WGNE-THORPEX-ECMWF workshop on Physically Based Stochastic

Document Sample
WGNE-THORPEX-ECMWF workshop on Physically Based Stochastic Powered By Docstoc
       workshop on
     Physically Based
Stochastic Parameterisation

June 11-14, ECMWF, Reading, England

         Tom Hamill (NOAA, WGNE)
       Tim Palmer (Oxford U, ECMWF)
      Istvan Szunyogh (THORPEX/PDP)   1
Reminder: we discussed at WGNE-25
   whether to co-sponsor this…

          back to
stochastic parameterisation

                  Why a workshop?
• My reason (there are others) : deterministic forecasts and (ensembles)
  contaminated by bias, (lack of spread).
• One cause may be inappropriate deterministic assumptions built into
  parameterisations (next slide).
• Stochastic parameterisation a hot topic, but progress could be faster.
• A workshop and its recommendations could help
   – Share recent knowledge.
   – Educate grad students, post-docs.
   – Through recommendations, provide official WMO, THORPEX,
     WGNE blessing to proposal writers as to importance of subject
   – Spur development of common data sets (e.g., CRM simulations)
     that could be used for training and validation.
   – Determine what sort of diagnostic techniques should become
    Plant & Craig
Cloud-resolving model run to
statistical equilibrium. Range of
convective mass flux responses
shown here and how they vary
as a function of the size of the
grid box.

Lesson: as we move to finer resolutions,
the deterministic assumptions
built into the parameterizations
becomes increasingly inappropriate.

       Why on “physically based”
      stochastic parameterisation?

• There are techniques in operations that act to
  increase spread in ensembles, but they cannot
  be defended from first principles.

• In iterative development of NWP systems, it’s
  important to get the right answer for the right
  reason, else future testing is complicated by
  building upon past ad-hoc work.
            Where WGNE can help
• Are there missing topics in proposed workshop (next slides)?
  Missing key questions?

• Help identify who to invite, both as speakers and as participants.
   – I have a preference for a moderately sized workshop, say 50
     people, with many post-docs and grad students. They are the
     ones who will probably be doing the work and can benefit the
     most. THORPEX/PDP disagreed somewhat, wanted more
     emphasis on active researchers.
   – I want to identify speakers that are good communicators as
     much as more than those who have extensive research

• Feel free to e-mail me your thoughts, tom.hamill@noaa.gov            7
  Proposed organization of workshop,
(3) Current developments in parameterising model
  uncertainty. [this session would be a longer one, intending to
  give the major players doing research a chance to describe
  their recent work]. Examples of major subject areas include:
  (a) Stochastically perturbed physical tendencies.
  (b) Stochastic backscatter.
  (c) Stochastic convective parameterisation.
  (d) Perturbed-parameter approaches.
  (e) Multi-parameterisation approaches.
  (f) Integrating stochastic differential equations.

  Proposed organization of workshop,
(4) Issues in stochastic parameterisations.
   (a) Balance and time/space covariances for stochastic noise,
   i.e., inappropriate column assumption - how to incorporate
   (b) Relative benefits of comprehensive approaches
   (stochastically perturbed physical tendencies) vs. process-by-
   process approaches
   (c) Methods of diagnosis and verification for stochastic
   techniques - how do we know if we're doing the right thing
   (d) Use of ensemble Kalman filters and other data assimilation
   techniques as a diagnostic tool

Proposed organization of workshop
(5) Breakout groups. Questions for breakout groups might
   (a) What data sets do we need to collate or build for training
   and validation, such as CRMs, prior relevant field experiment
   (b) What underlying principles should be followed in order for
   a stochastic method to be considered “physically based?”
   (c) Are there general diagnostic techniques that can be
   applied across the suite of parameterisations?
   (d) What recommendations should be made to funding
   agencies and NWP centres regarding this topic?

(6) Plenary.
         Workshop organization

• Please e-mail me, tom.hamill@noaa.gov

Proposed organization of workshop
(1) Overview. Also: statements of interest from WGNE, THORPEX, other
    organizations about their scientific interest and role. Review of ensemble
    prediction, existing efforts to deal with model uncertainty via multi-
    model, multi-parameterisation ensembles, statistical post-processing.
(2) Review of parameterisations (for each topic below, the invited speaker
    would describe the physical process, describe common parameterisation
    approaches, and describe where there are deterministic assumptions
    built into the parameterisation, what scales these deterministic
    assumptions are valid for). This session is background.
      (a) Land-surface
      (b) Boundary layer
      (c) Deep convection
      (d) Stratocumulus
      (d) Radiation
      (e) Microphysics
      (f) Turbulence, mountain drag
      [etc.]                                                                 12