Global Numerical Weather prediction: the role of convection Peter Bechtold 1 European Centre for Medium Range Weather Forecast Slide 1
Content Global picture of convection and methods to analyse waves (Hoevmoeller, Eliassen-Palm fluxes, wavenumber-frequency diagrams) Balances and generation of potential energy and its conversion into kinetic energy Madden-Julian Oscillation Quasi-equilibrium closures for CAPE Diurnal cycle of convection Slide 2
How to evaluate model (convection), how to trust the Analysis? Analysis increments SON 2011 -UKMO Slide 3
Day+5 Forecast errors: EPS ensemble mean vs. high-resolution Slide 4
Precipitation: SEEPS against other Centres 2010 &2011 2010/2011 & 2011/12 Slide 5
Time-series: Precipitation Slide 6
Precipitation climatology mean=2.67 mm/day mean=2.85 mm/day Slide 7
Occurrence of deep and shallow convection Slide 8
JJA Precip and SWnet errors uncoupled Slide 9
Convective Tendencies: total & shallow Slide 10
Tropical T tendency budgets rad cloud Slide 11
The global Lorenz Energy cycle da dt Generation Conversion NQ NQ Lorenz efficiency factor Net heating R [1 ( 1 1)] T ( 1 1) q P Slide 12
Generation rates Total Generation rate (W/kg) Generation rates maximum in upper tropical troposphere Generation rate - radiation Grid-scale conversion rate Radiation does not contribute to the conversion rates but to the generation rate, but even there has only at poles a positive contribution (cooling at cold places) but globally a negative contribution (as in Tropics it is cooling where it is warm) Steinheimer et al. 2008, Tellus Slide 13
Conversion rates and convection Grid-scale conversion rate (W/kg) Subgrid conversion rate Grid-scale has positive and negative contributions to kinetic energy conversion rate, maximum in upper-tropical troposphere Subgrid conversion rate - convection Convection so important because contribution always positive! Slide 14
Composite of the time-height sections of wavenumber 10 phase for q and Q. Glenn Shutts, 2006, Dyn. Atmos. Ocean 30 km At z~10 km, q and Q in phase 0 26 356 time (hours) Think of red/orange as warm regions in m=10 wave and dark shading represents convective warming Slide 15
Shallow water system and linear waves V U U e e G z m 2 y /2 ik ( x ct ) 0 0 (, ) 1 2y 2 V V y y e G z m : Hn( y) 2 2 c 2 y /2 0 ( ) 4 1 (, ) k (2n 1), ; 0,1,2,... c 2 k c gh n Kelvin wave, geostrophic c k gh General, Hermite Polynomials Modes alternate asymm./symmetric Dispersion relation G z m e e z/(2 Hs ) imz (, ) Re( ) see T. Matsuno. Quasi-geostrophic motions in the equatorial area. J. Met. Soc. Japan, 44:25-42, 1966. Slide 16
Wave number Frequency Spectra OLR Cy38r1 (2012) NOAA 90 50 25 12 8 Cy31r1 (ERAI) Slide 17
2D wave propagation with Eliassen-Palm fluxes Slide 18
General circulation and equilibrium in the Tropics Horizontal temperature fluctuations in the Tropics are small <1K/1000 km; and in the absence of precipitation the vertical motions(subsidence) tend to balance the cooling through IR radiation loss: w dθ/dz = dθ/dt_rad = -1-2 K/day => w ~ -.5 cm/s When precipitation takes place, heating rates are strong; e.g. 100 mm/day precip ~ energy flux of 2900 W/m2 or an average 30 K/day heating of the atmospheric column => w ~ 8.6 cm/s. However, this positive mean motion is composed of strong ascent of order w ~ 1 m/s in the Cumulus updrafts and slow descending motion around ( compensating subsidence ) Ro=NH/f with N the Brunt-Väisälä frequency and H the tropopause height, is the Rossby radius over which a perturbation spreads. In Tropics it is infinit as f->0, in the midlatitudes it is of O(1000 km). Therefore, daily weather forecasting is much more difficult in Tropics.. But contrary to middle-latitudes where predictability does not go beyond 14-days or so, Tropics have longterm predictability through intraseasonal variability (MJO) and SST coupling (ENSO) Slide 19
The MJO U850 U200 27 November 2011: Meteosat 7 + IFS Analysis Slide 20
YOTC: Hovmoeller of the OLR anomaly Slide 21
Progress in MJO prediction Slide 22
Correlations with T at 500 hpa for Phase 2/3 and forecast steps 12-36 dt/dt_conv 60W 0 60E 120 180 120W 60 Precip 60W 0 60E 120 180 120W 60 For energy transformations in MJO see also Yanai, Chen, Tung (2000), and Matthews et a. (1999) Slide 23 23 YOTC Asian Monsoon Symposium Beijing 16-20 May 2011 @
P (hpa) P (hpa) Difference in T-tendency: Convection over West Pacific - convection over Indian Ocean P (hpa) Dynamics (K/day) Conv (K/day) 50E 100 150 20W 50E 100 150 20W Cloud (K/day) Radiation (K/day) 50E 100 150 20W 50E 100 150 20W Slide 24
Correlations with T at 250 hpa for Phase 2/3 and forecast steps 12-36 dt/dt_conv 60W 0 60E 120 180 120W 60 Precip 60W 0 60E 120 180 120W 60 Slide 25 25 YOTC Asian Monsoon Symposium Beijing 16-20 May 2011 @
YOTC: OLR anomalies Slide 26
Effect of moisture sensitivity in convection scheme to MJO prediction q day+1 conv in 2007 day+5 dq/dt conv 2007-today day+1 day+5 see also Hirons et al. QJ RMS 2013 Slide 27
MJO initiation over Indian Ocean Take a long time series of filtered TRMM data and ERA-Interim reanalysis Identify MJO events, distinguish between primary, and successive, and separate from non-mjo convective events see also Ling et al. JAS 2013 to appear Slide 28
Anomalies in precipitation and 850 hpa wind Non-MJO day+3 day 0 day-3 day-6 day-9 day-12 day-15 Primary MJO significant easterly wind anomaly to the East Slide 29
day 0 day-6 day-12 day-18 day-24 P (hpa) Temperature anomalies Primary MJO Non-MJO significant cold anomaly in mid-troposph. 10-20 days ahead Slide 30
day 0 day-6 day-12 day-18 day-24 P (hpa) Specific humidity anomalies Primary MJO Non-MJO significant mid-tropospheric dry anomaly to the East Slide 31
Closure in Numerical Weather prediction Define (adiabatic) convective available potential energy and entraining density weighted CAPE=PCAPE compute its temporal derivative Slide 32
Closure in Numerical Weather prediction write prognostic equation formally as Define large-scale and convective contribution Slide 33
Closure in Numerical Weather prediction Need mass flux, can also estimate convective contribution from compensating subsidence term, M* first-guess mass flux (kg/m2 s) In diagnostic scheme formulate closure as or as quasi-equilibrium definition Slide 34
Closure Deep in Numerical Weather prediction Requires however specification of adjustment time-scale Diurnal cycle depending on cloud depth H, mean updraft velocity w and resolution n Slide 35
Closure Deep in Numerical Weather prediction Need to take into account the imbalance between deep convective motions and surface forcing, define: Adv+Rad+surf buoyancy flux Slide 36
Closure shallow in Numerical Weather prediction Define shallow as cloud depth<200 hpa, used twice as large entrainment as for deep. PCAPE integral singular for very shallow cloud, go back and use boundary-layer equilibrium only Use moist static energy h=cpt+gz+lq Could have also used q instead of h, or surface buoyance flux (as in previous slide) Slide 37
Diurnal cycle of Precipitation JJA: Amplitude (mm/d) TRMM CTL NEW Slide 38
Diurnal cycle of Precipitation JJA: Phase (LST) TRMM CTL NEW Slide 39
Diurnal cycle: Surface Energy Budgets TP=total precipitation SW=shortwave radiation SF&LF=sensible&latent heat flux Note: (i) shift in TP between CTL and NEW, (ii) TP in CTL in phase with SF+LF=wrong! (iii) for Europe LF>SF, Africa SF>LF Slide 40
How does diurnal Precip scale? TP=total precipitation HF=surface enthalpy flux BF=surface buoyancy flux NOTE: in NEW = revised diurnal cycle surface daytime precipitation scales as the surface buoyancy flux Slide 41
Composite diurnal cycle: Model vs Obs Slide 42
Closure and diurnal cycle over Sahel June 2012 Slide 43
Diurnal evolution of total heating profile -radiation congestus Deep convection Turbulent heat flux Shallow convection Slide 44
Diurnal cycle Sahel June 2012: IFS & CRM Q1-Qrad -Q1 Slide 45
Diurnal cycle & change in circulation: analyse soil moisture change in seasonal forecasts Slide 46 46
West-African Monsoon during AMMA campaign Daily mean precipitation [mm/day] August 2006: day+1 forecast 12 o N OBS (FEWS RFEv2) Forecast Precip in model is shifted too far South, and too much precip over Ocean Slide 47 47
Diurnal cycle: Impact on weather forecasts Slide 48
Wintry showers: radar & forecasts Slide 49
Conclusions Global picture of convection and methods to analyse waves (Hoevmoeller, Eliassen-Palm fluxes, wavenumber-frequency diagrams) Madden-Julian Oscillation: Big improvement obtained through convection (sensitivity to moisture, charge/discharge cycle, energy conversion, propagation?) quasi-equilibrium closures for CAPE: several possibilities from scaling arguments, but practice tells which is/are optimal Diurnal cycle of convection: Large impact not only one phase of precip but also on circulation (notably African summer monsoon0 and weather forecast Slide 50
Towards higher resolution Scalability in Computing Scaling the Globe: the small planet testbed (not shown) Slide 51
IFS grid point space: EQ_REGIONS partitioning for 1024 MPI tasks Each MPI task has an equal number of grid points ICON DWD G. Modzynski G. Zängl Slide 52
IFS model: current and future model resolutions IFS model resolution Envisaged Operational Implementation Grid point spacing (km) Time-step (seconds) Estimated number of cores 1 T1279 H 2 2013 (L137) 16 600 2K T2047 H 2014-2015 10 450 6K T3999 NH 3 2023-2024 5 240 80K T7999 NH 2031-2032 2.5 30-120 1-4M 1 a gross estimate for the number of IBM Power7 equivalent cores needed to achieve a 10 day model forecast in under 1 hour (~240 FD/D), system size would normally be ~10 times this number. 2 Hydrostatic Dynamics 3 Non-Hydrostatic Dynamics Slide 53
Sustained Exaflop in 2033? Slide 54
Exascale problem projections To run a T7999 L137 forecast (~2.5km) may require approximately 1-4 million processors (of current technology) to run in one hour At the same time 1-4 Million processors could run a 50 member ensemble of T3999 L137 in the same hour But first we have to be able to run a T3999 L137 forecast efficiently in one hour! Slide 55