Met Office Unified Model

© Crown copyright 2006 COSMO, 18-21 September 2006, Bucharest Page 1

Met Office Unified Model

Terry Davies

Dynamics Research


UM Dynamical Core

Non-hydrostatic formulation Hybrid-height co-ordinate Semi-Lagrangian advectionSemi-implicit predictor-corrector integrationC grid horizontal staggeringCharney- Phillips vertical staggering

Davies, Cullen, Malcolm, Mawson, Staniforth

White, Wood 2005 Quart. Journal Roy. Met. Soc


Vertical and Horizontal Grid Staggering

Horizontal staggering - Awakawa c-grid

No grid decoupling

Better geostrophic adjustment for wavelengths of grid size less than Rossby radius of deformation

Vertical staggering - Charney-Phillips

No computational modes

More consistent with thermal wind balance

Can have complications in coupling with boundary layer parametrization

W,

U,ρ

W,

Ui- 2

1 ,j

2

1Vi,j+

2

1Vi,j-

Ui+

2

1 ,ji,j


Physical Parametrizations

Edwards-Slingo Radiation (Edwards & Slingo 1996)

Mixed phase precipitation (Wilson & Ballard 1999)

New Boundary Layer + 38L (Lock et al 2000)

New GWD scheme + GLOBE orography

smoothed (Raymond filter)

Modern spectral database for gaseous absorption in the atmosphere + new H2O continuum -

flexible configuration Multiple scattering included Better optical properties for

clouds inc. non-spherical ice parametrisation








Physically based transitions between vapour, liquid, ice and rain

Ice content now a prognostic variable rather than diagnosed from cloud scheme








Allows for non-local mixing in unstable regimes

Scheme diagnoses 6 different mixing regimes in order to represent stable, well mixed and cumulus processes

Scheme includes boundary layer top entrainment parametrisation

Improved interaction with the convection scheme








Simplified scheme Expression for linear 2D flow

used to calculate total surface pressure drag

Gravity wave amplitudes proportional to depth of sub-grid mountains above blocked layer

Remainder of drag is attributed to flow blocking


Future Developments

Dynamical core improvements

More consistent treatment of moisture

Conserving semi-Lagrangian advection scheme

Variable resolution grid

Resolution increases - (70 levels, 40km)

New physical Parametrisations

New prognostic cloud scheme

New convection scheme


UM Operational Configurations

Global 40 kmN320L50640x481x50 63 km top150 million numbers

North Atlantic & European 12 km720x432x38 38 km top120 million numbers

Old UK 12 kmRetiring eventually

New UK 4 km288x320x38 38 km top35 million numbers


MOGREPSMet Office Global and Regional Ensemble

Prediction System

Ken Mylne

Ensemble Forecasting Manager


ECMWF Ensemble (EPS)

51 members Control (unperturbed) + 25 pairs formed by adding

and subtracting a perturbation TL255 Resolution (approx 80km) Designed for use beyond 48h Perturbations are linear combinations of

Forward and Evolved Singular Vectors Includes Stochastic Perturbations to model

physics


MOGREPS – The Met Office short-range ensemble

NAE

MOGREPS is on Operational Trial for 1 year from September 2005

Ensemble designed for short-range forecastingGlobal ensemble (~90km resolution, 38 levels)ETKF used within global ensemble to determine initial condition pertsRegional ensemble over N. Atlantic and Europe (NAE) at 24km resolution, 38 levels. Nested within global ensemble for initial and lateral boundary conditionsStochastic physicsT+72 global, T+36 regionalGlobal run at 0Z and 12Z. Regional run at 6Z & 18Z


Ensemble Creation – Analysis Perturbations

To achieve these desirable properties: Not sufficient to sample randomly

models have ~107 degrees of freedom - too many slow-growing directions!

Look for rapidly growing perturbations Singular vectors (ECMWF)

Error breeding (NCEP) New! Ensemble Transform Kalman Filter (Met

Office) Thanks to Craig Bishop and colleagues.


Singular Vectors (SVs)

SVs use linear adjoint of ECMWF model to identify fastest-growing directions in phase-space over the next 48 hours. SV perturbations scaled by forecast error statistics

at 48h - fast-growing so very small at initial time Perturbations also include Evolved SVs from

48h previously identify areas of greatest analysis uncertainty

(where model background is likely to be in error)


SV Perturbations

Each perturbation is a linear combination of: 25 NHem SVs 25 SHem SVs 25 Tropical

moist SVs targetted on Caribbean TCs


Error Breeding

Start with random perturbation - allow to grow in forecast Rescale bred mode to analysis errors (fixed climatological rescaling factor)

Use for perturbation in next cycle

Cycle “breeds” the rapidly growing modes in the analysis cycle

Toth and Kalnay (1997), MWR 125, 3297-3319


Ensemble Transform Kalman Filter (ETKF)

T+12 perturbed forecast

T+12 ensemble mean forecast

( - ) + =

( - ) + =

( - ) + =

( - ) + =

( - ) + =Transform matrix

Control analysis

Perturbed analysis

0.9 Pert 1-0.1 Pert 2-0.1 Pert 3-0.1 Pert 4-0.1 Pert 5


Perturbation Structures – Mean and spread PMSL

Spread tends to be concentrated around fronts and sharp gradients

Perturbation is non-zero everywhere (in contrast to SVs)


Stochastic physics

…. the quest to increase spread!

Buizza et al., MWR, 2004

All three systems are

under-dispersive!!


MOGREPS employs three schemes to address different sources of model error: Random Parameters (RP)

Error due to approximations in parameterisation

Stochastic Convective Vorticity (SCV) Unresolved impact of organised convection (MCSs)

Stochastic Kinetic Energy Backscatter (SKEB) Excess dissipation of energy at small scales

Impact is propagated to next cycle through the ETKF

Stochastic physics in MOGREPS


Parameter Scheme min/std/MaxEntraiment rate CONVECTION 2 / 3 / 5

Cape timescale CONVECTION 30 / 30 / 120

Rhcrit LRG. S. CLOUD 0.6 / 0.8 / 0.9

Cloud to rain (land) LRG. S. CLOUD 1E-4/8E-4/1E-3

Cloud to rain (sea) LRG. S. CLOUD 5E-5/2E-4/5E-4

Ice fall LRG. S. CLOUD 17 / 25.2 / 33

Flux profile param. BOUNDARY L. 5 / 10 / 20

Neutral mixing length

BOUNDARY L. 0.05 / 0.15 / 0.5

Gravity wave const. GRAVITY W.D. 1E-4/7E-4/7.5E-4

Froude number GRAVITY W.D. 2 / 2 / 4

The Random Parameters

Stochastic scheme for the UM


RP+SCV in MOGREPS

Page 25

SKEB

Stochastic Kinetic Energy Backscatter (SKEB) Based on original idea and previous work by Shutts (2004) Related to new scheme for ECMWF EPS

Aim: To backscatter (stochastically) into the forecast model some of the energy excessively dissipated by it at scales near the truncation limit

In the case of the UM, a total dissipation of 0.75 Wm-2 has been estimated from the Semi-lagrangian and Horizontal diffusion schemes. (Dissipation from Physics to be added later on)

Each member of the ensemble is perturbed by a different realization of this backscatter forcing

Page 26

SKEB

Streamfunction forcing:1

( , )2

DF K R

K.- Kinetic En.; R.- Random field;

D.- Dissipated en. in a time-step

R is designed to reproduce some statistical properties found with CRMs

Largest at the jets/storm track

Example: u increments at H500

Page 27

SKEB

Preliminary results: Positive increase in spread (comparable to that seen at ECMWF)

SKEB

RP+SCV

Increase in spread respect to an IC-only ensemble

500 hPa geopotential height



Products for the Risk Manager

Plot of ensemble spread Range of uncertainty

0%

100%

Prob

Probability graph for multiple severity thresholds

Example of use for risk management in offshore oil industry


2006-2009 plans

Forecast uncertainty information derived from EPS (July 2007)

Report on public understanding of probabilistic forecast information based on experiments at Exeter University (July 2007)

Ensemble surge prediction system trials (October 2007)

Report on predicting extreme deviations from ensemble mean using singular vector perturbations (March 2008)


2006-2009 plans

Probabilistic short-range first-guess warning system for severe weather (March 2008)

Verification report on first-guess warnings (March 2009)

50 km global ensemble, 12km regional ensemble (November 2009)

Verification report on the enhanced resolution ensembles (March 2010)

Report on potential benefit of a convection-resolving EPS (March 2010)


HIGH RESOLUTION DATA ASSIMILATION

Sue Ballard

Z. Li, M. Dixon, S.Swarbrick, O.Stiller and H. Lean

Met Office, JCMM, Reading University


Contents

Aim of high resolution convective scale system Prediction of flood risk, replace nowcasting system Detailed local weather 4km UK 2005, 1.5km ~ 2008-2010

Trial system – small domains

Data assimilation options

Assimilation issues Impact of data assimilation Impact of relative humidity and latent heat nudging 4DVar of cloud and precipitation Assimilation of radar doppler radial winds

Conclusions

Exploitation of high resolution observations


High Resolution Trial Model

1 km76 levelsResolved convection

4 km38 levelsMass-limited convection


Original HRTM Assimilation Options

12 km 3D-Var Data Assimilation With or without moisture and Latent Heat Nudging (LHN)

using AC scheme (referred to as MOPS data – moisture observation processing system)

i.e. spin up 4km, 1km from 12km T+1 each cycle.

4km 3D-VAR with continuous cycles with or without MOPS 1km with nudged reconfigured 4km increments using IAUWith or without LHN and moisture nudging using AC scheme

IAU – increments output from 3D-Var and fixed over time window

AC scheme – increments depend on latest model fields so vary with timestep through weighting factor and model evolution/impact of data


Plans - ongoing

4km MOPS - Hourly cloud, 15min precipitation, filtering, weights Background errors – lagged/unlagged, lengthscales Operational doppler radar winds – superobbing, errors, monitoring

Salford Univ and COST 731

Observations – Satellite Applications + Radar Group + Obs Radar reflectivity – observation operators, compare model and obs

Reading Univ Geostationary imagery – low level moisture, cloud top Radar refractivity – low level moisture (Reading Univ) Wind profiler humidity, ground based radiometer, cloud radar

Development of 3D-Var and 4D-Var for direct assimilation of cloud and precipitation

3D-Var MOPS cloud cover, precipitation rate – currently not resourced Cloudy radiance, PF physics, infrastructure


Moisture Observation Preprocessing

Resolution: 15km, 3 hours

(Testing 1 hour)

Surface reports

Satellite data Radar data

3D Cloud fraction

3D Relative humidity

Nudge model state

Precipitation5km smoothed to 15km HourlyTesting 15min


T-3 T-1 T+0 T+1 T+3

AC scheme/UM3D cloud fractionSurface rainrate

IAU3 hour f/c: background Hourly ModelOb

Next analysisPrevious analysisNudging RH & Latent heat

T+2T-2

Conventionalobservations

3D-Var (FGAT)

Obs window

3D-Var system including MOPS RH and LH nudging via AC scheme

6hr accumulations from13Z to 19Z 16/8/04 from 12UTC analysis

Rain rates at 14.30 UTC from 12UTC analysis

With 4km 3D-Var +MOPS radarSpun-up from 12km T+1


Impact of cloud and precipitation data

Radar1 hour accumulation

T+2 forecast 15min precip and hrly cloud

T+2 forecast No MOPS data

14UTC 25 August 2005 – CSIP IOP 18


Development of 4D-Var

4D-Var operational in global and 12km NAE models3D-Var operational in 12 and 4km UK models3D-Var being set-up/run for 1.5km model

4D-Var NAE has MOPS RH and latent heat nudging in outer loop

Developing direct assimilation of surface precipitation rates (accumulations?)Cloudy radiancesCloud top pressure

Starting to set up research 4D-Var at 4km resolution


Trial Results - NAE summer rainfall

t+9 3DVAR

t+9 4DVAR radar


Exploitation of high resolution data

Radar radial doppler winds

Initial development using Chilbolton research radar – single elevation

Now winds available from 2-4 radars in operational network – multiple elevations

Need to modify background errors to exploit high resolution information


Impact of S-band radial radar wind data- radial wind on 1deg scan elevation

12km Back-ground

Super-obbedRadarDopplerwind

½ Length scaleAnalysis

Reduced background wt


Impact of S-band radial radar wind data- radial wind on scan elevation

4km Back-ground

Super-obbedRadarDopplerwind

½ Length scaleAnalysis

Reduced background wt


Need to combine synoptic scale and high resolution analysis

3D-Var analysis in small area cannot capture synoptic scales

Need to somehow get synoptic scale information from larger area coarser resolution analysis

Can analyse different scales in different areasHowever won’t always have an up-to-date coarser resolution analysis eg 6 hourly with high resolution hourly

Needs further research


theta inc

12km analysis

“new” 4km analysis

/12km back

“new” 4km analysis

/4km back & half length

standard 4km analysis doesn’t see obs outside domain

Zero at boundary

4km short wave analysis 12km long waves

Problem of small domain Scale selective analysis


Conclusions

Need 1-4km to capture small scale severe events Need DA to overcome spin-up of explicit convection MOPS RH & LHN improves locations

Amounts sensitive to model formulation, weights, frequency of data

Ideally want to use assimilation of cloud and precipitation in 3D-Var and 4D-Var

Need to combine synoptic scale and convective analysis

Need improved background error covariances Need to move towards full analysis for 1km model


Plans - Future

Aim: 1hrly forecasts 0-6hours 1-1.5km NOWCASTS also 36 hour UK forecasts ?

Need to move towards full analysis for 1km model Ideally want 4D-Var and Ensembles - Start with 3D-Var +MOPS

need to build on experience with 4km and NAE Use reduced vertical resolution (and horizontal?) for

analysis? High resolution data Balance and background errors Variable resolution model Surface analysis – SST and soil mositure


Other Collaborations

DARC (Reading University) PhD – combining small scale and large scale in LAM

analysis Post-doc/lecturer – surface friction in control variable Post-doc/lecturer – non-linear evolution of Gaussian pdfs Post-doc – balance at convective scale Post-doc – wavelet transforms Post-doc and PhD – Ensemble Kalman Filter

EPSRC project – Peter Clark, Surrey, Aston FREE (Flood Risk from Extreme Events) – Reading

University, Met Office and CEH, Wallingford – assimilation of clear air doppler winds and humidity from refractivity


Current/planned data sources include:

Surface observations : now SYNOPs hourly, 69 stations every 10mins, full network every 10 mins by 2006-2009

AMDAR : now every 3 hours take off and landing can request hourly at extra cost

Geostationary imagery: every 15mins

Radar VAD profile: every hour user requirement, every 15mins potentially Radar radial doppler winds: every 15mins, 5 elevations range 125km now 2 radars

Rainfall rate analysis : now every 15minutes 12 radars potentially every 5 mins

Radar reflectivity : 12 radars every 5 mins , 5 elevations range 255km

Cloud cover analysis: every hour (potentially every 15mins)

Wind profiler: 5 sites every 30mins (poss 15mins)

GPS: every 15mins 70-150 sites in GB end 2006

Integrated system – wind profiler, microwave radiometer (1+2), cloud radar, ceilometer, GPS


Radar Network Coverage – 2006

1km resolution

2km

5km


2006-2009 plans

Deliver capability to assimilate operational Doppler radar winds (March 2006)

Higher frequency cycles at 1km resolution (March 2007)

Initialise high resolution model using high resolution observations but remaining consistent with large-scale model at synoptic scales (December 2007)

Enhancements to convective scale data assimilation (December 2008)

Met Office Unified Model

Documents

dvar mopsneed

dvar mopsradarspun

high resolution databalance

high resolution hourlyneeds

vertical resolution

models3dvar operational

uk models3dvar

analysis doesnt