Scores of ALADIN-FRANCE during the first semester of 2000

(more details Samuel Westrelin, Météo-France)

1.  Comments of contingency table - first semester 2000

Tables below are the cloud cover and precipitation contingency tables for 36 hours forecast range. Classes have been defined according to the EWGLAM rules.

These classes correspond to :

RR < 0.1 mm in 6 hours : no precipitations, 0.1 £ RR < 0.2 : light, 2.0 £ RR < 10 mm : moderate and RR ³ 10 mm : heavy. Units for classes intervals of cloud cover are octas.

In these contingency tables, the blue (respectively red) color means that the model forecasted a class of smaller (bigger) rank than observed, so it underestimated (overestimated) the parameter.

CLOUD COVER

The percentage of correct forecasts is only 49%. High cloud covers are much less often forecasted than observed (21% against 45%) majorly due to an overestimation of medium cloud covers. When the model predicts medium cloud cover class, in 47% of cases it is actually high one which is observed.

200001-200006

CLOUD COVER

MODEL :

ALADIN

00 H start

AREA :

FRANX01

Forecast Range

36 H

181 days

Forecast

0 - 2

3 - 6

7 - 8

SUM

0 - 2

15.4%

6.9%

0.4%

22.8%

Obs.

3 - 6

10.2%

17.9%

4.7%

32.7%

7 - 8

6.9%

21.9%

15.7%

44.5%

SUM

32.5%

46.7%

20.8%

43066

Correct :

49.0%

Rousseau :

0.23

Heidke :

0.25

PRECIPITATIONS

The percentage of correct forecasts is 74%. Globally there is a quite good agreement between the predictions of a class and the observations. A tendency to predict too often light precipitations when no precipitations have been observed appears (11% of cases). However the method of control can play a role in this sense because the observations which are representative of one location are actually compared to an average of precipitations over a mesh (what the model produces). When moderate class is forecast actually in 59% of cases no precipitations or low precipitations are observed and in 4% of cases they are heavy. For heavy precipitations, false alarm rate and non detection rate are both very high, around 80%.

200001-200006

PRECIPITATIONS

MODEL :

ALADIN

00 H start

AREA :

FRANX01

Forecast Range

36 H

181 days

Forecast

NO rain

LIGHT

MODERATE

HEAVY

SUM

NO rain

65.4%

10.8%

1.8%

0.1%

78.1%

LIGHT

5.6%

6.1%

2.5%

0.1%

14.3%

Obs.

MODERATE

1.2%

2.7%

2.8%

0.3%

7.0%

HEAVY

0.0%

0.2%

0.3%

0.1%

0.6%

SUM

72.2%

19.9%

7.3%

0.6%

45855

Correct :

74.4%

Rousseau :

0.36

Heidke :

0.36

2.  Comments of scores against SYNOP - first semester 2000

The scores (bias and root mean square error) plotted on the figures are calculated against SYNOP over the whole domain ALADIN-FRANCE. Scores are averaged over the first semester 2000, for each 6 hours forecast range.

We use about 200 synoptic stations over the domain.

MSLP

The RMS rises with forecast range from 0.7 to 2.1 hPa. The model underestimates this parameter from 12 H range till 48 H to reach -0.7 hPa.

CORRECTED TEMPERATURE

It is overestimated on afternoon (forecast ranges 12 and 36 hours) and underestimated on evenings and nights, about 0.5°.

WIND INTENSITY

Its underestimation is persistent and strongly enhanced on afternoons.

3.  Comments of precipitations maps - first semester 2000

Errors are strongly correlated to the orography : precipitations are often overestimated over the highest regions (Ardennes, North of the Alps, Massif Central, Vosges, Jura, eastern part of Pyrénées). We can also see an overestimation along the coasts from North (north west coast) to Normandie and from Britany to Vendée (west coast). Over the extreme south west, a large area of underestimation corresponds to big quantities of precipitations observed ; the model has difficulties to predict large amount of precipitatatios at a same location.

Observed 24 hours cumulated precipitations (from 06 UTC to 06 UTC) over France on first semester 2000.

Difference between 24 hours cumulated precipitations predicted by ALADIN (between 30 and 6 hours forecast ranges) and observed over France on first semester 2000.




Home