the convective-scale unified model: results from uk case studies

© Crown copyright 2005

The convective-scale Unified Model:Results from UK case studies

Richard Forbes (JCMM, Met Office)

October 2005


Talk Outline

1. The high resolution UM in action An example UK case this summer from CSIP.

2. How are we doing ? Verifying the high resolution UM convective rainfall.

3. Improving the model Examples of recent model developments



2. How are we doing ? Verifying the high resolution UM convective rainfall.

3. Improving the model Recent model developments.

Peter Clark, Humphrey Lean


HRTM Domains

Note that operational UK 4km model uses larger (whole UK) domain


CSIP

Alan Blyth, Keith Browning, Lindsay Bennett, Karl

Beswick, Karen Bozier, Barbara Brooks, Peter Clark, Fay Davies, Wendy Garland, Charles Kilburn, Darcy Ladd,

John Marsham, Cyril Morcrette, Emily Norton,

Doug Parker, Ed Pavelin, Nigel Roberts, Ann Webb.


Salford Doppler Lidar

Aberystwyth Wind Profiler

Leeds AWS

Leeds Sodar

Reading JCMM, Forecast Centre

Met Office Radiosonde

Met Office Radiosonde

Radiosondes

UMIST CessnaChilbolton Radars and Lidar

Met Office Unified Model Forecasts

Cyril Morcrette, University of Reading


CSIP IOP 18 – 25th August 2005

0700 Water Vapour

850 hPa w, 300 hPa height

1200



Network radar – 1/2/4 km Composite

09 UTC




10 UTC




11 UTC




12 UTC




13 UTC




14 UTC

© Crown copyright 2005 Page 17


MSG High Resolution Visible


CSIP IOP 18 – 25th August 2005 Observations

Chilbolton Rainfall Rate Timeseries

Chilbolton 1.5m Temperature Timeseries

8 K drop

Peak 40 mm/hr Sferics 11Z to 13Z


A convective-scale NWP System

Animation of surface rain rates for 12km, 4km, 1km and radar from 0800 UTC to 1500 UTC on 25/08/2005

UM 12km UM 4km UM 1km Radar

300 km


CSIP IOP 18 – 25th August 2005 – 12 UTC

4 km model

10 m wind and convergence Rainfall rate



4 km 12 km

Screen Temperature



4 km

DivergenceScreen Temperature



4 km 8hr f/c Radar

Surface Rainfall Rate



Visible Sat Image

4 km 8hr f/cHigh/Med/Low Cloud


Summary – CSIP case study

• Showed high resolution UM results for one convective case study this summer (25th Aug 2005)

• Secondary generation of convective storms by cold pools is an important process that needs to be captured by the model for a good forecast.

• A 12km resolution model is poor at representing this aspect of the dynamics, but 4km and 1km models with explicit convection are able to do so.



2. How are we doing ? Verifying the high resolution UM rainfall.


Nigel Roberts, Humphrey Lean, Peter Clark


Background - What do we want to know?

1km model – should improve precipitation forecasts

In some circumstances (e.g. strong orographic forcing) small scales can be relatively predictable, but most of the time small scales are less predictable.

Can a 1-km model provide more accurate and useful forecasts of rainfall events on the scales of river catchments?

On what scales should the output be presented?


Verification approach

Verify over different spatial scales using a conceptually simple approach. Fractions/probabilities from nearest neighbouring points.

Verify against radar – good spatial coverage. Stable network over UK.

Verify accumulations - smooth out temporal noise.

Use accumulation exceedance thresholds e.g. > 4 mm, > 8 mm ….


Radar 12 km forecast 1 km forecast

0.125 0.5 1 2 4 8 16 32 mm

The problem we face

0 100 km

Six hour accumulations 10 to 16 UTC 13th May 2003


Schematic example - different scales


1-km forecast Radar

0.125 0.5 1 2 4 8 16 32 mm

Six hour accumulations 10 to 16 UTC 13th May 2003


4 mm threshold, Fractions at grid scale (1 or 0)

Model Radar> 4 mm > 4 mm

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Fraction


Model Radar> 4 mm > 4 mm

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Fraction

4 mm threshold, Fractions within 35x35 km squares


Model Radar

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Fraction



Brier score for comparing fractions

Skill score for fractions/probabilities - Fractions Skill Score (FSS)

A score for comparing fractions with fractions


Graphical behaviour of the Fractions Skill Score


Summer 2004 Trial

• Run seven cases from 2004 period (mostly convective)

• For each case run 4 forecasts at 3 hour intervals

• Run one suite with 4km, 1km assimilation and a second initialising 4km, 1km from 12km analyses.

• Forecasts out to T+7 for 1km model

• Aggregate statistics over forecasts and cases.


HRTM Domains

Note that operational UK 4km model uses larger (whole UK) domain


Assimilation Configuration

• 12km 3d-Var, MOPS/LHN

• 4km 3d-Var (scale selective), MOPS/LHN

• 1km 4km increments, MOPS/LHN

• 3 hour cycles all models


Scores for 6 hour accums 1, 4 and 16mm thresholds

Solid: AssimDotted: Spinup

Blue 12kmGreen 4kmRed 1km

1mm / 6hr threshold

4mm / 6hr threshold 16mm / 6hr threshold


Area average rain rates over 2004 summer trial


Blue 12kmGreen 4kmRed 1kmBlack Radar


Scores for 1 hour accums 1 and 4mm threshold




Summary - Verification

• Verifying high resolution precipitation forecasts on the grid scale is not always very helpful given the chaotic nature of convection.

• A skill score for an area is found to be a useful measure of rainfall forecast performance.

• Verification from seven cases during the summer of 2004 shows there is increasing skill for higher rainrates as the resolution is increased.

• Bias in the precipitation is still an issue (too much at high resolution) but this is being addressed.



2. How are we doing ? Verifying the high resolution UM rainfall.


Carol Halliwell, Richard Forbes, Peter Clark, Terry Davies, Yongming Tang


Parametrization of sub-grid turbulent mixing

Carol Halliwell, Peter Clark, Richard Forbes


Parametrization of sub-grid mixing in the UM

Existing parametrizations in UM: In the vertical

Convection scheme 1D non-local boundary layer scheme

In the horizontal First order conservative operator with constant diffusion coefficient

For high resolution, require a 3D turbulence parametrization

First order scheme may be sufficient

We have implemented a variant of Smagorinsky-Lilly subgrid model.

Eddy-viscosity and eddy-diffusivity computed from resolved strain-rate, scalar gradients and certain prescribed length scales.


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

(0 is basic mixing length)

Smagorinsky-Lilly subgrid-turbulence scheme with Richardson number based stability factor


Impact of turbulence scheme on convective forecast (4th July 2005)

Reference With Turbulence Param.

1km UM 6 hour forecast surface rainfall rate.


Impact of turbulence scheme on convective forecast (4th July 2005)

Number of cellsReference With Turbulence

Histogram of cell sizes

Average cell size

Time →

Time →


Summary – Turbulent Mixing

3D sub-grid turbulent mixing parametrization introduced into the UM (based on Smagorinsky-Lilly).

Tested in idealised and real case studies and can have a very significant impact on convective initiation and evolution.

Reduces overprediction of small convective cells at 1km. Reduces excessive rain rates in larger storms.

Work is ongoing into most appropriate formulation for different resolutions, and enhancing the scheme (e.g. stochastic backscatter).


Variable Resolution Grids

Yongming Tang, Peter Clark, Terry Davies


Variable Resolution

• An alternative approach to 1-way nesting.

• Grid varies from coarse resolution at the outer boundaries smoothly to a uniform fine resolution in the interior of the domain

• Benefits close to hires domain boundary, e.g. reduces spin-up of convection at inflow boundaries

UniformHigh Res

zoneVar-Res 2Var-Res 1

UniformCoarse Res 1

UniformCoarse Res 2

Typically, there are 3 regions, and inflation ratio R1 = R2 = 5~10%

R2R1


May 3 2002 Case - Variable Resolution Model

Rainfall at 14 UTC. The three regions of the variable resolution domain are shown


Summary – Variable Resolution

Variable resolution grid capability implemented in the UM.

Tested in idealised and real case studies with a nesting ratio of 1 : 4 and results look promising.

Currently working on the model parametrizations to make them depend appropriately on the local grid-length in different parts of the domain (e.g. grid-length dependent convection scheme).


Summary


Summary

1. 4km UM now operational for the UK (since May 2005)

2. Performed many case studies with 1km/4km models over UK as well as idealised studies (diurnal cycle of convection) and other regions of the world (Alps, Africa, NewZealand)

3. Convective rainfall is of particular interest (for flooding) and 4km/1km models show skill in forecasting higher rain rates (better than 12 km model)

4. But …there are still improvements to be made !!!

Including missing processes in the model (e.g. turbulence, radiation on slopes, microphysics, improved representation of urban areas, other surface characteristics, lakes, snow)

Different approaches to modelling (e.g. variable resolution)

Understanding high resolution processes (convective initiation – CSIP)

Data assimilation (3D/4DVAR ?)

New applications (air quality)


The End


A convective-scale NWP System

Surface rainfall rate (mm/hr) at 13:00 UTC on 04/07/2005 from the 1km UM and radar.

UM 1km UM 1km on 5km radar grid

Radar 5km

300 km


Scores for 1 hour accums 10% and 1% threshold



Threshold is rainrate for which 10% of rainy grid points are higher

Threshold is rainrate for which 1% of rainy grid points are higher



Radar

4 km

12 km

1 km

Rainfall Accumulations over 300x300 km box



Reflectivity

Doppler Radial Velocity

1227 UTC120 RHI



10 m wind and convergence Rainfall rate



2 gridlengths

Cross section through 4 km model gust front Wind speed along section


Incoming solar radiation on orographic slopes

James Manners


Surface radiation interactions

Real situation is complex:

• Sloped surfaces, shadowed regions, reflection in valleys...

• Non-isotropic diffuse radiation... …and this is just the SW.

Current 2-stream parametrization:

• SW and LW up and down.

• Direct SW incoming at angle of Sun

• Assumes flat surface.


Orographic features important to radiation

• Oliphant et. al. 2003, ‘Spatial variability of surface radiation fluxes in mountainous terrain’

• Characteristics in order of importance: slope aspect, slope angle, elevation, albedo, shading, sky view factor, leaf area index

• The most important factor is the area presented by each grid-box to the incoming direct SW radiation


Grid-box mean slope aspect and angle

Slope aspect: Slope angle:


4km Mesoscale Unified Model: 8 hr forecast

Extra direct SW surface flux: Temperature difference at 1.5 metres:


4km Mesoscale Unified Model: 16 hr forecast

Extra direct SW surface flux: Temperature difference at 1.5 metres:


Summary – Orography and Radiation

Included slope aspect and angle into the incoming direct short-wave radiation scheme.

Tested in UM with grid resolutions ranging from 60km (global) to 1km (over the southern UK).

At high resolution, small (0.5K) surface temperature changes resulting from the scheme can lead to differences in convective initiation and evolution.


Scales of Motion & Predictability

Hail shaft

Thunderstorm

Front

Extratropical

CycloneSpace Scale

Lifetime

Predictability?

10mins 1 hr 12hrs 3 days

30mins 3 hrs 36hrs 9 days

1000km

100km

10km

1km

MCS


Boscastle accumulations

the convective-scale unified model: results from uk case studies

Documents