© Crown copyright 2007 Page 1
Physics for
‘High Resolution’ UM Configurations
Peter Clark
Met Office (Joint Centre for Mesoscale Meteorology, Reading)
© Crown copyright 2007 Page 2
Talk Outline
1.Current status
2.Microphysics
3.Turbulence
4.Q&A
5.Urban surface exchange
6.Radiation
© Crown copyright 2007 Page 3
‘High Resolution’ Aims & objectives
Prediction of individual major storms over 1-3 h timescale. Fine scale DA – nowcasting.
Useful (statistical) prediction of storm characteristics over 24 h timescales.
Organisation and triggering better than parametrization can achieve.
Improvement of forecast characteristics particularly affected by surface forcing:
rainfall, visibility and fog, extreme wind, applicability to the urban environment
Intermediate scale model (4 km) operational since early 2006. UK coverage. 3h cycle 3DVAR+Latent Heat Nudging/Moisture Observation Processing
System.
Development of a new convective scale NWP model and data assimilation configuration with a grid length of about 1km.
1.5 km ‘on-demand’ 450x450 km Dec 2006. 1.5 km UK 2009. Ensembles 2011+
© Crown copyright 2007 Page 4
Current status – 1 & 1.5 high resolution models
Substantial experience running at 1 km. Now implemented 1.5 km. 76 levels (2x38), 50 s timestep. (Will be implementing 70 level set soon).
Enhanced microphysics (see later).
Standard BL scheme+del-4 horizontal diffusion, no convection scheme. (See later)
Standard MOSES II 9-tile surface exchange (ITE 25 m land-use over GB) + anthropogenic heat source. New ‘two-tile’ urban scheme under test.
Radiation called every 5 min (6 timesteps) + radiation on slopes.
Variable resolution working and likely to be adopted for 2009 implementation.
© Crown copyright 2007 Page 5
Microphysics
© Crown copyright 2007 Page 6
High Resolution Microphysics
Operational Unified Model
Wilson and Ballard (1999)
“Cloud Resolving” Models
Diagnostic
© Crown copyright 2007 Page 7
UM Physics Status for Convective Scale
Enhanced microphysics available from UM 6.0
Bulk, single moment formulation Switches enable choices:
Single ice prognostic with diagnostic split between snow/ice.
Prognostic/diagnostic rain.
Graupel/no graupel Most evaluation done with intermediate scheme
Tested in idealised model, especially GCSS LBA diurnal cycle (Grabowski et al. 2006).
Major benefits of prog. rain for ‘seeder/feeder’ orographic enhancement.
Cloud liquidwater
Water vapour
GraupelIce
crystalsSnow
aggregatesRain
Cloud liquidwater
Water vapour
Ice crystals +
Snow aggregates
Rain
© Crown copyright 2007 Page 8
Diurnal Cycle Case Study
Data from TRMM LBA observational campaign (Rondonia, Brazil)
Initialisation from representative single profile at sunrise (07:30 am local time). Diurnally varying surface fluxes. Bicyclic model domain.
Intercomparison of CRMs (GCSS Deep Convection WG Case 4, Grabowski et al. 2006).
Focus on development of convection in first 6 hours. Observed onset of precipitation is ~10:30 (3 hours after sunrise).
Plan view of model
surface rain rate 6 hours after sunrise (1.30pm local
time).
Average rainrates
through the diurnal cycle from TRMM-LBA radar.
© Crown copyright 2007 Page 9
UM Reference
GCSS TRMM-LBA Diurnal Cycle
UM with enhanced microphysics
Timeseries of vertical profiles of hydrometeor water contents
Comparison with CRM – possibly excessive glaciation
© Crown copyright 2007 Page 10
CSIP IOP 18 – 25/08/2006
Modis Terra 1125 UTCRadar 1130 UTC
Cloud streets from coast
Squall line
1146-1148 UTC
3GHz Radar
© Crown copyright 2007 Page 11
Unified Model 1.5km Domain
•360x288 gridpoints•76 Vertical Levels•Nested in UK 4km model•Initial and LBC operational 06 UTC 12 km ‘UK Mesoscale’ •No additional DA
© Crown copyright 2007 Page 12
1.5 km L76 UM Forecast 13 UTC
Cross Section
© Crown copyright 2007 Page 13
Microphysics sensitivity 11 UTC
Control
With graupel
No rain evaporation
Wind speed Potential Temp
White contours=Cloud fraction
Dashed line=Freezing level
HeterogeneousNucleation only at T<-40C
© Crown copyright 2007 Page 14
Future plans - microphysics
Improvements to vertical transport (especially graupel).
Improved timestep dependence – especially Bergeron-Findeisen.
Two moment single ice to replace single moment ice and snow.
Minor updates to process rates.
© Crown copyright 2007 Page 15
Turbulence
© Crown copyright 2007 Page 16
Turbulence at 1 km
Current forecasting capability of UM at 1 km horizontal resolution uses ‘standard’ non-local 1D BL plus fixed (del-4) horizontal diffusion.
Works well but not perfectly. Anticipate need for 3D scheme, but highly asymmetric grid. Starting point is Smagorinsky-Lilly approach: horizontal and vertical
diffusion function of Richardson no., shear and a mixing length that scales with grid length.
Tested robustness of the UM dynamics and implementation of scheme by comparing genuine large-eddy simulation with the Met office Large-eddy model (which has been thoroughly tested at this limit).
Dry CBL Cu-capped BL (BOMEX equilibrium trade cumulus case)
Tested appropriate choice of scheme at ~1 km using idealised diurnal cycle and real cases.
© Crown copyright 2007 Page 17
Problems with initiation and shallow cumulus
MSG High Res Visible 1 km Cloud-top temperature
1 km precipitation rate
CSIP IOP 12 28/07/2005
Cirrus
Cloud streets
Radar (5km)
We have a consistent problem of precipitation from explicit ‘shallow’ cumulus.
© Crown copyright 2007 Page 18
Subgrid turbulence scheme in UM
2mSf Ri 2
h hSf Ri
jiij
j i
uuS
x x
1/ 2
2
, ,1,3
1/ 2
2ij iji j
S S S
22 20 0
1 1 1
k z z
Smagorinsky-Lilly subgrid-turbulence scheme with Richardson number (Ri) based stability factor
where
Mixing length scale
Wind shear
Stability function (unstable)
1/ 2(1 16 )mf Ri 1/ 21(1 40 )
0.7hf Ri and
0 sC x where
© Crown copyright 2007 Page 19
Dry CBL idealised model set up
Met Office Unified model in idealised mode: bi-periodic domain prescribed forcings e.g. surface fluxes and geostrophic winds
Dry convective boundary layer case: prescribed surface heat flux of 300Wm-2
initial mixed layer up to 1km with overlying stratification Domain 5kmx5kmx5km resolution 50 m in horizontal for both. Refined vertical grid near the surface
for UM. Comparison with Met Office Large-eddy model in the same configuration.
Smagorinsky model: Mixing length = Cs D where D is the horizontal grid length- significantcontrol of sub-grid dissipation. Lilly ’69 derives a value of Cs=0.17 for a homogeneous inertial sub-range. In practical large-eddy simulation Cs is adjusted in the region of this value. A value of Cs=0.23 is used in the control UM and LEM simulations.
© Crown copyright 2007 Page 20
UM/LEM comparison at 50 m resolution
W at 1 km snapshotUM Cs=0.46 UM Cs=0.23
UM Cs=0.115 LEM
•UM works at 50 m resolution•Requires Cs smaller than LEM•Cu-capped BL acceptable
•More variability•Within range of other models
© Crown copyright 2007 Page 21
UM Simulations
Reference:1D vertical non-local boundary layer scheme.Constant horizontal diffusion.
3DSL “3D” Smagorinsky-Lilly local turbulent mixing
scheme with Cs=0.23.
Series of sensitivity simulations with variations to mixing length (Cs) and combinations of the above.
© Crown copyright 2007 Page 22
Sensitivity to grid resolution (Surface rainrate)
REFERENCE
Increasing delay of first rain and overshoot with decreasing resolution
“3D” Smagorinsky scheme
reduces overshoot significantly and reduces variation of delay with res.
200m “3D” Smagorinsky scheme is close to 200m CRM (within uncertainty)
1km reference run has the first rain at the same time as the 200m UM and CRM
3DSL Cs=0.23
© Crown copyright 2007 Page 23
Increasing delay of first rain and overshoot with decreasing resolution
“3D” Smagorinsky scheme reduces overshoot significantly and reduces variation of delay with res.
200m “3D” Smagorinsky scheme is close to 200m CRM (within uncertainty)
1km reference run has the first rain at the same time as the 200m UM and CRM
Sensitivity to grid resolution (Hydrometeor Content)
REFERENCE
3DSL Cs=0.23
© Crown copyright 2007 Page 24
Impact of vertical mixing
Increased vertical mixing in the boundary layer leads to earlier convective initiation
All UM runs have constant horizontal diffusion K=1430
© Crown copyright 2007 Page 25
Impact of vertical mixing
Increased vertical mixing in the boundary layer leads to earlier convective initiation
All UM runs have constant horizontal diffusion K=1430
© Crown copyright 2007 Page 26
Impact of horizontal mixing
Increased horizontal mixing in the boundary layer leads to later convective initiation
All UM runs have the non-local boundary layer scheme in the vertical
© Crown copyright 2007 Page 27
Impact of horizontal mixing
Increased horizontal mixing in the boundary layer leads to later convective initiation
All UM runs have the non-local boundary layer scheme in the vertical.
ConstDiff Coefficient: K=1430.
Max Diff for Cs runs: K=2086.
© Crown copyright 2007 Page 28
Implications for sub-grid turbulence param.
Results are a subtle balance of horizontal mixing (delays initiation) and vertical mixing (promotes initiation).
For 1km grid resolution, the results suggest: The non-local scheme is appropriate for vertical mixing in the boundary layer.
There is a need for increased mixing of convective updraughts in the free-troposphere to reduce the overshoot. A shear/stability dependent approach is more physical than constant coefficient diffusion.
For 200m grid resolution, the results suggest: The shear/stability dependent approach of the Smagorinsky-Lilly scheme is
more appropriate than the non-local scheme.
The model is close to convergence (from earlier comparison with 100m resolution simulations).
© Crown copyright 2007 Page 29
Impact of turbulence scheme on convective forecast (CSIP IOP18 - 25th Aug 2005)
Horiz Cs=0.075 Horiz Cs=0.10 Horiz Cs=0.15
Reference Satellite IR and RadarSatellite (Visible) MODIS
© Crown copyright 2007 Page 30
Convective cell statistics (CSIP IOP18)Sensitivity to turbulence scheme
Model data is area-averaged to 5km radar grid
Reference
Radar
Cell Area (>2 mm/h)
Cell Number (>2 mm/h)
Radar
Reference
© Crown copyright 2007 Page 31
Convective cell statistics (CSIP IOP18)Sensitivity to turbulence scheme
Model data is area-averaged to 5km radar grid
Average convective cell size as a function of rainrate threshold Average number of convective cells
as a function of rainrate threshold
ReferenceRadar
Radar
Reference
© Crown copyright 2007 Page 32
Convective cell statistics (CSIP IOP18)Sensitivity to turbulence scheme
Reference run has too many, too small convective cells compared to the observations, particularly at low rain rates.
Simulations with horizontal turbulence scheme have cell sizes closer to observed, particularly as the horizontal mixing is increased (higher Cs).
Simulations with the horizontal turbulence scheme have cell numbers closer to observed, particularly at lower rain rates (<4 mm/hr) but still have too many cells with higher rain rates (> 4mm/hr).
(Note, the 8mm/hr threshold is dominated by the main organised squall line in the radar and is not representative.)
The model still does not have enough stratiform rain around convective cores.
© Crown copyright 2007 Page 33
Summary
3D sub-grid turbulent mixing parametrization introduced into the UM (based on Smagorinsky-Lilly). UM works as LES (50 m).
At ~ 1 km use hybrid approach combining the 1D non-local boundary layer scheme with aspects of the 3D scheme.
Tested in idealised and real case studies and can have a very significant impact on convective initiation and evolution.
Reduces over-prediction of small convective cells at 1.5km. Reduces excessive rain rates in larger storms.
© Crown copyright 2007 Page 34
Future plans - turbulence
‘Blended’ BL and (moist) 3D turbulence.Mixed turbulence/large
eddy behaviour in BLSmagorinsky outside (?)
Stochastic backscatter. Initially based on
Weinbrecht/MasonExtensions for shallow
Cu?
Mix
ing
leng
th
∆x
Turbulence scale
© Crown copyright 2007 Page 35
Questions & Answers
© Crown copyright 2007 Page 36
Urban Surface Exchange
© Crown copyright 2007 Page 37
Urban surface exchange in the UM
The UM uses a ‘tile’ surface exchange scheme, including an ‘urban’ tile.
The urban tile is quite crude: Enhanced roughness. Enhanced drainage. Modified albedo. Urban ‘canopy’ to
represent thermal inertia of buildings.
Anthropogenic heat source
Nocturnal heat island in 1.5 km forecast– 05/07/2006 00 UTC
© Crown copyright 2007 Page 38
Impact of anthropogenic heat flux
23 Cases
London Weather Centre Remote Rural
With AHF
No AHF
© Crown copyright 2007 Page 39
Urban Canyons
Troof
Twall1 Twall2
Tfloor
Troof
Tcanyon
Tcanyon
Tcanyon
•Negligible roof<>canyon coupling.•Single canyon temperature.•Implies two-tile simplification.
•Resistance measurements (Barlow/Harman)•Resistance model (Harman)•Radiation model and two surface simplification (Harman)•Two-tile surface only (Best) UM•Two-tile with radiation model single column UM (Harman)•Further work, full UM (Porson)
© Crown copyright 2007 Page 40
Surface-only tests
Martin Best
© Crown copyright 2007 Page 41
Next Steps
Fully integrated two-tile model in UM. Parameter provision – different approaches.
Validate in surface only model Model intercomparison.
Impact on mesoscale flow. Boundary layer development through urban/rural/urban
transition.
Revisit momentum transport (and scalar) – move away from effective roughness treatment.
© Crown copyright 2007 Page 42
Radiation
© Crown copyright 2007 Page 43
Radiation
Edwards-Slingo radiation scheme has been modified to include slope aspect and angle in direct solar radiation part. (Dominant terms, based on Oliphant et al 2003).
Significant impacts on screen temperature but very difficult o demonstrate impact on forecast.
© Crown copyright 2007 Page 44
Scientific question. Spatial impact of model changes.
Single case
Change to vertical levels
Roberts, 2007, MWR (In press)
© Crown copyright 2007 Page 45
Scientific question. Spatial impact of model changes.
Single case
Modelling radiation on slopes
© Crown copyright 2007 Page 46
Questions & Answers