Upload
brice-moody
View
219
Download
2
Tags:
Embed Size (px)
Citation preview
Statistical PostprocessingStatistical Postprocessingof LMof LM
Weather ParametersWeather Parameters
Ulrich DamrathVolker Renner
Susanne TheisAndreas Hense
25.09.2003COSMO - 5. Annual Meeting
Overview
- Introduction
- Short Description of Neighbourhood Method
- Verification Results
- Calibration
- Concluding Remarks
25.09.2003COSMO - 5. Annual Meeting
Assumption: LM-forecasts within a spatio-temporal neighbourhood are assumed to constitute a sample of the forecast at the central grid point
Neighbourhood Method
25.09.2003COSMO - 5. Annual Meeting
Definition of Neighbourhood II
Size of Area
Formof Area
LinearRegressions
hs
25.09.2003COSMO - 5. Annual Meeting
Products
Probabilistic Information• Quantiles• (Exceedance probabilities for certain threshold values)
„Statistically smoothed” fields• Quantiles for p = 0.5• Expectation Values
25.09.2003COSMO - 5. Annual Meeting
LM Total Precipitation [mm/h] 08.Sept.2001, 00 UTC, vv=14-15 h
Original Forecast Expectation Values
25.09.2003COSMO - 5. Annual Meeting
LM Total Precipitation [mm/h] 08.Sept.2001, 00 UTC, vv=14-15 h
Original Forecast Quantiles for p=0.9
25.09.2003COSMO - 5. Annual Meeting
LM T2m [oC] 08.Sept.2001, 00 UTC, vv=15 h
Original Forecast Quantiles for p=0.5
25.09.2003COSMO - 5. Annual Meeting
Direct model output of the LM for precipitationat a given grid point
25.09.2003COSMO - 5. Annual Meeting
VerificationDataLM forecasts; 1.-15.09.2001; 00 UTC starting time1 h values; 6-30 h forecast timeall SYNOPs available from German stationscomparison with nearest land grid point
NM-Versionssmall: 3 time levels (3 h); radius: 3s ( 20 km)medium: 3 time levels (3 h); radius: 5s ( 35 km)large: 7/3 time levels (7/3 h); radius: 7s ( 50 km)
Averagingsquare areas of different sizestemperatures adjusted with -0.65 K/(100 m)
25.09.2003COSMO - 5. Annual Meeting
RMSE(T) for VariousArea Averages
1.1
1.2
1.3
1.4
original 3x3 5x5 7x7 9x9 11x11
[K]
RMSE(T) for 3 NM-Versions(0.5-Quantiles)
1.1
1.2
1.3
1.4
small medium large
[K]
25.09.2003COSMO - 5. Annual Meeting
Reliability Diagram T
10
20
30
40
50
60
70
80
90
10 20 30 40 50 60 70 80 90
small
25.09.2003COSMO - 5. Annual Meeting
Richardson, D.S., 2001: Measures of skill and value of ensemble prediction systems, their interrelationshipand the effect of ensemble size. Q.J.R.Meteorol.Soc., Vol.127, pp. 2473-2489)
25.09.2003COSMO - 5. Annual Meeting
Reliability Diagram T
10
20
30
40
50
60
70
80
90
10 20 30 40 50 60 70 80 90
small
large
25.09.2003COSMO - 5. Annual Meeting
FBI for Various Area Averages
0.0
0.5
1.0
1.5
2.0
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
3x3
5x5
7x7
9x9
11x11
25.09.2003COSMO - 5. Annual Meeting
TSS for Various Area Averages
05
10152025303540
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
3x3
5x5
7x7
9x9
11x11
25.09.2003COSMO - 5. Annual Meeting
HSS for Various Area Averages
0
5
10
15
20
25
30
35
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
3x3
5x5
7x7
9x9
11x11
25.09.2003COSMO - 5. Annual Meeting
ETS for Various Area Averages
0
5
10
15
20
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
3x3
5x5
7x7
9x9
11x11
25.09.2003COSMO - 5. Annual Meeting
LOR for Various Area Averages
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
3x3
5x5
7x7
9x9
11x11
25.09.2003COSMO - 5. Annual Meeting
FBI for 3 NM-Versions (Exp. Values)
0.0
0.5
1.0
1.5
2.0
2.5
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
small
medium
large
11x11
25.09.2003COSMO - 5. Annual Meeting
ETS for 3 NM-Versions (Exp. Values)
0
5
10
15
20
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
small
medium
large
11x11
25.09.2003COSMO - 5. Annual Meeting
FBI for 3 NM-Versions (0.5-Quantiles)
0.0
0.5
1.0
1.5
2.0
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
small
medium
large
11x11
25.09.2003COSMO - 5. Annual Meeting
ETS for 3 NM-Versions (0.5-Quantiles)
0
5
10
15
20
25
0.1 0.2 0.5 1 2 5
Precipitation Threshold in mm/h
original
small
medium
large
11x11
25.09.2003COSMO - 5. Annual Meeting
Reliability Diagram RR(0.1-Quantile >= 0.1 mm/h)
10
20
30
40
50
60
70
80
90
10 20 30 40 50 60 70 80 90
small
medium
large
25.09.2003COSMO - 5. Annual Meeting
Reliability Diagram for 0.9-Quantiles of RR
02468
1012141618
0.1 0.2 0.5 1 2 5
Minimum Value in mm/h
small
medium
large
25.09.2003COSMO - 5. Annual Meeting
Reliability Diagram for 0.9-Quantiles of RR
0
5
10
15
20
25
30
0.1 0.2 0.5 1 2 5
Minimum Value in mm/h
small
medium
large
chance
25.09.2003COSMO - 5. Annual Meeting
Based on Richardson (2001) we derive a relation:
p' = p' (p,M,,)
M: effective sample size,related to the parameters of the beta distribution (r,s)
Calibration (addressing the effect of limited sample size only)
25.09.2003COSMO - 5. Annual Meeting
Preliminary Result of Calibration
Eliminationof and
Artificial choice:M=10
Reliability diagram of 1 h accumulated precipitation in the period 10.-24. July 2002(LM forecasts starting at 00 UTC; small version of neighbourhood method)
25.09.2003COSMO - 5. Annual Meeting
[mm/h]
Relative value of precipitation forecasts (original forecast and 90 % quantiles)for users, whose cost/lost ratio is equal to 10 % for the threshold valuesgiven by the abscissa
original forecast
smallmediumlarge
25.09.2003COSMO - 5. Annual Meeting
Concluding Remarks
“Statistically Smoothed” Fields
For temperature no mean advantage is to be seen in comparison with simple averaging
The results for precipitation are difficult to judge upon; proper choice amongst the various possibilities is still an open question
Reliability Diagrams
Possible improvement by calibration remains to be explored
The results for the lower quantiles of precipitation indicate the need for improving the model (reduce the overforecasting of slight precipitation amounts)
Higher quantiles of precipitation seem to be valuable already without calibration
25.09.2003COSMO - 5. Annual Meeting
Concluding Remarks (ctd.)
• Results of the neighbourhood method will be compared to those of more extensive experimental ensemble integrations
• Calibration of quantiles will hopefully be further developed
• The method will be tested on the 2.8 km LM version and compared to other/simpler methods. Application in combination with a small ensemble (LAF) will be investigated
• The program has been implemented at MeteoSwiss and may be made available to other COSMO members
Outlook