratiform rainfall in CY25R1 over central Great Plaines that is replaced by a smooth distribution of convective rainfall in new cycle (upper left picture)

“Grid-point storm” – American problem (5)Effect on first guess errors (12h forecast-Analysis)(thanks to F. Grazzini and G. v. Grijn): 

“Grid-point storm” – American problem (5) Effect on first guess errors (12h forecast-Analysis) (thanks to F. Grazzini and G. v. Grijn) Idem previous Example, but for a case in North America: previous oper. Cycle left, new cycle on the right Example of large First Guess 200 hPa Mass (isolines) and Wind (arrows) errors (Analysis Increments) over North America - in model cycle 25R1 (operational prior to 14 January 2003) compared to cycle 25R4 (oper since 14 January 2003) with improved convective trigger

Data usage/rejection in Analysis:- aircraft observations at 200 hPa Jet-level: 

Data usage/rejection in Analysis: - aircraft observations at 200 hPa Jet-level 25R1 25R4 Large spread. Obs Wind speed > Forecast: resolved ascent act as solid obstacle

“Grid-point storm” – American problem (4)Effect on first guess errors (12h forecast-Analysis): 

“Grid-point storm” – American problem (4) Effect on first guess errors (12h forecast-Analysis) Idem previous example but for a case on South America (Buenos Aires region)

“Grid-point storm”–American problem (6)Global first guess Z and Wind errors at 200 hPa for May 2002: 

“Grid-point storm”–American problem (6) Global first guess Z and Wind errors at 200 hPa for May 2002 CY25R1=old CY25R4=new

“Grid-point storm” – American problem (7)Effect on Forecast: 

“Grid-point storm” – American problem (7) Effect on Forecast Evolution of 500 hpa height and forecast errors with new cycle (left) and previous cycle (right) : 200 hPa wind errors (arrows), surface pressure errors (blue and red isolines). Note that the largest 24h forecast errors in the Northern Hemisphere are over North America ahead of the 500 hPa trough. 7 Mai 2002

“Grid-point storm” – American problem (8)Effect on Forecast: 

“Grid-point storm” – American problem (8) Effect on Forecast Idem previous example, but here large forecast error growth occurs on day 3 associated with the 500 hpa trough over US, …… then by day 5 the forecast for Europe is strongly degraded. All problems are not yet solved with new cycle (left side) – convectively unstable situations are always difficult, but the future new humidity analysis will further reduce CAPE – better stability of forecast 28 Mai 2002

Effect on Spurious Cyclone (over/under) development related to insufficient stabilization in Analysis/Forecast: 

Effect on Spurious Cyclone (over/under) development related to insufficient stabilization in Analysis/Forecast

Effect on Tropical Cyclone Forecast (1)– Gerald v.d Grijn: 

Effect on Tropical Cyclone Forecast (1) – Gerald v.d Grijn

Effect on Tropical Cyclone Forecast (1)– Gerald v.d Grijn: 

Effect on Tropical Cyclone Forecast (1) – Gerald v.d Grijn Significant smaller bias in intensity w.r.t. previous season.

Effect on Tropical Cyclone Forecast (2)– JJA verification Gerald v.d Grijn: 

Effect on Tropical Cyclone Forecast (2) – JJA verification Gerald v.d Grijn But strong ‘slow’ bias!

Effect on Scores (1): Wind Nh: 

Effect on Scores (1): Wind Nh

Tropical Forecast Biases and Convection (1)Effect on Forecast: 

Tropical Forecast Biases and Convection (1) Effect on Forecast Forecasts of tropical atmosphere are naturally very sensitive to any changes in the convection scheme On the longer term (10-20 days) the tropical atmosphere is in radiative convective equilibrium, so that the detrainment of water substance by the convection significantly affects the upper-tropospheric temperature and moisture biases The upper-tropospheric wind biases are also strongly affected by the entrainment coefficient in the momentum flux formulation – “cumulus friction” The convergence/precipitation in the ITCZ are strongly affected by the deep convection, but equivalent important is the representation of shallow convection in the subtropics determining the moist low-level flow in the Tropics Furthermore, statistics on tropical variability (cyclones and Madden-Julian oscillation) are also strongly affected by the convection parameterization …….

Tropical Forecast Biases and Convection (2)upper tropospheric zonal mean T biases for Day+1 and Day+2 forecasts with different model cycles: 

Tropical Forecast Biases and Convection (2) upper tropospheric zonal mean T biases for Day+1 and Day+2 forecasts with different model cycles CY25R4=new CY25R1=old

Tropical Forecast Biases and Convection (3)upper tropospheric zonal mean V-wind biases: 

Tropical Forecast Biases and Convection (3) upper tropospheric zonal mean V-wind biases CY25R4=new CY25R1=old

Tropical Forecast Biases and Convection (3)moisture biases with new humidity Analysis (since Oct 03): 

Tropical Forecast Biases and Convection (3) moisture biases with new humidity Analysis (since Oct 03)

Tropical Convection (5)Validation of cloud top heights using HIRSFrédéric Chevallier: 

Tropical Convection (5) Validation of cloud top heights using HIRS Frédéric Chevallier In midlatitudes too many clear (sunny) days

Tropical Climate (1)3 ensemble 4 months T95 integrations:Rainfall rate against TRMM: 

Tropical Climate (1) 3 ensemble 4 months T95 integrations: Rainfall rate against TRMM

Tropical Climate (3)3 ensemble 4 months T95 integrations: TOA SW – comparison with ERBE. Trade cumuli better but still problem with marine Sc: 

Tropical Climate (3) 3 ensemble 4 months T95 integrations: TOA SW – comparison with ERBE. Trade cumuli better but still problem with marine Sc 25R4 25R1

Tropical Climate (4)3 ensemble 4 months T95 integrations -getting Sc right -by M. Köhler: 

Tropical Climate (4) 3 ensemble 4 months T95 integrations -getting Sc right -by M. Köhler 25R4 New Low CC Total CC

Tropical Climate (5)3 ensemble 4 months T95 integrations –influence of Aerosols on African Monsoon: 

Tropical Climate (5) 3 ensemble 4 months T95 integrations –influence of Aerosols on African Monsoon

MJO Analysis (1) – F. VitardComparison Analysis – monthly ensemble Forecasts: EOFs of velocity potential at 200 hPa along the 5N-5S equatorial band: 

EOF1 and EOF2 from analysis EOF1 and EOF2 from MonthlyFC MJO Analysis (1) – F. Vitard Comparison Analysis – monthly ensemble Forecasts: EOFs of velocity potential at 200 hPa along the 5N-5S equatorial band Average over 2002 +Jan/Feb 2003 monthly forecasts

MJO Analysis (3) – F. VitardExample of Comparison of Analysis and one control T255 Forecast – in this case Forecast useful at 20-day range: 

MJO Analysis (3) – F. Vitard Example of Comparison of Analysis and one control T255 Forecast – in this case Forecast useful at 20-day range

MJO Analysis (4) – F. VitardEvolution of Anomaly correlation and RMS for Control T255, ensemble mean, and persistence of anomaly (from 30 Forecasts): 

RMS Error Anomaly Correlation Up to day 20, Forecast is more valuable than Climatology – average over tropical band from 5N-5S MJO Analysis (4) – F. Vitard Evolution of Anomaly correlation and RMS for Control T255, ensemble mean, and persistence of anomaly (from 30 Forecasts)

Humidity Analysis (1) – E. V. HolmInfluence of humidity analysis on stability (CAPE): 

Humidity Analysis (1) – E. V. Holm Influence of humidity analysis on stability (CAPE) In present humidity analysis background errors have a bias towards moist values – non-linear analysis with skewed distributions corrects this problem

Diurnal Cycle of Convection (1) Mean rainfall rate for February 2002 from T511 24-72h forecastsand from TRMM (product 3B43): 

Diurnal Cycle of Convection (1) Mean rainfall rate for February 2002 from T511 24-72h forecasts and from TRMM (product 3B43)

Diurnal Cycle of Convection (2)Evolution of surface fields for south-eastern Amazonia bassin for February 2002 from ensembles of 72 h T511 forecasts and comparison with surface rainfall observations during LBA.: 

Diurnal Cycle of Convection (2) Evolution of surface fields for south-eastern Amazonia bassin for February 2002 from ensembles of 72 h T511 forecasts and comparison with surface rainfall observations during LBA.

Diurnal Cycle of Convection (3)Monthly mean Soundings for Manaus (Amazonia) for February 2002 from dialy 72h forecasts at T511 with different versions of convection schemes and comparison with ERA40: 

Diurnal Cycle of Convection (3) Monthly mean Soundings for Manaus (Amazonia) for February 2002 from dialy 72h forecasts at T511 with different versions of convection schemes and comparison with ERA40

Diurnal Cycle of Convection (4)Hovmoeller Diagrams for February 2002 from an ensemble of daily forecasts at T511 (25R1+25R4/26R1), and comparison with TRMM (3B42): 

Diurnal Cycle of Convection (4) Hovmoeller Diagrams for February 2002 from an ensemble of daily forecasts at T511 (25R1+25R4/26R1), and comparison with TRMM (3B42) Observations show maximum rainfall intensity for Africa and South America at 15 LST. Cycle 25R1 produces maximum precip at 9 LST, compared to 12 LST with the new cycle.

Diurnal Cycle of Convection (5) Same as previous Hovmoeller diagrams but from single 40-day T159 forecasts and comparison with TRMM (microwave+infrared): 

Diurnal Cycle of Convection (5) Same as previous Hovmoeller diagrams but from single 40-day T159 forecasts and comparison with TRMM (microwave+infrared) The results for the long integrations are quasi-identique but smaller precip amplitude due to some dry moisture bias

Slide41: 

Importance of shallow convection (1): Surface energy fluxes

Slide42: 

Importance of shallow convection: (2) Frequency of occurrence from IFS climate runs shallow deep

Importance of shallow convection: (3) Mean profiles of Theta and q over tropical Pacific with (red) and without (blue) shallow convective parameterisation. : 

Importance of shallow convection: (3) Mean profiles of Theta and q over tropical Pacific with (red) and without (blue) shallow convective parameterisation.

Influence of entrainemnt on 72h forecasted tropical Soundings for Manaus and Nauro: : 

Influence of entrainemnt on 72h forecasted tropical Soundings for Manaus and Nauro: slight Dilemma: Dry and warm or moist but too cold

Resolution dependency: T_adj, dtConvective Mass flux Pdfs at 1.2 km for T159, T255, T511: 

Resolution dependency: T_adj, dt Convective Mass flux Pdfs at 1.2 km for T159, T255, T511

Resolution dependency: T_adj, dtConvective Mass flux Pdfs at 4.9 km for T159, T255, T511: 

Resolution dependency: T_adj, dt Convective Mass flux Pdfs at 4.9 km for T159, T255, T511

Stability:options for solution of advection equation: 

Stability: options for solution of advection equation Sub-timestepping: easy, only one routine to change. Tests yet only in T255 forecasts. Scores neutral to slightly better, but results sensitive to how one updates T, q (with total physical tendency – or only convective tendency) Implicit solution Semi-Lagr solution How to call (define) in general the different physical processes (clouds,radiation etc.) on the lagrangien trajectory

Stability: implicit advection: 

Stability: implicit advection if ψ = T,q => Only bi-diagonal system For u,v, and tracer initialise:

Stability: SL advection: 

Stability: SL advection if ψ = T,q Nota: if S=T then S cancels out apart from melting term + evaporation below cloud as not taken into account in updraft

Stability: Choice of entrainment(consider model “drift” somewhere above cloud base): 

Stability: Choice of entrainment (consider model “drift” somewhere above cloud base) With organized and turbulent entrainment/detrainment rates defined as one obtains With Mb obtained non-linearly (microphysics) through a CAPE closure in the present model version the normalized specific humidity tendency is used instead

Summary and current activities: 

Summary and current activities Current optimized massflux convection schemes already provide accurate and efficient forecasts (provide reasonable stabilization)– not easy to do better ! But still need to further reduce intense grid-scale precipitation events, and some model biases (moisture, Hadley circulation). Don’t forget importance of shallow convection Problems that still persist concern the diurnal cycle (12 LST against 15 LST) and problems in representing the amplitude and propagation of the MJO, distinction between Sc and shallow Cu Numerics are an integral part of every scheme Any fundamental new approaches should only be developed if they can be consistently and efficiently used in data assimilation procedure (noisy fields, linear assumptions etc.) Our immediate plans are fully implicit/SL solution, and at a bit longer term optimisation of convection scheme parameters using either adjoint or simply Jakobians to globally minimize variation with respect to radiosondes Implement Tracer transport and assimilation of CO2

What else is going on ……: 

What else is going on …… A new PBL scheme is close to implementation by M. Köhler A. Tompkins is working on a statistical cloud scheme+prognostic ice equation Assimilation of sattelite derived rain rates using full adjoint of simplified cloud scheme (P. Lopez), and further work on humidity Analysis (E. Holm)