Convection parametrization development in the Met Office Unified Model. Steve Derbyshire COST Summer School, Croatia 2013

Transcription

1 Convection parametrization development in the Met Office Unified Model Steve Derbyshire COST Summer School, Croatia 2013

2 Thanks to Colleagues Anna Maidens, Sean Milton, Rachel Stratton, Martin Willett, Andy Brown, Adrian Lock, Cyril Morcrette, Alison Stirling, Ben Shipway Coauthors on the EUROCS humidity paper (see later) Members of the COST project Students!

3 Contents 1. Approach and methodology 2. The EUROCS humidity case 3. Adaptive detrainment and its impact 4. Further analysis and issues 5. Summary and future directions Crown copyright Met Office

4 1. Approach and methodology Crown copyright Met Office

5 A pessimistic view of science.. Crown copyright Met Office Source:Anonymous

6 A complex problem! Ambient atmospheric state due to various processes Growth of convective clouds Vertical discretization and other engineering

7 We need to bring together: 1. Observations Process studies Data-model confrontation in Numerical Weather Prediction 2. Theory Of convection Of dynamical interactions with convection 3. Computation Climate / NWP analysis Process studies Cloud Resolving Models, LES etc. All of this is challenging - some degree of conflicting evidence is normal!

8 GASS/GCSS < GEWEX < WCRP < WMO GEWEX Cloud Systems Study from 1992 Cloud-Resolving Modelling a key tool Process study methodology cannot be static! Evolves to address current NWP/climate concerns. Progression from ~ Basic CRM validation (eg vs ARM site) Basic evaluation of fluxes/tendencies under large-scale forcing Greater large-scale context Understanding feedbacks/coupling with large scale More explicitly multiscale approaches (eg CASCADE) Renewed attention to the smallest scales / microphysics More sophisticated data sources e.g. Cloudsat, TWP-ICE

9 Classic GCSS (Deep Convection Working Group) methodology CRMs Forcing Observations Evaluation Evaluation Development SCMs Forcing Evaluation Development Forcing Forcing Analysis NWP Forecast Evaluation (consistent errors?) CAPT/Transpose AMIP better NWP better climate? Climate model Jon Petch

10 MetO CRM formulation for completeness Anelastic dynamics (optional; remains unified with LES) TVD advection (also optional) 3-phase double-moment bulk microphysics Simple on-off condensation Smagorinsky turbulence (with stability effects) Edwards-Slingo radiation (optional) Resolution flexible: x normally ~1km or better (100m?) Cyclic horizontal BCs (domain length typically 100km) Other CRMs are available!

11 Observed or idealized cases? Obs case studies (BCs/forcing as observed ) E.g. use of ARM, TOGA, TWP-ICE data etc. under EUCREM, EUROCS, GCSS Spot-checks on CRM and SCM performance Idealized studies (experiments) Seek conceptual progress for parametrizations (not simply fine-tuning) Therefore test sensitivity to individual parameters Emphasis on cloud ensemble properties

12 E.g. (i) organized case C. Marshall 2D slice through 3D simulation of organized system (see also various papers by M.Gray on MCSs)

13 E.g. (ii) ensemble convection EUROCS humidity case 50km domain/250m hor. resolution (plan)

14 Setup & Boundary Conditions Principle of Consistent Testbeds for parallel SCM-CRM runs (allows some latitude) There are legitimate alternative choices of how to set up the lateral or surface boundary conditions (LBCs/SBCs) Different choices may be consistent but give different sensitivities (review if very small or very large) LBCs SBCs

15 Setup & Boundary Conditions E.g. when we prescribe SBCs we have choices: prescribe fluxes <w T >, <w q >? or means <T>, <q>? or surface energy-balance model? There s virtue in simplicity, but want to obtain ballpark realistic sensitivities This all applies equally to idealized or real cases

16 Large-scale forcing CRMs are usually set up with cyclic (periodic) lateral boundary conditions Any systematic flux into the domain due to large-scale advection is handled as an additional large-scale forcing terms S θ (z,t) (=source for θ) etc. For an observed case these may be derived from a budget analysis It s desirable to test sensitivity to domain size

17 LS forcing or feedback? Large scale forcing [S θ etc.] distributes budget impact horizontally (origin in budget studies) Forms a consistent testbed with CRMs/SCMs However we increasingly recognize that convection feeds back on large scales (i.e. forcing not really constant) cf. 2-column studies feedback may affect sensitivity so we want to understand convection in context

18 LS feedback schemes because convection can change the large scales We now have choices of large-scale feedback schemes for process studies Weak Temperature Gradient approach (after Sobel & Bretherton): LS w acts to hold θ v fixed. Alt (simpler): nudge mean profiles on timescale t n small t n : ~semiprognostic, large t n : ~ classical forcing Used in EUROCS humidity case with t n =1hour Both allow forcing to change in response to convective activitiy

19 2. The EUROCS humidity case Sensitivity of moist convection to environmental (free-tropospheric) humidity Crown copyright Met Office

20 EUROCS humidity case (part of wider EUROCS cloud project) Participants: Steve Derbyshire (Met Office), Isabelle Beau (Meteo-France), Peter Bechtold (ECMWF), Jean-Yves Grandpeix (LMD), Jean-Marcel Piriou (Meteo-France), Jean- Luc Redelsperger (Meteo-France), Pedro Soares (University of Lisbon) Acknowledgment: EU contract EVK2 CT Ref: D et al. (QJRMS 2004)

21 Parcel ascent - adiabatic Classical convection theory considers adiabatic ascents (no entrainment) Adiabatic ascent determined essentially by near-surface T,q (schematic) z environment parcel curve (adiabatic) θ v

22 Parcel ascent - entraining Entraining parcel loses buoyancy in proportion to loss of moist static energy, so the slope of the parcel curve changes Approximately this change in slope varies like ε(q e -q sat ) z parcel curve (entraining) environment (schematic) θ v

23 Range of ε or RH values Change in slope varies like ε(q e -q sat )= ε(1-rh)q sat so high-rh or low-ε parcel curves are nearly adiabatic z parcel curves (entraining) environment (schematic) θ v

24 Observed associations between convective type/ intensity in TOGA/COARE and mid-tropospheric relative humidity From Johnson (1997) in ECMWF Workshop Proceedings (humidity profiles composited by convection type)

25 Setup and forcing for EUROCS humidity case: Sea surface specified with SST such that target θ across BL=1K. LS Forcing specified as relaxation of mean profiles to target profiles (1hr) Run to statistical equilibrium

26 Participating models CRMs: Met Office CRM (Derbyshire) CNRM CRM (Redelsperger) SCMs: ARPEGE-CLIMAT (Beau) ARPEGE-NWP (Piriou) LMD (Grandpeix) ECMWF (Bechtold, Soares; Tiedtke or Bechtold convection) Met Office (Derbyshire)

27 Spatial structure of convection in CRM EUROCS humidity case 50km domain/250m hor. resolution (plan)

28 MetO CRM CRM CNRM 90% 90% 25% 25% Updraught mass flux intercomparison of 2 Cloud-Resolving Models and several Single-Column Models IFS-MNH SCM ARP-NWP SCM Each plot shows 4 humidity subcases with the 4 different humidities (25%,50%,70%,90%). IFS-Tiedtke SCM MetO SCM Are the results of the 2 CRMs consistent? Do the SCMs follow the CRMs?

29 r (=column relative humidity) Observational study, Bretherton et al. JClim 2004

30 Precip vs. RHt in EUROCS humidity case (note 1.5mm/hr=36mm/d)

31 CRM robustness Sensitivity tests to vertical/horizontal resolution (x2), domain size (x2), diagnosis time

32 How good do CRMs need to be? To support parametrization progress CRMs need to be: better than current SCMs (but not perfect at any finite time) able to shed light on scalings/structural/theoretical issues Increasing internal checks on non-microphysical issues some indications of convergence ~250m horizontal resolution similarly we can check domain size impacts Microphysics clearly still a key issue especially in borderline situations (congestus etc.) test sensitivity in runs we are using for parametrization Use field experiments e.g. RICO (Abel & Shipway 2007)

33 EUROCS humidity case/motivation for adaptivity Mass flux profiles in a range of free-tropospheric humidities resolution) Cloud-Resolving Model SCM (UM4.5) 90% 90% 25% 50% 70% 25% Derbyshire et al., QJRMS (2004) EUROCS Special Issue [Numbers are for humidity parameter RHt ]

34 3. Adaptive detrainment and its impact Crown copyright Met Office

35 Met Office Unified Model

36 Temperature biases in NWP 5- day forecast (pre-2005) Warm bias in most of tropical troposphere but cold bias at the top of convection (top too low?)

37 Relative Humidity biases in 5-day forecast tests Substantial dry bias in inner-tropical upper troposphere (up to 10% RH in places)

38 Precipitation and wind spinup illustrating data available from NWP analysis Sean Milton

39 Wind spin-up in 5-day forecasts

40 EUROCS humidity case/motivation for adaptivity Reminder: Mass flux profiles in a range of freetropospheric humidities resolution) Cloud-Resolving Model SCM (UM4.5) 90% 90% 25% 50% 70% 25% Derbyshire et al., QJRMS (2004) EUROCS Special Issue [Numbers are for humidity parameter RHt ]

41 Ensembles and detrainment I experimented with toy Arakawa-Schubert-like sub-ensembles, evaluating aggregate properties Main generic impact seemed to be from partial detrainment governed by distribution shapes, e.g. ~(C-θ v ) γ

42 Updraught buoyancy pdfs in CRM (EUROCS case)

43 Adaptive partial detrainment Take the above distributions parametrized by γ and assume convective elements detrain when they become neutrally buoyant This leads to an extension of the Gregory- Rowntree algorithm for forced partialdetrainment δ f of the form: where () * denotes evaluation before partial detrainment and R det =(γ+1)/(γ+2)

44 25% RHt=90% Re-evaluation with SCM in EUROCS case - clearly able to do smoother adaptive detrainment - other issues (see later)

45 Adaptive Detrainment Relative Humidity impacts (Initial NWP tests by Sean Milton/Martin Willett)

46 Adaptive Detrainment Temperature impacts

47 Observations Initial climate tests J.Rodriguez 1-year annual mean precipitation (1990) Adaptive mod seems to improve Tropical W. Pacific with benefits to Pacific wind-stress HadGAM1 control HadGAM1 with adaptive detrainment

48 Impact on wind-stresses (Rachel Stratton)

49 4. Further analysis and issues Crown copyright Met Office

50 Further CRM diagnosis In CRMs we can do more detailed analysis of the updraughts, downdraughts etc. by conditional sampling E.g. select buoyant cloud updraughts Also can analyze entrainment, detrainment etc. either directly or by budget analysis Approaches pioneered by P.Siebesma and others

51 Core (BCu) area fractions in CRM Note mass-flux adaptation to humidity seems to come from buoyant cloudy core area a u, not w u RHt=90% 90% 25%

52 Core (BCu) updraught velocities Almost independent of RH! Peak in mid-layer almost coincident with buoyancy peak

53 Core (BCu) buoyancy Limited sensitivity to RH Peak in mid-layer coincident with buoyancy peak 25%

54 CRM-diagnosed ε and δ (From bulk q t budget following Swann 2001 after Siebesma 1995; also comparing to G-R=Gregory-Rowntree 1990) (cf. Siebesma 1998) 25% 90% 25%

55 Further changes to ε Increased entrainment together with adaptive detrainment can capture a shallow regime 90%

56 5. Summary.. and the future? Crown copyright Met Office

57 Improving convection parametrization is challenging but important for weather and climate prediction Given the difficulty of the challenge, we need to bring together obs, theory and computational results Note observational information comes in through NWP as well as process studies

58 We have followed a GCSS-style strategy of making systematic use of Cloud Resolving Models Important to think about the CRM setup and forcing in relation to large-scale context WTG? Relaxation? EUROCS humidity intercomparison showed the impacts of mid-tropospheric humidity For us this motivated a change called adaptive detrainment

59 Adaptive detrainment summary Deliberately simple, incremental and based on simple conceptual model of ensemble pdfs Seems to give considerable benefit in Greory- Rowntree scheme Adaptivity allows scope for further development of this scheme Can we explain the impacts? Exercise for the student!

60 Total heat and parametrization sensitivity Total heat thinking (Emanuel-Neelin): tropical convection broadly response to accumulation of h=c p T+Lq+gz, exported largely via Convective updraught ω ( h / p) dp Possible explanations for excessive convection: Convection exports insufficient h for given mass-flux (cf. Neelin s Gross Moist Stability)? Cloud layer BL Precipitating downdraught Surface transfer ~ c h U h DD too weak? Or carry insufficiently low h back into the BL (because saturated)? (cf. Raymond BLQ) Winds too strong? c h too high? Sea too warm? (cf. Emanuel)

61 Further plans for convection in the Met UM include.. More work on entrainment Convective memory Convective organization Stochastic options Learning from other schemes Etc. etc...