Saltzman B. (editor) Anomalous Atmospheric Flows and Blocking

Подождите немного. Документ загружается.

314

SIMMONS

the

0.6

correlation coefficient reached on average about

day later in the

forecast period over Europe than over North America. However, the extent

of these differences is not confirmed by examination of another measure of

skill, the standard deviation of forecast error, normalized by persistence, for

the 850-mbar wind.

definitive specification of regional differences in fore-

cast accuracy over the extratropical Northern Hemisphere is thus difficult to

give on the basis of such objective verification.

Forecasting for the tropics and immediate subtropics undoubtedly poses

particular problems. There is not only in general a severe deficiency in data

coverage, but also strong sensitivity to aspects of the data assimilation and

parameterization schemes of the forecasting system. In addition, the strong

persistence of some regional circulations emphasizes difficulties.

an ex-

ample, the upper panel of Fig.

shows the absolute correlation of the 850-

mbar vector wind for a region covering India and part of Southeast Asia for

July 1983. Overall. the numerical forecasts barely improve on a persistence

forecast for the summer monsoon flow, although individual cases of quite

accurate forecasts

transient behavior over several days can

found. The

verification for Northwest Africa also shown in Fig.

indicates a much better

performance relative to persistence, although the absolute correlation of the

forecast evolves similarly for the two regions. Some further discussion of

tropical forecast accuracy is given in Section

The accuracy of forecasts for a given month

season can vary substan-

tially from year to year even on a hemispheric domain. Extreme cases for the

extratropical Northern Hemisphere, comparing anomaly correlations for

January 1982 and 1983 and for July 1983 and 1984, are presented in Fig. 5.

Although changes in the forecasting system over the two intervening I-year

periods may have contributed to the differences, particularly between 1983

and 1984, the tests of the various changes that were made indicate that most

of the differences seen in Fig.

are not due to development of the model or

data assimilation. Rather, differences

predictability of up to

days at the

0.6

level of anomaly correlation appear to be mostly a consequence of

differences in the circulation patterns from month to month. The extent to

which this is due to a fundamentally higher predictability, due to a greater

sensitivity to data coverage

analysis techniques,

due to a greater sensi-

tivity to systematic model deficiencies,

certain synoptic situations

un-

clear.

The

results shown in Fig. 5 are confirmed by corresponding plots

the

standard deviation of forecast error. However, it is also found that standard

deviations for persistence forecasts show some of the same variability. In Fig.

we plot a measure of forecast accuracy based on the standard deviation of

the model forecast normafized by the standard deviation of persistence, for

the same months as in Fig. 5.

For

the two Januaries there appears little

NUMERICAL PREDICTION

315

1.0

0.8

0.6

0.4

OPERATIONA-

FORECAST

INDIA

o.21---.--

N.W.

AFRICA

Days

FIG.

Monthly means of the absolute correlation of 850-mbar vector wind for July 1983 as

functions of forecast range. Results are shown for both operational numerical forecasts and

persistence forecasts for two regions (6"-33"N, 72"-102"E, a; 12"-36"N, 2I0W-12"E, b)

spanning the tropics and subtropics.

316

SIMMONS

Days

FIG.

Monthly-mean anomaly correlations

height

for

1OOO-

200-mbar

levels and the

extratropical

Northern Hemisphere

functions

of forecast

range

for

January

1982

and

1983,

and July

1983

and

1984.

0.2

8 9

Days

FIG.

Fig.

but

for

a measure of forecast accuracy based

standard deviation normal-

ized by persistence.

'\.<

JANUARY

I I

318

SIMMONS

difference in forecast quality out to day 4, according to this measure, al-

though the superiority of 1983 still becomes evident in the second half of the

forecast range. July 1984 appearsclearly better than July 1983 beyond day

Thus although differences are not quite

pronounced for a measure nor-

malized by the skill of a persistence forecast, they are nevertheless substantial

in the medium range.

THE

PREDICTION

BLOCKING

AND

CUTOFF

Lows

Spells of above- and below-average predictability are also found within a

monthly time scale. Examples have been discussed by Bengtsson and Sim-

mons

(

1983), and here we present a more recent case, November 1983, as

basis for choice of specific examples

forecasts involving blocking and the

formation of cutoff lows.

Figure 7 shows height anomaly correlations for 3-, 7-, and

0-day forecasts

performed from initial dates within November 1983. Particularly evident in

the 7-day forecasts is a slow variation in forecast skill over the course of the

month, with below-average performance at the beginning of the month and

in the final week, and an accurate spell during the second and third weeks.

This variation is discernible in the 3-day forecasts, and more

at day

10.

1.0

I.//I.l,,ll/ll

-021

Initial

date

forecast

November

1983

FIG.

Anomaly correlations of height for

1000-200

mbar and the extratropical

Northern

Hemisphere

for

3-,

7-,

and

IMay

forecasts performed from initial dates within the month

November

1983.

NUMERICAL PREDICTION

319

Small daily variations in the accuracy of individual forecasts at

day

are seen

to be amplified at

day

10.

illustrate the range of accuracy

large-scale forecasts within this

month, 5-day mean maps of 500-mbar height from analyses and two fore-

casts are shown in Fig. 8. The forecast from

November shown in the

left-hand column is classified as the most accurate of the month according to

the anomaly correlations at days

and 10 shown in Fig.

The right-hand

forecast plot shows the result from the forecast from 26 November, which

ranks as the poorest on the hemispheric scale.

The forecast from

November is successful in its representation of most

large-scale features present over the final 5 days of the forecast period. It has

captured the retrogression of the block initially located over western Europe,

the enhancement of the closed low to the northeast, and the decay of the low

over northern Canada. Conversely, major error is seen in the forecast from

26 November. In particular, its evolution has produced a predominant zonal

wavenumber 2 pattern in the high-latitude height field rather than the strong

wavenumber

pattern which occurred in reality. The forecast is successful in

its formation of a cutoff low in the European sector, but for practical appli-

cation suffers from an important error in the position of this low, and in the

eastward extension of the zonal jet over the Atlantic.

It is of interest from the viewpoint of systematic model errors (to be

discussed further in Section 8) to examine in more detail the development of

the latter features of the forecast from 26 November. Figure 9 shows the

initial 500-mbar height analysis for the European/Atlantic region, together

with the corresponding maps for the 5-day forecast and its verifying analysis.

At the 5-day range a pronounced ridge extends over western Europe, and the

cutoff low has just formed to the southeast in both the forecast and reality,

although it is already seen to be located further east in the forecast. To the

west, the forecast exhibits a characteristic error in that the trough over the

eastern Atlantic has a phase lag at southern latitudes, and fails to produce a

weak cutoff west of the Iberian Peninsula.

Figure 10 shows the 6-day forecast and verifying analysis. As in idealized

studies of mature baroclinic waves (e.g., Simmons and Hoskins, 1978) the

tilted eastern Atlantic trough in the forecast has decayed and enhanced

westerly flow in this sector, whereas in reality a weak cutoff remains, with

high pressure over southern England. In addition, the major Mediterranean

cutoff has drifted slightly eastward in the forecast, rather than becoming

established over the south of Italy.

Grernaas (1983) has carried out an extensive study of forecasts for the

European/Atlantic area for the years 1980 and 198

He identified 20 spells

lasting

days or longer in which the anomaly correlation coefficient

500-mbar height for the European area indicated above- or below-average

ANALYSIS

8-13

NOV

1W.W

'ipi

ANALYSIS

16.21

NOV

FORECAST 16-21 NOV

ANALYS!&23-28 NO!=+

,so+

NALYSIS 1-5

DEC

FORECAST 1-5

DEC

FIG.

Mean

500-mbarheightanalysesfortheperiods8-13

November(upperleft), 16-21

November (middle

left),

23-28 November

(upper

right),

and

1-5

December (middle

right).

Corresponding means

for

November and

December are also shown

for

the forecasts

from

and 26 November,

respectively.

The contour interval

dam.

320

FIG.

The 500-mbar height analysis for 12

GMT

26 November (upper) and

December

(middle), and the 500-mbar height field forecast for

December (lower) from initial conditions

for 26 November. The contour interval is 6 dam.

322

SIMMONS

FIG.

10.

The

500-mbar

height field

for

December

(upper)

and the corresponding

6-day

forecast

(lower) from 26 November.

forecast quality. These spells were typed synoptically, and it was discovered

that the best scores were found for situations characterized by some form of

blocking, with poorest results in prevailing zonal situations. Mean scores for

day

in blocking spells were comparable with those for day

in cases of

zonal flow.

More specifically, Gr~rnaas

(

1983)

reports above-average forecast scores

when persistent large-scale synoptic features such as blocking and cutoff lows

are present in the initial analyses,

predicted within the first

days of the

forecast. The poorer performance in zonal situations reflects not only phase

errors of traveling disturbances, but

also

errors (which have a systematic

element) in the cyclone tracks and in the rate of filling of cyclones. The

NUMERICAL

PREDICTION

323

systematic errors appear to inhibit the formation of blocking later in the

forecast period and give rise to a tendency for the cyclonic activity on

the western side of a ridge or block to break down that feature. An indication

of this has already been discussed with respect to the forecast from 26 No-

vember 1983.

Situations in which the development

blocking is accurately predicted

may also be used to examine the mechanisms and interactions involved, and

the features of the forecasting system which are of crucial importance. This

can be achieved by controlled numerical experimentation, and case studies

examining a range of sensitivities to such features as orography, model

resolution, and parameterizations have been reported by Bengtsson (198 l),

Ji and Tibaldi (1983), Tibaldi and Buzzi (1983), and Tibaldi and Ji (1983).

6. DEVELOPMENTS

PREDICTIVE SKILL

Miyakoda

al.

(

1972) suggested that the results of their first comprehen-

sive trial of2-week predictions, using a hemispheric model, might

taken as

a benchmark for future comparisons. Their ensemble mean anomaly corre-

lations of 500-mbar height for the extratropical Northern Hemisphere, based

on 12 forecasts from January cases taken from the years 1964 to 1969, are

presented in Fig. 11, together with corresponding results from the opera-

tional ECMWF forecasts for the December to February period for the years

1982/1983 and 1983/1984. Included in the ECMWFresultsis the month of

January 1983, which we have already shown to have been one of high

predictability with the then-operational gridpoint model, but results with the

new spectral model and envelope orography for January 1984 are only

slightly poorer than achieved in the operational forecasts for the previous

January, and

Fig.

1 1

thus appears to be reasonably representative of current

ECMWF skill for the winter months.

Although a number

other qualifying remarks could be made about the

comparison presented in Fig. 11, it is nevertheless clear that a substantial

improvement has taken place over the past

10-

years in our ability to

predict at least the larger scales of motion over the extratropical Northern

Hemisphere. The improvement is confirmed by comparison of limited-area

verifications of operational forecasts over Europe

(L.

Bengtsson, personal

communication). Figure

shows a large increase in accuracy in the very

early part of the forecast, suggesting an important contribution from a more

accurate specification of the initial state. Overall, the more recent forecasts

reach the 0.8 level of anomaly correlation almost 2.5 days later, and they also

exhibit a somewhat smaller error growth beyond this time, as measured

either by anomaly correlation or by standard deviation. Differences between