Summary of Microscale Modelling Workshop held on
27th March 2003, 10:30 - 16:00, at the
UK Met Office .
Presentations
1. Introduction: Stephen Mobbs.
(link to presentation).
Background to the project and an analysis of the rationale,
project objectives, issues to be addressed. The project timetable was
presented. Copies of the
original project specification approved by UWERN were distributed,
but it was appreciated that there was now a need to incorporate recent
scientific progress.
2. Workshop structure: Alan Gadian.
(link to presentation).
Workshop outline, programme for the day, general issues and what are we
trying to achieve.
3. Equation Sets: Andy White.
(introductory slide). Specific analysis of the continuity equation
with reference to fully compressible set, anelastic set (many variants),
pressure set and hydrostatic pressure set. A tabular summary, with specific
reference to acoustic modes is provided.
(link to presentation).
4. Validity of anelastic and other equation sets as inferred from
normal-mode analysis: Nigel Wood.
(link to presentation). This work will appear in the QJRMS, but a
preprint version is temporarily available
(link to preprint). Discussion of the importance of use of complete
compressible equation sets for linearised solutions was presented for
different scales. For meso-scale applications, pseudo in-compressible and
compressible set were satisfactory; for NWP full equation set is necessary.
5. Blasius and orography issues: Andy Brown
(link to presentation). Discussion of the Blasius grid structure, UM
grid structure and scale dependent decay.
6. High order non-oscillatory schemes for advection-reaction
equations: Tito Toro
(link to presentation) Examination of the reaction advection equation
and the ADER approach. An extension to a multi-dimension approach,
with comparison with numerical solutions was given.
7. Data Assimilation: some thoughts: Ian Roulstone
(link to presentation). Some thoughts on Data Assimilation, high
resolution modelling. General Data Assimilation issues and those
specifically related to the Microscale Model were discussed.
8. Microscale Modelling from an Engineering perspective:
Rex Britter & Bill Dawes
(link to presentation). N.B. slides not in the original order .
This talk presented the need and relevance for a building scale model. Aspects
of Dispersion modelling from the Roy. Met Soc Policy Statement, comments
relating to COST 615, EU model evaluation group, the Saturn project and
current confidence predictions from the QNET-CFD newsletter. Willing
to assist with a Building scale model development.
9. Issues about a fine scale model Mike Cullen (given by AG):
(link to presentation). What is the model for? Data assimilation
requirements? Developments of the UM and what can and can't be covered by
the UM.
10. Experience of modelling atmospheric flows with a commercial CFD
package: Christiane Montavon:
(link to presentation). Examples of the CFX models and what was achieved
over a relatively short time scale was presented. Examples of applications
from ventilation of a railway station to foehn effects and flow over isolated
hills was given.
Discussion
Questions were raised during each talk, and for a prolonged session
at the end of the afternoon. The main comments are collated below.
Specific discussion areas in the workshop
What is the model for?
This issue was raised throughout the workshop in the presentations
and in the discussion. It was agreed that there
was a need to create a model to look at local scale flows over steep
terrain and also a model for use in urban type conditions.
the need for an urban type model was expressed by both the
engineering and the Met Office communities. The need for a model
that could be applied for flow over steep orography was agreed.
How "steep" was not resolved. These two objectives could
not be achieved by the same model. There was an opinion presented
that there should be two possible models, if resources allowed.
Equation set?
Significant input from Piotr Smolarkiewicz and Martin Miller, provided
useful input from non-UK sources. The presentations provided an important
foundation for the discussion. The advantages of the compressible
equation set were proposed by several of the speakers. The use of
an anelastic set several was also feasible. Some suggested that perhaps
the Lipps - Hemler anelastic set could be equally used.
However, many suggested that comparability with the UM equation set would
be of advantage. There are issues of having to cope with sound waves,
e.g. by solving an elliptic equation as in the UM, or using some
time splitting scheme as in RAMS etc. The advantages of the
Laprise hydrostatic pressure co-ordinate system were presented.
However, there was a consensus feeling that the UM equation set would
the current preferred option.
UM development and use for a microscale model.
A need for the microscale model was variable horizontal (or nesting) and
vertical grids. It was stated that this was in line with UM developments
and that there would be development of the UM in this direction.
The ability of the pressure solvers to cope with steep terrain
following co-ordinates was discussed. It was he view of of several
people that the pressure solvers could cope with steep slopes , but
not overhangs. It was suggested that investigations were made to
test the applicability of existing pressure solvers.
Data assimilation.
Discussion of the needs for data assimilation was discussed. The role
of DA as opposed to "coupling". For a steep terrain model, DA would
be of great advantage, and would usefully link in to Met Office
future developments. If the UM / new dynamics equation set were
used this would be of advantage.
Grid systems.
Limitations were seen in current grid systems. For an urban code,
unstructured grids, adaptive grids, are options and further discussion
is required. For a steep terrain model, development of variable
horizontal grids are envisaged. (see notes below)
. . . . . . . ---------------------------------------------------
Overall summary and actions following the workshop
1. There would be a web collection of the presentations and
conclusions of the meeting produced (this document).
2. An "urban scale" model and "steep terrain" model were not compatible.
If both objectives are to be attained, then separate models would
need to be constructed. The feasibility of this will be discussed at
the next workshop.
3. Email sub-groups, managed by Alan Gadian, would be convened to
look at specific aspects of
the project. These groups would examine the following topics. The groups
are to be formed from those wishing to participate. Suggested topics are
listed below, but are not meant to be exclusive. Participation,
and structure of email groups is to be arranged and are only suggestions
drawn from the workshop discussion.
Urban Code
Cfd codes. The use of suitable codes or approaches would be assessed.
How much resource is required? Can existing
resources (e.g. work of M. Hubbard) be utilised for a non-adaptive
finite volume approach? Can one be applied or developed to the
UM equation set. Should an unstructured model be utilised?
Is this aspect of the research suitable for a EPSRC joint sion?
How could such a project be achieved?
Email Group?
Development of an urban scale model.
SB, RB, XC, BD, AG, AB, MH ?????
Steep terrain code
The workshop suggested the "New Dynamics" equation set
of the UM would be the way forward for a microscale model. Is this
agreed by the group (e.g. rather than the Laprise / Lipps Hemler set)
Should a cartesian based structure be used? Can the UM code be
adapted to do this and what specific work needs to be done.
Are there any modifications to the equation set required?
What effort is required to make this a portable system, and
can a work structure be defined.
Generalised Coordinate System (vertical & horizontal). Can
variable horizontal grid be incorporated? What is required to achieve
this.
Email Group?
AW , NW, AS, TD , AB , AG , SV , SM, RK, PM , SG , SM , JT ????
Data Assimilation. It was agreed that the model setup need to
include the demands of data assimilation, rather than just simple coupling.
What controls need to be included in a microscale model. The group
would define these for the next workshop.
Email Group?
IR, RK , PM , CS , AG ?????
Numerical Techniques. This includes all aspects of
model numerics, eg pressure solvers and advection. schemes.
The workshop considered these issues would become more important in
later stages of the project, and were not critical at this stage. Is
this agreed and what criteria / methods should be used. If using the UM
as a basis code, would it be useful to test elements; e.g. pressure solvers
Email Group?
NW, JT, SV, TT, PM , AG ?????
4. Test Cases. The generation of set test cases should be considered
by the working groups, and ideas discussed at future workshops. Specific ideas
have been generated in the
pre-workshop background, the
pre-workshop agenda
and in the "additional submissions" section below.
5. The project was designed to include aspects of UWERN activity, and
current and future holders of grants would be approached to develop
specific areas. This would be discussed by university groups
and for the future UWERN / JIF proposal.
6. The next workshop would be held in the autumn to address
conclusions of the sub-groups, and progress made.
. . . . . . . ---------------------------------------------------
Additional submissions
A. Tito Toro.
CARTESIAN CUT CELLS
The Cartesian Cut Cell (CCC) approach offers some advantages over the mature
Unstructured Gridding (UG) approach. Note that CCC is also unstructured. CCC
can deal with arbitrarily complex geometries. In 2D the "boundary fidelity" is
optimal; in 3D some compromises are necessary. The degree of automation of CCC
Bl with arbitrarily complex geometries. In 2D the "boundary fidelity" is
optimal; in 3D some compromises are necessary. The degree of automation of CCC
is claimed to be higher than that of UG; the cost of grid generation is lower
for CCC. Also, I believe CCC is more suitable for Adaptive Mesh Refinement
(AMR) applied in both space and time. Some attempts have been made to include
adaptation in time in the UG approach, but I am not aware of these being
successful.
I have some experience in CCC as applied to three-phase (3 velocity vectors)
problems including combustion and moving boundaries in 2D axi-symmetric very
complicated configurations. I'm very satisfied with the performance of the whole
approach.
General remarks:
A research tool (in the form of a mathematical model, numerical method, grid
generation system, initial, boundary conditions, closure relations, etc.) is
different from an application tool. This is specially true in weather
forecasting, where efficiency is fundamental. This is not so for research
different from an application tool. This is specially true in weather
forecasting, where efficiency is fundamental. This is not so for research
purposes, where "accuracy" should be an overriding priority.
B. John Thuburn
General Comments:
It seems unlikely
that we'll be able to come up with a single model that will do
everything everybody wants, but maybe two models between them
will do most things most people want.
Certainly I think the New Dynamics should be one of the models
considered.
With conversations with Terry Davies, the New Dynamics could be adapted
to have vertical no-flow boundaries, i.e. to cope with
cliffs, buildings etc. He thought it shouldn't be too much work.
If that can be done then the modified New Dynamics on its own
might satisfy most needs. This would have the advantage of
short-circuiting a lot of model development.
One limitation of the New Dynamics is that its schemes are not
inherently conservative. That might not matter for the overall
accuracy for the kinds of problems to be looked at, but it would
affect the ability to diagnose closed budgets, e.g. of moisture.
(We recently had quite a job closing the moisture budget in the
current LEM, which is supposedly conservative.)
My main comments are on testing models, which I've been thinking
about on and off over the years. While everyone else has been
devising their favourite test cases, I've been trying to
advocate that first we need to decide what is the purpose of
any test case and what is it trying to test. Below
is my latest version of my evolving thoughts on the subject.
Not all are completely relevant to the microscale model, but
many are.
Issues that proposed test cases might address
(1) Detect coding errors
(2) Domain and gridding:
Poles and other grid inhomogeneities / anisotropies;
Bottom boundary (may be non-trivial e.g. for theta-coords with
mass-less layers);
Top boundary? It's probably impossible to treat the top boundary
"correctly" in a model with a finite number of layers, so
test cases should be designed so that the top boundary isn't
an issue, or test the effectiveness of any proposed "sponge"
layer.
(3) Advection:
Horizontal;
Vertical.
[Includes things like "age of air", spurious numerical mixing,
behaviour of tracer-tracer correlations,...]
(4) Conservation:
Mass of air; mass of tracer; angular momentum; energy.
Arguably, whether or not we demand conservation of potential
enstrophy (and possibly even energy?) should depend on
whether
or not there is a cascade to small scales;
Material conservation of potential temperature and potential
vorticity,
tracer mixing ratio, (equivalent potential temperature in
the moist case); non-trivial if these are not advected or
used as coordinates).
[E.g. examine correlations of these quantities with passive
tracers
initialised to look the same; compare with trajectory
calculations]
(5) Rossby wave propagation:
Horizontal and vertical;
Zero and non-zero phase speeds.
(6) Gravity wave propagation:
Horizontal and vertical;
Zero and non-zero phase speeds;
Sensitivity to time step, especially with semi-implicit schemes.
(7) Acoustic wave propagation (for non-hydrostatic models):
Horizontal and vertical;
Zero and non-zero phase speeds;
Sensitivity to time step, especially with semi-implicit schemes.
(8) Balance:
Maintenance of an initially balanced flow (straightforward e.g.
for
steady zonally symmetric flow, but less obvious for more
complicated flows where balance is a fuzzy concept)
[e.g. monitor mean square divergence tendency?];
Include case in which nonlinear terms are important part of
balance;
Adjustment to balance, bearing in mind application to climate
models
where convection and gravity wave drag schemes, for example,
force at the grid scale.
For non-hydrostatic models consider quasi-hydrostatic balance as
part of the overall balance.
(9) Orography:
Pressure gradient term;
Orographic resonance.
(10) Cascades to unresolved scales:
Layerwise 2D flow;
Frontogenesis;
(Fully 3D turbulence? smallest scale eddies are more energetic in
this case, so a subgrid parameterization might be essential,
whereas in 2D the smallest scales are essentially passive
and
we just need sufficient dissipation to soak up the cascade.)
[E.g. monitor PV field, energy and enstrophy spectra, look for
grid-scale noise.]
(11) Non-hydrostatic effects:
e.g. gravity wave reflection.
(12) Buoyancy-driven flow:
E.g. Buoyant bubble (may be tricky if it generates 3D
turbulence);
Other buoyancy-driven flow relevant to climate, e.g. cold surge
type flow?
(13) Elliptic solvers:
e.g. Poisson equation; Helmholz equation.
(14) Computational modes:
Could be associated with horizontal or vertical grid or time
stepping. I'm not sure how one could systematically test for
them (perhaps by forcing on the grid scale for one time step?
careful analysis of the discrete equations?)
(15) Convergence with improving time and space resolution.
(16) Cost / efficiency.
(17) For adaptive models one could add a list of issues related to
the adaptivity.
C. Other comments(Mike Cullen)
Since people had reservations about the cost of maintaining any model,
I would have thought that using 2 models was even more a problem.
Urban code:
A key issue is to be able to maintain
a state of rest in hydrostatic balance.
Steep terrain code: I suspect the new dynamics can do this job, with
extra terms included in the coordinate transformation. If not, then a
different sort of grid structure has to be used and we get the urban
code.
Generalised coordinate systems:
The real question is whether terrain
following coordinates can be used in the steep terrain model. Need to
establish when this approach ceases to be viable. This is best
addressed by test cases, but then need reliable reference solution to
check against.
Data assimilation: I would expect standard techniques to work, with
some multivariate coupling to maintain closeness to hydrostatic
balance when appropriate.
Numerical techniques: Either anelastic or compressible models need a
pressure solver. The new dynamics uses an efficient solver for
non-symmetric problems, so can cover the requirement if the grid is
structured. Solvers for unstructured grids should be available from
enginerring type codes. I doubt whether advection schemes are an
issue, given that quasi-monotone upwind schemes are now in general
use. These are certainly appropriate for fine scales. Conservation is
only an over-riding issue for high speed flows, not our case.
D. Other Comments (2)
Test cases are a good point for collaboration. i
Variable horizontal grid for the UM is a key element
of the research and development plans for the MO and so there
could be no problems on that front (other than possibly time scales).
Attendees:
Alan Burns ,
Rex Britter ,
Andy Brown ,
Xiaoming Cai ,
Peter Clark ,
Terry Davies ,
Bill Dawes ,
Alan Gadian ,
Sue Gray ,
Brian Golding ,
Roy Kershaw ,
Paul Mason ,
Christiane Montavon ,
Stephen Mobbs ,
Martin Miller ,
Ian Roulstone ,
Chris Smith ,
Piotr Smolarkiewicz ,
Andrew Staniforth ,
John Thuburn ,
Tito Toro ,
Simon Vosper ,
Neils Wedi ,
Andy White ,
Nigel Wood.
Unable to attend:
Stephen Belcher , Ken Carslaw, Ian Castro, Mike Cullen , Martin Jukes,
Doug Parker , Ian Renfrew, Glenn Shutts