15
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Probabilities from COSMO-2 derived with the neighborhood method COSMO General Meeting 18 – 21 September 2007 Athens, Greece Pirmin Kaufmann, MeteoSwiss

Probabilities from COSMO-2 derived with the neighborhood method

Embed Size (px)

DESCRIPTION

Probabilities from COSMO-2 derived with the neighborhood method. Pirmin Kaufmann, MeteoSwiss. COSMO General Meeting 18 – 21 September 2007 Athens, Greece. Neighborhood method. Probabilistic forecast from deterministic model - PowerPoint PPT Presentation

Citation preview

Page 1: Probabilities from COSMO-2 derived with the neighborhood method

Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss

Probabilities from COSMO-2 derived with the neighborhood

method

COSMO General Meeting

18 – 21 September 2007

Athens, Greece

Pirmin Kaufmann, MeteoSwiss

Page 2: Probabilities from COSMO-2 derived with the neighborhood method

2 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Neighborhood method

• Probabilistic forecast from deterministic model• Represents probability related to local-scale spatial and

temporal model uncertainty and predictability• Important especially on small scales (e.g. thunderstorms)

• Method does not represent uncertainty in synoptic forcing (complement to, not substitute for EPS)

Page 3: Probabilities from COSMO-2 derived with the neighborhood method

3 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Original shape of the neighborhood

• Ellipsoidal • Original shape of

Theis et al. (2005)• Spatial radius

decreases with increasing distance in time: Space-time dependency

x

y

t

Page 4: Probabilities from COSMO-2 derived with the neighborhood method

4 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Shape of the neighborhood

• Cylindrical• Assuming spatial and

temporal uncertainty are independent

• True in weak ambient winds, not true for strong ambient winds

• Equal weights for all grid poins

x

y

t

Page 5: Probabilities from COSMO-2 derived with the neighborhood method

5 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

0

0.5

1

-30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30

spatial radius

wei

gh

t

Linearly fading weights

• Problem: Circles around singular high model values • Idea: Smooth edges• Introduce linear fading of weights (similar to relaxation)• Adds sponge layer around cylindrical neighborhood

Large, medium, small neighborhood

Page 6: Probabilities from COSMO-2 derived with the neighborhood method

Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss

Cases from „Windbank“ Simulations1995 – 1999

Threshold 35 mm / 12 h

Page 7: Probabilities from COSMO-2 derived with the neighborhood method

7 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1995-07-12 (Case A) Temporal radius

rt=6rt=3

rt=1obs

Page 8: Probabilities from COSMO-2 derived with the neighborhood method

8 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1995-07-12 (Case A) Spatial radius

rxy=15rxy=10

rxy=5obs

Page 9: Probabilities from COSMO-2 derived with the neighborhood method

9 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1995-09-13 (Case E) Temporal radius

rt=6rt=3

rt=1obs

Page 10: Probabilities from COSMO-2 derived with the neighborhood method

10 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1995-09-13 (Case E) Spatial radius

rxy=15rxy=10

rxy=5obs

Page 11: Probabilities from COSMO-2 derived with the neighborhood method

11 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1999-10-25 (Case L)Temporal radius

rt=3

obs rt=1

rt=6

Page 12: Probabilities from COSMO-2 derived with the neighborhood method

12 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

1999-10-25 (Case L)Spatial radius

obs rxy=5

rxy=15rxy=10

Page 13: Probabilities from COSMO-2 derived with the neighborhood method

13 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Recent case: 8/9 August 2007 Flood

Page 14: Probabilities from COSMO-2 derived with the neighborhood method

14 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Objective probabilistic verification

• Verification needs to be done, but can it be done?• High density rain gauge network: Only 24 h resolution• Network with 10 min rainfall data: Only coarse spatial

resolution• Optimal radius: Quality measure (e.g. BSS) increases with

increasing radius, „an optimal neighborhood size cannot be found at all“ (experience of S. Theis; Theis 2005) • Result for 7 km COSMO resolution -> still valid at 2 km ?

• Sparseness of data: Extreme precipitation • occurs only in few cases• is often limited in spatial extent

Page 15: Probabilities from COSMO-2 derived with the neighborhood method

15 COSMO General Meeting 2007, Athens, Greece

[email protected] (presented by [email protected])

Neighborhood method

• Main influence is spatial radius rxy and fading zone rf.

• Only minor changes with temporal radius rt.

• Consequence of accumulated precipitation values• Should we treat the time dimension differently

altogether? -> Maximum over time• Current settings at MeteoSwiss (precipitation):

• COSMO-2: rxy=15, rf=15, rt=3

• COSMO-7: rxy= 5, rf= 5, rt=3

• Need of high resolution data for verification: Resolution of rain gauges insufficient, either spatial or temporal

THE END