Vol.8 No.l Advances in Atmospheric Sciences February 1991

How much Numerical Products Affect

Weather Forecasting

Xia Jianguo (-~_~) National Meteorological Center, Beijing 100081

Received April 5, 1990; revised June 21, 1990

ABSTRACT

The paper shows how much improvement can be achieved in weather forecasting by using N W P products. And

for weather element forecasts, the types and number of N W P products highly impact on the quality o f MOS forecasts

and other utilities.

Numerical products have been extensively used in making weather forecasts issued to the public in China. NMC Beijing runs its own models to make NWP products�9 It also gets N W P products from other centers (e.g. ECMWF, U S - N M C , and JMA). These products are daily transmitted to weather forecast offices across China�9 The products arc used as the guidance to weather forecasters or for making MOS. Now NWP products have improved weather fore- casting at many forecast offices,especially for day+2 to day+5 forecasts�9

Now the question to be discussed here is that how much improvement is achieved by us- ing numerical products�9

Firstly, Fig. 1 shows that the percentage of correct forecasts for heavy rain made by MOS techniques is 7 to 12%, higher than that by man -m ad e ones. Note that these forecasts are for day +4 and for day +5. That indicates that for medium range, NWP-based forecasts outclass the ma n -ma de one.

Percentage (%) of correct forecasts = (C / /7 ) x 100

where: C is the total number of correct forecasts; F is the total number of heavy rain forecasts for the area; The threshold for the correct heavy rain forecasts is: l) rainfall: at least one station receives 25 mm or more rainfall within 24 hours in the

area; or two stations receive l0 mm or more. 2) valid time: it is allowed that heavy rain occurs one day earlier or later than valid time.

ao- �9 77 . 76 xo |

?s- ~ /

70- * 7O

iS- iS N]UI-IIt4DE

so- . l l o / /

s s - / so - . / 4 1

I S -

. . . . ,~�86 ......................................... Ol CORLlll CT ~ m JOLX AUGUIT Z911 FO~CASYB

Fig. 1. Percentage of correct forecasts for heavy rain made in Xian city in West Ch|tla, Solid

line for MOS forecast is based on products from ECMWF, dash line for m a n - m a d e oriel.

108 Advances in Atmospheric Sciences Vol.8

7a h r -

60 h r -

4 s h r -

36 h r -

24 h~'-

l Z h r -

0 h r - ~u ADV~J~CZ

i ......... I ......... I . . . . . . . . . . . . . . . . . . . .

1 9 5 0 8 1 9 0 0 8 1 9 7 0 8 s i n c e 2083

Fig. 2. Cold wave warnings with 80 % confidence of occurrence issued to the public 24 hours in

advance in 1950s, 36 hours in advance in It)60s and 1970s. 60 hours in advance since 1983.

Secondly, Fig.2 shows that the cold wave warnings issued to the public have been 60 hours in advance since 1983. That is one day earlier than in 1970s. The major reason for the

progress is that the numerical products come to forecast offices and forecasters have had the

skill to use them. For example, a cold wave hit most area of China in later November of 1987 and caused

widespread high winds (50 to 70 km per hour) and sharp drop in temperature (10-20 degrees centigrade of dropping), as well as heavy snow and freezing rain. 60 hours in advance, N M C Beijing issued the cold wave warning to the public on radio and TV. Two and half days later

the observations proved that the cold wave warning was successful. The major reason for the success is the weather forecasters employed the flow pattern forecasts both from ECMWF

and our own models. For space saving, the pictures showing the cold wave are not given here. Thirdly, a comparison has been made for the heavy rain forecasts by two different fore-

cast offices. Fig.3 shows Threat Scores(TS) given to the forecast offices. The forecast office of Xi 'an city was the winner of the better Threat Score (0.33), because the forecasters working

there are skillful in using numerical products and taking guidance of MOS. The worse TS (0.13) came to the other forecast office whose forecasters made weather forecasts mainly by

using traditional methods. Threat Score = C / (F + O - C)

C." Number of correct heavy rain forecasts.

F. Total number of heavy rain forecasts. O: Total number of observed heavy rain days.

The next topic to be discussed is whether or not the types and number of numerical products affect the quality of MOS forecasts. Fig.4. shows the comparison of the root mean

square errors (RMSE) at 500-hPa height forecasts made by models (~i" E C M W F and N M C Beijing. It was found that E C M W F model performed better than N M C Beijil~g model.

0 . 4 0 -

0 . 3 5 -

..... I .... I

a . 2 s -

o . l o -

o~ ~.5- 0 . 1 3

0 . 1 0 -

0 . 0 5 -

o . o o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

TKImt,? s c o l ~ Z l ~ q THK OTnI~R ONZ

Fig. 3. Threat Scores for heavy rain forecasts made in Xi'an city and the other one in the

sumlner of 1988.

No. 1 Xia Jianguo 109

8 0 -

7 0 -

6 0 - 5 S . 3

5 0 - " / /

40 - / 3 4 . 4

3 0 - 2 5 . 8

2 0 - 14 . g

1 0 -

7 2 . 4 IOIC B B Z J Z H G . /

/ 6 4 , 6

5 1 . 6 ~ /

3 8 . 3 ~

3~48g (m} . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D A Y + I DAY+2 DAY+3 DAY4-4 DAY+5

EglGlY

Fig. 4. Root mean square errors (RMSE, in m) at 500 hPa height forecasts covered the Northern

Hemisphere averaged over 12 months o f 1988, solid line for E C M W F model, dash line for N M C

Beijing model.

3 , . 0 0 - 2 . g 5 - 1 . 0 0 1 . 0 1 ~ BJP, B Z D

2 . 5 0 - 2 . 5 3 DBZJZMG B A 8 ~ 2 1 8

2 , 0 0 -

1 . 5 0 -

1 . 0 0 - I t S ( e ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CODZ, 8JU~BON D A Y + 2 D A T + 3 D A Y + 4 D A Y + 5

3 . 0 0 -

2 , 5 0 - 2 , 3 S 2 .311 2 . 4 5 I~ ~MIl lP ~IA fIZD

2 , 0 0 - ~ 1 , 9 2 B B Z J Z ] r G BIJUS~D 1 . I G ? ~ "

1 . 5 0 -

1 . 0 0 - I l X l t e ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ~ . . . . . .

l l ]kJUI 8 3 & 8 O N D A Y + 2 D ~ Y + 3 D A Y + 4 D A Y + S

Fig. 5. Mean absolute errors of MOS min imum temperature forecasts for cool season (above, Oct.

1988--Mar . 1989). Warm season (below, Apr. 1988--Sep. 1988). For day+2 and day+3 based on

N M C Beijing model products. For day+3 to day+5 based on E C M W F model products transmit-

ted to Beijing. All are averaged over 260 cities across China.

NMC Beijing issues MOS temperature (max. and min.) forocastsfor day+l to day+5 au- tomatically once every day. The MOS forecastsfor day+l to day+3 are based on our own model, for day+3 to day+5based on ECMWF model products we get from GTS (Global Transmission System).

Here are the verifications of MOS minimum temperature forecastsagainst observations ( see Figs.5 and 6 ).

Note that the MOS minimum temperature forecasts for day+3 are based on both ECMWF and NMC Beijing models. It was found that the ECMWF-based has larger Mean Absolute Errors (2.95~ and 2.35~ than the Beijing-based (2.53~ and 1.92~

Fig.6 shows the correlation coefficients between MOS forecasts and observations. Now the question is that why ECMWF model performs better than NMC Beijing model

while the ECMWF-MOS minimum temperature forecasts for day+3 are worse than the Beij- ing-MOS.

The answer to the question is that the model products available from ECMWF are much fewer than those from NMC Beijing. Table 1 shows the model products received from the two centers for mid- and high-latitudes.

Only three types of forecast fields for mid- and high-latitudes are available through in- ternational transmission line from ECMWF. For making MOS, many additional fields were

110 Advances in Atmospheric Sciences Vol.8

0 . 9 0 -

0 . 0 S -

0 . 8 0 -

0 . 7 8 -

0 . 7 0 -

0 . 6 8 -

O . S O - I~r

ODOL 8]ULSOM D & u

0 . 0 0 - 0 .00 . .~ .

0 . 8 5 -

0 . 8 0 -

0 . 7 5 -

0 . 7 0 -

0 . 1 5 -

0 . 0 0 - r

WAR]E 8EAHOJl DAT+2

0 . 8 0

" ~ ' 0 . 0 4 8 R Z J Z ~ ~ R D

0 . 6 4 . 0 " 6 4 - - " " ' - 0 . 8 2 B~mllP BI~SKD

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DAY+3 DAT§ ~ Y ~ S

~ 0 . 8 7 B B Z J Z ~ 8A83D

0 " 7 3 ~ 0 . 7 1 ~ 0 . 7 0 ECXlrF BA81D

DAY~3 D~Y+4 DAY+S

Fig. 6. Correlation coefficients between observations and MOS minimum temperature fore-

casts. The rest same as in Fig.5.

derived from the three, e.i. the 1000/500 thickness, the geostrophic vorticity, the U - and V-geostrophic wind components. But the ECMWF-based MOS forecasts are still less accu- rate than Beijing-based MOS. It indicates that the types and number of numerical products have a strong impact on the quality of MOS forecasts and other utilities.

Table 1. A List of Model Forecast Fields Available from Two Centers

ECMWF Beijing Beijing H 500-hPa H 500 hPa Q 0.9 Sigma P Sea Level P Sea Level U 500 hPa T 850-hPa T 0.9 Sigma U 0.7 Sigma

H 0.7 Sigma U 0.9 sigma H 0.9 Sigma V 500 hPa T 500 hPa V 0.7 Sigma T 0.7 Sigma V 0.9 sigma Q 500 hPa P Ground Q 0.7 Sigma T Sea Lc',cl

We can come to a conclusion from the fact stated above that to make good we,',~her fore- casts, meteorologists not only need advanced numerical models but also need adequate types of numerical products available. It is reasonable that NMC Beijing must develop its own me- d ium-range NWP model, even though during the comingyears the model can not perform as well as those with higher resolution and perfect physics. But it can offer much more NWP products routinely than those transmitted from most of other countries and centers.

REFERENCE

ECMWF Annual Nurllerical Weather Prediction Progress Report, NWPP Report Series 0989), WMO, No.15,

41-61.