isabel franco trigo climatology and interannual ...idlcc.fc.ul.pt/pdf/trigo_climatology_2006.pdf ·...

Isabel Franco Trigo

Climatology and interannual variability of storm-tracksin the Euro-Atlantic sector: a comparison between ERA-40and NCEP/NCAR reanalyses

Received: 19 January 2005 / Accepted: 18 July 2005 / Published online: 23 December 2005� Springer-Verlag 2005

An objective methodology is applied to ERA-40(European Centre for Medium-Range Weather Fore-casts 40-year Reanalysis) and NCEP/NCAR (NationalCenters for Environmental Prediction/National Centerfor Atmospheric Research) reanalyses, to build twostorm-track databases for the Euro-Atlantic sector(85�W–70�E; 20�N–75�N), spanning the period Decem-ber 1958–March 2000. The technique uses the full tem-poral (6-hourly) and spatial resolutions (1.125� and 2.5�regular grids, for ERA-40 and NCEP/NCAR, respec-tively) available. It is shown that the strong discrepan-cies in the number of storms in each dataset (higher forERA-40) result from differences in the resolution of thefields subject to the storm detecting/tracking algorithm,and also from the characteristics of the integrationmodels and assimilation schemes used for each reanal-ysis. An intercomparison of ERA-40 and NCEP/NCARstorm-tracks is performed for spatial distribution, andmain characteristics, of the overall cyclone populationand of a class of severe storms—explosive cyclones.Despite the discrepancies in storm numbers, both rea-nalyses agree on the main cyclone activity areas (for-mation, minimum central pressure, and lysis). The mostpronounced differences occur where subsynoptic systemsare frequent, as these are better resolved by ERA-40data. The interannual variability of cyclone counts,analysed per intensity classes and for different regions ofthe domain, reveals reasonable agreement between thetwo datasets on the sign of trends (generally positive innorthern latitudes, and negative in the Azores-Mediter-ranean band), but discrepancies regarding their strengthin the most southern areas, where the mismatches

between ERA-40 and NCEP/NCAR detected lows aregreatest.

Keywords Storm-tracks Æ Cyclones Æ Climatevariability Æ ERA-40 Æ NCEP/NCAR

1 Introduction

Extra-tropical cyclones, their paths and intensity, havelong been the subject of climatological studies. Thepassage of storms strongly impacts on local weather inthe mid-latitudes. Thus, changes in the dominant storm-tracks, in the frequency, or intensity of cyclones maysignificantly affect surface climate, particularly the pre-cipitation regimes. The earlier compilations of storm-tracks, which include the Euro-Atlantic region, werebased on the visual identification of cyclones, andrespective paths, in synoptic charts (e.g. Petterssen 1956;Klein 1957; Hayden 1981; Agee 1991). The use of grid-ded data, either from observations or available fromdifferent reanalysis projects, gave way to a wide numberof studies applying automated (e.g. Alpert et al. 1990;Blender et al. 1997; Trigo et al. 1999; Sickmoller et al.2000; Hodges et al. 2003; Hanson et al. 2004), or semi-automated (e.g. Gulev et al. 2001) schemes to detect andtrack storms. Other types of methodologies have alsobeen developed to identify storm-tracks. One of the mostcommonly used consists in the analysis of the root meansquare of band-pass filtered fields (generally 1000 or500 hPa geopotential height), retaining variabilitywithin the typical synoptic scales (between 2 and 6, or8 days). Although this methodology has been widelyapplied in the analysis of storm-track variability andtheir relation with large-scale circulation patterns (e.g.Lau 1988; Rogers 1997; Trigo et al. 2000; Sickmolleret al. 2000), it lacks direct measures of cyclone intensityand density.

The analysis of cyclone characteristics and interan-nual variability in the Euro-Atlantic sector presented

Submitted to Climate Dynamics in December 2004

I. F. TrigoDepartamento de Fısica, Faculdade de Ciencias,Instituto de Meteorologia, Lisboa,Centro de Geofısica da Universidade de Lisboa,Lisboa, PortugalE-mail: [email protected]: +351-217-500977

Climate Dynamics (2006) 26: 127–143DOI 10.1007/s00382-005-0065-9

here is based on an automated lagrangean method,which identifies and follows individual lows. The cyclonedetecting and tracking algorithm, first developed byTrigo et al. (1999) for the Mediterranean region, isapplied to geopotential data at 1000 hPa from two 42-year reanalyses datasets: those recently available fromthe European Centre for Medium-Range WeatherForecasts (ECMWF) 40-year Reanalysis (ERA-40); andfrom the National Centers for Environmental Predic-tion/National Center for Atmospheric Research(NCEP/NCAR) Reanalysis project. In both cases, thereanalyses are obtained as analysis of past observations(from conventional and satellite data) using a fixed data-assimilation system, providing the best availableconsistent datasets. Although the quality and coverageof the observations may vary considerably in time, theconsistency of the assimilation scheme guarantees somedegree of homogeneity of data, useful for climatologicalpurposes.

Hodges et al. (2003) and Hanson et al. (2004) havealready carried out intercomparisons of storm-tracksobtained from different reanalyses, namely the NCEP/NCAR reanalyses; and the first version of ECMWFreanalyses, available for 15 years starting in 1979(ERA-15), complemented by the ECMWF operationalanalysis up to 2000 or 2001. In both studies, thecyclone tracking is performed using fields withthe same temporal (6-hourly) and spatial (2.5·2.5�)resolutions. It is well known that the output of featuretracking algorithms is influenced by the resolution ofthe fields used (e.g. Blender and Schubert 2000), andthus, using common spatial and temporal resolutionsmakes the intercomparison of different databaseseasier. However, the smoothing the ERA-15 (orERA-40) fields from its original resolution (about1.125�·1.125�) to that of NCEP/NCAR (2.5�·2.5�)may filter out smaller scale systems, impacting on areaswhere subsynoptic systems prevail, such as the Medi-terranean and Middle East regions (e.g. Trigo et al.2002). In contrast with previous works, here the storm-tracking algorithm is applied to 1000 hPa geopotentialheight fields, at the highest resolution available for eachreanalysis. The intercomparison between the two sets ofstorm-tracks is not focused on the absolute number ofdetected lows, which will be inevitably higher in thehigher resolution dataset, but takes into consideration:(i) the spatial distribution of cyclogenesis, cyclolysis,and of the positions where storms reach their maxi-mum intensity; (ii) the statistical distribution of thecyclones’ main characteristics, including lifetime, andintensity; (iii) the inter-annual variability and trends inthe last 40 years for different categories of stormintensity. Finally, the direct impact of data resolutionon the ERA-40—NCEP/NCAR discrepancies is alsoanalysed, through the comparison of ERA-40 storm-tracks with those that would be obtained using a2.5�·2.5� coarser version of those reanalysis data.

Since very intense cyclones, as well as changes in theseverity of storms may have huge impacts on local

weather, this study also includes the intercomparison ofexplosive cyclones, or bombs (e.g. Sanders and Gyakum1980; Sanders 1986; Lim and Simmonds 2002), detectedin the ERA-40 and NCEP/NCAR datasets. The analysisof cyclones’ interannual variability is performed forintensity categories, defined based on the maximumdeepening rate adjusted to a reference latitude. Thetrends obtained for those classes, and for differentregions within the studied domain, are then comparedfor two reanalyses.

2 Data and methodology

2.1 Re-analyses data

This study is focused on the comparison of two storm-track databases over the Northern Atlantic—Europeanarea, obtained using the ECMWF 40-year reanalyses(ERA-40) and the reanalyses provided by the NCEP/NCAR. In both cases, the storm detection and trackingis performed over 1000 hPa geopotential fields, for theEuro-Atlantic sector, defined here as the area 85�W–70�E, 20�N–75�N. The data are available 6-hourly, andthe total studied period consists of December–Marchseasons between December 1958 and March 2000. TheDecember–March months not only correspond to themost active period of the year in terms of storminess inthe Northern Hemisphere, but also tend to present acoherent interdecadal variability over the Atlantic Re-gion (e.g., Osborn et al. 1999). The ERA-40 (NCEP/NCAR) 1000 hPa height used in this study is availableon a 1.125�·1.125� (2.5�·2.5�) regular grid.

The ERA-40 and NCEP/NCAR reanalyses are pro-duced using a consistent model and assimilation schemethroughout the whole reanalysed period. The ERA-40project uses a spectral model with T159 truncation and60 vertical levels, between the surface and 0.1 hPa, withabout 13 levels within the boundary layer. The hori-zontal and vertical resolutions have been improved whencompared with the previous ECMWF reanalyses (ERA-15), available for the 1979–1993 period only. The use ofoptimal interpolation in ERA-15 has also been replacedby 3-dimensional variational analysis (3D-Var) in themain assimilation system, reinforcing the spatial andtemporal consistency of the new ERA-40 reanalyses (e.g.Simmons 2001). The NCEP/NCAR reanalyses are alsobased on a spectral model, with a 3D-Var assimilationscheme used throughout the whole reanalysed period.However, the model is integrated at a lower resolution ofT62, with 28 vertical levels between the surface andabout 3 hPa, being about five levels within the boundarylayer (ftp://wesley.wwb.noaa.gov/pub/reanal/random_-notes/model).

The model-assimilation systems used for ERA-40 andNCEP/NCAR reanalyses present several differencesregarding not only the spatial resolution, but also thephysical parameterizations (e.g. boundary layer, soil and

128 Trigo: Climatology and interannual variability of storm-tracks in the Euro-Atlantic sector

surface fluxes, radiation and clouds), and the assimilateddata. However, the resolution and the orography rep-resentation are likely to be among the aspects that willinfluence the most cyclone detecting and tracking. Inboth cases the model orography uses the mean orogra-phy at the respective horizontal scales of about 125 km(ERA-40), and 210 km (NCEP/NCAR), and the ERA-40 model also includes a parameterization of subgridorographic variability. Further details about the modeland assimilation schemes used for the ERA-40 may befound in Simmons (2001), and for the NCEP/NCARreanalyses in Kistler et al. (2001) and Kalnay et al.(1996).

Satellite data become available for assimilation intoboth reanalysis systems from 1979 onwards. Impactstudies show that the northern hemisphere is not greatlyaffected by the use of satellite retrievals, in contrast withthe Southern Hemisphere, where conventional observa-tions are sparser (Kistler et al. 2001; Andrae et al. 2004).Since one of the objectives of the present study is fo-cused on the inter-annual variability of storm-tracks inthe Euro-Atlantic area, the longer 42-year commonperiod between 1958 and 2000 will be used here.

2.2 Storm detecting and tracking scheme

A common objective scheme has been applied to the twosets of reanalysed 1000 hPa geopotential data to obtainthe ERA-40 and the NCEP/NCAR storm-track data-bases. The cyclone identification and tracking algorithmconsists essentially on the detection of minima in geo-potential height fields that are eligible for possible stormcentres, and on their tracking by looking for the nearestneighbour in consecutive charts. Such methodology hasbeen widely applied during the last years, both to nearsurface geopotential and to mean sea level pressure fields(e.g. Blender et al. 1997; Serreze et al. 1997; Trigo et al.1999; Sickmoller et al. 2000; Hanson et al. 2004). It iswell known that low troposphere vorticity (e.g. at850 hPa) is less influenced by strong background flows,and thus more suitable for the tracking of smaller scalesystems (e.g. Sinclair 1997). However, vorticity fields aremuch noisier, particularly at the ERA-40 spatial reso-lution; an automated algorithm, similar to that usedhere, would have to be applied to a smoother or coarserversion of ERA-40 voriticity fields (e.g. Sinclair 1997;Hodges et al. 2003).

The 1000 hPa height local minima, identified over a3·3 grid point area, are considered possible cyclonecentres if: (i) the minimum is not higher than 165 gpm(corresponding to a sea level pressure of about1020 hPa); and (ii) the geopotential gradient, averagedover an area of about 10002 km2, is at least 4.5 gpm/100 km (c.a. 0.55 hPa/100 km). Although the algorithmis applied directly to fields of geopotential height at1000 hPa, the cyclone intensity measures will be analy-sed here in pressure units. This allows an easier com-parison with results from previous works, where

minimum pressure, pressure deepening rates, or pressuregradients are generally used. Thus storm variables re-lated with 1000 hPa geopotential (central height, orgradient) are converted to the corresponding mean sealevel pressure values through the hydrostatic balancecondition, assuming a constant surface density(1.23 kg m�3). The thresholds mentioned above wereadjusted empirically for the Mediterranean Region(Trigo et al. 1999, 2002), and subsequently verified todetect and track cyclones over the whole area studiedhere (e.g. Hanson et al. 2004).

The cyclone tracking is based on a nearest neighboursearch in the previous field, within an area defined byimposing thresholds to the maximum cyclone speed of300 km/6 h in the westward direction and of 660 km/6 hin any other. If no cyclone centre is found within thatarea, then cyclogenesis is assumed to have occurred.

Finally, two additional thresholds are applied to thewhole lifecycle of the storms considered in the followinganalysis, in order to eliminate weak systems with littleimpact on local weather: (1) a minimum lifetime of 24 his required; (2) the minimum 1000 hPa geopotentialheight reached by each centre, throughout its lifecycle,must not exceed 82.6 gpm (corresponding to a sea levelpressure of about 1010 hPa).

3 Cyclone climatology

3.1 Spatial distribution

The spatial distribution of the genesis, minimum centralpressure reached by each centre, and cyclolysis are pre-sented in Fig. 1 and 2 for ERA-40, and NCEP/NCARreanalyses, respectively. The terms cyclogenesis andcyclolysis refer to the first and to the last position,respectively, detected for each cyclone. This correspon-dence is reasonably good except near the western (east-ern) border of the domain, where a significant fractionof the high density of cyclogenesis (cyclolysis) events islikely to correspond to high number of storms entering(leaving) the area of study.

The impact of the data resolution on the results ofstorm detecting and tracking algorithms has alreadybeen discussed before (e.g. Blender and Schubert 2000;Zolina and Gulev 2002; Hodges et al. 2003). In thepresent study, both datasets have the same time sam-pling (6-hourly), but different spatial resolution—ERA-40 grid of 1.125�·1.125� against a coarser 2.5�·2.5�NCEP/NCAR grid. Cyclone frequencies, correspondingto the counting of all the positions of each low, cannotbe directly compared for the two datasets; if cyclonetrajectories are not interpolated to common spatial andtemporal resolutions (e.g. Zolina and Gulev 2002), suchcomparison inevitably leads to higher frequencies for thefiner grid, simply because this has more potential pointsto be accounted for. To avoid this problem, Fig. 1 and 2show cyclone counts, i.e., each system is taken into ac-

Trigo: Climatology and interannual variability of storm-tracks in the Euro-Atlantic sector 129

count only once during its lifecycle, fixing either theposition when it is first detected (Fig. 1a, 2a), when itsminimum central pressure is reached (Fig. 1b, 2b), orwhen it is last detected (Fig. 1c, 2c); these values arecomputed per 5�·5� grid box, normalised to the corre-sponding area at 50�N (about 200·103 km2). The den-sity maps of cyclogenesis, together with the spatialdistribution of the location of minimum central pres-sure, and of cyclolysis (i.e., last detected positions) give agood picture of the storm-track distribution in the Euro-Atlantic sector.

Both reanalyses datasets present good agreement ofcyclogenesis spatial patterns (Fig. 1a, 2a). The mostfavourable areas for the formation of new systems in themid-latitudes winter are generally associated with strongsurface baroclinicity, mainly linked to land-sea or sea-icetemperature contrasts, and to local orography (e.g.Trigo et al. 2002; Petterssen 1956). Accordingly, cyclo-genesis maxima are found: (i) near the eastern coast ofthe United States; (ii) to the east of Greenland, whichexhibits one distinct maximum in the NCEP/NCARreanalyses, and two in the ERA-40 dataset; (iii) in theMediterranean Basin, with one maximum over the Gulfof Genoa corresponding to cyclogenesis to the lee of theAlps (Trigo et al. 2002), and another in the neighbour-hood of the Black Sea; (iv) near the Caspian Sea. These

results are generally in line with previous works onnorthern hemisphere storm-tracks, despite the differ-ences in datasets and/or methodologies used (e.g. Sick-moller et al. 2000; Gulev et al. 2001; Hodges et al. 2003;Hanson et al. 2004).

The ERA-40 map presents considerably more centresand details than that of the NCEP/NCAR, particularlyover areas where subsynoptic cyclones tend to prevail.This is the case near the Arctic, with multiple maximaaround Greenland found in the ERA-40 reanalyses, andin the Mediterranean, where the cyclogenesis maximumover the Aegean Sea is not well depicted in the NCEP/NCAR dataset. The Black Sea cyclogenesis found in theERA-40 dataset (e.g. Alpert et al. 1990; Tayanc et al.1998) is displaced further south, to the Anatolian Pla-teau. As will be further discussed in Sect. iv, the dis-crepancies between the two reanalyses storm datasets areparticularly pronounced over the Mediterranean andMiddle East sectors. The relatively small-scale cyclonesfrequent in those areas make them particularly sensitiveto the integration resolution of the model used in thereanalysis process (Hodges et al. 2003).

As previously seen, the spatial distribution of mini-mum central pressure (Pmin, Fig. 1b, 2b) and of cyclol-ysis (Fig. 1c, 2c) for ERA-40 and NCEP/NCARdatasets show very similar patterns, despite the pro-

a b

c

Fig. 1 Spatial distribution of a cyclogenesis (i.e., first detected position for each low), b positions where lows reach their minimum centralpressure, and c cyclolysis (i.e., last detected positions), obtained for ERA-40 reanalysis. The values correspond to the average number ofevents per DJFM season, detected per 5�·5� area normalised for 50�N


nounced differences in the total number of events de-tected. The genesis maximum identified in Fig. 1a and 2aoff the eastern coast of America, over the Gulf Stream,seems to move northeast wards, spreading into the ra-ther diffuse maxima over central Atlantic in Fig. 1b and2b for Pmin, and in Fig. 1c and 2c, for lysis density. Overthe northern part of the Atlantic, both reanalyses pres-ent high rates of Pmin and lysis densities over the Lab-rador Sea, to the west of Iceland, and near theScandinavian coast. Although not shown, those stormsmay have different origins, such as cyclones, which enterthe study domain from the American continent, cyclonesthat form off Cape Hatteras, or cyclones that are alsogenerated near the Greenland coast. Concerning theMediterranean region, the Pmin and cyclolysis distribu-tions present maxima, which tend to be displaced to theeast of those found in the genesis maps. Most winterstorms generated within the Mediterranean tend tomove along the northern coast, where surface barocli-nicity is high, being often reinforced as they move to theAegean sub-basin, or to the Black and Caspian Seas (e.g.Alpert et al. 1990; Trigo et al. 2002).

Figure 3 presents the spatial distribution of explosivecyclogenesis. Following the criterion used by Sandersand Gyakum (1980), storms are classified as explosivecyclones, or bombs, when the central pressure, geo-strophically adjusted to 60�N, decays at least 24 hPa in24 h, i.e.:

dpdtadj

¼ dpdt

sinuRef

sinu� �24 hPa=24 h ð1Þ

where u is the storm latitude, and uRef is 60�N. Lim andSimmonds (2002), following Sinclair (1995) consider-ations on the evolution of cyclones’ central pressures,suggest that the criterion for explosive cyclones shouldtake into account the deepening rate of relative pressurewith respect to its background climatology. This is toavoid the detection of spurious bombs, which in factcorrespond to storms rapidly moving (polewards) tolower pressure areas. The condition described in (i) willbe, however, maintained here, since the sensitivity of thebomb classification to the use of pressure or relativepressure is much smaller in the northern hemisphere,where meridional pressure gradients are less pronouncedthan in the Southern Hemiphere (Lim and Simmonds2002).

Figure 3a and b shows good agreement between thetwo datasets in the predominant formation areas ofexplosive cyclogenesis. The strong thermal contrast be-tween the northern American continent in the winterand the relatively warm Gulf Stream waters give way tothe most favourable area for the occurrence of bombs.The formation region then extends to the mid-Atlanticand to Greenland and Iceland in the north. Althoughnot shown, the density of Pmin positions exhibit twomain maxima, in both reanalysis datasets, one to the

a b

c

Fig. 2 As in Fig. 1, but for NCEP/NCAR reanalysis


southwest of Greenland over the Labrador Sea, and thesecond between Greenland and Iceland, over the Den-mark Strait. These are in accordance with the mainpaths of bombs in the Atlantic (Fig. 4), showing either anortheastward direction, or moving along the northernAmerican coast, deflecting further north towards theLabrador Sea or to the Canadian coast. Despite being apredominantly oceanic phenomenon, explosive cyclonesare rare in the Mediterranean Sea. These cases (such asthe four Mediterranean tracks in the ERA-40 diagramsof Fig. 4 for, but absent in the NCEP/NCAR maps),usually associated with extremely severe weather in theregion, are mostly detected in the ERA-40 reanalysesonly.

Figure 4 shows all bomb-tracks detected in the ERA-40 and in the NCEP/NCAR datasets for two 5-yearperiods, namely 1966–1970 and 1990–1994. Since thoseare representative of two periods dominated by oppositephases of the North Atlantic Oscillation (NAO, e.g.Hurrel 1995; Trigo et al. 2004), Fig. 4 shows a goodsummary of paths in the whole studied period. Thus,

Fig. 4 puts into evidence the main bomb-corridors, andalso gives insight about the inter-annual variability ofbomb-tracks during the last decades. Beside the path-directions mentioned in the previous paragraph, morezonaly-oriented tracks are also detected in the 60 s,when negative NAO pattern prevailed during the winter(e.g. Dickson et al. 2000). By then, more cases ofexplosive cyclogenesis are found at lower latitudes in theAtlantic (e.g. between 30�N and 40�N). During the 90 s,when positive NAO dominated (e.g. Dickson et al.2000), the bomb-tracks are more confined to the eastand northeastern Atlantic. The interannual variability ofstorms, including bombs, in the Euro-Atlantic will befurther discussed in Sect. iv.

3.2 Storm characteristics

Despite the difference in cyclone counts previously de-scribed, the statistical distributions of cyclones’ charac-teristics have several features in common. Panels (a) and

a

b

Fig. 3 Number of explosivecyclogenesis events obtained fora ERA-40 and b NCEP/NCARreanalysis. The valuescorrespond to the averagenumber of cases per DJFMseason, detected per 5�·5� areanormalised for 50�N


(b) of Fig. 5 show the distributions of minimum centralpressure (Pmin), taking into consideration the 1010 hPaupper threshold mentioned in Sect. 2.2, for all detectedsystems (black bars), and for explosive cyclones (greybars). The remaining histograms correspond to totalcyclone duration (Fig. 5c, d), and minimum pressuretendency in 6 h, geostrophically adjusted to 60�N (d p/dt_min; Fig. 5e, f). It should be noted that positive valuesof d p/d t_min are obtained when Pmin occurs at the firstdetected position.

Although Pmin distributions have similar medianvalues of 993 and 990 hPa (see Table 1) for ERA-40 andNCEP/NCAR, respectively, the former dataset tends todetect a higher fraction of weak systems. This leads toincreasing frequencies with central minimum pressure upto about 1005 hPa, while the NCEP/NCAR distributiontends to peak around 1000 hPa. The long left tail of thedistribution is similar between the two datasets, al-though the ERA-40 reanalyses also pick slightly moreextreme intense cases (typically deeper than 940 hPa),than the NCEP/NCAR. The Pmin histograms of explo-sive cyclones exhibit distributions closer to the normal,with medians around 964 and 965 hPa (Table 1) for theERA-40 and NCEP/NCAR, respectively. The range ofPmin values detected in the ERA-40 dataset is slightlywider than those found for the NCEP/NCAR. The 95%percentiles, corresponding to explosive cyclogenesis de-

tected at lower latitudes, are of the order of 987 and985 hPa, respectively, while the 5% percentiles are about943 and 945 hPa for the ERA-40 and NCEP/NCAR,respectively (Table 1).

Accordingly, with the higher frequency of weakerlows detected in the ERA-40 dataset discussed above,the frequency of systems living up to 2 days is about 8%higher than that in the NCEP/NCAR data (Fig. 5c, d).In accordance with previous works (Serreze et al. 1997;Trigo et al. 1999; Gulev et al. 2001; Hanson et al. 2004),the distribution of cyclones duration obtained for bothdatasets peaks around 2 days, decaying rapidly thenumber of systems lasting longer than 3 days; theaverage duration of the detected cyclones is of the orderof 2.2 days and 2.5 days, for the ERA-40 and NCEP/NCAR reanalyses, respectively. The lifetime of bombsexhibit distributions that tend to be more symmetricaround the mean, in strong contrast with the highlyskewed ones obtained for the overall cyclone popula-tion. Smaller scale cyclones are likely to be better re-solved in the higher resolution ERA-40 reanalyses.Accordingly, the frequency of shorter living bombs isalso slightly higher in the ERA-40 data, with an averageduration of 3.8 days, than in the NCEP/NCAR, with anaverage of 4.2 days.

The distributions of maximum deepening in 6 h, i.e.,d p/d t_min, are shown in Fig. 5e and f. As expected (e.g.

a b

c d

Fig. 4 Detected tracks of explosive cyclones for two 5-year periods, obtained for the two reanalysis datasets under study: a 1966–1970 forERA-40; b 1966–1970 for NCEP/NCAR; c 1990–1994 for ERA-40; d 1990–1994 for NCEP/NCAR


920 940 960 980 1000 10200

5

10

15

Fre

quen

cy

(%

)

Pmin

(hPa)

NCEP/NCAR__DJFMCyclonesBombs

1 2 3 4 5 6 7 8 9 100

10

20

30

40F

requ

ency

(

%)

Duration (days)

–24 –20 –16 –12 –8 –4 0 40

5

10

15

20

25

Fre

quen

cy

(%

)

dp/dtmin

(hPa/6h)

920 940 960 980 1000 10200

5

10

15

Fre

quen

cy

(%

)

Pmin

(hPa)

ERA 40__DJFM

1 2 3 4 5 6 7 8 9 100

10

20

30

40

Fre

quen

cy

(%

)

Duration (days)

–24 –20 –16 –12 –8 –4 0 40

5

10

15

20

25

Fre

quen

cy

(%

)

dp/dtmin

(hPa/6h)

a b

c

e f

d

Fig. 5 Statistical distributions of minimum central pressure (hPa), total cyclone duration (days), and minimum pressure tendencygeostrophically adjusted to 60�N (hPa/6 h) for ERA-40 (panels a, c, and e, respectively), and NCEP/NCAR data (panels b, d, and f,respectively). The black (grey) bars correspond to the overall (explosive) cyclones population


Serreze et al. 1997; Gulev et al. 2001), they exhibit longleft tails, with maximum observed frequencies between�4 and �2 hPa/6 h, respectively. The left tails of thesedistributions correspond to a large extent, to explosivecyclogenesis, particularly for values below about�10 hPa/6 h. Regarding such extreme cases, the ERA-40 distribution presents lower d p/d t_min values, thanthe NCEP/NCAR, with 5% percentiles of the order of�19 and �16 hPa/6 h, respectively (Table 1).

The distributions of storm characteristics vary con-siderably within the studied domain. As an example,Fig. 6 shows the distributions of Pmin, lifetime, and dp/dt_min for storms reaching their intensity peak (mini-mum pressure) within the northern Atlantic (85–20�W;55�–75�N), and within the Mediterranean (20�W–45�E;32.5–52.5�N) sectors. For simplicity, only ERA-40 re-sults are shown, since the contrasts between the twoareas obtained using the NCEP/NCAR database aresimilar. The northern Atlantic Pmin distribution peaks atlower pressures, in line with results obtained in previousworks (e.g. Gulev et al. 2001). The estimated medianvalue is about 986 hPa, while that obtained for theMediterranean is of the order of 1001 hPa. The Pmin

distribution of the former area exhibits a considerablylonger left tail, which results in a higher standard devi-ation (16 hPa in the Atlantic, against 9 hPa obtained forthe Mediterranean). The higher spread of the distribu-tion in the Atlantic region when compared with theMediterranean is also generally observed for other stormcharacteristics. Notably the d p/d t_min histograms(Fig. 6e, f), which exhibit very similar median values(�3.9 and �3.2 hPa/6 h for the Atlantic and Mediter-ranean, respectively), show considerably more cases ofvery rapid deepening in the Atlantic. There the fractionof storms with rates below �11 hPa/6 h (values geo-strophically adjusted to 60�N) reaches 10%, comparedwith only 1% in the Mediterranean area. The averagelifetime estimated for the northern Atlantic (Mediterra-nean) is of the order of 2.5 days (2.1 days). Althoughthese values are in line with those obtained in otherstudies (e.g. Gulev et al. 2001), the total storm durationmight be slightly underestimated since the tracking is

performed over a limited area and not over the entirehemisphere.

3.3 Impact of data resolution on storm-tracking

It is generally accepted that cyclone tracking method-ologies applied to datasets with different resolutionsyield different tracks. To distinguish the discrepanciesbetween the ERA-40 and NCEP/NCAR datasets, whichcan be accounted for by the different spatial resolutions,from those that result from intrinsic characteristics ofthe datasets, the cyclone detection and tracking algo-rithm, was applied to a degraded version of the ERA-40fields. The ERA-40 geopotential height at 1000 hPais re-sampled to the NCEP/NCAR 2.5�·2.5� grid,using the nearest grid point from the original1.125�·1.125�grid. The track comparison methoddeveloped by Blender and Schubert (2000) was thenapplied to the cyclone tracks obtained from the coarseERA-40 fields and to the NCEP/NCAR storm-tracks(test sets), using the results from the ERA-40 originalgrid as the reference set. The method is based on thecomputation of a metric distance between the storm-paths found in the reference and test sets, allowing theestimation of the number of matched tracks (Npairs) inthe reference and test storm databases. Simple statisticsmay then be derived, such as: (i) the probability ofdetection (or matching probability) as the ratio betweenNpairs and the number of tracks in the respective test set(Ntest); (ii) the fraction of missing tracks (compared herewith the paths in the reference set, Nref), given by the(Nref - Npairs)/Nref ratio.

The results obtained for the matching probability andthe percentage of missing tracks are summarised in Ta-ble 2 for the comparison between the original ERA-40dataset with its lower resolution version (ERA-Lo-wRes), and with the NCEP/NCAR data, respectively.Those values are estimated twice, first taking into con-sideration all the storm-tracks under analysis in thisstudy, and then considering explosive cyclones only. Thecomparison between the two ERA-40 resolution ver-

Table 1 5%, 50% (median), and 95% percentiles of minimum central pressure (Pmin), minimum pressure tendency in 6 h, adjusted to60�N (d p/d t_60�), and cyclone duration

ERA-40 NCEP/NCAR

Whole cyclone population

5% Median 95% 5% Median 95%

Pmin (hPa) 960 993 1008 959 990 1008Duration (day) 1 1.8 5.0 1.3 2.0 5.8dp/dt_60� (hPa/6 h) �11.3 �3.4 +0.7 �10.9 �3.6 +0.4

Explosive cyclones5% Median 95% 5% Median 95%

Pmin (hPa) 943 964 987 945 965 985Duration (day) 1.5 3.5 7.0 2.8 4.0 7.8dp/dt_60� (hPa/6 h) �19.1 �11.2 �7.9 �15.8 �10.7 �7.8

The values are estimated for the overall population, and for explosive cyclones, for ERA-40 and NCEP/NCAR reanalyses, respectively


920 940 960 980 1000 10200

5

10

15

20

25

30a b

dc

e f

Fre

quen

cy

(%

)

Pmin

(hPa)

AtlanticN__DJFM

1 2 3 4 5 6 7 8 9 100

10

20

30

40

Fre

quen

cy

(%

)

Duration (days)

–24 –20 –16 –12 –8 –4 0 40

5

10

15

20

25

30

Fre

quen

cy

(%

)

dp/dtmin

(hPa/6h)

920 940 960 980 1000 10200

5

10

15

20

25

30

Fre

quen

cy

(%

)

Pmin

(hPa)

Mediterranean__DJFM

1 2 3 4 5 6 7 8 9 100

10

20

30

40F

requ

ency

(

%)

Duration (days)

–24 –20 –16 –12 –8 –4 0 40

5

10

15

20

25

30

Fre

quen

cy

(%

)

dp/dtmin

(hPa/6h)

Fig. 6 Statistical distributions of minimum central pressure (hPa), total cyclone duration (days), and minimum pressure tendency (hPa/6 h) obtained using ERA-40 reanalyses, for two regions within the studied domain: the Northern Atlantic, ranging form 85 to 20�W, andfrom 55 to 75�N (panels a, c, and e, respectively); and the broad Mediterranean region, ranging from 20�W to 45�E, and from 32.5� to52.5�N (panels b, d, and f, respectively)


sions produces statistics very similar to those obtainedby Blender and Schubert (2000), for similar decrease inthe spatial resolution of the reference dataset. Theprobability of finding matching tracks in the ERA-Lo-wRes dataset is of the order of 0.84, only slightly higherthan that obtained for the NCEP/NCAR data. Thepercentage of missing tracks, however, is considerablyhigher in the NCEP/NCAR case. One of the reasons forsuch discrepancies in the estimation of missing tracksmay be associated with the total number of storm-tracksfound in each dataset, which in the case of ERA-Lo-wRes (NCEP/NCAR) is about 80% (50%) of the thosefound in the ERA-40 original resolution.

The results shown in Table 2 put into evidence thatthe discrepancies between the two reanalysis datasetscannot be attributed only to the different resolutions ofthe fields used for the cyclone detecting and trackingalgorithm. As mentioned before (and also in Hodgeset al. 2003; Hanson et al. 2004), the resolution of themodels integrated for each reanalysis is likely to play animportant role in resolving systems in the ERA-40 data,which are not represented in the NCEP/NCAR. Thatmight be particularly the case of relatively small-scalecyclones, lows remaining relatively week throughouttheir lifecycles, or of secondary centres within complexsystems. This seems to be confirmed by the results ob-tained from the comparison of explosive cyclones, whichare intense systems, often larger than the average size forthe whole cyclone population (Lim and Simmonds2002). Accordingly, the probability of matching ishigher, close to 0.9 for ERA-LowRes and NCEP/NCAR, and the fraction of missing tracks is also lowerin both cases, than that found for the overall storms.However, some of the bomb-tracks missing in the low-resolution datasets correspond to Mediterranean (sub-synoptic) cases, which are only detected in the fullresolution ERA-40 dataset.

4 Interannual variability

4.1 General overview

The discrepancies between the two storm-track data-bases involve two main factors—(i) the spatial resolutionupon which the detecting and tracking algorithm is ap-plied; (ii) the intrinsic differences between the two rea-nalyses, including the resolution of the two models used,the observational data used, and the assimilation

schemes. In the previous section, we have seen that al-though both factors have to be taken into account, theresolution of the model integrated for ERA-40 reanal-ysis is likely to be an important contributor for thehigher number of storm-tracks obtained using thatdataset. Here, we will analyse and compare the inter-annual variability of cyclone activity within each data-base.

Figure 7 shows the decadal relative trends of cyclonenumbers aggregated within 9� long·9� lat cells for theERA-40, and within 10� long·10� lat cells for theNCEP/NCAR dataset. The geographical position ofthose boxes, which correspond to 8·8 (4·4) grid pointsat the ERA-40 (NCEP/NCAR) spatial resolution, issufficiently close to allow the intercomparison of the twofields. The cells representing trends in cyclone numberssignificant at less than 10% (using the Mann–Kendalltest for the null hypothesis of no trend, e.g. Hirsch andSlack 1984) are also indicated.

Although the intensity and significance of the trendsobtained for ERA-40 and NCEP/NCAR are not iden-tical, the overall patterns are very similar. Both showtwo long, nearly zonal, bands with opposite behaviours;one exhibiting positive trends spanning from the Lab-rador Sea to Scandinavia, and the second, showing adecrease in the number of cyclones, ranging from themid-Atlantic around the Azores, to Central Europe/Northern Mediterranean, up to the Caspian Sea. Thelocation of the main storm paths along the northernAtlantic is highly modulated by the North AtlanticOscillation (NAO), e.g. Lau (1988), Rogers (1990, 1997),and Trigo et al. (2004). Thus, much of the changes in thefrequency of cyclones shown in Fig. 7 for the Atlantic,Europe and Scandinavia are clearly linked to the well-documented trends in the phase of the NAO. During thestudied period, the NAO remained predominantly neg-ative until the 1970 s (e.g. van Loon and Williams 1976;Hurrel 1995; Dickson et al. 2000), while the followingdecades were dominated by a sharp increase in the NAOindex, which maintained a predominantly positive phasethrough most of the 1990 s (e.g. Hurrel 1995; Dicksonet al. 2000; Marshall et al. 2001). Accordingly, the pat-terns shown in Fig. 7 suggest a northward shift of thestorm-tracks in the Euro-Atlantic sector (also identifiedby, e.g. Schmith et al. 1998; Ulbrich and Christoph 1999;Sickmoller et al. 2000; Dickson et al. 2000). A directconsequence is the sharp decrease in the number of cy-clones affecting the Iberian Peninsula, and also Centraland Southern Europe, leading to the decline in winter

Table 2 Probability of matching tracks, and fraction of missing tracks (%) obtained when the storm-tracks detected in the full resolutionERA-40 data are compared with: storms in the ERA-40 lower resolution (2.5·2.5�) version (ERA-LowRes); the NCEP/NCAR storms

All cyclones Explosive cyclones

ERA—LowRes NCEP/NCAR ERA—LowRes NCEP/NCAR

Matching_probability 0.84 0.82 0.90 0.87Tracks_missing (%) 34% 58% 29% 51%

The results are presented for the overall and explosive cyclone population


precipitation over parts of Europe and the Mediterra-nean (e.g. Trigo et al. 2000; Alpert et al. 2002; Xoplakiet al. 2004; Trigo et al. 2004).

The signs and location of the decadal trends of cy-clone numbers shown in Fig. 7 are fairly coincident forthe ERA-40 and NCEP/NCAR databases; the area ofnorthern Africa is among the few exceptions, althoughthis sector is characterised by relatively low cyclonefrequencies (Fig. 1, 2). Nevertheless, the intensity andsignificance of those trends may differ significantly be-tween the two data sets. The trend pattern over CentralEurope/Northern Mediterranean is rather homogeneousin the NCEP/NCAR case, with constant values of theorder of �12% to �14% per decade between the Gulf ofBiscay and the Black Sea. The ERA-40 exhibits a clear

distinction between the Western and the Eastern(northern) Mediterranean, with decadal trends of theorder of �13% to �19% over Iberia and the Gulf ofBiscay, and �9% around the Black Sea. The NCEP/NCAR—ERA-40 contrast is even stronger over theAsian sector, between the Black and Caspian Seas.There the relatively smooth negative trends of the ERA-40 dataset are replaced by decreasing rates, which mayreach values of the order of �25% for the NCEP/NCAR. Although such discrepancies between the twosets of storm-tracks are not as marked over high lati-tudes, the positive trends obtained between Greenlandand Scandinavia present slightly higher values for theERA-40 (around 10–12%), than for the NCEP/NCARcase (of the order of 6%–9%).

a

b

Fig. 7 Decadal relative changes in DJFM cyclone numbers, aggregated within 9�·9�, and 10�·10� grid boxes, for a ERA-40, and bNCEP/NCAR data, respectively. Boxes with significant trends, at less than the 10% level, are marked with filled triangle, for positive, and withfilled square, for negative trends


4.2 Storm frequency time-series

The analysis of the interannual variability of storm-tracks in the two datasets will now be focused on thespecific areas shown in Table 3. These were definedtaking into account the spatial distribution of cyclonetrends shown in Fig. 7, ensuring fairly homogeneousregions in terms of changes during the last decades. Acyclone is considered to be part of one of those areas ifits minimum central pressure is reached within that re-gion. This classification now guarantees that each stormis accounted for only once, and also that it is consideredduring its intensity peak.

Figures 8 and 9 show the time-series of the totalnumber of storms per December–March season (NCYC,hereafter), which reach the minimum central pressurewithin the respective area (top curve in each panel), forERA-40 and NCEP/NCAR, respectively. The remainingcurves correspond to the seasonal number of stormswith d p/d t_min (minimum pressure tendency in 6 h,geostrophically adjusted to 60�N) below the 50%(NCYC_50) and 25% (NCYC_25) percentiles, respec-tively, estimated for each region separately. The maxi-mum deepening rate is generally directly linked to thegeneration of local vorticity maxima, and thus to thestorm severity; in a sense, using percentiles of d p/dt_min also works as an extension of the bomb criterion,mentioned earlier.

Despite the discrepancies in storm numbers (whichvary greatly among the studied regions), the coherenceof the interannual variability of corresponding time-series from the ERA-40 and NCEP/NCAR datasets isfairly consistent, with the exception of the Asian region.The significance of the trends of each represented curve,obtained through the Mann–Kendal test (e.g. Hirschand Slack 1984) is also shown in Table 3. As a whole,the ATL-S (Fig. 8b, 9b) area presents rather weak neg-ative trends in the ERA-40 database, with statisticalsignificances no lower than 10%. In the NCEP/NCARdata, such negative tendencies are even weaker. Thecorrelation between corresponding time-series from the

two reanalyses is fairly high, ranging from 0.67 to 0.69for the NCYC _50 and NCYC_25 time-series, respec-tively, increasing to 0.75 for the total number of storms(NCYC).

The NCYC, NCYC _50, and NCYC_25 time-seriesobtained for the ATL-N (Fig. 8a, 9a) and Med-N(Fig. 8c, 9c) regions exhibit very similar interannualvariability between the two reanalyses. The positive(negative) trends found for ATL-N (Med-N) are highlysignificant, with p-levels lower than 7%; only the higherintensity cases in the Med-N area, represented here byNCYC_25 time-series, show a weaker tendency in ERA-40 data, significant only at the 12% level. Accordingly,the correlations between corresponding NCYC andNCYC_50 time-series are of the order of 0.83 and 0.74(0.90 and 0.74), respectively, for the ATL-N (Med-N).The pronounced changes found for the ATL-NNCYC_25 in both reanalyses are likely to explain theirhigh correlation coefficient, of 0.87, while the higherdiscrepancies found for the Med-N NCYC_25 result in alower correlation of 0.58.

The most intense systems in the Med-N area, par-ticularly those which develop over the MediterraneanSea are frequently of subsynoptic scale (e.g. Alpert andNeeman 1992; Trigo et al. 1999, 2002), and thus alsomore sensitive to the resolution of the reanalysed data.The high discrepancies between the Asian storm time-series (Fig. 8d, 9d) in the two reanalysis data are alsolikely to be a consequence of the typical scales of thesystems affecting that area. These are also frequentlymesoscale cyclones (e.g. Alpert and Neeman 1992), andthus difficult to be fully detected and characterised in theNCEP/NCAR coarser database. The NCEP/NCARtime-series for the Asia region present negative trends,although only the NCYC is statistically significant at lessthan 10%. The ERA-40 time-series have very differentbehaviours; the total number of storms has beendecreasing, although not significantly, during the lastfour decades, while the time-series of the most intenseevents (both NCYC_50 and NCYC_25) present signifi-cant positive increases, at less than 5%.

Table 3 Signs of the linear trends in the 1959–2000 period and, below, the respective statistical significance according to the Mann–Kendal test, obtained for ERA-40 and NCEP/NCAR storm numbers, within the areas indicated

Area Trends

ERA-40 NCEP/NCAR

NCYC NCYC_50 NCYC_25 NCYC NCYC_50 NCYC_25

Atlantic-South (ATL-S) 85–20�W (-) (-) - · · (-)37.5–50�N (12%) (11%) 10% (14%)

Atlantic-North (ATL-N) 85–20�W + + + + + +55–75�N <1% <1% 2% 3% 7% 2%

Northern Mediterranean (MED-N) 20�W–40�E - - (-) - - -37.5–52.5�N 2% 4% (12%) <1% 1% <1%

ASIA 40–70�E · + + - · ·25–52.5�N 2% 2% 9%

NCYC represents the time-series of the number of lows reaching their minimum central pressure within each area; NCYC_50 (NCYC_25)time-series correspond to the fraction of NCYC, which have minimum pressure tendencies below the 50% (25%) regional centile


1960 1970 1980 1990 20000

50

100

150a

c d

bATLANTIC–NORTH

No.

of S

torm

s / D

JFM Total

p50%

p25%

1960 1970 1980 1990 20000

10

20

30

40

50

60ATLANTIC–SOUTH

No.

of S

torm

s / D

JFM

Total

p50%

p25%

1960 1970 1980 1990 20000

20

40

60

80NORTH–MED

No.

of S

torm

s / D

JFM

Total

p50%

p25%

1960 1970 1980 1990 20000

10

20

30

40

50

60

70ASIA

No.

of S

torm

s / D

JFM

Total

p50%

p25%

ERA 40

Fig. 8 Time-series, and respective linear trends, of the number of storms, which reach their minimum central pressure, within the aNorthern Atlantic (85�–20�W; 55�–75�N), b Southern Atlantic (85�–20�W; 37.5�–50�N), c Northern Mediterranean (20�W–40�E; 37.5�–52.5�N), and d Asia (40�–70�E; 25�–52.5�N) region. The three time-series in each panel correspond to: all identified cases (total); stormswith a minimum pressure tendency lower than the regional median (p50%), and lower than the regional 25% percentile (p25%), for theERA-40 database

1960 1970 1980 1990 20000

50

100

150a

c d

bATLANTIC–NORTH

No.

of S

torm

s / D

JFM

Total

p50%

p25%

1960 1970 1980 1990 20000

10

20

30

40

50

60ATLANTIC–SOUTH

No.

of S

torm

s / D

JFM

Total

p50%

p25%

1960 1970 1980 1990 20000

20

40

60

80NORTH–MED

No.

of S

torm

s / D

JFM

Totalp

50%p25%

1960 1970 1980 1990 20000

10

20

30

40

50

60

70ASIA

No.

of S

torm

s / D

JFM

Totalp

50%p25%

NCEP/NCAR

Fig. 9 As in Fig. 8, but for NCEP/NCAR reanalysis


5 Discussion and conclusions

A comparison of two 42-year databases of storm-tracks,derived from ERA-40 and NCEP/NCAR reanalyses hasbeen performed here, taking into consideration severalaspects, such as: (i) the spatial distribution of cyclo-genesis and tracks of the overall cyclone population, andof explosive cyclones or bombs; (ii) the distribution ofstorm characteristics, namely, total lifetime, minimumpressure, and maximum deepening rate; (iii) the inter-annual variability of storm frequency within the studiedarea. The determination of storm trajectories is based onthe search for minima in 1000 hPa geopotential heightfields, which are then tracked by searching the nearestneighbours in the next chart; a set of empirical thresh-olds for cyclone minimum gradient, maximum centralheight, maximum velocities, and minimum duration(24 h) are also imposed, as described in Sect. 2.2, andfollowing the storm-tracking algorithms used by Blenderet al. (1997), Serreze et al. (1997), Trigo et al. (1999), andHanson et al. (2004).

In recent years, a relatively small number of authorshave focused their research on the inter-comparison ofstorm-tracks derived from different reanalysis datasets,including NCEP/NCAR, and the previous ECMWFreanalysis, ERA-15 (e.g. Hodges et al. 2003; Hansonet al. 2004). Here, an updated version (with improvedresolution of the integration model, and improvedassimilation scheme) of the ECMWF reanalyses,available for a longer period (ERA-40) is used, allow-ing the inter-comparison to be performed for aboutfour decades, between 1958 and 2000. The cyclonetracking is performed using the full resolution1000 hPa height available for each reanalysis dataset,i.e., 6-hourly fields at 1.125�·1.125� (2.5�·2.5�) forERA-40 (NCEP/NCAR). The differences in the spatialresolution of geopotential fields lead to discrepancies inthe two derived sets of storm-tracks. It has beenshown, however, that factors other than field resolutionhave to be taken into account. The fraction of ERA-40tracks, which are missed if a coarser version (2.5·2.5�)of that dataset is used, is about 34%; this value rises to58% if the NCEP/NCAR data are used instead. Thehigher resolution of the model integrated for ERA-40,including the better resolved model orography, is likelyto favour the detection of relatively small-scale centres,which may not be resolved in the NCEP/NCAR data.Some of these centres may be of sub-synoptic scales, orcorrespond to secondary centres within complex sys-tems. However, their detection is important, as theirinterannual variability may have strong impacts onlocal weather, namely as main contributors to thespatial distribution, or decadal trends observed inprecipitation. Moreover, it is shown that the highernumber of storms detected in the ERA-40 dataset doescorrespond to wider distribution ranges of cyclonecharacteristics, particularly of those regarding intensity.Although the distributions of minimum central pres-

sure and maximum deepening rate exhibit similarshapes for both datasets (Fig. 5), the storm populationobtained from ERA-40 includes both deeper andweaker cases, than the overall population of NCEP/NCAR cyclones.

Despite the discrepancies in the total number ofstorms detected in ERA-40 and NCEP/NCAR reanal-yses, the main patterns of the genesis, maximum deep-ening position, and lysis are in good agreement (Fig. 1,2). Both datasets depict well the location of the maincyclogenesis maxima (Fig. 1a, 2a), namely: off the east-ern coasts of the United States and Greenland, in theAtlantic; the western basin of the Mediterranean Sea;the high activity area over the Caspian Sea. However,ERA-40 spatial distributions of cyclogenesis, maximumdeepening position, and cyclolysis present more centresand details, particularly where subsynoptic systems tendto be more frequent. As a result, areas such as the Ae-gean Sea (in the Eastern Mediterranean Basin), theBlack Sea, the Middle East (region to the south-east ofthe Caspian Sea), and the Artic exhibit multiple maxima,which are either absent, or very weak in the NCEP/NCAR reanalyses. On the other end of the spectrum, thetwo datasets show a better agreement in the detection ofexplosive cyclones (Fig. 3; Table 2). These are deepsystems, which occur predominantly over the AtlanticOcean, and which are often larger than the overallpopulation (Lim and Simmonds 2002); the smaller scaleMediterranean bombs are far less frequent, and aremostly detected in ERA-40 data (Fig. 4).

The major features of cyclone trend maps (Fig. 7)obtained for the last four decades are common to ERA-40 and NCEP/NCAR reanalyses. These include twozonal bands, one exhibiting an overall increase ofstorms, ranging from the Labrador Sea to Scandinavia,and a second band, dominated by decreasing cyclonecounts, which ranges from the Azores to Central Europeand the Mediterranean. Such pattern, suggesting anorthward shift of storm-tracks (also found by Schmithet al. 1998; Ulbrich and Christoph 1999; Sickmoller et al.2000), mirrors the fluctuations observed during the last40-year of the main mode of atmospheric variability inthe Euro-Atlantic sector—the NAO (e.g. Hurrel 1995;Marshall et al. 2001).

A closer inspection of cyclone count time-series,analysed by intensity classes and for different regionswithin the studied domain (Fig. 8, 9), reveals reasonableagreement between ERA-40 and NCEP/NCAR datasetson the signs of trends, but discrepancies regarding theirstrength. Over the Northern Atlantic (Fig. 8a, 9a),however, both reanalyses point towards an increase inthe number of storms, associated to increasing fre-quencies of moderate and of intense cyclones. These arein line with observed precipitation increase, of bothmean values and intensity rates, in parts of northernEurope, particularly in the British Isles, and in Nordiccountries (Førland et al. 1998; Osborn et al. 2000).

The stronger mismatches in the storm frequencytime-series are found for the so-called Asia region of the


domain (to the east of 40�E), for which ERA-40 indi-cates a strong increase of moderate-to-intense storms,despite the overall decline suggested by the NCEP/NCAR data (Fig. 8d, 9d; Table 3). The comparison ofstorm frequencies in the Northern Mediterranean(Fig. 8c, 9c) reveals good agreement in the variability ofthe total number of cyclones, with high significanttrends. The decline of the most intense cases, however, isconsiderably less pronounced for the ERA-40 (about6.5% per decade), than for the NCEP/NCAR reanalyses(of the order of 23% per decade). Only in the second halfof the studied period (from 1979 onwards) do the ERA-40 intense storms have a significant decrease (at p<6%),as also reported in Trigo et al. (2000) for the ‘‘wet sea-son’’ months, between October and March. The impactof such trends on the local weather of the Mediterraneanand Middle East, regions particularly sensitive to globalclimate change (Hulme et al. 1999), is likely to be ex-tremely high. There are several studies indicating ageneral decline in winter precipitation, particularly in thenorthern Mediterranean Basin (e.g. Trigo et al. 2000;Alpert et al. 2002; Xoplaki et al. 2004), likely to beassociated with the decrease in storm frequency for thatarea. However, there is growing evidence of an increaseof precipitation intensity in parts of Europe, includingGreat Britain (Osborn et al. 2000), and the Mediterra-nean (Brunetti et al. 2001a, b; Alpert et al. 2002). Al-though the non-significant decline in intense storms forthe Northern Mediterranean (and increase in the Asiaregion) obtained for ERA-40 fits better to these findings,than the NCEP/NCAR respective time-series, otherfactors are likely to play important roles, such as aprobable higher moisture atmospheric content driven byglobal warming (Osborn et al. 2000; Brunetti et al.2001a).

There is little doubt that the inter-annual variabilityin the location and intensity of storm-tracks has hugeimplications on local weather. The study presented hereshows that ERA-40 and NCEP/NCAR reanalyses agreewell on track changes, which can be directly linked tolarge scale variability modes, such as the NAO. Dis-crepancies arise when comparing frequency and inten-sity of smaller scale low centres. Further investigation ontheir link with recent changes in rainfall, and rainfallintensity, may be useful in understanding possibleundergoing changes in precipitation regimes in manyparts of Europe.

Acknowledgement The NCEP/NCAR Reanalysis data was kindlyprovided by the Climatic Research Unit. The author is grateful toRicardo Trigo for his helpful comments. This work was supportedby the Portuguese Science Foundation (FCT) through projectVAST (Variability of Atlantic Storms and Their impact on landclimate) POCTI/CTA/46573/2002.

References

Agee EM (1991) Trends in cyclone and anticyclone frequency andcomparison with periods of warming and cooling over theNorthern Hemisphere. J Clim 4:263–267

Alpert P, Neeman BU (1992) Cold small-scale cyclones over theeastern Mediterranean. Tellus 44A:173–179

Alpert P, Neeman BU, Shay-El Y (1990) Climatological analysis ofMediterranean cyclones using ECMWF data. Tellus 42A:65–77

Alpert P, Ben-Gai T, Baharad A, Benjamini Y, Yekutieli D, Co-lacino M, Diodato L, Ramis C, Homar V, Romero R, Mi-chaelides S, Manes A (2002) The paradoxal increase ofMediterranean extreme daily rainfall in spite of decrease in totalvalues. Geophys Res Lett 29:DOI 10.1029/2001GL013554

Andrae U, Sokka N, Onogi K (2004) The radiosonde temperaturebias correction used in ERA-40 ERA-40 Project Report SeriesNo. 15, Reading, MA, pp 1–34

Blender R, Fraedrich K, Lunkeit F (1997) Identification of cyclonetrack regimes in North Atlantic. Q J R Meteorol Soc 123:727–741

Blender R, Schubert M (2000) Cyclone tracking in different spatialand temporal resolutions. Mon Weather Rev 128:377–384

Brunetti M, Colacino M, Maugeri M, Nanni T (2001a) Trends inthe daily intensity of precipitation in Italy from 1951 to 1996.Int J Climatol 21:299–316

Brunetti M, Maugeri M, Nanni T (2001b) Changes in total pre-cipitation, rainy days and extreme events in Northeastern Italy.Int J Climatol 21:861–871

Dickson RR, Osborn TJ, Hurrell JW, Meincke J, Blindheim J,Adlandsvik B, Vinje T, Alekseev G, Masloske W (2000) TheArtic Ocean response to the North Atlantic Oscillation. J Clim13:2671–2696

Førland EJ, Alexandersson H, Drebs A, Hanssen-Bauer I, VedinH, Tveito OE (1998) Trends in maximum 1-day precipitation inthe Nordic Region. DNMI Report 14/98 klima, NorwegianMeteorological Institute, Oslo, pp 1–55

Gulev SK, Zolina O, Grigoriev S (2001) Extratropical cyclonevariability in the Northern Hemisphere winter from the NCEP/NCAR reanalysis data. Clim Dyn 17:795–809

Hanson CE, Palutikof JP, Davies TD (2004) Objective cycloneclimatologies of the North Atlantic—a comparison between theECMWF and NCEP reanalyses. Clim Dyn 22:757–769; DOI10.1007/s00382-004-0415-z

Hayden BP (1981) Secular variation in Atlantic Coast extratropicalcyclones. Mon Weather Rev 109:159–167

Hirsch RM, Slack JR (1984) A non-parametric trend test for sea-sonal data with serial dependence. W Resour Res 20(6):727–732

Hodges KI, Hoskins BJ, Boyle J, Thorncroft C (2003) A com-parison of recent reanalysis using objective feature tracking:storm tracks and easterly waves. Mon Weather Rev131:2012–2037

Hulme M, Barrow EM, Arnell NW, Harrison PA, Johns TC,Downing TE (1999) Relative impacts of human-inducedclimate change and natural climate variability. Nature397:688–691

Hurrel JW (1995) Decadal trends in the North Atlantic Oscillation:regional temperatures and precipitation. Science 269:676–679

Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, GandinL, Iridell M, Saha S, White G, Woolen J, Zhu Y, Chelliah M,Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropolewski C,Wang J, Leetmaa A, Reynolds R, Jenne R, Joseph D (1996)The NCEP/NCAR 40-year reanalysis project. Bull Am Mete-orol Soc 77:437–471

Kistler R, Collins W, Saha S, White G, Woollen J, Kalnay E,Chelliah M, Ebisuzaki W, Kanamitsu M, Kousky V, van denDool H, Jenne R, Fiorino M (2001) The NCEP-NCAR 50-yearReanalysis: monthly CD_ROM and documentation. Bull AmMeteor Soc 82:247–267; DOI 10.1175/1520-0477(2001)082

Klein WH (1957) Principal tracks and mean frequencies of cyclonesand anticyclones in the Northern Hemisphere. Research Paper40, U.S. Weather Bureau, pp 1–60

Lau NC (1988) Variability of the observed midlatitude storm tracksin relation to low-frequency changes in the circulation pattern.J Atmos Sci 45:2718–2743

Lim E-P, Simmonds I (2002) Explosive cyclone development in thesouthern hemisphere and a comparison with northern hemi-sphere. Mon Weather Rev 130:2188–2209


Marshall J, Kshnir Y, Battisti D, Chang P, Czaja A, Dickson R,Hurrell J, McCartney M, Sravan R, Visbeck M (2001) NorthAtlantic climate variability: phenomena, impacts and mecha-nisms. Int J Climatol 21:1863–1898

Osborn TJ, Briffa KR, Tett SFB, Jones PD, Trigo RM (1999)Evaluation of the North Atlantic Oscillation as simulated by acoupled climate model. Clim Dyn 15:685–702

Osborn TJ, Hulme M, Jones PD, Basnett TA (2000) Observedtrends in the daily intensity of United Kingdom precipitation.Int J Climatol 20:347–364

Petterssen S (1956) Weather analysis and forecasting, vol I: Motionand motion systems. McGraw-Hill, New York, pp 1–428

Rogers JC (1990) Patterns of low-frequency monthly sea levelpressure variability (1899–1986) and associated wave cyclonefrequencies. J Clim 3:1303–1310

Rogers JC (1997) North Atlantic storm track variability and itsassociation to the north Atlantic oscillation and climate vari-ability of northern Europe. J Clim 10:1635–1647

Sanders F (1986) Explosive cyclogenesis in the West-Central NorthAtlantic Ocean, 1981–84 Part I: composite and mean behavior.Mon Weather Rev 114:1781–1794

Sanders F, Gyakum JR (1980) Synoptic-dynamic climatology ofthe ‘‘bomb’’. Mon Weather Rev 108:1589–1606

Schmith T, Kaas E, Li T-S (1998) Northeast Atlantic winterstorminess 1875–1995 reanaysed. Clim Dyn 14:529–536

Serreze MC, Carse F, Barry RG, Rogers JC (1997) Icelandic lowcyclone activity: climatological features, linkages with theNAO, and relationships with recent changes in the northernhemisphere circulation. J Clim 10:453–464

Sickmoller M, Blender R, Fraedrich K (2000) Observed wintercyclone tracks in the northern hemisphere in re-analysed EC-MWF data. Q J R Meteorol Soc 126:591–620

Simmons AJ (2001) Development of the ERA-40 data assimilationsystem. In: Proceedings of the workshop on reanalysis, EC-MWF, Reading, UK, pp 11–30

Sinclair MR (1995) A climatology of cyclogenesis for the SouthernHemisphere. Mon Weather Rev 123:1601–1619

Sinclair MR (1997) Objective identification of cyclones and theircirculation, intensity and climatology. Weather Forecast12:595–612

Tayanc M, Karaca M, Dalfes HN (1998) March 1987 cyclone(blizzard) over the Eastern Mediterranean and Balkanregion associated with blocking. Mon Weather Rev 126:3036–3047

Trigo IF, Davies TD, Bigg GR (1999) Objective climatology ofcyclones in the Mediterranean Region. J Clim 12:1685–1696

Trigo IF, Davies TD, Bigg GR (2000) Decline in Mediterraneanrainfall caused by weakening of Mediterranean cyclones. Geo-phys Res Lett 27:2913–2916

Trigo IF, Bigg GR, Davies TD (2002) Climatology of cyclogenesismechanisms in the Mediterranean. Mon Weather Rev 130:549–569

Trigo RM, Pozo-Vasquez D, Osborn TJ, Castro-Diez Y, Gamis-Fortis S, Esteban-Parra MJ (2004) North Atlantic Oscillationinfluence on precipitation, river flow and water resources in theIberian Peninsula. Int J Climatol 24:925–944

Ulbrich U, Christoph M (1999) A shift of the NAO and increasingstorm track activity over Europe due to anthropogenic green-house gas forcing. Clim Dyn 15:551–559

van Loon H, Williams J (1976) Connection between trends of meantemperature and circulation at the surface 1 Winter. MonWeather Rev 104:365–380

Xoplaki E, Gonzalez-Rouco JF, Luterbacher J, Wanner H (2004)Wet season Mediterranean precipitation variability: influence oflarge-scale dynamics and trends. Clim Dyn 23:63–78; DOI10.1007/s00382-004-0422-0

Zolina O, Gulev SK (2002) Improving accuracy of mappingcyclone numbers and frequencies. Mon Weather Rev 129:748–759


isabel franco trigo climatology and interannual ...idlcc.fc.ul.pt/pdf/trigo_climatology_2006.pdf ·...

Documents