Scores on ALADIN-FRANCE during the first quarter of 1999 :   Comments on Contingency Tables

(more details francis.pouponneau++at++meteo.fr )

Change of the computation zone for ALADIN-FRANCE score

We have calculated the score of ALADIN-FRANCE model over FRANX01 grid since 1st march 1999.

An error in TEMPERATURE score of ALADIN-FRANCE

We calculate the score of different models against SYNOP over FRANCE. We use the nearest grid point over land interpolation. We noticed in previous communication that the bias of ALADIN-FRANCE model (monthly averaged) is more important (0.5 °K) than bias of ARPEGE model.

Actually, there was a bug in the nearest grid point over land interpolation. For five stations, the nearest grid point is on sea and was involuntarily selected for the control.

We then removed this five stations and obtained new scores : the gap between ARPEGE and ALADIN-FRANCE doesn't exist.

COMMENTS ON CONTINGENCY TABLES ON FIRST 1999 QUARTER

Next tables are cloud cover and precipitation contingency tables for the 36H forecast range of ALADIN-FRANCE model (0 hours initial data set). Classes are used in the verification of NWP models in EUROPE (EWGLAM). Contingency tables are here computed over France.

Three skill scores have been calculated over the whole table :

Contingency tables for 6 hours precipitation periods are computed with classes :

In contingency tables, blue color means an underestimation, red one represents an overestimation.

NEBULOSITY:   9901-9902 (59 days),   MODEL: ALADIN,   AREA: FRANP01-France,   Range: 36 H

Obs/Forecast

0 - 2

3 - 6

7 - 8

SUM

0 - 2

9.6%

7.6%

0.3%

17.5%

3 - 6

6.9%

18.1%

5.7%

30.7%

7 - 8

3.7%

24.7%

23.4%

51.8%

SUM

20.2%

50.3%

29.4%

7686


Correct: 51.1%
 


Rousseau: 0.23
 


Heidke: 0.26
 

The percentage of correct is 51.1%. The high nebulosity class (7-8) is more often observed than forecasted : 51.8% against 29.4% . The medium nebulosity class is more often forecasted than observed ( observed in 30% of cases but forecasted in 50% of cases). Heidke and Rousseau indexes are low (0.23 et 0.26).

PRECIPITATIONS:   9901-9902 (59 days),   MODEL: ALADIN,   AREA: FRANP01-France,   Range: 36 H

Obs/Forecast

NO rain

LIGHT

MODERATE

HEAVY

SUM

NO rain

58.6%

10.0%

1.9%

0.2%

70.7%

LIGHT

7.7%

8.0%

3.2%

0.1%

19.0%

MODERATE

1.5%

3.6%

3.9%

0.4%

9.4%

HEAVY

0.2%

0.1%

0.4%

0.2%

0.9%

SUM

68.0%

21.7%

9.4%

0.9%

8883


Correct: 70.7%
 

Rousseau: 0.38
 

Heidke: 0.38
 

The percentage of correct is more than 70.7%. "Light precipitation" class has been more often forecasted (21.7%) than observed (19%) whereas "NO rain" class has been more often observed (70.7%) than forecasted (68.0%).




Home