SElYrEMBER 1994 TIBALDI ET AL. 1971Northern and Southern Hemisphere Seasonal Variability of Blocking Frequency and PredictabilityS. TIBALDI, E. TosI, A. NAVARRA,* AND L. PEDULLIAtmospheric Dynamics Group, Department of Physics, University of Bologna, Bologna, Italy (Manuscript received 28 May 1993, in final form 21 December 1993)ABSTRACT Seven years of analyses and forecasts from the operational archives of the European Centre for MediumRange Weather Forecasts have been analyzed to assess the performance of the model in forecasting blockingevents. This paper extends the previous work by Tibaldi and Molteni to the other seasons of the year and to theSouthern Hemisphere. The dataset covers the period from I December 1980 to 30 November 1987 and consistsof 500-hPa geopotential height daily analyses and the 120 corresponding forecasts verifying on the same day, adataset commonly known as the "Lorenz files." Local blocking and sector blocking have been defined as inTibaldi and Molteni, using a modified version of the Lejenas and Okland objective blocking index. The results broadly confirm the conclusions previously reached for the winter season alone, extending theirvalidity to the rest of the year and, mutatis mutandis, to the other hemisphere. The main observational differencebetween blocking in the two hemispheres is in the number of preferred locations: Atlantic and Pacific blockingin the Northern Hemisphere, and only one broad region in the Southern Hemisphere, around 180- longitude.Forecasting the onset of blocking events is in general a task that the model finds difficult, whereas if theintegration starts from an already blocked initial condition, the performance of the model is usually better. Thepoor observational data coverage in the Southern Hemisphere is likely to produce initial conditions affected bylarger errors, making the correct forecast of the onset of a blocking event an even more difficult task than it isin the Northern Hemisphere. In the Northern Hemisphere, although the dynamical characteristics of Atlantic andPacific blocks are inferred from the respective model errors to be different, their detrimental effects on forecastperformance are similar in the two cases.1. Introduction Despite major advances in operational mediumrange numerical weather predictions, forecasting errorsproduced by numerical models still appear to be animportant manifestation of model inadequacies. Theseare the sum of a "random" component and of a systematic part, the so-called climate drift (e.g., Arpe1990; Tibaldi et al. 1991 and references therein). Sucherrors are known to be synoptically related to the prevailing weather type (Grosswetterlage), and Tibaldiand Molteni (1990, hereafter referred to as TM) haveinvestigated the relationship between model behavior(and therefore errors) and blocking in the winter seasonof the Northern Hemisphere, using European Centre forMedium-Range Weather Forecasts (ECMWF) operational archives of analyses and forecasts. They essentially documented the inability of the ECMWF model* Current affiliation: CNR-IMGA, Modena, Italy. Corresponding author address: Prof. Stefano Tibaldi, Gruppo Dinamica Atmosferica, Dipartimento di Fisica, Via Irnerio 46, 40126Bologna, Italy.tO properly forecast blocking onset beyond a few forecast days, and how this reflected in fewer blocking episodes being simulated by the model than in the realatmosphere. This inability, shown by forecasting models, to enterinto a blocked state is also a limiting factor of paramount importance for the extended-range forecasts(Tracton et al. 1989; Tracton 1990; Miyakoda and Sirutis 1990; Brankovic and Ferranti 1992), where theconsequences of such errors are amplified by the longerintegration time. An improved understanding of the relationship between blocking forecast failures andmodel formulation and errors would therefore have aneven larger impact on extended-range dynamical forecasts. In this study, the analysis on blocking operationalpredictability performed in TM for the winter seasonof the Northern Hemisphere (NH) will be extended tothe other seasons of the year and then to the SouthernHemisphere (SH). The data used are ECMWF analyses and forecasts of 500-hPa geopotential height fieldsfrom 1980 to 1987. The reader interested in a more upto-date analysis of the ability of the ECMWF operational model to forecast blocking (in the NorthernHemisphere winter only) and in particular to a quanc 1994 American Meteorological Society1972 MONTHLY WEATHER REVIEW VOLUME 122Nil ANALYSIS (DEC-FEB 1981 fo 1987)NH ANALYSIS (MAR-MAY 19811o 1987)3O2sI a '104 '90 ~0 ~0 ~' '5'0' '$'0' 9'0 1~0' '1~0' '1~0 '20' '210' '270LONGITUDE90 60 50 0 50 60 90 120 151:) 180 210 240 270 LONGITUDE30, 25v 200~ 15z-- 10DAY $ FORECAST (DOT: VER. ANALYSIS)DAY 3 FORECAST (DOT: VER. ANALYSIS)50 -90 -60 -$0 :e0 f50 60 90 120 150 180 210 240 270 -90-60 -50 0 50 60 90 120 150 180 210 240 270 LONGITUDE LONGITUDE50DAY 10 FORECAST (DOT: VER. ANALYSIS)DAY 10 FORECAST (DOT: VER. ANALYSIS)... ./'......:..:' O__ ~ ~01 L ~ ~ 1~ ' 1~'0' m 6'0 m ' ~'0 ' 11~0' 11~01 11 ~ 0m 12; O' 12~ O' '270 __~0--60 J ~ 0 0 ~0 60 '0 J~ 0 ~SO 180 ~10 ~40 ~70 LONGITUDE LONGITUDE FIG. l. Percentage frequency of blocked days in the Northern Hemisphere as a function of longitude computed foreach day in the analyses (first row), day 3 forecasts (second row.), and day 10 (third row): first column winter (DJF),second column spring (MAM), third column summer (JJA), and fourth column autumn (SON).titative evaluation of. the recent improvements is referred to Tibaldi et el. (1993). Additionally, Tibaldi(1993) contains a brief account of the potentials ofblocking diagnostics as a tool to assess climate generalcirculation model (GCM) performances. The organization of the dataset and the analys~sprocedures are described in section 2. Section 3 isdevoted to the description of NH blocking frequencyand its longitudinal variations both in observationsand forecasts. Sections 4 and 5 describe the model'sSEPTEMBER 1994 TIBALDI ET AL. 1973NH ANALYSIS (JUN-AUG lgS1 to lg87)NN ANALYSIS (SEP-NOV lgs! f0 lg87)-60d i - , - ~ r ,--~O 50 60 go 120 150 180 210 240 270 -90 -60 -50 0 50 60 90 120 150 180 210 240 270 LONGITUDE LONGITUDEDAY 3, FORECAST (DOT: VER. ANALYSIS)DAY 3 FORECAST (DOT: VER. ANALYSIS)3OE,~ 2~1 g~ 2og~o ~0o.~1~ -90 -60 -`10h0 50 60 90 120 150 180 210 240 270 -90 -60 -`10 0 30 t;OgO 120 I$0 NtO 210 240 270 LONGITUDE LONGITUDE30E2 200DAY 10 FORECAST (DOT: VER. ANALYSIS)DAY 10 FORECAST (DOT: VER. ANALYSIS) - '.. ./....f\.....: ; i"-: : '~....'~ ..'05. ~ '"" :'~ -gO -60 -`10 I ..": ~., -..,: - ?.. '"0 `10 60 gO 120 150 180 210 240 270 -90 -60 -$0 0 ,10 60 90 1:20 150 180 210 240 270 LONGITUDE LONGITUDEFIg. l. (Continued)ability to represent blocking, with particular attentionto onset and duration. Section 6 analyzes the possibleeffects of blocking on objective forecast skill scores.Section 7 is devoted to a separate analysis of observed and predicted blocking in the Southern Hemisphere. A summary and conclusions are contained insection 8.2. Description of the dataset and of the analysis procedure The database for this study consists of daily 500-hPageopotential height analyses and the corresponding day1 to day 10 forecasts. For each day, 11 fields are available: analysis and day 1 to day 10 forecasts, all veil1974 MONTHLY WEATHER REVIEW VOLUME 122' TABLE 1. This table shows the total number of blocking days(which satisfy the criterion of three adjacent longitudes blocked in asector) and blocking cases (which satisfy also the requirement of theminimum duration of 5 days) for the Euro-Atlantic (EA) and Pacific(PAC) sectors~Blocking days Blocking casesSeason EA PAC EA PACWinter 281 285 27 25Spring 357 182 29 11Summer 194 204 14 20Autumn 209 99 18 8lying on the same day but started from progressivelylagging initial conditions. Such an arrangement of analysis and forecast fields is commonly known as a "Lorenz files" dataset. The total record includes sevencomplete years, from 1 December 1980 to 30 November 1987. Seasons have been defined somewhat arbitrarily as 90-day periods, the NH winter covering the90 days starting on 1 December [December, January,and February (DJF)], NH spring from 1 March[March, April, and May (MAM)], NH summer from1 June [June, July, and August (JJA)], and NH autumnfrom 1 September [ September, October, and November (SON)]. Since there are 11 records relative to eachday, the complete dataset consists of 2800 days. A buffer zone of 10 days is added at the end of each 90-dayseason to allow for blocking events straddling acrossseasons. The original .data consisted of global fields,projected on spherical harmonics coefficients truncatedat triangular truncation 40 (T40). For the present analysis the data have been projected on a regular latitudelongitude grid (3.75- x 3.75-) and only the regionsnorth of 20-N for the Northern Hemisphere and southof 20-S for the Southern Hemisphere have been considered. The ECMWF operational data assimilation schemeand' forecasting model have constantly been modifiedduring the 7-yr period under consideration. Changeshave been as major as a complete revision of the physical parameterizations and the transformation fromgrid-point to spectral numerical schemes. Tibaldi andMolteni considered the problem of the impact of thesemodifications on the representation of blocking in theNorthern Hemisphere winter, .performing the diagnostic analysis separately for the different periods in whichthe model was not changed, and their conclusion wasthat the model development during these 7 years didindeed produce some improvement in the ability "torepresent blocking at an advanced state during the forecast." This impact, however, was found not to be substantial. Recently, Tibaldi et al. (1993) have shown thatthe operational model has indeed measurably improvedits performance in winter during the most recent years. A local and instantaneous blocking index, based onthe TM modification of the original index by Lejenasand Okland (1983), will be used throughout this work.The index and the blocking criteria are extensively discussed in TM; therefore, only a brief definition, for thesake of completeness, will be given here. For theNorthern Hemisphere (the modifications necessary toapply the index in the Southern Hemisphere will bediscussed in section 7), two values of the geopotentialheight gradient are evaluated at each longitude:GHGS = z(~b0) - z(~bs) (2.1) 4'0 - q's GHGN = z(~b~) - z(~bo) ~b~ - ~b0 ' (2.2)where ~b~ = 80-N + A, ~bo = 60-N + A, ~bs = 40-N+ A, and A = -3.75-, 0% +3,75-. A given longitudeis then defined to be blocked at a certain instant in timeif the following conditions are both satisfied for at leastone of the three values of A: GHGS > 0 (2.3) -10m GHGN < (2.4) degrees of latitude ' Condition (2.4) was added by TM to the original Le jen~is and Oldand index--that is, condition (2.3) alone--to ensure that cutoff lows anomalously dis placed to the north were not counted as blocks. This criterion is different from other objective defi nitions of blocking to be found in the literature (e.g., Dole and Gordon 1983; Lau 1983) but satisfies the re quirements of time and space locality useful in this type of analysis. However, a local and instantaneous index such as this is able to identify blockinglike structures but has to be supplemented by further conditions, re flecting the synoptic requirements of spatial extension and time duration, which distinguish a transient block inglike flow pattern from a true blocking event. Such conditions will be developed and added in the forth coming section, where the concept of a blocked sector will be introduced. Lejenas and Okland (1983) and Ruti (1992) have shown that the index is quite consis'tent with synoptic assessments of blocking events. Moreover, the main objective of this paper is to per form a relative comparison between observations and forecasts, and therefore, the detailed nature of the blocking indicator should not affect the major conclu sions of the work.3. Northern Hemisphere observed blocking patterns, frequency, and distribution The results of the computation of the simple blocking local index on the complete observational databasein the Northern Hemisphere are shown in Fig. 1. Thetop panels [(a)-(d)] show the observed percentagefrequency of blocking as a function of longitude for thefour different seasons. The quadrants of preferredSEPTEMBER 19948O TIBALDI ET AL.PERCENTAGE OF BLOCKING DAYS - EURO ATLANTIC BL.197570soj40 7 DEC JAN80 /FEB MAR APR MAY JUNJUL AUG SEP OCT NOV Dt~CPERCENTAGE OF BLOCKING DAYS - PACIFIC BL.?060$0401o- / 0DECbJAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV Dt~CFIG. 2. Percentage of blocked days for each decade of the year: (a) Euro-Atlantic blocking, and (b) Pacific blocking.winter blocking are well evident in panel a and correspond to the synoptically known Euro-Atlantic (EA)and Pacific (PAC) blocking regions, separated by twogaps around 100- and 255-E, where the circulation flowappears to be predominantly zonal, according to theindex. A minor maximum can also be detected between50- and 70-E but it is difficult to separate it clearly fromthe major adjacent EA peak. The two major blockingpeaks reach a similar maximum value of approximately20%, indicating that winter blocking has an approximately equal frequency of occurrence in the two Pacificand Euro-Atlantic sectors. These two major blocking features persist throughspring. The EA maximum blocking frequency increases up to 25% and shifts from around 15- to 30-E,whereas the Pacific peak weakens considerably, reducing its maximum amplitude to about one-half of its winter value. An even deeper gap separates the peaks ataround 90-E, while the Great Plains separation ataround 250-E is lifted somewhat. The picture is different in summer. The EA blockingfeature has lost coherence and strength, the previousstrong peak being broken into many weaker features,among which it is difficult to identify a coherent structure. The central Asian gap (around 90-E) is harder todistinguish, and a noisier distribution appears in thatarea, while the gap over the Great Plains and the NorthAmerican continent is now more evident. The Pacificarea shows little difference with respect to the springsituation. The previous distinctive winter pattern appears now in disarray to the point that a clear definitionof lEA and PAC blocking areas would be impossiblebased on the data of this season alone. Although autumn is the season showing the loweroverall blocking frequency, during this period the EApeak starts rebuilding and the PAC sector is almostcompletely quiescent. The strong geographical localization, particularlyevident in winter and spring, together with the arguments on the nature of the objective index introducedat the end of section 2, suggest that an analysis basedon "sectors" rather than on isolated longitudes shouldbe more appropriate. As in TM, two sectors are thenidentified, Euro-Atlantic and Pacific, with the following longitudinal limits (note that the limits of the EuroAtlantic sector have been slightly modified with re1976 MONTHLY WEATHER REVIEW VOLUME 122 F~G. 3. Signatures of the regimes--that is, differences between mean observed fields of500-hPa height in the ensembles of "blocked" and "zonal" days, for the Euro-Atlanticsector [panels on the left, (a)-(d)] and for the Pacifi~ sector [panels on the right,(h)]. From top to bottom, winter, spring, summer, and autumn. The isolines greater orequal to zero are continuous, negative isolines are dashed, and the distance between isolinesis 30 m.SEPTEMBER 1994 T1BALDI ET AL. 1977 F~G. 4. Model day 10 500-hPa geopotential height total systematic errors (SEs, model ensemble mean minus observedensemble mean ) computed separately for each season: (a) winter, (b) spring, (c) summer, and (d) autumn. The i solinesgreater than or equal to zero are continuous, negative isolines are dashed, and the distance between isolines is30 m.spect to TM, where, they were 28-W and 32-E)~Euro-Atlantic: 22.5-W and 45-E; and Pacific: 152-Eand 136-W. A sector is then considered to be blockedif three or more adjacent longitudes within the sectorare blocked according to the previous index definition.The sector definition is kept fixed for all seasons. In addition to this criterion on spatial extension, afurther constraint on duration in time is needed to approach the synoptic concept of blocking "a la Rex."This has been done by requiring, for a blocking event(or episode) to be cataloged, that a sector is blockedfor at least five consecutive days. This is a totally arbitrary limit, which stems from the compromise ofhaving tree long-lasting blocking events at~d a sufficient number of them to constitute a minimal statisticalbase. The results of the application of these further conditions on the observed database are collected in Table1. The largest amount of blocked days (but also events)for the Euro-Atlantic sector occurs in spring, whereasin the Pacific sector, winter appears to be the preferredseason for blocking. The seasonal march of blockingfrequency is shown in detail in Fig. 2 for both sectors.Blocking in the Northern Hemisphere is thus confirmedto be characterized by a marked seasonal cycle; in fact,EA blocking by itself would define only two extendedseasons: a blocked winter-spring, and a zonal summer-autumn. In the Pacific, blocking seems to be a1978 MONTHLY WEATHER REVIEW VOLUME 122a --IFORECAST DAYFORECAST DAYb~o-.o ! 2 ~ ~ ~ 7 b ~ ,b ' o ~ ~ ~ , ~ ,~ .~ ~, ,FORECAST DAY FORECAST DAYFiG. 5. Total and correct (shaded) forecasts of b|ocldng in the Euro-Atlantic sector: (a) winter, (b) spring, (c) summer, and (d) autumn.dphenomenon mostly confined to winter, with a weaksecondary peak in summer. The seasonal cycle of bothsectors appears to have higher-frequency structures superimposed, with relatively isolated maxima in Marchand May for EA and in September and July for PAC.It is likely that such structures are an artifact of thelimited sample of the database and of the particulardefinition of the blocking index. As a means to check the sector analysis procedure,the sector analysis was performed on the gap regionSEPTEMBER 1994 T I B A L D I E T A L. ' 1979FORECAST DAYa iIFORECAST DAYbo~o~oz ~o.CSOdZd ':1~ ~ ~ ~ ~ ~ ~ ,~ ' ~, ~ ~ ~ ~ ~ ~ ~ ~ ,~ FORECAST DAY FORECAST DAY FiG. 6. As in Fig. 3 but for the Pacific sector.evident in Fig. la between 70- and 130-E, which is aregion where an absolute minimum of blocking episodes is to be expected. The total number of these"background" blocking events can be taken as an estimate of a threshold value above which a signal shouldbe detectable from the background noise. This "confidence threshold limit" could be conservatively estimated at the frequency value of 15%, which meansthat all blocking peaks in Fig. 2 are well above thisconfidence level for all seasons, thereby indicating thatthe definition of blocked sector appears to be consistent.1980 MONTHLY WEATHER REVIEW VOLUME 122-5~-4 a []- -I= EURO-ATLANTIC BL. -[]= PACIFIC BL.FORECAST DAY -- [].- = EURO-ATLANTIC BL.[~= PACIFIC BL. FORECAST DAY-3C[] - - - - [] - -I= EURO-ATLANTIC BL.[]= PACIFIC BL. - w__1_ I i -- [] d- = EURO-ATLANTIC BU[]= PACIFIC BL.FORECAST DAY FORECAST DAYFIG. 7. Mean error in blocking intensity in the Northern Hemisphere:(a) winter, (b) spring, (c) summer, and (d) autumn. To highlight synoptic patterns corresponding to sectorblocking regimes, the following procedure has been applied. Data have been classified according to whether aday is zonal, Pacific blocked, or Atlantic blocked. Daysblocked in both sectors at the same time have not beensingled out in a separate category (a "global blocking"regime) due to their small overall number; therefore,they appear in both blocked classes. Anomaly maps ofthe blocked regimes have then been constructed by cornpositing (i.e., ensemble averaging) the maps of the twoblocked categories and subtracting out the composite ofall zonal days (the "zonal" regime). Figure 3 showsNH 500-hPa mean maps of the anomalies correspondingto the two blocked regimes for the four seasons. Figure 3 documents a noticeable yearly cycle of intensity and pattern shape of blocking signature in boththe Atlantic and Pacific Sectors. The winter regimeanomalies [panels (a) and (e) for Euro-Atlantic andPacific sectors, respectively ] show the familiar blocking highs. Lows appear to the north and south of themain blocking high, whereas the residual high over theopposite hemispheric sector results mostly from thoseblocking episodes occurring at the same time over bothsectors (the "global" blocks). The main feature of thespring anomalies [panels (b) and (f)] is the relativeprevalence of the Atlantic peak over the Pacific: TheAtlantic signature also shows up in the Pacific blockingregime much more than in the opposite case. An exSEPTEMBER 1994 TIBALDI ET AL. 1981a[]: -URO-ATLANTIC BL. [] [][7I]= PACIFIC lqL. []FORECAST DAY[]= EURO-ATLANTIC BL.[]= PACIFIC BL.4 S 6 7 8 9 10FORECAST DAY12-4'-8 mC[]: EURO-ATLANTIC 8L. [][]= PAC~F)C BL. [] [] [] [] [] [] []! [] [][]: EURO-ATLANTIC BL.J7ql= PACIFIC BL.mira[]1 2 $ 4 5 6 7 8 9 I0 1 2 3 4 6 7FORECAST DAY FORECAST DAYFIG. 8. Mean error in blocking phase (longitude) in the Northern Hemisphere: (a) winter, (b) spring, (c) summer, and (d) autumn.dJ9 10amination of the time sequences of the blocking indicesshows that the long duration of the Atlantic blockingepisodes during spring makes global blocking rathermore probable than average. Summer and autumnanomalies [panels (c), (g) and (d), (h) of Fig. 3, respectively] show similar features.4. Short- and medium-range forecasts of blocking in the Northern Hemisphere Before analyzing the model's ability in representingblocking, and as a preliminary measure of generalmodel performance, we begin by showing here 500hPa geopotential height model systematic errors--thatis, the ensemble mean of forecast minus analysisfields--according to season. The results are shown inFig. 4, where it is evident that winter is the season withthe largest amplitude systematic errors. It can also benoted that the winter systematic error of the model hasa synoptic structure reminiscent of (but opposite insign) the superposition of the synoptic structures ofEuro-Atlantic and Pacific sector blocks (Figs. 3a and3e). Additionally, a distinctive feature of the summersystematic error (SE) is the strong positive maximumin correspondence of the left exit of the Asian jet anda rather featureless and weak negative SE elsewhere. We now turn to the computation of the objectiveblocking index on all forecast fields, therefore provid1982 MONTHLY WEATHER REVIEW VOLUME 1221087642 1000000 000 000 0000 0~0000I0111 010 011 0000 OOlOlO III~1111 ~11 ~11 ~1~ IIII1~II101 III 011 ~1~1 IIIIII IIIIIIII III III IIII IIIIII IIIoon a10987 654..521O0001 OlOI OI 0000 00000 00000 IO001 k0000 OeO0 O0 O00e eeooo eooeo oooooeooo eeoo oo eoao oeeeo ooeoeoooo oooo no onnn nnnnn nonoo0000 0010 I0 0110 01110 IllO0Olll 0000 I0 Oleo onnnn onmno onnmoIIII IIII II IIII IIIII IIIII IIIII10 g87 654321O1098 5 4000 '00 I000 [] 0 O0 []IlO OI 0000 0 0 O0 0000 O0 I000 0 [] O0 []000 O0 0000 0 [] O0 0OeO OO 0000 - 0 eO 0010 O0 O000 '1 0 O0lee O0 olon - o oo -III II IIII - - II -C21OO0 IO0 000 I~0 O0 0 0000oo coo ooo noo no o eaeoll 010 lib lib lO 0 IIIIII I10 III III II 0 IIIIII III III III II - IIIIdFiG. 9. Success/failure of forecast of blocking onset in the Euro-Atlantic sector: (a) winter, (b) spring,(c) summer, and (d) autumn. A black square represents a correct forecast, and a white square represents afailed forecast at a given forecast time. Forecast time along the ordinate and different blocking cases alongthe abscissa, grouped by different years.ing a measure of model performance. A detailed analysis of the behavior of the ECMWF operational forecasting model in winter alone can be found in TM. Thatwork has been extended here to the other seasons but,for reasons of brevity, comments will be restricted tosome selected forecast times (mostly day 3 and day 10,respectively, representative of the short and mediumrange) and to a selection of diagnostic quantities. In the middle and bottom panels of Fig. 1 [(e) - (h)and (i)-(1)], the longitude dependence of the localblocking index for the forecasts at day 3 and day 10 isshown, together with the corresponding verification diagram (dotted). The forecasts appear to underestimatethe frequency of blocking days in all seasons and forall forecast times, with the possible exception of shortrange forecasts during autumn (September-November). This effect appears to increase with increasingforecast time, with the underestimation already evidentat day 3 and well evident at day 10. A further analysisof the number of correctly predicted blocked days versus forecast time (Figs. 5 and 6, respectively, for EAand PAC sectors) reveals that the decrease of correctlySEPTEMBER 1994 TIBALDI ET AL. 1983109876542 1010 9 8 7 6 54' $ O~00 000~ 0~0 0000 000 0000 O00O~00 000~ DO0 0000 O~O 0000 0000000 0~0~ 000 0000 OBO OOeO 0000~00 0~0~ 000 0000 0~0 000~ 0000000 000~ I01 0101 000 001~ 0000I00 0000 ~00 000~ I10 0011oo o ooo oooo ooo - ooo oolo oOO 0 ~00 !O00 0OI 0 000 IO00 Iol o ooo ~ooo oII I I~1 IO~1 -II ] III IIII -b10 98 7 6 5 4 3 2 1 O10 98765432 1OOOO 000 DO00 O0 O0 000 000000 000 0000 De O0 000 000000 000 ~000 O0 O0 000 000000 00~ 0000 O0 O0 000 O00000 DO0 0000 O0 O0 000IOO O00 oooa Oo oo ooo OOOIO0 000 OOOO ~0 O0 IOJ 000moo ooo oooo oo no nee ooocl o o o ooo []o ! O 0 000 00 0 0 0 nO0 00 0 0 0 OeeO 0I I - I III -- - I I III -d BLOCKING CASESF[o. 10. As in Fig. 7 but for the Pacific sector.predicted blocked days is, regardless of the season, this maximum is attained as an estimate of the phase.very fast (in some seasons/sectors almost linear, but It is then possible to derive some more precise quanin the worse cases approximately exponential), con- titative assessment of the model behavior and performfirming the results of Miyakoda and Sirutis (1990), ance as a function of forecast time. Figure 7 shows,who found a similar behavior of blocking forecast season by season [panels (a)- (d)], the error in blockskill in a study concerning a number of 30-day fore- ing intensity (difference between the maximum valuecasts with the Geophysical Fluid Dynamics Labora- of GHGS computed over the observed and forecastedtory model, fields) as a function of forecasting time averaged overTurning now to sector diagnostics, we can assume all correctly predicted blocking episodes in each sector.the maximum value of the quantity GHGS [Eq. (2.1), There is a general tendency to underestimate blockingsection 2] inside the blocked sector as an estimate of intensity, modulated by a large seasonal variability ofthe intensity of the block, and the longitude at which the model behavior. The exception is provided by EA1984 MONTHLY WEATHER REVIEW VOLUME 122q : aFORECAST DAYb2 3 4 5 6 78 9 10FORECAST DAYz0o,0z0 0 ~ - d~: ~ (/-) <(o 0-q- (2) ~ : ~ ~o~* ~ ~ O 7 ; -- o - ~ i i i 0 1 5 6 ~ ~ 9 10 0 1 2 3 4 5 6 7 8 9 10 FORECAST DAY FORECAST DAYF~. ] ]. Numbe~ o[ co~-cfi~ p~edicted onsets o[ blockin8 in the Earn-Atlantic secto~ as aJunction of incm~sin8 [o~ccast time: (a) wimc~, (b) sp~nS, (c) summe~, and (d) ~atumn.blocking during autumn, for which the intensity is overestimated up to forecast day 6. Spring seems to providethe overall best model performances in forecastingblocking intensity, both for the Atlantic and the Pacificsectors, while both extreme seasons (winter and summer) show a similar behavior, with PAC blocking being ill treated by the model more than its EA counterpart well into the medium range. Phase errors (differences in the longitude of themaximum value of GHGS between analysis and forecasts) are shown in Fig. 8. Large drifts of opposite signsincreasing throughout the forecast time are evident inwinter and summer, corresponding to a systematicmodel eastward shift of Atlantic blocks and westwardshift of Pacific blocks. Phase errors seem to be negligible in comparison during transition seasons, confirming the comparatively better skill of the model in theseseasons already evident in the blocking intensity diagrams of Fig. 7. Since these are mean (systematic) errors, a degree of cancellation between positive and negative errors is taking place. Root-mean-square errors ofboth phase and amplitude (not shown) are approximately 40%-50% larger than their correspondingmean values. The large fluctuations evident in forecast errors ofboth blocking intensity and phase from forecast days 4and 5 onward are most likely due to the progressivelydecreasing number of correctly predicted blocks asforecast time increases, which in turn impoverishes thesample on which forecast errors are computed. The importance of this fact will become even more evident inthe next section.5. NH blocking onset and blocking duration It is commonly accepted that forecasting blockingonset is a difficult task for operational prediction models (e.g., Gmnas 1983 ). The unstable nature of the process of entering'the blocking state, which has oftenbeen advocated in theoretical blocking studies, wouldSEPTEMBER 1994 TIBALDI ET AL. 1985 a012345678910 FORECAST DAYbFORECAST DAY 0UZ0 6 7 8 910 0 1 2 3 4 5 6 7 8 910 FORECAST DAY FORECASI DAY Fio. 12. As in Fig. 9 but for the Pacific sector.in fact justify such difficulties, but, on the one hand,the precise physical nature of the instability process hasnever been satisfactorily and uniquely (if possible)identified and, on the other hand, precise quantitativeevidence of the above-mentioned forecasting difficultyhas started only recently to be brought forward (e.g.,TM; Tracton et al. 1989; Miyakoda and Sirutis 1990).It appears therefore useful to try to quantify such difficulties and their dependence upon seasons. Blocking onset is therefore defined as the transitionbetween a zonal day and a blocked day in a given sector, and it becomes possible to verify how far back intime it is possible to go before the ability of modelingsuch a transition as an initial value problem vanishes.Figure 9 attempts to do just this for the Euro-Atlanticsector: each blocking episode is represented by a column. In each column, a black square indicates that theforecast started n days (n being the value along the yaxis of each panel) before the onset date of the event(the first blocked day) was blocked, and a hollowsquare that it was not. Day 0, being the analysis, isobviously indicated as blocked for all cases. The totalnumber of blocking episodes in each season (the number of columns in each of the four panels) correspondsto the numbers indicated in the third column of Table1. The episodes are ordered chronologically along thex axis, and columns are grouped in subsets to indicatethat they took place in the same year. There are usuallyseven subsets in each panel, as many as the number ofyears in the dataset. If they are less than seven (e.g.,Pacific blocking in spring), this indicates that there areyears with no blocking episodes in a given season. The model displays, in the EA sector, a large case-tocase variability of skill, composed of both intraseasonaland interannual contributions, but no clear systematicseasonal cycle or secular trend is apparent. Good andbad cases appear to be fairly uniformly scatteredthroughout the seasons and the years. The situation forthe Pacific sector, shown in the companion Fig. 10, isalso rather similar, with no apparent trends well evident.1986 MONTHLY WEATHER REVIEW VOLUME 122 BLOCKING CASES FIG. 13. Fo[-cast of blocking duration in the Euro-Adantic sector. Black bars represent blocking durationin the analysis, and white ba~s represent blocking duration in the forecast. Black-topped b~s mean forecastedblocks shorter than observed, and white-topped bars mean forecasted blocks longer than observed. Totallyblack bars mean that the block duration has been correctly predicted: (a) winter, (b) spring, (c) summer,and (d) autumn. Figures 11 and 12 try to give a summary of this information by counting the total number of successfulblocking onset forecasts as a function of forecast time.There is an approximately exponential decrease in thenumber of correctly predicted onsets in both the Atlantic and the Pacific sectors. The decrease appears to bemore rapid in winter and less so in spring. About 50%of the onsets are correctly forecasted four days aheadin spring but only 30% in winter and 14% in summer.Similarly, for the Pacific case, about 40% of the casesare forecasted correctly four days ahead in winter and25% in summer, the two seasons containing, in thissector, the largest number of cases. We now turn our attention to blocking maintenance,that is, the ability of the model to predict blocking duration. This is shown in Figs. 13 and 14. We restricthere the analysis to those forecasts starting on the veryfirst day of the blocking event--that is, containing theSEPTEMBER 1994 TIBALDI ET AL. 1987,,,,o, 111 ~IIi II, [ I I I I I I [ I I I I I I I I I I I I I I I I I [ I I I I I I I I [ I I I10 1110 g 8 7 6 5 421I I I I I I I I I I [ I [ I I I I I I I I I I I [ I I [ I111087654.~21iI1110987652 1 BLOCKING CASES14. As Jn Fig. 9 but for the Pacific sector.blocking already in the initial conditions. In the twofigures, black and white bars represent, respectively,the duration of the observed blocking event in the analysis and in the forecast. Bars (events) with a white topindicate a forecast of a blocking episode longer thanobserved, while bars with a black top indicate the opposite. Correct forecasts, as far as blocking duration isconcerned, are indicated as completely black bars.Events that last longer than 10 days are grouped together and indicated as 11-day events. The overall model perfot~nance is good, and a number of outstanding successes with relatively long-lasting blocks are apparent. Furthermore, the model appears to be measurably more successful at predictingthe correct duration of Euro-Atlantic blocks than ofPacific blocks, although this difference is probably statistically not significant, due to the limited sample ofevents. A quantitative evaluation of the statistical significance of the difference between the rms error ofblocking duration in the two sectors is made very difficult by the large number of blocks with durationlonger than 10 days, for which computing a durationerror is not possible. However, histograms of blockingevent duration can be computed to verify the qualitative1988 MONTHLY WEATHER REVIEW VOLUME 122 ANALYSIS$ 4 5 6 7 8 9 10BLOCKING DURATION (DAYS) b ANALYSIS FORECAST "3 4. 5 6 7 8 9 10 11BLOCKING DURATION (DAYS)ANALYSISFORECASTCSANALYSISFORECAST H rt rlrt '6 ' nr: rl F1IIJ I I t J I I I I I I I $ 4 5 6 7 8 9 10 1~1 .'5 4 5. 6 7 8 9 10 1~1 BLOCKING DURATION (DAYS) BLOCKING DURATION (DAYS)dFIG. 15. Distribution of blocking duration in the Euro-Atlantic sector:(a) winter, (b) spring, (c) summer, and (d) autumn.impression (Figs. 15 and 16). It is possible to note thatlong (>10 days) Atlantic blocks are captured by themodel better than long Pacific blocks, but for blocks ofintermediate duration a mixed situation prevails. These two figures provide some additional information about the ability of the model to exit blockscorrectly. Much more attention has usually been devoted to the onset forecasting problem--that is, thezonal-to-blocking transition--but less attention is usually paid to the corresponding problem of blocking-tozonal transition. Figures. 15 and 16 show that often themodel fails to exit a blocking condition at the correcttime, persisting in a blocked state that is no longer observed. It is interesting to point out how this fact, together with the information contained in Figs. 5 and 6,affects the interpretation of the results shown in Fig. 1.The cumulative measure of model blocking frequency,computed irrespective of correct verification of theforecast [as the curves of Fig. 1, panels (e)-(l), elfectively are], hides the fact that, by day 10, more than50% of the events that contribute to the frequency donot verify correctly (see Figs. 5 and 6). If, for example,Fig. li were redrawn using only correctly verified 3- or5-day forecasts, the already low model peaks in correspondence of Atlantic and Pacific blocking would befurther reduced, leaving little signal.6. Blocking and forecast skill in the NH Another very interesting way to study the impact ofblocking on medium-range predictions is to stratify objective model skill according to the observed prevailingweather regime. This allows an objective, quantitativeevaluation of common, but often unproven, beliefs thatmake blocking a favorable condition for numericalforecasting. Before going into a more detailed analysisof the results, it is perhaps useful to recall the heuristicbasis for the above-mentioned belief.SEPTEMBER 1994 TIBALDI ET AL. 1989ANALYSISFORECAST3 4 5 6 7 8 9 10 11BLOCKING DURATION (DAYS)a '~- ANALYSIS FORECAST O.(..) ~. 3 4. 5 6 7 8 9 10 11 BLOCKING DURATION (DAYS) ANALYSIS C ._ _rg.R_E_g.A_S_T... !"$ 4 5 6 7 8 9 10 11 BLOCKING DURATION (DAYS)ANALYSISFORECASTd~ ~ , 1~13 4 5 8 7 8 9 10BLOCKING DURATION (DAYS)FIG. 16. As in Fig. 13 but for the Pacific sector. Numerical models have, for the most part of theirearly history, been affected by the tendency (or, rather,systematic error) to show too much of a persistent dynamical behavior and, correspondingly, too little high~and low-frequency variability. This generic behavior,common to many, if not all, early numerical forecastingmodels, has produced the belief that persistent atmospheric regimes (of which blocking is an outstandingexample) favor numerical weather prediction, in thesense that they go in the direction of minimizing theeffects of model systematic error. If this simplistic argument were entirely correct, one would expect objective model performance measure to reflect this byshowing measurable differences in ensemble means ofmodel skill scores stratified according to the regimeprevailing in the initial conditions (ICs). More precisely, one would expect forecasts starting fromblocked ICs to be on average more skillful than forecasts starting from zonal ICs. A third group of forecasts, which would score last in this list, is the oneformed by 10-day forecasts, during which a significantregime transition takes place. Given the unstable natureof the regime transition, and on the basis of the resultsdiagnosed so far, it can be safely assumed that theseforecasts should be the least successful. To determine forecast skill, the choice of an objective measure is required. The one used here will beanomaly correlation coefficient (ACC) of 500-hPageopotential height computed over the relevant sectorsof the Northern Hemisphere (Pacific or Euro-Atlantic). The limitations of this objective skill score arewell known (e.g., Palmer and Tibaldi 1988; Murphy1991 ). This measure of skill, however, was preferredto the more usual rms error because of its sensitivity toflow shape and structures rather than to its absolutegeopotential height values, a characteristic that appearsdesirable during blocking situations. The climate usedto compute anomalies is the corresponding entire 7-yrseasonal climate. To stratify forecasts according to theGrosswetterlage prevailing during the forecast period,1990 MONTHLY WEATHER REVIEW VOLUME 122lOO90807o6o5040302o lOa123456 891o FORECAST D~loo-i~b9 O- ~....~.~' '"'8070604030- ZONAL ..... BLOCKING TRANSITIONFORECAST DAY--"-, 100~ 90Z0 80~ 70._lrY 60-~0 50 40 30'O 20BLOCKING ' ". '....100 9080 7060504030 20 10ZONAL '"" ..d1 4 8910 1234 678910 FORECASTD~ FORECASTD~ FIG. 17. Anomaly correlation coefficient of 500-hPa height between analysis and forecast as a functionof forecast day in the Euro-Atlantic sector, computed on three different ensembles of cases, depending onthe prevailing mean circulation: zonal (continuous line), blocked (short dashed line), and transition (dottedline). (a) winter, (b) spring, (c) summer, and (d) autumn.an objective criterion is also needed. Here, the following was chosen, solely on the grounds of reasonabilityand ease of implementation. Forecasts started on theday of blocking onset, 1, and 2 days later have beenclassified as the "blocking ensemble," forecasts started4, 5, and 6 days before blocking onset have been classified as the "transition ensemble," and all the forecastswith no blocking events throughout the integration period have been classified as the "zonal ensemble." Figures 17 and 18 summarize the results of this analysis. The only cases where the above-mentioned simplelogic (best forecasts during blocking situations, next bestduring zonal conditions, worst when a zonal-to-blockingtransition occurs) is completely borne out by model behavior are winter and autunm Pacific blocking and summer Euro-Atlantic blocking. A second type of modelbehavior is observed during Euro-Atlantic blocking inspring and winter and for Pacific blocking in spring andsummer--that is, best forecasts during zonal conditions,next best during blocking, and worst during transitions.The Pacific sector during summer behaves similarly,with zonal situations scoring better than ~both blockingand transitional regimes. The only period during whichtransition forecasts score best is autunm in the EuroAtlantic sector. In this last case, however, relative differences are very small. Additionally, autumn is a seasonduring which blocking in the Atlantic is relatively weakand rare. In general, the model has more difficulties if atransition occurs within forecasting range during the seasons when blocking activity is vigorouS. A conclusion emerging from the foregoing results isthat, while the zonal-to-blocking transition is a seriousproblem for numerical model 'forecasting during allseasons and in both NH sectors, at least as far as ACCis concerned, serious forecast failures are also commonduring both persistent blocking and persistent zonal sitSEPTEMBER 1994 TIBALDI ET AL. 19910Z100 9080 70 60 50 4O 20 10BLOCKING2 4 5 7 8 9 10 FORECAST DAYa100 9080 7060 50 4030 20 10ZONAL ~'~'~"BLOCKING1 2 $ 4 5 7 8 g 1 FORECAST DAYb0Z100 9080 70 60 50 4030 20 10 C"~r~7,A~ifZA"FORECAST DAY FXG. 18. As in Fig. 15 but for the Pacific sector.~oo- d90 - '"2"~%~.a.~. 80 70 60 50 4o- ZONAL .............. ~ $0 BLOCKING ~m~mmm~m~=~2o - TRANSITION - I t ~ i i i m I I 2 3 4 5 6 7 8 9 10 FORECAST DAYuations. This attempt to diagnose a possible blockingskill relationship does not provide any additional information as far as the origins of errors in the two different regimes. It is possible to speculate that smaller,more rapid, baroclinic scales could dominate error patterns during zonal regimes, while larger, slightly slowerand more barotropic processes would dominate thegeneration of forecast errors during persistent blockingsituations, including failures to correctly maintain theobserved blocked structure. This problem is in need offurther investigations, possibly based on more detailedcase studies, and will be the subject of future work.7. Blocking in the SH Before discussing the extension of the analysis to theSH, a description of the modifications to the index definitions appropriate for the SH is needed. These affectthe latitudes at which geopotential height gradients areevaluated. The need for such a change is explained bythe fact that blocks in the Southern Hemisphere areusually located at lower latitudes than Northern Hemisphere blocks (see Taljaard 1972; Coughlan 1983; Lejen~is 1984; Okland and Lejen~is 1987). To define the index in the Southern Hemisphere, thegeopotential height gradients GHGN and GHGS (middle and high latitudes, respectively) are again evaluatedat each longitude point of the grid as GHGS = z(qbs) - Z(qbo) (7.1) GHGN = z(~bo) - Z(qbs) qbo - qb~ ' (7.2)where qb~ = 35-S + A, ~b0 = 50-S + A, rbs = 65-S+ A, and A = --3.75-, 0-, +3.75o.1992 MONTHLY WEATHER REVIEW VOLUME 122BSH ANALYSIS (JUN-AUG 1981 fo 1987)SH ANALYSIS (SEP-NOV 1981 fo 1987)30 a2520. A15tO-5-0 .,..,..,..,..,.~ -90 -60 -30 0 30 60 90 120 150 180 210 LONGITUDE240 270 -90 -60 -30 0 30 60 90 120 150 180 210 240 270 LONGITUDE;~02520DAY 3 FORECAST (DOT: VER. ANALYSIS)eDAY 3 FORECAST (DOT: VER. ANALYSIS)90 60 30 0 30 60 90 120 150 180 210 240 270 90 60 $0 0 30 60 90 120 150 180 210 240 270 LONGITUDE LONGITUDEDAY 10 FORECAST (DOT: VER. ANALYSIS)DAY 10 FORECAST (DOT: VER. ANALYSIS)30~ 25v~jZ 20'0w 15z~ io.0o, $ 0 /........:"~....-90 -60 -30 0 30 60 go 120 150 180 210 240 LONGITUDE ..,,," "~:"..~.270-90 -60 -30 0 30 60 90 120 150 180 210 240 270 LONGITUDE FIG. 19. Percentage frequency of blocked days in the Southern Hemisphere as a function of longitude computed for each dayin the analyses (first row), day 3 forecasts (second row), and day 10 forecasts (third row): first column winter (JJA), secondcolumn spring (SON), third column summer (DJF), and fourth column autumn (MAM).S~PTE~aE~ 1994 T1BALDI ET AL. 1993SH ANALYSIS (DEC-FEB 1981 to 1987)SH ANALYSIS (MAR-MAY 198t fo 1987)5O~ 25.z 20'w2:~0to ~o~m 5i 0 -go -60 -50Cd~0 60 ,0 J~o ,so ~0 ~,0 ~0 ~70-,0'-~0-~0 ~' io 6'0' ','0' '~0 '~0' ~0' ~i0 '~io '~70 LONGITUDE LONGITUDEDAY $ FORECAST (DOT: VER. ANALYSIS)DAY 3 fORECAST (DOT: VER. ANALYSIS)30g 25 2Ow~C~ 15~ io. . .... ..L....o~ ....:" '"' 2': 5 ., ..~; '~ .~.... 0% , - ~ .... i ~ . ~ . . ~ . , ~ , , -~0 -$0 -~0 :. ,......'::0 50 60 gO '120 150 180 210 240 270 -go -60 -50 0 ,~0 60 gO '120 150 180 210 24.0 270 LONGITUDE LONGITUDE30 25.z 20I.~ 15ZzDAY 10 FORECAST (DOT: VER. ANALYSIS)DAY 10 FORECAST (DOT: VER. ANALYSIS)--~O'-~o'L~d j ~'0 60 ~,0 ,20 ,do ~80 z,o 2.,0 2?0 so ~0 .~0 0 ~0 60 ~0 ?20 ,50 ,& 2,0 2.0 vo LONGITUDE LONGITUDEFiG. 19. (Continued)1994 MONTHLY WEATHER REVIEW VOLUME 122TABLE 2. AS for Table I but for the Australiansector of the Southern Hemisphere. Season Blocking days - Blocking casesWinter 249 21Spring 190 15Summer 184 17Autumn 222 18 A given longitude is then defined to be blocked at acertain instant in time if the following conditions areboth satisfied for at least one value of A:GHGN > 0GHGS <-10m(degree of latitude)(7.3)(7.4)Condition (7.4) was added (as for the NH case) toensure that cutoff lows anomalously displaced to thesouth were not counted as blocks. The result of the local and instantaneous index calculation on the analyses in our database is shown in thetop panels of Fig. 19 [(a)- (d)]. The season definitionis of course specular with respect to the one used forthe Northern Hemisphere: winter indicates the JJA period, spring is SON, summer is DJF, and autumn isMAM. As for the Northern Hemisphere, for'the Southern Hemisphere, the TM modified version of the Lejen~is and Okland index behaves in a way very similarto the original index, as can be seen from a comparisonof Fig. 19 with Fig. 2 of Lejen~is (1984). Compared to their Northern Hemisphere counterparts, Southern Hemisphere blocking events show asimpler picture. They are overall much less frequentand do not show strong seasonal dependence. Duringwinter, there is only one area, very wide in longitude,of maximum frequency, extending from 150-E to70-W. This broad area has two relative maxima. Thefirst maximum, localized between 150- and 210-E, hasa frequency of occurrence above 10% and remainsabove this level for the whole year, with very smallseasonal variations of intensity. This sector can be identified with the Australia-New Zealand blocking region. The other maximum, from 70- to 100-W, has afrequency of occurrence between 5% and 10%, is localized east of the South American coast, and can beidentified as the Andes blocking area. With the marchof the seasons, the blocking frequency in the Andesarea decreases and reaches its minimum value in summer, when the signal is indistinguishable from thebackground noise. In autumn there are signs of an increasing blocking activity and the frequency starts torise again. It is worth noting here that, with the possibleexclusion of the winter period, when the blocking activity has its maximum value, the frequency of blockingoccurrence in the Andes area, as it appears from ourindex, is very low and does not appear to give an important contribution to the global blocking activity ofthe Southern Hemisphere. As in the case of the Northern Hemisphere, the geographical localization of blocking suggests an analysisthat considers sectors instead of isolated longitudes.Following the same criteria used before, an Australiansector has been defined with the longitudinal limits150-W and 150-E. The sector definition is kept fixedfor all seasons. The results of this sector blocking definition are summarized in Table 2, which contains the number ofblocking days identified in the Australian sector, andin Fig. 20, which shows the seasonal march of sectorblocking days frequency. The further application of thecondition on time duration (blocking cases defined asepisodes longer than 5 days) gives the results containedin the second column of Table 2. From the table it isevident that about 30% of the days are blocked overthe year (see Fig. 20), and as previously noted, only aslight seasonal trend is detectable with a certain prePERCENTAGE OF BLOCKING DAYS - AUSTRALIAN BLOCKING8O60 .5030 ~ -:20 ~ /~ DEC JAN FEB MAR APR MAYJUN JUL AUG SEP OCT NOV DECFIG. 20. Yearly cycle of the percentage of blocking days (for each decade) for Australian sector blocking.SEPTEMBER ] 994 , i ~': ~ ~ ," t a ',,, i ....... ~ .~-~r-~ / -~ ~ - ~ ~ ~ ~ / ...... ~.~ '~ 4-;/~7",. '_-- ...... X , - . . . o ~ ...... , ~. ----~ ...... ........ ', 5:'d d, .. ~: " ).:... ,., / ', : )~ / , '.~ , ~./ , ,, ~ .... ~ ~ ~1995 FiG. 21. Signatures of observed blocking regimes for the Australian sector of the Southern Hemisphere--that is,differences between mean observed fields of 500-hPa geopotential height in the ensembles of "blocked" and "zonal"cases referring to the Australian sector. Panels (a)-(d) are winter, spring, summer, and autumn, respectively. Theisolines greater than or equal to zero are continuous, isolines less than zero are dashed, and the distance between isolinesis 30 m.dominance of winter, both in terms of blocking daysand the number of cases. In parallel to what we did for the NH, we show herethe signature of the Australian blocked regime in Fig.21. The differences between the observed fields at 500hPa in the ensembles of "blocked" and "zonal" daysare shown for the different seasons [winter-autumn,panels (a) - (d), respectively]. As was the case for NHblocking signatures, the shape of the patterns appearsynoptically realistic, which justifies a posteriori thechoices made to define the objective index. As alreadynoted, SH blocking shows less of a seasonal cycleand slightly weaker features (at least for those seasons, like winter, which have the strongest blocking inthe NH).a. Short- and medium-range forecast of blocking in the SH The results of the application of the local and instantaneous index to the predicted fields (day 3 and day10 forecastsl as for the Northern Hemisphere) areshown in the middle [(e)- (h)] and bottom [(i)-(1)]panels of Fig. 19, respectively. The dotted line in allpanels refers to the observations and has been superimposed to facilitate the comparison. A general trend,detectable in all seasons (and already noted for theNH), is the decrease of the blocking frequency withthe increase of the forecast period. Already at day 3 theblocking frequency starts to decrease and, particularlyin winter and spring, its maximum shifts to the east. At1996 MONTHLY WEATHER REVIEW VOLUME 122 ~, 5 6 7 8 9 I0 0 1 2 3 ~, 5 6 7 8 9 10FORECAST DAY FORECAST DAYCFORECAST DAYdFORECAST DAYFIG. 22. Total and correct (shaded) forecasts of blocking in the Australian sector: (a) winter, (b) spring, (c) summer, and (d) autumn.day 10 the simulated blocking frequency becomes solow to be almost indistinguishable from the backgroundnoise. The number of correctly predicted blocked daysversus forecast time is shown in Fig. 22 and does notdiffer substantially for the corresponding figures (Figs.5 and 6) presented for the Northern Hemisphere. Also,in the case of the Australian blocking there is an almostexponential decrease in the number of correct forecasts.SEPTEMBER 1994 TIBALDI ET AL. 1.9973 i2-3[],, AUSTRALIAN BL. - AUSTRALIAN BL. I , ,1 23 4 5 6 7 8 9 10 1 2 3 4 5 6 7 FORECAST DAY FORECAST DAY9 10 2~-2[] []c 3 2v[] [],,, AUSTRALIAN BL. [] AUSTRALIAN BL. [] ~ ! ~1 23 4 5 6 7 8 9 10 1 2 3 4 5 6 7 FORECAST DAY FORECAST DAYdFIG. 23. Mean error in blocking intensity in the Southern Hemisphere:(a) winter, (b) spring, (c) summer, and (d) autumn. [][]8 9 10 The systematic errors in the strength and position ofthe correctly predicted blockinglike structures are presented in Fig. 23 and Fig. 24, respectively. The meanerror (mean over the correctly forecasted days) in theblocking index indicates a general underestimation ofthe amplitude of the blocks. The forecast of the positionof the block has a clear eastward shift only during winter. For all the other seasons the tendency toward aneastward shift is characteristic only of the first forecastdays. A feature common to both the amplitude and theposition of the forecasted blocks is a return of forecastskill after forecast day 7. This return of skill has beenobserved also in the Northern Hemisphere blocking,and there is no simple and unambiguous explanationfor it. A possible explanation could be that only verypersistent, and therefore predictable blocks, can beforecasted 7 or more days in advance, and so the statistics of these forecast times is representative only ofthe long-lasting episodes. At this stage, however, thisremains only a speculation.b. Blocking onset and blocking duration in the SH The performance of the ECMWF model in forecasting blocking onset is shown in Fig. 25, which summarizes the results obtained for the Australian sector.The prediction of blocking onset in the Southern Hemisphere as an initial value problem appears to be a difficult task for the ECMWF model, even at very shortrange. This appears to be the most evident differencebetween the model's behavior in the two hemispheres.Neglecting, for the moment, medium-range forecasts,for which correct forecasts are almost completely absent, in the short range, 1-3 days, the success rate is1998 MONTHLY WEATHER REVIEW VOLUME 12212-8- AUSTRALIAN BL.1 2 3 4 5 6 7 FORECAST DAY8 9 1012'-8--1AUSTRALIAN BL.2345678 FORECASTDAY9 1012-8- AUSTRALIAN BL. 12C 8 ~,4 O ' U.I U.I -8 - AUSTRALIAN BL. , 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 FORECAST DAY FORECAST DAYFIG. 24. Mean error in blocking phase (longitude) in the Australian sector of the Southern Hemisphere: (a) winter, (b) spring, (c) summer, and(d) autumn.much lower than that found for Northern Hemisphereblocking. In fact, the success rate for day 1 forecasts isonly 70% (50 successes over 71 cases), and drops to42% (30/71) for day 2 forecasts. The correspondingrates for the Northern Hemisphere Atlantic blockingare 88% (77/88) for day 1 and 60% (53/88) for day2, and for Pacific blocking, 80% (51/64) for day 1 and63% (40/64) for day 2. There is no appreciable seasonal cycle in the behavior of the model. Although it is always difficult to put forward essentially unsubstantiated interpretations, the most likelyreason for the different performance of the model inthe two hemispheres could be found in the differentavailability of observations, which strongly influencesthe accuracy of the initial conditions, so that betteranalyses in the Northern Hemisphere lead, on average,to a better operational predictability of blocking events. The performance of the ECMWF model in predicting the duration of Southern Hemisphere blocking isshown in Fig. 26; it is evident that in general the modelunderestimates the duration of the blocking episodes.In 66% of the cases the predicted block is shorter thanobserved; in 24% of the cases the predicted duration iscorrect, while predicted blocks last longer than observed only in the remaining 10% of the cases. Moreinformation about the behavior of the model in predicting the duration of the blocking episodes can bederived from inspection of Fig. 27, where the histograms of the number of episodes as a function of duration are shown. Moving toward longer block durations, it is possible to note a decrease in the number ofpredicted cases, which is higher than the observed decrease.c. Forecast skill and blocking in the SH Observed differences between blocking in the twohemispheres are usually ascribed to geographical difSEPTEMBER 1994 TIBALDI ET AL. 199910 9'~ 4 3 2 1 O109876 5 4 $2 1O1098765432 1O10 987 6543O00 000 0000 O0 00~- 0~ 000 0~0~ O~ DO0- 00 000 0~0~ O0 000 00000~0 00~ 0~00 O0 OOO ~OOOO0~0O0O0~0O~aO0 0000 O0 OOO 000 0O0 0000 O0 000 000 0O0 0000 O0 000 000 0O0 0000 O0 ~00 000 []O0 0000 O0 000 000 0O0 0000 O0 O~O 000 0O0 00~0 O0 000 000 0~0 O~O O0 ~00 0~ 0b0~ 0 O0 0000 000 O0- 00 0 O0 0000 OO~ O0- 0~ 0 O~ 0000 00~ OODO0 [] O0 0000 000 O0000 0 O0 0000 00~ O0000 0 O~ 0000 ~00 O0- 00 0 O~ OOO0 000 O0- 00 0 O~ 000~ ~00 O0O0 J 0 O~ ~0~ ~ OlooDO C~O0 000 O0 O~O O0000 000 O~ 000 O0000 000 OO ~OO O00~0 000 O0 DO0 O0000 OOJ O0 000 O0000 ~OO O0 000 O~0~0 00~ O0 000 O~000 DO0 O0 000 J~- OO 000 ~0 ~ O0O0 O0 []O0 O0 0O0 O0 []OO O0 0O0 O0 0 O~II II JII IiBLOCKING CASESd Flo. 25. Success/failure of forecast of blocking onset in the SH Australian sector: (a) winter, (b) spring,(c) summer, and (d) autumn. A black square represents a correct forecast, and a white square represents afailed forecast at a given forecast time. Forecast time along the ordinate and different blocking cases alongthe abscissa, grouped by different years.ferences (e.g., mountains and land-sea contrasts) andto dynamical differences (e.g., the different role playedby ultralong planetary waves). The results presentedand discussed above point toward an additional important difference between the behavior of the ECMWFmodel in the two hemispheres, which reflects an increased difficulty (compared to the situation in the NH)to forecast, even at short range, blocking onset in theSouthern Hemisphere. A possible explanation, as suggested above, to justify this lower SH predictability of blocking onset hasbeen the lower accuracy of the atmospheric analysisused as an initial condition. This lower accuracy, inturn, is most likely due to lack of observations. If thisinterpretation is correct, we should expect, in theSouthern Hemisphere, a lower forecast skill than in theNorthern Hemisphere not only for blocking onset butfor all other weather regimes as well.2000 MONTHLY WEATHER REVIEW VOLUME 12211109 8 76 52 1897654BLOCKING CASES FIG. 26. Forecast of blocking duration in the Australian sector. Black bars represent blocking duration inthe analysis, and white bars represent blocking duration in the forecast. Black-topped bars mean forecastedblocks shorter than observed, and white-topped bars mean forecasted blocks longer than observed. Totallyblack bars mean that the block duration has been correctly predicted: (a) winter, (b) spring, (c) summer,and (d) autumn, The objective forecast skill in the two hemispherescan be evaluated comparing the ACC composited according to different weather regimes shown in Figs. 17and 18, referring to NH Euro-Atlantic and Pacificblocking, respectively, with Fig. 28, referring to SHAustralian blocking. The differences between the twohemispheres stand out clearly. In the Northern Hemisphere, appreciable differences during "blocked,""zonal," and "transition" regimes are evident, whilein the Southern Hemisphere the behavior of the modeldoes not show any significant systematic difference between the various regimes. The comparison also confirms the generally lower forecast skill in the SouthernHemisphere compared with the Northern Hemisphere,both for the short and medium ranges. The rate of decrease of the ACC typical of the day 2-day 6 periodof the SH is attained in the NH seldom and only afterday 6. Such results are consistent with two alternativeSEPTEMBER 1994 TIBALD1 ET AL. 2001ANALYSISFORECAST n __,$ 4 5 6 7 8 9 10 11BLOCKING DURATION (DAYS)ANALYSISFORECAST"3 4 6 7 8 9 10BLOCKING DURATION (DAYS)b0~0,~-_ANALYSISFORECAST:I r; ;~I q ~ . , I I I II / 6 7 8 9 lO11BLOCKING DURATION (DAYS)C (D 2 9ANALYSISFORECASTd"$ 4 5 6 7 8 9BLOCKING DURATION (DAYS)FIG. 27. Distribution of blocking duration in the Australian sector: (a) winter, (b) spring, (c) summer, and (d) autumn.interpretations: either this regime classification is unable, in the SH, to show any measurable predictabilitydifference between the regimes or the predictive capabilities of the forecasting system in the SH (observational network, data assimilation system, and forecasting model) are insufficient to show the subtle differences in predictability among the weather regimes.8. Summary and conclusions The analysis performed on 7 years of ECMWF operational Lorenz files--that is, 500-hPa geopotentialheight data, both analyzed and predicted--has revealed a significant seasonal variability of the frequency of blocking episodes in both the Euro-Atlanticand Pacific sectors of the Northern Hemisphere, as theyare detected by an objectively defined synoptic-typeindex. Such an index (a modification of the originalLejen~s and Okland index) allows a separate analysisof blocking predictability in the two preferred NorthernHemispheric sectors--that is, Euro-Atlantic and Pacific. The winter-spring extended season emerges asthe most active period for Atlantic blocking, whereasblocking in the Pacific sector is more probable duringthe more extreme seasons of winter and, to a lesserdegree, summer. For the Northern Hemisphere, the model performance also shows some considerable seasonal dependence, with the best predictive results obtained for theAtlantic and Pacific sector blocking during spring.However, the model exhibits difficulties in representingblocking with the correct frequency and amplitude thatare rather uniformly distributed through the year. Inparticular, there is a general tendency to underestimateblocking intensity, with the notable exception of autumn in the Euro-Atlantic sector. Spring appears to be2002 MONTHLY WEATHER REVIEW VOLUME 1220Z100 908O 706050403020 10 aZONAL ...;.,.., ,,~~'"'":..~... BLO.....C~ "'"-:.,.~ "%.-,-.1 2 3 4 5 6 7 8 9 1 0 'FORECAST DAY100 9080 70 60 50 40 20 10 b Oo - -o -__BLOCKING ~TRANSITIONFORECAST DAY100908O70'605040$02010 CTRANSITIONFORECAST DAYt00- d9080 70' '"..............:.. 60504030- ..B.L.O.C..KJ.Ng .... "..... "-... 10' FOR[CAST DAY Fig. 28. ACC between analysis and forecast as a function of forecast day. The ACC is computed only for the Australiansector, separately for the three categories of mean circulation: zonal (continuous line), blocked (short dashed line), andtransition (dotted line). (a) winter, (b) spring, (c) summer, and (d) autumn.the period of the year during which blocking intensityis best predicted, in both sectors of the Northern Hemisphere. The model performance is in general slightlybetter over the Atlantic sector than it is over the Pacific,indicating that the physical mechanisms responsible forAtlantic and Pacific blocking might be at least partiallydifferent. Large phase (longitude) errors of opposite sign andincreasing with forecast time are evident in modelingthe evolution of blocking, corresponding to a modeleastward shift of Atlantic blocks and to a model westward shift of Pacific blocks. Such phase errors, however, almost vanish during transition seasons (springand autumn) to become very large in winter and summer, in agreement with Tibaldi and Molteni's (1990)results. The preceding conclusions on the adequacy ofthe model's blocking climatology are drawn withoutdistinguishing, for a given forecast day, between verified and unverified blocking occurrences. It is quitepossible that such results could be modified by removing unverified (wrongly forecasted) blocks from thedataset. This, however, would impoverish the databasetoo much, and even more so for late forecast days. Predicting blocking onset is confirmed to be a diffi~ 'cult task for the model, with no clear seasonal cycle orlong-term trend, at least in the 7-yr period analyzedhere, but see the results of Tibaldi et al. (1993) on morerecent winter data. The performance of the model onblocking maintenance, on the contrary, is on averagegood, with a number of outstanding successes in predicting long-lasting blocks and almost as many overpredictions of blocking duration as underpredictions.SEPTEMBER 1994 TIBALDI ET AL. 2003 Regarding blocking and objective skill scores, thecommonly expected situation of best forecasts duringblocking situations, next best during zonal conditions,and worst when a zonal-to-blocking transition occursis actually verified only in winter and autumn Pacificblocking and summer Euro-Atlantic blocking. A second, common case is that forecasts during zonal andblocked situations are equally easy (or equally difficult) to make, while zonal-to-blocking transitions areby far the most difficult situations to predict (the casefor the Euro-Atlantic sector in spring and winter andfor the Pacific sector in spring and summer). The shape and intensity of the synoptic structurestypical of blocking (the blocking signatures) have beenshown to possess an evident seasonal cycle in both sectors of the NH; this seasonal cycle is much weaker inthe SH. Model SEs are described by horizontal spatialstructures that are similar (but with the opposite sign)to those typical of the blocking signature. This is confirmed to be more evident during the NH winter season.During the other seasons, SEs appear to be weaker inintensity. We have also assessed the skill of the model in shortand medium-range forecasts of blocking in the Southem Hemisphere. The comparison of results obtainedfor the Northern Hemisphere shows that the behaviorof the model is very similar in the two hemisphereswith only one important difference: the skill in the prediction of the blocking onset, even for short-range forecasts, is much lower in the Southern Hemisphere thanit is in the Northern Hemisphere. The lower accuracyof the initial analyses (due to the comparative lack ofobservations in the Southern Hemisphere) can partiallyaccount for this result. This idea is strengthened by factthat the overall forecast skill in the Southern Hemisphere is confirmed to be lower than in the other hemisphere. Acknowledgments. This work has been partially sup ported by CNR of Italy, Progetto Finalizzato "Sistemi Informatici e Calcolo Parallelo," and by CEC, EPOCH Programme (C21C Project). The contributions of F. Molteni and R. Mureau of ECMWF in terms of pro viding data, assistance, and useful exchanges of ideas are gratefully acknowledged. F. D'Andrea provided useful comments on an early version of the manuscript. REFERENCES Arpe, K., 1990: Impacts of changes in the ECMWF analysis-fore casting scheme on the systematic error of the model. Proc. ECMWF 1989 Seminars on Ten Years of Medium-Range Weather Forecasting, Reading, Berkshire, United Kingdom, ECMWF, 69-114.Brankovic, C., and L. Ferranti, 1992: Seasonal integrations with re alistic boundary forcing. ECMWF Workshop on New Develop ments in Predictability, Reading, Berkshire, United Kingdom, ECMWF, 305-333.Coughland, M. J., 1983: A comparative climatology of blocking ac tion in the two hemispheres. Aust. Meteor. Mag., 31, 3-13.Dole, R. M., and N. D. Gordon, 1983: Persistent anomalies of the extratropical Northern Hemisphere wintertime circulation: Ge ographical distribution and regional persistence characteristics. Mon. Wea. Rev., 111, 1567-1586.Gronaas, S., 1983: Systematic error and forecast quality of ECMWF forecasts in different large-scale flow patterns. ECMWF Semi nar/Workshop on Interpretation of NWP Products, Shinfield Park, Reading, United Kingdom, ECMWF, 161-206.Lau, L.-C., 1983: Mid-latitude wintertime circulation anomalies ap pearing in a 15-year GCM experiment. Large Scale Dynamical Processes in the Atmosphere, B. J. Hoskins and R. P. Pearce, Eds., Academic Press, 111-125.Lejen~is, H., 1984: Characteristics of Southern Hemisphere blockingas determined from a time series of observational data. Quart.J. Roy. Meteor'. Soc., 110, 967-979.--, and H. Okland, 1983: Characteristics of Northern Hemisphere blocking as determined from a long time series of observational data. Tellus, 35A, 350-362.Miyakoda, K., and J. Sirntis, 1990: Subgrid scale physics in 1-month forecasts. Part II: Systematic error and blocking forecasts. J. Atmos. Sci., 118, 1065-1081.Murphy, A. H., 1991: Forecasts verification: Its complexity and di mensionality. Mon. Wea. Rev., 119, 1590-1601.Okland, H., and H. Lejen~is, 1987: Blocking and persistence. Tellus, 39A, 33-38.Palmer, T. N., and S. Tibaldi, 1988: On the prediction of forecast skill. Mon. Wea. Rev., 116, 2453-2480.Ruti, P., 1992: La rappresentazione delle situazioni di blocco nei modelli di circolazione generale dell'atmosfera. Ph.D. thesis, University of Bologna. [Available from the Department of Physics, University of Bologna, Via lrnerio 46, 40126 Bologna, Italy.]Taljaard, J. J., 1972: Synoptic meteorology of the Southern Hemi sphere. Meteorology of the Southern Hemisphere, Meteor. Mon ogr., No. 13, 139-211.Tibaldi, S., 1993: Low-frequency variability and blocking as diag nostic tools for global climate models. Proc. NATO Advanced Research Workshop on Prediction of lnterannual Climate Vari ation& NATO-ASI, 173-182.--., and F. Molteni, 1990: On the operational predictability of blocking. Tellus, 42A, 343-365. , T. N. Palmer, C. Brankovic, and U. Cubasch, 1991: Extended range predictions with ECMWF models: lnfluence of horizontal resolution on systematic error and forecast skill. Quart. J. Roy. Meteor. Soc., 116, 835-866. - P. Ruti, E. Tosi, and M. Maruca, 1993: Operational predicta bility of winter blocking: An ECMWF update. Proc. ECMWF Seminars on Validation of Forecasts and Large-Scale Simula tions over Europe, Reading, Berkshire, United Kingdom, ECMWF.Tracton, M. S., 1990: Predictability and its relationship to scale in teraction processes in blocking. Mon. Wea. Rev., 118, 1666 1695. , K. Mo, W. Chen, E. Kalnay, R. Klister, and G. White, 1989: Dynamical extended range forecasting (DERF) at the National Meteorological Center. Mon. Wea. Rev., 117, 1604-1635.