J.-F. Geleyn, D. Banciu, R. Bro žková and L. Gerard

1. Entrainment as the main controling parameter of convective activity in the 3MT moist-physics unifying scheme. Trials for an extension towards a 'cold-pool mechanism’-oriented ‘memory of entrainment’ J.-F. Geleyn, D. Banciu, R. Brožková and L. Gerard (with strong reference to ideas of J.-M. Piriou) 26/3/09, Prague, CHMI

2. A few words about 3MT Goals Main choices Implementation characteristics Grey-zone results

3. Why 3MT? • (i) Attacking, within a long-term perspective, the challenge of the horizontal scales (δx ~ 5 km) where precipitating convection is neither fully resolved nor likely to be correctly parameterised in a ‘classical’ way. • (ii) Insisting on stable (for longer t) and cost-efficient algorithmic solutions (bulk model for the drafts for instance). • (iii) Having a ‘NWP-controlled’ progress (novelty AND ascending compatibility). • (iv) Modularity-flexibility as the essential tool to obtain a multi-scale character (being able to swap and/or tune the ‘processes description’ without touching the structuring algorithmic). • (v) Using a prognostic orientation for reconciliation of ideas about complex microphysics and mass-flux-type parameterisation (neither CRM nor QE).

4. 3MT, the acronym • Three ideas/concepts: • Modular, because of the ALARO-0 effort made in order to stay compatible with a general phys-dyn interfacing while searching proximity with the AROME concepts (R. Brozkova, I. Stiperski, J.-F. Geleyn, B. Catry, D. Banciu); • Multi-scale, because a great deal of the architectural constraint comes from the ‘grey-zone’ oriented work, initiated in 2001 by L. Gerard; • Microphysics & Transport, to underline the decisive catalysing role played by the central proposal of J.-M. Piriou’s PhD work, made in 2004.

5. Microphysics AND Transport (M-T) • It is the basic idea behind all what follows. • Allows to rethink around two trivial facts: • ‘Detrainment’ here = ‘Entrainment’ somewhere else. • ‘Cloud+precipitation microphysics’ is anything but instantaneous (fall speed of drops ~ propagation speed of convective structures). CONSEQUENCES • 3MT gets away from assumptions of a stationary cloud (neither in size nor in properties). • 3MT fully takes into account the fact that microphysics has a rather long lag-time and is not only happening ‘within the drafts’.

6. One key equation • If we set • Mc from two independent prognostic equations for c and up • up as constant (in the cloud) along the vertical => 2D-only closure • E from ‘something’ (see later in the presentation) • Then D cannot be ‘parameterised’ (overdetermination otherwise) and it is obtained from all other computations. Piriou et al. (2007) showed that it is in fact mainly constrained by the microphysical activity => a justification for using the M-T decomposition.

7. Time- and space-scale issue • Basically, 3MT is a way to do ‘as if’ deep convection was resolved but without needing to go to scales where this is true. • This is thanks to: • Prognostic and diagnostic ‘memory’ of convection; • A unique micro-physical treatment beyond all sources of condensation. • But, owing to the peculiar role of entrainment in the M-T concept, this requires to better understand what one means when ‘specifying’ entrainment in one way or the other.

8. 3MT

9. The nice sides … • NWP orientation: bulk mass-flux but fully prognostic handling of the mass-flux AND of the 2D closure. • With M-T, the 2D closure and a prognostic equation for the mass-flux, no need to parameterise anymore detrainment. • Facility to work on ‘modularity for flexibility’. • One single microphysical-type computation, except for the condensation/re-evaporation, the latter being obtained from the sum of a ‘resolved’ contribution and of a ‘convective’ one. • Lot of freedom for a complex fully prognostic micro-physics => more ‘memory’ of past convective events. • The ‘cold-pool’ effect’s parameterisation should happen quite naturally in this framework.

10. But … • The handling of the ‘cascade’ (neither sequential nor parallel treatment of individual contributions) is not always easy: • Avoiding ‘double-counting’ for updraft and downdraft closure is not trivial; • The sedimentation aspect of the downdraft impact must be treated heuristically; • In order not to be forced to iterate expensive computations, one must make judicious choices about which information to pass or not to pass to the next time-step (and on how to best use it). • Not enough effort was devoted to the closure formulation, especially in view of its ‘multi-scale’ impact. • For a ‘deep’ framework, a vertically constant area fraction for drafts is OK; but this does not hold anymore in the ‘shallow’ case.

11. Modularity for flexibility. Where? • Updraft (and downdraft): • Ascent computation (including the prognostic handling of an in-ascent vertical velocity); • Closure (leading to a 2D field of active area fractions, also prognostic); • M-T equations => (non-negociable). • Microphysics: • General organisation (explicit need of fall-speeds, non-advective sedimentation, geometrical overlap’s impact, protection of convective condensates at the ensuing time steps, …) => (non-negociable); • Type of sedimentation’s handling; choice of the computation (or set-up) of fall-speeds; • Auto-conversion type processes; • Collection type processes; • Evaporation, Sublimation, Melting, Freezing of falling species. • ‘Resolved’ adjustment: • Choice of critical relative humidity; • Formulation of cloud-cover; • Condensation/Evaporation rate => (separation from the other items isnon-negociable; but the choice of the formulation is free).

12. Unique vertical loop for the microphysics, made purely ‘local’ thanks to ‘PDF-based sedimentation’ (Geleyn et al., Tellus-A, 2008) This ‘microphysics’ is sandwiched between up- et downdrafts’ computations (Gerard, QJRMS, 2007) Sequential handling of both condensation sources, but summing of their inputs for a unique‘microphysics’ call Prognostic, barycentric and conservative phys-dyn interfacing (Catry et al., Tellus-A, 2007) Mcu/d(p)=-u/d(p).u/d(p), bothprognostic (Gerard & Geleyn, QJRMS, 2005) Convective equations in Microphysics-Transport form (Piriou et al., JAS, 2007) 3MT, the backbone

13. And it works quite like expected … • The next viewgraph presents the structure of 6h cumulated precipitation amounts over the Alps for 4 forecasts valid at 18h range for the initial date 21/6/06 00UTC: • Two with 9km mesh-size (left column) and two with 4.7km mesh-size (right column); • Two with ALARO-0-minus-3MT (upper row) and two with ALARO-0 [i.e. with 3MT] (lower row). • One notices the strong ‘grey zone syndrome’ when trying to apply the ‘classical diagnostic convection scheme’ at the 4.7km mesh-size (upper-right diagramme). • On the contrary, when using the 3MT ensemble of novelties, the precipitation patterns ‘scale’ correctly from one resolution to the other (bottom row).

14. Scalability of precip. patterns

15. Operational implementations • At ~9km mesh-size: • Cz (4/6/08) • Si (16/6/08) • Sk (19/8/08) • At soon (already in LAEF set-up) • Interest in Hr, Ro, Pt, Tr • At ~4.5 km resolution • Be (15/1/09) • Interest in No

16. Now about entrainment(no need to speak about detrainment in 3MT; it is ‘decided’ by microphysics) Role with respect to the novelties of 3MT Empiricism of the ‘diagnostic’ set-up  value of its ‘tuning’ A prognostic alternative? Some model-run diagnostics

17. Two constatations • Apart from the autoconversion time scales for microphysics, the parameters controlling the entrainment rate for the convective ascent’s computation were the only ones which needed retuning whan introducing 3MT in ALARO-0: • Doubling of the min and max entrainment rates (to respectivly 0.00005 and 0.0016 m-1) • Dividing by 1.5 the sensitivity to ‘free ascent buoyancy’ (which regulates the maxmin adjustment) • Diminishing by a factor 2.5 the account taken of ‘ensembling entrainment’ • All this goes in the direction of higher entrainment rates (more realistic but still below ‘measurements’)

18. Standard ‘static’ entrainment Standard ‘static’ entrainment qv ‘Doubled’ Entrainment Max / Min rates => 3MT choice ‘Doubled’ Entrainment Max / Min rates => 3MT choice STATIC 3MT T Entrainment matters more for convection in 3MT than in the equivalent ‘static’ scheme

19. Total Total Total Total qv BEFORE AFTER T Reduction of the difference to the ‘static’ reference via the retuning (3 aspects in it)

20. ‘Diagnostic’ / weak entr. ‘3MT’ / weak entr. ‘3MT’ / strong entr. Reduction of the difference to the ‘static’ reference via the retuning (3 aspects in it) Detr. Entr.

21. Structure of the ‘static’ entrainment rate computation (Gerard & Geleyn, 2005) Very heuristic but ‘fine tuned’ by more than 10 years of operational use

22. Cold pool mechanism • Basic entrainment is by nature strong and prevents immediate deep penetration of convective clouds • But the evaporation of first convective precipitation amounts creates density currents, cold pools and thus ascent-favourable conditions at the edge between the latter ones • Quickly raising plumes entrain relatively less (a mechanism already parameterised by ‘’ in our ‘static algorithm’) • So the idea is to encompass the effect in an hysteresis-like link between past evaporation of precipitations and current entrainment rates; the ‘time-lag’ aspect calls for a prognostic algorithm, with all the associated risks and difficulties. • In his PhD thesis, Piriou proposed and tested the above in a demo-fashion. 3MT in principle allows to go to a more concrete test.

23. Equations of J.-M. Piriou’s proposal

24. 3MT-linked change of variable of L. Gerard downdraft is treated prognostically like its updraft equivalent and ‘integrates’ all effects of previous evaporation (but resolved + convective, since the distinction disappears in 3MT)

25. Improvements by D. Banciu • Replacing f(p/psurf) by an adimensional scaling (inspired by Met-Office’s LES estimates) in (-surf) • ‘Observed’ value 3.5. Our min/max bracket [0.3-3.] • NB:this quite useful scaling will also be used for the presentation of the results • Quitting the linear relationship • Advecting  and reinitialising it to the ‘static’ tuning when no convection is present

26. Hysteresis-control Mironov-type term Our current equations downdraft is treated prognostically like its updraft equivalent and ‘integrates’ all effects of previous evaporation (but resolved + convective, since the distinction disappears in 3MT)  is a fully prognostic quantity Tuning => cp=5000 s E=2cp

27. First results with ‘prognostic entrainment’ Diagnostic 3MT computation. Histogramme at all model levels of the (admimensional) product of the entrainment rate by the height above the ground The same but for the ‘best’ tuning (so far) of Piriou’s proposal for a treatment of entrainment with cumulated influence of the past evaporation rates. => Correct histogramme structure but too little convective activity!

28. First results with ‘prognostic entrainment’ Updraft Mass Flux Divergence Diagnostic entrainment Prognostic entrainment Downdraft Mass Flux Divergence

29. (Preliminary) lessons • Prognostic entrainment can mimick the ‘well-tuned’ statistical structure of the ‘static’ case • The space- and time-averaged mass fluxes agree quite well after this ‘tuning’ • But the frequency of convective activation is diminished => more intermittency for the ‘prognostic’ case • Unfortunately, we did not get closer to ‘observed’ entrainment rates (our upper bound is still just below the ‘truth’) => the problem stays!

30. Two ways to look at entrainment (1/3) • First definition: the positive contribution to the vertical change of mass-flux with height • Second definition: the rate of mixing of ‘cloud’ and ‘environment’ conservative properties • In 3MT, both physical processes, though not completly disconnected, are really treated separately => one may try and diagnose them independently of each other

31. Case without the ‘’ parameterisation (forget the units) E1 E2 Two ways to look at entrainment (2/3) General agreement, except where the ‘plumes’ start, but details differ, in fact rather significantly for an horizontal average

32. Case with the ‘’ parameterisation (forget the units) E1 E2 Two ways to look at entrainment (3/3) Distinction between the two regimes is even more marked but details elsewhere match better (let us have a closer look)!

33. Tuning not so bad! ‘Pseudo mass-flux’ integrals of E1 and E2(comparison ‘diagnostic’ vs. ‘prognostic’) Diagnostic Diagnostic Prognostic Prognostic

34. E1 E1 E2 E2 Hidden constraint? Better parallelism ‘Pseudo mass-flux’ integrals of E1 and E2(comparison of the two methods) Diagnostic Prognostic

35. Conclusions • With the additional degrees of freedom of 3MT, the role of entrainment is increased • Retuning of the ‘static’ entrainement rates without touching the algorithm delivers what we believe to be a good reference target for the ‘prognostic entrainment scheme’ • The latter delivers some first promising results (more freedom, better dissociation between the two sources of ‘ascent’) • We are very sorry that we did not manage to push the assesment any further before the workshop !