© crown copyright met office met office experiences with convection permitting models humphrey lean...
TRANSCRIPT
© Crown copyright Met Office
Met Office Experiences with Convection Permitting Models
Humphrey Lean MetOffice@Reading, Reading, UK
Nowcasting Workshop, Boulder, Oct 2011
© Crown copyright Met Office
Timeline of MO convection permitting models
• Met Office has been experimenting with convection permitting versions of UM since 2001 (NH UM version).
• UK 4km model in operational suite since April 2005.
• “On demand” 1.5km model (9 domains) from Dec 2006
• UKV 1.5km model from Nov 2009.
• Extended range UK 4km (global downscaler) from Dec 2010
• Nowcasting Demonstration system from June 2012
• Convective ensemble (2.2km) from June 2012
© Crown copyright Met Office
UKV Model
• Runs out to T+36 4 times a day.
• 3 hour 3dvar assimilation cycle with nudging of radar reflectivity.
• 1.5km over most of UK, variable res to 4km at edge of domain
• Similar configuration to low res models except:
No convection scheme
Smagorinsky turbulence
Prognostic rain
© Crown copyright Met Office
UKV Domain
744(622) x 928( 810) points
1.5x1.5
1.5x4
1.5x4
4x1.5 4x1.5
4x4
4x4
4x4
4x4
Variablezone
Inner 1.5km domain covers most of UK.Gridlength increases to 4km at edge.
© Crown copyright Met Office
UKV Model from 03UTC 19/11/2009
© Crown copyright Met Office
Carlisle Flood - Observed and Forecast Accumulations
Roberts, Forbes + EA
12 km
4 km 1 km
Hand analysis of gauges and radar
12 km 1 km
Model Orography
Why High resolution?Benefits from more detailed orography 1:Orographic Rain
© Crown copyright Met Office
Forecast visibility at 12 UTC 10/12/2003 from 18 UTC 09/12/2003
12km L38 (part domain) 1km L76
Visibility (m) at station height, synoptic observations (km)
Rachel Capon and Peter Clark
Why High resolution?Benefits from more detailed orography 2: Fog in valleys
© Crown copyright Met OfficeUrban heat islands in UKV
Heatwave temperature (00 UTC 19th July 2006)
Why High resolution?Benefits from more detailed land use: Urban Heat Islands
© Crown copyright Met Office
However biggest benefit is expected to be in representation of Convection.
• Explicit convection means losing problems associated with parameterisation at these gridlengths.
• Also represent related features which are important (convergence lines etc).
© Crown copyright Met Office
3rd May 2002Scattered convection case
History 1: No convection scheme
Nigel Roberts
© Crown copyright Met Office
History 2: Smagorinsky turbulence
• At early stage in research into km scale models found that horizontal diffusion was needed to reduce gridscale structure.
• Also discovered that applying too much uniform horizontal diffusion had detrimental effect on convective initiation (delay).
• This was motivation for using Smagorinsky turbulence (only apply once shear built up).
• Currently UKV uses 2D Smagorinsky with BL mixing in vertical.
© Crown copyright Met Office
Example of rainfall forecastsSquall line southern England
14 UTC 1st July 2003T+7 forecast
12km 4km
1km Radar
© Crown copyright Met Office
Rainfall Accumulations 12-18 UTC 16th August 2004
12 km 4 km NIMROD radarForecasts from 03 UTC
Peak Accumulations >60mmOn 4 km gridPositional error and false alarm
Boscastle Flood
Peter Clark
© Crown copyright Met Office
Snow Showers penetrating inland
• Well known problem with parameterised convection is showers not penetrating far enough inland.
25th Nov 2010 Snow showers coming in on NE wind gave significant snow in NEEngland
© Crown copyright Met Office
Snow Showers penetrating inland24 hour precip accumulation (mm) 25th Nov 2010
1km radar UKV (1.5km) NAE (12km)Operational models
© Crown copyright Met Office
How about objective verification?
Need to take care with standard gridpoint scores!
April to Oct 2010
Equitable Threat Score (ETS)
Using gauges
M Mittermaier, N Roberts & S Thompson submitted to Met Apps
UKV 1.5 km
UK 4 km
NAE 12 km
Global ~25 km
© Crown copyright Met Office
Predictability Issues
• Use of 1.5km model does NOT automatically mean that we can issue forecasts with 1.5km accuracy.
• Small scales less predictable so individual showers not predictable more than a few hours ahead (unless driven by larger scale feature such as orography or convergence line).
• Consequences for: 1. Sensitivity testing of convective scale systems except in extreme cases one case is meaningless 2. Verification of models scale selective techniques 3. Interpretation/presentation of forecasts Avoid presenting unpredicable information. Move to probabalistic presentation.
© Crown copyright Met Office
Skill depends on the scale you look at
Nigel RobertsNigel RobertsRoberts and Lean MWR 2008
© Crown copyright Met Office
Summary FSS scoresUKV vs NAE
Forecast
Ranges
Percentage of times UKV better minus percentageof times its worse.Background colour gives indication of statistical significance(green >95%).
Marion Mittermaier and Matthew Trueman
© Crown copyright Met Office
Summary FSS scoresUKV vs UK4
Forecast
Ranges
Percentage of times UKV better minus percentageof times its worse.Background colour gives indication of statistical significance(Green >95%).
Marion Mittermaier and Matthew Trueman
© Crown copyright Met Office
Problems with representation of convection• UKV does improve on representation of
convection in lower resolution models with convective parameterisations.
• However problems remain:
Peak rain rates often too great.
w too large (up to 15m/s in UK)
Cell properties very dependent on mixing settings.
Too much gridscale structure (esp in w).
• Know in principle that deep convection very under-resolved at 1.5km.
• Evidence that behaviour at 1.5km still dominated by gridlength (convection “permitting” rather than “resolving”)
© Crown copyright Met Office
Gridscale structure in 750 hPa w13UTC 12/05/2010
4km 1.5km 500mEmilie Carter
© Crown copyright Met Office
Compare gridlengths down to 100m
4km 1.5km
500m 100m Radar (1km)
Emilie Carter
Features continue to get smaller
12 UTC from 06 UTC run 7th Aug 2011
Areas shown are 80x80km(whole domain of 100m)
© Crown copyright Met Office
Need to find out how to do best wecan at 1.5km.
• Mixing (horizontal and vertical)
• Shallow convection scheme
• Standard convection scheme (as in UK4)?
• Stochastic backscatter
• Microphysics
• Other?
© Crown copyright Met Office
Effect of vertical mixing at 1.5km
Radar Control Smag+Vert mixing
• Vertical Mixing has big effect on no of cells
© Crown copyright Met Office
Need to constrain model set up with observations
• DYnamical and Microphysical Evolution of Convective Storms (DYMECS). Hogan et. al. U of ReadingTrack cells with Chilbolton research radar and build up statistics of properties of convection. (runs from now through summer 2012).
• COnvective Precipiatation Experiment (COPE).Blythe et. al. with Met Office.Development of convection in SW England. (Peninsular convergence lines)(field programme summer 2013).
For both of these will compare to UM at gridlengths between 100m and 4km.
© Crown copyright Met Office
Rainfall (mm/hr)
Rainfall (mm/hr)
Reflectivity (dBZ)
Reflectivity (dBZ)
Chilbolton RadarRadar Composite PPI Composite
Grey: 5dBZ isosurfaceRed: 30dBZ isosurface
Thorwald Stein
© Crown copyright Met Office
Conclusions
• Met Office has gathered much experience with convection permitting versions of the UM.
• Current models are used for longer time ranges than nowcasting.
• Many benefits seen for representation of convection and other phenomena
• Problems still remain with representation of convection which are being addressed
• Need to consider predictability issues for verification and interpretation of forecasts.
© Crown copyright Met Office
Thank you for listening.Any questions?