© Crown copyright Met Office Vicky Pope input from Brian Golding, Nigel Roberts, Clive Pierce, Peter Buchanan May 2012 Met Office work and future research.

© Crown copyright Met Office© Crown copyright Met Office

Vicky Pope input from Brian Golding, Nigel Roberts, Clive Pierce, Peter Buchanan

May 2012

Met Office work and future research relevant to flooding

© Crown copyright Met Office

Carlisle Flood 2005 - Observed & Forecast Accumulations

12 km

4 km 1 km

Hand analysis of gauges and radar

12 km 1 km

Model Orography


Coupling to hydrological models

Flood warning level

Flow forecasts using 1 km model rainfall is of similar quality to those using observed rain rain gauges.

Nigel Roberts, Richard Forbes (MetO) Steve Cole & Bob Moore (CEH) Wallingford Dan Boswell (EA) Northwest Met. Apps., In Press

- Observed flow- Raingauge- 1km model- 4km model- 12km model

7th Jan 2005 Caldew

9am

3pm


Insufficient ensemble size leaves gapsConstructing a probability forecast

12 member 2.2km resolution ensemble forecast


Constructing a probability forecast

12 member 2.2km resolution ensemble forecast

Probability of rain in period around the time of interest


User requirement

• High resolution ensemble precipitation forecast with a range of ~2 days for pluvial & fluvial flood prediction and warning

• ≤ 2 km resolution

• rapid update nowcast

• 50-100 members


• Forecasts in blend weighted according to skill

Extrapolation nowcast + MOGREPS-UK + noise

MOGREPS-R + noise

• Noise

Represents uncertainty in timing / location of precipitation

Adds unresolved detail to MOGREPS-R forecasts

Allows large ensembles to be produced cheaply

• Behaviour of blended forecasts

Seamless and look like radar or MOGREPS-UK forecast

Variability between ensemble members is confined to small scales initially but gradually extends to larger scales

Blending models with Nowcasts


Enhancing our rainfall prediction capability

Current

• Met office Global and Regional Ensemble Prediction System – MOGREPS:

• 60km and 18km

• UKV single forecast

• 1.5km

• Ensemble Nowcasts

New

• UK Ensemble Prediction System – MOGREPS-UK

• 2.2km 12 member ensemble -

• Blending with ensemble Nowcasts


Enhancing our rainfall prediction capability

Current

• Met office Global and Regional Ensemble Prediction System – MOGREPS:

• 60km and 18km – reliable day 2 county scale severe weather

• UKV single forecast

• 1.5km – minimal day 1 city scale pluvial forecast severe weather

• Ensemble Nowcasts – reliable short range (a few hours)

New

• UK Ensemble Prediction System – MOGREPS-UK

• 2.2km 12 member ensemble – reliable day 1 probabilistic city scale severe weather

• Blending with ensemble Nowcasts – reliable rapid update of ensemble


SEPA

Area Flood Warning Teams:•Flood Detection (local)•Flood Forecasting (local)•Flood Warning

Real time dataFEWS

Met Office

Services to Government:•Forecasts•Heavy Rainfall Warnings•Radar/forecast output

Public Weather Service:•Forecasts•NSWWS•PWS Advisors

CCA Responders – Flood Response

Flood ForecastingCentre

NSWWS:AlertWarning

Flood Warning

Forecasts and Alerts under contract

Flood Forecasting Service

•Flood Vigilance•Flood Forecasting (strategic)•Flood Detection (strategic)•Flood Advice•R&D

UKCMFR&DConsistency of productsX-border coordination

Flood GuidanceStatement


Questions


Convection-permitting 2.2 km ensemble from Spring/Summer 2012

Embedded within MOGREPS-R ensemble members (18 km)

MOGREPS-UK

36-hour forecasts

12 members

6-hour cycling (time-lag two cycles)

Downscaling – 18km initial conditions

No high-resolution initial perturbations or ‘forecast error’ perturbations to start with


Sector / Capability Enabler Option 2009 2010 2011 2012 2013 2014

Reliable Day 2 county scale forecast of severe weather

16 km 48hr NAE ensemble

C

● ● ● ● ● ● ● ● ●

Minimal Day 1 city scale pluvial forecast of severe weather

Single 1.5 km model

C

● ● ● ● ● ● ● ● ●

Reliable Day 3-4 county scale forecast of severe weather

12 km T+120 NAE ensemble

C

● ●

Reliable Day 1 probabilistic city scale forecast of severe weather risk

1.5 km (now 2.2km) ensemble

C

● ●

Hourly 12-hour nowcasts of severe weather

Hourly 1.5km nowcast

C

Reliable Day 1 probabilistic town scale forecast

800m ensemble

C

High Performance Computing upgrade benefits forecast

█ insufficient computer power, █ research, █ real-time trialling, • operational



Flood warning level

Flow forecasts using 1 km model rainfall is of similar quality to those using observed rain rain gauges.

Nigel Roberts, Richard Forbes (MetO)Steve Cole & Bob Moore (CEH) WallingfordDan Boswell (EA) NorthwestMet. Apps., In Press


55 mm 55 mm 96 mm

Highest 6-hour totals

x x x

MOGREPS-R output: 18 km model (top) UKV 6-hour accumulations:1.5 km model (bottom)

Computed on 4.5km grid – Changgui Wang Nigel Roberts

30/10/08


Convective-scale ensembles

w

950 hPa

Ottery-St-Mary storm

○ 1.5km model storms


All pixels exceeding critical thresholds 1.5km model

Computed on 4.5km grid – Changgui Wang

‘Extreme’ threshold for surface water flooding

1 in 30 years

1 in 10 years

Nigel Roberts


Ottery-St-Mary flood event Probability of getting the top 1% or 5% of rainfall amounts. Peak values exceeded extreme rainfall thresholds. Produced from 24 forecasts from the 1.5 km UKV model. X marks Ottery.


Probability of an ‘extreme’ event within 36x36km squares

Traditional clustering method found wanting – weighted sampling should be better

Nigel Roberts


How about other situations

5-6th September 2008

Morpeth flood

Northumberland

Probability of exceeding 100 mm in 18 hours within 10 miles of each pixel

Frontal, orographically enhanced rain

More predictable

Likewise – Cumbria floods

Probability of more than 100 mm of rain in 18 hours within 10 miles of any location


Morpeth flood event 5-6 Sept 2008

Probability of exceeding 50mm in 17 hours

UKV 24 members 2.2km 24 members

Neighbourhood 13.5 x 13.5 km (3x4.5km pixels)


Using the ensemble for probabilities


Maximum 3hr rainfall total within 10 and 25km from Edinburgh for each UKV2.2. member.

How many ‘extreme’ totals near to Edinburgh ?

40 mm in 3 hours - surface water flooding likely (especially urban) Halcrow report – used for Extreme Rainfall Alerts


Cartoon time

training notestraining notes

Impact

Useful forecast? Probabilities can be tricky

Communication Partnership


Case study – showers over northern France

Radar Visible satellite

NAE (12 km) T+8 4 km T+11 UKV (1.5km) T+11

8 UTC

10/11/10

Much better realism – especially showers / thunderstorms


Does higher resolution give more skilful forecasts?

Apparently not! What’s going on?

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

Marion Mittermaier


Misplaced by ~35 km

Misplaced by ~30 km

(better)

NAE 12 km compared to 4 km

4 km

NAE

1.5 km is better again

M Mittermaier, N Roberts & S Thompson submitted to Met AppsMarion Mittermaier

90th percentile

T+24 6-hour accumulation


Schwartz et al 2009 Oklahoma Univ + NOAA

2-month period

Next day forecasts hourly accumulations

- a big ask!

What about in the US?

Big storms, unhelpful topography, difficult thresholds

100 km75 km 125 km

150 km 200 km >300 km


New Met Office 12-member 2.2 km ensemble

Case study – three of the members T+24

Embedded within MOGREPS-R 18 km ensemble (Bowler et al 2008)

Giovanni Leoncini , Changgui Wang, Sarah Beare, Neill Bowler


Full ensemble - what now !


Insufficient ensemble size leaves gaps




Probability of rain in period around the time of interest


Outline

• Long term progress: Verification of surface pressure

• Recent progress: Carlisle, Cockermouth & Cornwall

• Routine rainfall verification

• The Fractions Skill Score


Long term progress in reducing forecast errors

RMS surface pressure error over the NE Atlantic



International comparisons12-month running mean RMS error of Northern

Hemisphere MSLP in Pa

100

150

200

250

300

350

400

2003

01

2004

01

2005

01

2006

01

2007

01

2008

01

2009

01

2010

01

2011

01

Met OfficeECMWFUSAFranceGermanyJapanCanada


Carlisle Flood 2005 - Observed & Forecast Accumulations

12 km

4 km 1 km

Hand analysis of gauges and radar

12 km 1 km

Model Orography



Flood warning level

Flow forecasts using 1 km model rainfall is of similar quality to those using observed rain rain gauges.Nigel Roberts, Richard Forbes

(MetO)

Steve Cole & Bob Moore (CEH) Wallingford

Dan Boswell (EA) Northwest

Met. Apps., In Press


Cumbria flood 2009: 2-day forecast Probability of exceeding 50mm in 24hrs

http://www.ffc-environment-agency.metoffice.gov.uk/


Cumbria flood 2009: Extreme Rainfall Alert to emergency services

Issued 2131 UTC 18/11/2009

http://www.ffc-environment-agency.metoffice.gov.uk/


UK4 vs UK1.5 accumulations for Cumbrian floods 2009

The Cornish Floods, November 2010: 8-hour rainfall totals

RadarUK1.5UK4



The ‘Morpeth Flood’, 06/09/2008Prototype 1.5 km forecast

12 km L50

1.5 km L70

‘Morpeth flood’06/09/2008

Provisional NCIC 3 day totals

Morpeth

0600 UTC


Probability maps for the Ottery storm, obtained by post-processing a UK1.5 ensemble allowing

for small scale positional uncertainty

UKV – all 24 members + neighbourhood approach to each

Nigel Roberts


Precipitation verification summary

• Existing scores sensitive to frequency. Significance gets worse the higher the threshold.

• New scores allow for positional uncertainty & represent skill better, but are sensitive to radar biases.

• Current scores at low thresholds match timeline document & HPC case.

• Predictability depends on precip mechanism. More frontal & orographic events in 2007-9 enhanced the scores.

• Benefits of 4km convincingly demonstrated. Benefits of 1.5km smaller, but model still being optimised.


Lead time

Model/grid QPF Capability: space/time scales represented accurately

Contribution to flood prediction

0-1hr Radar/STEPS1km/2km

Rate, accumulation & type for existing storms. 5km / 10min

Distributed small catchment models with response > ~2hrs; Lead time too short for pluvial.

1-3hr STEPS2km

Rate, accumulation, type for existing storms. Prob of exceeding threshold accum.15km/ 40min

Small catchment models > 400km2. Pluvial flooding areas.

3-12hr STEPS/UK42km/4km

Accumulation & type. Prob of exceeding threshold accum. 40km/ 2hr

Catchment models >1600 km2. Alert of pluvial / flash floods in smaller catchments.

12-36hr UK4/ MOGREPS-R4km/18km

Accumulation & type. Prob of exceeding threshold accumulations. 100km/ 6hr

Catchment models >5000 km2; early warning major river floods. Alert smaller catchment floods.

36-96hr Global UM + ECMWF EPS40km/50km

Prob of exceeding threshold accumulations. 250km/ 1day

Probability of large scale floods in EA region

4-10day ECMWF EPS50km

Prob of wet spell. 500km/ 2days England & Wales alert of large scale floods

1-4week Monthly ensemble

Accumulation anomaly quintiles. 1000km/ 1week

UK alert to possibility of wet spell

1-4mon Seasonal ensemble

Accumulation anomaly quintiles. 3000km/ 1month

UK alert to possibility of wet season


ETS T+18-24 >4mmmonthly

0.0000

0.0500

0.1000

0.1500

0.2000

0.2500

0.3000

0.3500

0.4000

0.4500

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Jan-

05

Jul-0

5

Jan-

06

Jul-0

6

Jan-

07

Jul-0

7

Jan-

08

Jul-0

8

Jan-

09

Jul-0

9

Jan-

10

Jul-1

0

Jan-

11


ETS 6hr accum > 4mm3yr accum values

0.150.160.170.180.19

0.20.210.220.230.240.250.260.270.280.29

0.30.310.320.330.340.350.360.370.380.39

0.4

Jan-

04

May

-04

Sep-0

4

Jan-

05

May

-05

Sep-0

5

Jan-

06

May

-06

Sep-0

6

Jan-

07

May

-07

Sep-0

7

Jan-

08

May

-08

Sep-0

8

Jan-

09

May

-09

Sep-0

9

Jan-

10

May

-10

Sep-1

0

Jan-

11

6

12

18

24

30

36

42

48


ETS 6hr accum > 16mmmonthly

0

0.05

0.1

0.15

0.2

0.25

Dec-0

6M

ar-0

7Ju

n-07

Sep-0

7Dec

-07

Mar

-08

Jun-

08Sep

-08

Dec-0

8M

ar-0

9Ju

n-09

Sep-0

9Dec

-09

Mar

-10

Jun-

10Sep

-10

Dec-1

0

UK4 T+3-9

NAE T+6-12

NAE T+0-6


Scale dependence of T+24 FSS 4mm threshold, NAE vs UK4


Skilful scale for >4mm in 6hrNAE vs UK412m running mean


Skilful scale for top 10% accumsNAE vs UK412m running mean


Forecasting is like a digital camera

• If a weather system is like a face:

• One pixel has one colour

• Many pixels are needed to define a face

• Many more are needed to distinguish a particular face

• The UK 1.5km grid model:

• Has one value of temperature, humidity etc. in each 1.5km grid box

• Represents recognisable weather features with 5x5 pixels (7.5km x 7.5km)

• Requires more pixels to distinguish an average from an extreme feature

• Initially, is able to distinguish average features on a scale of about 30km x 30km (“city scale”) up to about 12 hours ahead, but there is more uncertainty in the location of extreme convective storms


Forecasting is like a game of billiards

• If a weather forecast is like a game of billiards:

• A one day forecast is like hitting the white to pot the red

• A three day forecast is like hitting the white to hit another ball to pot the red

• A five day forecast is like hitting the white to hit another ball to hit another ball to pot the red

• …

• Each impact makes the end result harder to control, especially when balls are close to each other

• In meteorology, each new weather system that develops will make the forecast less predictable

• We can estimate how much by comparing several slightly different forecasts


Background• WWRP FDP Sydney 2000

• Short Term Ensemble Prediction System, 20022006Bowler, N. E., Pierce, C. E., and A. W. Seed, 2006. STEPS: A probabilistic precipitation forecasting scheme which merges an extrapolation nowcast with downscaled NWP. Q. J. R. Meteorol. Soc., 132, 2127–2155.

• Blending convective scale NWP with ensemble nowcasts, 2008-2012


Blending algorithm formulation – modelling of errors

• Radar observation errors

Z-R and VPR errors

Radar-gauge error covariance model

Errors modelled as additive perturbations to the extrapolation cascade

• Nowcast errorsLagrangian evolution of the extrapolation nowcast on cascade levels

Advection errors

• NWP forecast errorsDomain average estimate of skill on cascade levels


Achievements software

• Produces a range of seamless control and ensemble products

Extrapolation nowcast + one or more NWP precipitation forecasts e.g. UKV + MOGREPS-R

Downscaling NWP precipitation forecasts e.g. MOGREPS-R

• Makes configuration changes to NWP models transparent to users


Achievements - scientific

• Better understanding of performance characteristics of NWP models

• Two configurations of blended forecast evaluated

Extrapolation nowcast + UKV + noise

MOGREPS-R + noise

• Blended forecasts demonstrate the following capabilities

Seamless evolution

Skilful relative to components

Ensembles well calibrated

© Crown copyright Met Office Vicky Pope input from Brian Golding, Nigel Roberts, Clive Pierce, Peter Buchanan May 2012 Met Office work and future research.

Documents

ensemble nowcasts slide

met office work

ensemble members

flood detection local

member ensemble reliable

mogrepsr forecasts

pm slide

forecast accumulations