weather regimes and their connection to the winter

INTERNATIONAL JOURNAL OF CLIMATOLOGY

Int. J. Climatol. 25: 33–50 (2005)

Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/joc.1101

WEATHER REGIMES AND THEIR CONNECTION TO THE WINTERRAINFALL IN PORTUGAL

J.A. SANTOS,a,b* J. CORTE-REALa,c and S.M. LEITEa,b

a Geophysics Centre, University of Evora, Evora, Portugalb University of Tras-os-Montes and Alto Douro, Vila Real, Portugal

c ICAT, University of Lisbon, Lisbon, Portugal

Received 15 March 2003Revised 9 August 2004Accepted 9 August 2004

ABSTRACT

Wintertime rainfall over Portugal is strongly coupled with the large-scale atmospheric flow in the Euro-Atlantic sector.A K-means cluster analysis, on the space spanned by a subset of the empirical orthogonal functions of the daily meansea-level pressure fields, is performed aiming to isolate the weather regimes responsible for the interannual variability ofthe winter precipitation. Each daily circulation pattern is keyed to a set of five weather regimes (C, W, NAO−, NAO+ andE). The dynamical structure of each regime substantiates the statistical properties of the respective rainfall distributionand validates the clustering technique. The C regime is related to low-pressure systems over the North Atlantic thatinduce southwesterly and westerly moist winds over the country. The W regime is characterized by westerly disturbedweather associated with low-pressure systems mainly located over northern Europe. The NAO− regime is manifestedby weak low-pressure systems near Portugal. The NAO+ regime corresponds to a well-developed Azores high withgenerally settled and dry weather conditions. Finally, the E regime is related to anomalous strong easterly winds andrather dry conditions. Although the variability in the frequencies of occurrence of the C and NAO− regimes is largelydominant in the interannual variability of the winter rainfall throughout Portugal, the C regime is particularly meaningfulover northern Portugal and the NAO− regime acquires higher relevance over southern Portugal. The inclusion of the Wregime improves the description of the variability over northern and central Portugal. Dry weather conditions prevail inboth the NAO+ and E regimes, with hardly any exceptions. The occurrence of the NAO+ and the NAO− regimes isalso strongly coupled with the North Atlantic oscillation. Copyright 2005 Royal Meteorological Society.

KEY WORDS: winter precipitation; rainfall regimes; large-scale circulation; Portugal; K-means

1. INTRODUCTION

Rainfall is of the utmost importance, as it is a primary variable in most hydrological models. Owing to theMediterranean-type climatic conditions that prevail in Portugal, winter precipitation is the determining factorin the budgets of the hydrological cycle. Moreover, wintertime rainfall amounts play a key role in triggeringdrought episodes, because although natural and socio-economic systems are already prepared to cope withthe Mediterranean summer dryness, they are not prepared for a lack of winter rainfall. In addition, from astatistical viewpoint, the study of summer rainfall in Portugal is quite complex owing to its sporadic characterand irregularity.

Changes in the large-scale atmospheric flow have an important impact on the winter precipitation variabilityover the Mediterranean basin (Corte-Real et al., 1995a; Dunkeloh and Jacobeit, 2003; Fernandez et al., 2003).Furthermore, the development and movement of synoptic weather systems that originate over the NorthAtlantic explain most of the winter precipitation variability over western Europe (Murphy, 1999). Many earlier

* Correspondence to: J.A. Santos, Universidade de Tras-os-Montes e Alto Douro, Departamento de Fısica, Quinta dos Prados, Apartado1013, 5000-911 Vila Real, Portugal; e-mail: [email protected]

Copyright 2005 Royal Meteorological Society

34 J. A. SANTOS, J. CORTE-REAL AND S. M. LEITE

studies also developed relationships between large-scale atmospheric circulation and precipitation over Iberia(Zorita et al., 1992; Corte-Real et al., 1995b; Rodo et al., 1997; Esteban-Parra et al., 1998; Rodriguez-Pueblaet al., 1998; Rocha, 1999; Romero et al., 1999; Gonzalez-Rouco et al., 2000; Munoz-Dıaz and Rodrigo, 2003;Sumner et al., 2003). Cyclones located to the west of the British Isles clearly favour winter precipitation overPortugal through strong advection of maritime air masses along their southern flank (Ulbrich et al., 1999;Wibig, 1999; Goodess and Jones, 2002). Furthermore, changes in the frequencies of occurrence of a numberof large-scale atmospheric circulation patterns in the Euro-Atlantic sector have a strong impact on Portugueserainfall (Corte-Real et al., 1995b, 1998, 1999; Trigo and DaCamara, 2000).

The strong wintertime connection between several climate variables over Europe and the North Atlanticoscillation (NAO) is widely recognized (Hurrell and van Loon, 1997; Qian et al., 2000a,b; Hurrell et al.,2001; Trigo et al., 2002). The NAO pattern in the mean sea-level pressure (MSLP) field displays a renowneddipolar structure, which is characterized by one ‘centre of action’ located over the Iceland region and another,of opposite sign, located near the Azores (Wallace and Gutzler, 1981; Barnston and Livezey, 1987; Hurrell,1996). The NAO is frequently measured by a two-station index (Jones et al., 1997; Rogers, 1997; Pozo-Vazquez et al., 2001; Goodess and Jones, 2002), and the Gibraltar–Iceland (G–I) index (Jones et al., 1997)is particularly suitable for monitoring the winter NAO (Osborn et al., 1999).

Owing to the importance of winter precipitation in Portugal, and making an allowance to the studies referredto above, the main aims of the present work concern the identification, characterization and intercomparisonof the different weather regimes specifically associated with Portuguese winter rainfall. Section 2 describesthe data and methods used to achieve the latter purposes. The results are presented and discussed in Section 3.The main conclusions are ultimately summarized in Section 4.

2. DATA AND METHODOLOGY

The recent-past atmospheric conditions were monitored using the National Centers for EnvironmentalPrediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis data (Kalnay et al., 1996).The present study employed the daily fields of the MSLP, geopotential height at 500 hPa, zonal and meridionalwind components at 850 hPa and specific humidity at 850 hPa. All these datasets are defined over a 2.5°

latitude × 2.5° longitude grid. The time period was chosen to begin in the International Geophysical Year(1957) and to end in 1998, and the winter season corresponds to the months of December, January andFebruary.

In order to relate the frequencies of occurrence of the different weather regimes with the NAO, the G–Iindex was used. This index was provided by the Climatic Research Unit, University of East Anglia, Norwich,UK (http://www.cru.uea.ac.uk/). The historical time series of the gridded monthly precipitation produced bythe latter unit (Hulme, 1992, 1994; Hulme et al., 1998) were also used in this study. The geographical locationsof the three grid boxes (P1, P2 and P3) that cover the entire mainland of Portugal are shown in Figure 1. Thisgridded precipitation dataset provides full coverage of the interannual variability of the winter precipitationover Portugal. The relationships between the statistical properties of station and gridded data were analysed indetail by Osborn and Hulme (1997). In addition, the gridded observed precipitation is generally more reliablethan the reanalysed precipitation, though high correlations are found between them (Wilby and Wigley, 2000).

An automatic classification scheme was developed so that each day was keyed to a particular weatherregime. The weather regimes were isolated by a K-means cluster analysis on a subspace spanned by a subsetof empirical orthogonal functions (EOFs) of the daily MSLP. With this procedure, both a prefiltering and adata reduction were achieved. The K-means algorithm starts with a preset number of clusters K and thenmoves objects between clusters with the goal of, first, minimizing the variance within clusters and, second,maximizing the variance between clusters. The classification of weather regimes by the K-means clusteranalysis has been used extensively (e.g. Solman and Menendez, 2003).

The pressure anomalies were previously weighted so that every grid point had contributions to the covarianceestimates proportional to its representative area (the anomalies at each grid point were multiplied by thesquare root of the respective latitude’s cosine). A principal component analysis (PCA) was then applied to the

Copyright 2005 Royal Meteorological Society Int. J. Climatol. 25: 33–50 (2005)

PORTUGUESE WINTER RAINFALL AND WEATHER REGIMES 35

Figure 1. Map with the geographical locations of the daily precipitation stations (Porto, Braganca, Lisbon and Beja) and of the gridboxes (P1, P2 and P3) of the gridded monthly precipitation

covariance matrix of the winter daily MSLP weighted-anomalies defined within a Euro-Atlantic sector withthe following coordinates: (25–60 °N; 30 °W–10 °E). Next, the first few principal components (PCs) wereused as input variables for the clustering analysis.

The classification was carried out on the daily MSLP fields due to the wide range of applications provided bythis variable when compared with other candidates, such as the geopotential height at 850 hPa. First, the dailyMSLP fields are more widely available (both in space and in time) and more reliable than other competitivevariables, which enables the extension of the same statistical approach, based on the same climate variable, toareas or time periods where other variables are unavailable or unreliable. Second, most numerical models arereasonably skilful in reproducing the daily MSLP fields, which allows a straightforward expansion of the studyto model data. Third, this variable has already been applied successfully in previous classification proceduresin the same geographical area (e.g. Corte-Real et al., 1998; Trigo and DaCamara, 2000). It is still worthemphasizing that a combination of variables is not possible in this particular case, since the input variablesin the K-means method are the non-standardized PCs. Therefore, a combination of different variables withdifferent physical dimensions is not appropriate. On the other hand, the standardization gives all the PCs thesame weight, which is an unrealistic assumption.

Aiming to identify weather regimes clearly related to the rainfall over Portugal, an information measure,Inf, was defined as a function of a previously specified threshold, thr, and of the number of clusters K in thefollowing manner:

Inf(thr,K) =K∑

i=1

|nri(thr) − prni |

where nri(thr) is the number of rainy days within the ith cluster with a rainfall amount beyond the threshold

thr; ni is the total number of days within the same cluster; and pr is the probability of a rainfall total abovethe specified threshold in all time realizations of the full series. For each specific threshold, thr, the optimalnumber of clusters (the number of clusters that best differentiates between the precipitation regimes withamounts above and below the specified threshold) maximizes the function Inf(thr, K). A simpler version ofthis information measure was previously developed by Corte-Real et al. (1998).

The dependence of the solution on the seeding of the K-means algorithm (Michelangeli et al., 1995) wastaken into account by selecting the seeding that maximizes the information measure for a given threshold andfor a preselected number of clusters. The seeding that maximizes the 1 mm threshold is the most appropriate



choice, since this threshold enables a clear differentiation between wet and dry days. It is still worth notingthat other classifications based on Lamb’s weather types (Trigo and DaCamara, 2000; Goodess and Jones,2002) are not specifically developed taking into account local precipitation amounts. Therefore, these latterweather types are not optimal rainfall regimes.

The mathematical literature related to this method commonly refers to a ‘pseudo-F ’ test, based on theanalysis of variance, as a proper indicator of the optimal number of clusters. Nevertheless, this measure isless suitable when the aim is to identify weather regimes that discriminate the values of a particular time seriesassociated with a specific climate variable (e.g. precipitation). Despite the different nature of these measures,the present analysis revealed that the solution that maximizes the information measure nearly maximizes the‘pseudo-F ’ statistic.

The gridded dataset only comprises monthly data, and so the time series of the daily rainfall recorded atPorto, Braganca, Lisbon and Beja were thereby used to compute the information measure. These daily timeseries were provided by the European Climate Assessment project (Klein Tank et al., 2002). The geographicallocations of these four stations are depicted in Figure 1. Herein, the stations of Porto and Braganca representthe conditions over northern Portugal, and Lisbon and Beja represent the conditions over southern Portugal.

Although the daily time series and the gridded precipitation are obtained from different sources, the totalwinter precipitation recorded at each station is highly correlated with the time series in the nearest grid boxes(Table I). All these coefficients are statistically significant at the 1% significance level, which is also clearproof of the dominance of the large-scale forcing of winter precipitation throughout the country; the localand regional processes are not strong enough to weaken these correlations.

3. REGIMES

The first three orthogonal modes cumulatively represent about 80% of the total variance and are statisticallysignificant according to North’s rule-of-thumb (von Storch and Zwiers, 1999) and to other widely used criteriabased on Monte Carlo approaches (Prohaska, 1976; Overland and Preisendorfer, 1982; Leite, 1991).

Cluster analysis was applied afterwards in order to identify the weather regimes in the current atmosphericstate. The selection of the optimal number of clusters (weather regimes) was undertaken by computingthe aforementioned information measure for each of the daily precipitation time series in Portugal (Porto,Braganca, Lisbon and Beja). The information measures (for several threshold values) were expressed as apercentage of their absolute maximum value, which is reached when the number of clusters equals the samplesize. Therefore, each information measure exhibits an upward trend and the first pronounced local maximumindicates the optimal number of clusters for the corresponding threshold.

A detailed analysis of the scores of these information measures suggested that five is the appropriatenumber of weather regimes at the four stations. As an illustration, the information measure at Porto has aclear local maximum for five clusters, and for all but the 5 mm threshold (Figure 2). Additionally, the resultsrevealed that the information measure for all thresholds is almost invariant when the number of retained PCsis higher than three (higher PCs have lower fractions of variance represented). Consequently, following both

Table I. Correlation coefficients (r × 100) between the single-station winter precipitation at Porto, Braganca, Lisbon andBeja and the gridded precipitation in the P1, P2 and P3 grid boxes. All these correlations are statistically significant at a

significance level of 1%. The highest coefficient for each station is given in bold

r × 100

Porto Braganca Lisbon Beja

P1 95 91 72 68P2 89 93 91 91P3 68 80 93 97



Figure 2. Scores of the information measure (Inf) for several daily rainfall amounts (mm) recorded at Porto as a function of the numberof clusters. The information measures are expressed as a fraction (as a percentage) of its absolute maximum value (Infmax)

the parsimony principle and the statistical significance of the orthogonal modes, only the first three PCs wereretained as input variables for clustering purposes.

In order to get an integrated view of the dynamical structure of each weather regime, and to stress thedifferences between them, composites of the following daily climate variables were performed: MSLP,geopotential height at 500 hPa, relative vertical vorticity at 850 hPa and the total humidity advection at850 hPa. The last two variables were computed using both the zonal and meridional wind components(vorticity and advection) and the specific humidity at 850 hPa (advection). As will be shown below, thedifferences in the spatial features of the composites referred to above for each regime constitute a validationof the clustering technique.

These composites can actually be interpreted as typical circulation patterns of each weather regime and willhenceforth be presented. The frequencies of occurrence of the weather regimes and some relevant statisticsare also listed in Table II.

3.1. Cyclonic regime

The cyclonic regime (C regime) is the least frequent (14%; Table II). However, the high interannualvariability can occasionally turn it into the dominant regime. From a physical viewpoint, the C regime isassociated with a high density of cyclonic systems, mainly located just to the west of the British Isles, whereas

Table II. Frequencies of occurrence (FO) of each weather regime (WR) and some relevant statistics of the daily rainfallobserved at Porto (P), Braganca (Br), Lisbon (L) and Beja (B) computed separately for each weather regime (C, W,NAO−, NAO+ and E) and for the total sample of days (climate). The time period ranges from the 1957–58 winter to

the 1997–98 winter

WR FO (%) Probability of arainy day

(R ≥ 1 mm) (%)

Contribution for thetotal precipitation (%)

Mean rainfall on arainy day

(relative values)

P Br L B P Br L B P Br L B

Climate — 44 34 36 31 — — — 10.3 6.7 7.6 6.5C 14 79 74 71 61 34 40 37 35 15.6 (1.5) 11.4 (1.7) 11.8 (1.5) 9.2 (1.4)W 19 71 50 43 33 31 24 16 14 10.9 (1.1) 5.9 (0.9) 5.2 (0.7) 4.0 (0.6)NAO− 17 58 48 60 55 21 25 30 35 10.1 (1.0) 7.6 (1.1) 8.7 (1.1) 8.3 (1.3)NAO+ 28 26 16 15 11 11 8 9 8 5.6 (0.5) 2.8 (0.4) 4.4 (0.6) 3.7 (0.6)E 22 11 8 14 14 3 3 8 8 4.4 (0.4) 2.2 (0.3) 5.2 (0.7) 4.9 (0.8)



the Azores high is almost absent (Figure 3(a)). The anomaly composite of the geopotential height at 500 hPalevel shows dynamically consistent negative anomalies over the North Atlantic (Figure 3(d)).

Strong cyclonic (positive) vorticity over the same area was also found (Figure 3(e)), which is related toa high density of extratropical cyclonic systems. The vorticity values over Portugal are nearly zero, whichmeans that neither significant large-scale air rising nor strong baroclinicity occur over Portugal; the rainfallgeneration is indeed linked to travelling frontal systems (e.g. cold fronts) that usually extend to latitudes to the

(a) (b)

(c) (d)

(e) (f)

Figure 3. Absolute composites of the reanalysed daily MSLP for the C regime, in the winters from 1957–58 to 1997–98, for: (a) allwinter days; (b) rainy days (R ≥ 10 mm) at Porto; (c) rainy days (R ≥ 10 mm) at Beja (contour intervals of 4 hPa). Correspondinganomaly composites of the: (d) geopotential height at the 500 hPa level (contour interval of 30 gpm); (e) vertical vorticity at the 850 hPalevel (contour interval of 5 × 10−6 s−1); (f) total humidity advection at the 850 hPa level (contour interval of 10 g kg−1 m s−1 andthe arrows represent the anomalies in the zonal and meridional components of the total advection). Shaded areas represent absolute

anomalies above 20 g kg−1 m s−1 (statistically significant at least at a confidence level of 99%)



(a) (b)

Figure 4. Scatter plots between the winter daily rainfall on a rainy day and the wind direction at Porto in the (a) C and (b) W regime.The average wind direction and the average rainfall on a rainy day for the full time series are also shown (hollow squares)

south of Portugal in winter. In fact, the humidity advection anomalies highlight their key role in explaining theoccurrence of rainfall (Figure 3(f)). Strong humidity advection is associated with westerly and southwesterlywinds from the North Atlantic that occur over Portugal along the southern flank of the cyclonic systems,which have a high moisture content, particularly over northern and central Portugal. The dominance of thelarge-scale southwesterly and westerly winds during the wet days of this regime is apparent in the scatter plotof the wind directions on a rainy day at Porto (Figure 4(a)).

The latter physical description is confirmed by well-above-average rainfall probabilities throughout Portugal(Table II). Furthermore, the C regime gives the largest contributions to the total winter rainfall; they rangefrom 34 to 40% (Table II). The precipitation rates are also well above the climate mean values; the averagerainfall on a rainy day is of about 1.5 times its climatic mean. These properties suggest that the C regimeplays a leading role in governing the winter rainfall variability over Portugal, despite being the least frequentregime. This assumption will be more objectively quantified in Section 3.7.

The rain-generating conditions can be overridden because not all days within a regime are actually rainydays. This can be overcome by conditioning the composites (conditional composites) to the occurrenceof a rainfall amount in excess of 10 mm, recorded at a specific station. Owing to Portuguese climateheterogeneity, the conditional composites for rainfall at Porto (northern Portugal) and at Beja (southernPortugal) are presented here. Although both conditional composites (Figure 3(b) and (c)) are very similar tothe total composite (Figure 3(a)), the most remarkable rain-generating patterns tend to have a southeastwardlydisplaced low-pressure centre. In fact, the closer the low-pressure centre is to Portugal (weaker Azores high),the more favourable the conditions are to the rainfall occurrence.

3.2. Westerly regime

In the westerly disturbed weather regime (W regime), the Iceland low is southeastwardly displaced and theAzores high undergoes a southward displacement (Figure 5(a)). The MSLP values north of the British Islescan be more than 20 hPa below the mean values (not shown). The typical ridge of the Azores high over Iberiais absent or very weakened, and a more westerly flow prevails. A dynamically coherent enhancement of thenorth–south contrast in the mid-tropospheric flow is also noteworthy (Figure 5(d)). This feature is related toanomalously strong cyclonic vorticity near the British Isles (Figure 5(e)), which reflects an anomalously highdensity of extratropical cyclonic systems over this region. The vorticity over Portugal is slightly negative,which means that the rainfall generation in the W regime, as in the C regime, must be linked to travellingfrontal systems that extend as far to the south as Portugal. As for the C regime, humidity advection has aleading role in explaining the rainfall occurrence (Figure 5(f)). The high humidity advection anomalies arenow associated with westerly and northwesterly winds (Figure 4(b)) along the southern flank of the cyclonicsystems centred over northern Europe, and their effects are particularly clear over northern Portugal; thestrong north–south gradient over Portugal is worth emphasizing. For instance, the probability of occurrenceof a rainy day is about 71% at Porto and it is only 33% at Beja, a value slightly higher than the climate meanvalue of 31% (Table II). Moreover, the relative average rainfall at Porto (1.1) is narrowly higher than its



(d)

(f)

(b)

(e)

(c)

(a)

Figure 5. As Figure 3, but for the W regime

climate mean value, whereas the rates are lower than average at all the other stations. Although the W regimeis more frequent than the C regime (19% against 14%), the contributions of the W regime for the winterrainfall at all stations are much lower than the corresponding contributions of the C regime; its contributionsare much higher over northern Portugal (Porto, 31%) than over southern Portugal (Beja, 14%).

The rain-generating conditions are clearly connected to a pressure trough that extends from the British Islestowards Iberia and the western Mediterranean for both northern and southern Portugal (Figure 5(b) and (c)).As expected, the pressure trough is deeper for Beja than for Porto, and the presence of the Azores high,almost absent in the C regime, is another noticeable feature of the W regime.



(a) (b)

(c) (d)

(e) (f)

Figure 6. As Figure 3, but for the NAO− regime

3.3. NAO− regime

The third regime is linked to the negative phase of the NAO (NAO− regime). Its frequency of occurrence is17% (Table II) and its typical pattern is remarkably different from the patterns of the aforementioned regimes(Figure 6(a)). Both the Iceland low and the Azores high are particularly weak; over the Iceland-low regionthe average anomalies are positive and higher than 16 hPa and over the whole of Iberia the anomalies arenegative and higher than 8 hPa. This north–south inverted dipolar structure clarifies its connection with thenegative phase of the NAO. The correlation coefficient between its frequency of occurrence and the G–I indexis −0.78 (statistically significant at the 99% confidence level). The rainfall probabilities nearly fall withinthe range from 50 to 60% (Table II). Although the mean precipitation on a rainy day is higher over northernPortugal, the relative precipitation rates are higher over southern Portugal (1.3 at Beja). Also, the contribution



of the NAO− regime to the total winter precipitation is higher over southern Portugal (35% at Beja and 21%at Porto). The anomalies in the mid-tropospheric flow (Figure 6(d)) are a result of the more southerly pathof the westerlies, and anomalous strong cyclonic vorticity that occurs over Iberia (Figure 6(e)). Therefore,during the NAO− regime the occurrence of weak cyclones (relatively low vorticity anomalies) in the vicinityof Iberia is apparent, with an average vorticity maximum just off the Portuguese coast. The near-averagevalues of humidity advection over Portugal suggest that the rainfall generation in the NAO− regime mustbe, in large part, attributed to large-scale air rising and baroclinic instability over Portugal (Figure 6(f)). Thecyclonic circulation over Portugal is also plainly resolved (Figure 6(f)).

In contrast to the scatter plots of the wind direction on a rainy day in the C and W regimes (Figure 4),the scatter plot for the NAO− regime (Figure 7) displays a high scattering of the rainy days with the large-scale wind direction, though the rainfall at Porto is more likely to occur in the presence of westerly winds.Therefore, the cyclones located to the north of Portugal (westerly winds at Porto) establish the highest rain-generating conditions over northern Portugal, whereas over southern Portugal the rain-generating conditionsare almost independent of the cyclone’s position.

The conditional composites emphasize the inverted dipolar structure (Figure 6(b) and (c)). On a smallerscale, the key rain-generating systems tend to be located near the station, but with their cores located tothe north (mesoscale westerly winds). Therefore, the low-pressure systems located off the Portuguese coast(small core in Figure 6(c)) are more relevant to the rainfall occurrence at Beja than at Porto. Conversely,low-pressure systems located to the north of Iberia are more significant to the rainfall occurrence at Porto.A meridional pressure gradient (MPG) between the centres of the inverted dipole was computed. Thegeographical coordinates of the poles are (62.5 °N, 11.25 °W) and (40 °N, 11.25 °W). Only less than 3%of the days at both stations have MPG scores below average, and at Beja the heaviest rainy days tend to berelated with higher scores (Figure 8). These latter outcomes corroborate the high relevance of the inverteddipolar structure in the rainfall generation under the NAO− regime. This relationship between the MPG andrainfall is not observed in the other regimes.

(a) (b)

Figure 7. Scatter plots between the winter daily rainfall on a rainy day and the wind direction (a) at Porto and (b) at Beja in the NAO−regime. The average wind direction and the average rainfall on a rainy day for the full time series are also shown (hollow squares)

(a) (b)

Figure 8. Scatter plot between the daily precipitation (a) at Porto and (b) at Beja and the north–south pressure gradient (MPG). Thevertical solid line corresponds to the mean gradient (−16.3 hPa)



(b)(a)

(d)(c)

(f)(e)

Figure 9. As Figure 3, but for the NAO+ regime

3.4. NAO+ regime

The NAO+ regime is strongly linked to the positive phase of the NAO and is the most common regime, witha frequency of occurrence of 28% (Table II). The correlation coefficient between its frequency of occurrenceand the G–I index is 0.75 (statistically significant at the 99% confidence level). The Azores high is enhancedand presents a well-defined ridge over Iberia (Figure 9(a)). As a result of the sinking and divergence of thesurface winds during this regime, Portugal usually experiences very settled and fairly dry weather conditions.

The mid-tropospheric ridge over western Europe is enhanced and the westerlies follow wavy paths(Figure 9(d)). The opposite dipolar structures of the anomalies between the NAO− and the NAO+ regimeare quite clear (Figures 6(d) and 9(d)). The anomalous high anticyclonic vorticity over Portugal is in clear



dynamical conformity with the above-stated dynamical features (Figure 9(e)). The easterly and northeasterlyadvection anomalies reinforce the dry and settled weather conditions (Figure 9(f)).

The probabilities of rain clearly fall behind the climatic mean values for every station (Table II). Theaverage rainfall on a rainy day is about one half its climatic mean value, and the contributions to the winterrainfall are only nearly 10% throughout the country.

A more detailed inspection of the few rainy days in this weather regime led to the conclusion that they areessentially due to mesoscale processes (e.g. cold fronts), which are not properly resolved by the coarse gridof the dataset and are smoothed out by daily means. The majority of the rainy days are connected to cycloneslocated far to the north of Portugal, but with well-developed frontal systems that briefly interrupt the Azoreshigh (Figure 9(b) and (c)). In some cases, the mean fields are indeed indistinguishable from others related todry weather conditions. Furthermore, despite the occurrence of precipitation over Portugal, both the Azoreshigh and the Iceland low are generally strong (positive phase of the NAO).

3.5. Easterly regime

Lastly, in the easterly regime (E regime), is the presence of high pressure over the western European basinand the British Isles that are unprecedented in the other regimes (Figure 10(a)). This is the driest regime andoccurs on roughly 22% of the winter days (Table II). During this regime the mid-tropospheric ridge overwestern Europe is strongly enhanced (Figure 10(d)).

The vorticity is anomalously low near the British Isles, near average over northern and central Portugaland slightly higher than average over southern Portugal (Figure 10(e)). The easterly wind anomalies weakenthe advection of maritime air masses over the country (Figure 10(f)), which further enhance the dryness ofthis regime. The prevailing easterly winds over Portugal justify the name of this regime.

The rainfall probabilities over northern Portugal are much lower than in the NAO+ regime (Table II).Nevertheless, despite the lack of rainy days, the rainfall probabilities and the mean rainfall on a rainy dayover southern Portugal are slightly higher than in the NAO+ regime. In actual fact, the E regime explains 8%of the total winter rainfall at Beja (the same value as for the NAO+ regime), whereas, over northern Portugalit explains only 3% of this value. This can be explained by occasional cyclonic systems located southward orsouthwestward of Portugal that can lead to heavy rainfall events and to severe storms over southern Portugal,despite being extremely rare events. The presence of these cyclones is manifested in the conditional patternfor Beja by the weak pressure minimum over Madeira (Figure 10(c)).

Sporadically, when an eastward moving trough over North Africa (along the southern flank of the high-pressure system) reaches the Atlantic waters it develops and can be accompanied by an upper level cold-corecyclone, supplying the required conditions for the generation of strong local convective rainfall (Tullot,1983). This Iberian–African trough plays an essential role in the rainfall generation over northwestern Africa(Knippertz et al., 2003; Knippertz, 2004). With the purpose of identifying these cold-core lows in the E regime,the conditional composites of the geopotential height anomalies at 500 hPa and of the vorticity anomalies at850 hPa for a rainy day (R ≥ 10 mm) at Beja (most affected station) were computed. The enhancement ofthe negative mid-tropospheric anomalies over the Madeira area is a manifestation of the equivalent barotropicstructure of these lows (Figure 11(a)). Furthermore, the high vorticity anomalies southwestward of Portugalplainly confirm the presence of these systems during rainy episodes in the E regime (Figure 11(b)). However,it must be kept in mind that these events are extremely scarce and their differentiation within an automatedclassification scheme, with a small number of regimes, is virtually impossible.

3.6. Intercomparison of the regimes

Even though the previously described weather regimes effectively discriminate rainy days from dry days,a discrimination of the rainfall amounts must also be accomplished. The probabilities of exceeding a stateddaily rainfall amount within each weather regime can highlight this issue.

At Porto, the C regime is clearly the moistest regime (Figure 12(a)). In fact, the decrease in its distributionfunction is slow, which means that intermediate and heavy rainfall days are relatively frequent; about 40%of the rainy days have rainfall values over 15.0 mm. For instance, the recorded maximum daily rainfall



(a) (b)

(d)(c)

(e) (f)

Figure 10. As Figure 3, but for the E regime

(84.4 mm) occurred in this regime. The NAO− and W regimes present similar distributions that are also veryclose to the distribution of the full time series. For the NAO+ and E regimes, the frequencies of occurrenceare all clearly below average. For both regimes, about half of the rainy days have rainfall values equal orbelow 2.0 mm, and only roughly 10% of the days have rainfall values above 15.0 mm.

At Beja, the C regime is less dominant than at Porto (Figure 12(b)); the decrease in the distribution functionis steeper than at Porto and is actually very similar to the corresponding curve of the NAO− regime. Thedistribution function of the W regime is always below the mean curve at Beja, which explains the differencesin its statistics at Porto and at Beja (Table II). The NAO− regime plays a greater role over southern Portugal,whereas the W regime plays a more secondary role.



(a)

(b)

Figure 11. Conditional composites for the E regime of the (a) geopotential height anomalies at 500 hPa and of the (b) vorticity anomaliesat 850 hPa on a rainy day (R ≥ 10 mm) at Beja

(a) (b)

Figure 12. Probabilities (as a percentage) of exceeding a stated daily precipitation, F (R), on a rainy day (a) at Porto and (b) at Bejafor the entire time series (solid curves) and for each regime separately

The C regime at Porto is responsible for the occurrence of most of the stormy days with rainfall amountsabove 40 mm (Figure 13(a)). For intermediate rainfall events (15–40 mm) the C and W regimes show similarcontributions. The NAO− regime has lower contributions in almost all classes. Although the contributionsof the C regime are similar at both stations, the contributions of the NAO− regime are superior at Beja(Figure 13(b)). At Beja, the W regime is only significant in the occurrence of light rainfall values (lesser than15 mm). The NAO+ and E regimes still have relevant contributions for light precipitation; about 40% (Porto)or 30% (Beja) of the days with rainfall amounts below 5 mm are keyed to these two regimes. The rainfallextremes at Porto are related to the C, W and NAO− regimes, whereas only the C and NAO− regimes arerelevant in the occurrence of extremes at Beja. As was already noted, the E regime can lead to sporadic heavyrain events over southern Portugal related to the development of cold-core lows near Madeira. However, atBeja, the high contribution (50%) of this regime in the class centred at 47.5 mm can be misleading, sinceit corresponds to a single E regime event that evidently does not have an important impact on the winterrainfall as a whole; it is a rather abnormal event.



(a) (b)

Figure 13. Histograms of the relative contributions (as a percentage) of each weather regime for the occurrence of a rainfall amountwithin a specific precipitation class (a) at Porto and (b) at Beja

3.7. Interannual variability

With the aim of quantifying the relevance of each regime in the interannual variability of the winterprecipitation over Portugal, a linear regression model was individually fitted to the winter precipitation ateach grid box (P1, P2 and P3). The frequencies of occurrence of the C, W and NAO− regimes for eachwinter were considered as predictors, because these regimes are clearly connected to winter daily precipitationover Portugal and the introduction of the other two regimes (NAO+ and E) brings redundancy to the models.Moreover, the W regime does not give a significant contribution to the precipitation modelling in the P3 box,and was thus removed from the corresponding model.

The application of conventional hypothesis testing to the coefficients led to the conclusion that all thecoefficients are statistically significant at the 99% confidence level. In addition, the F -test highlights a stronglinear relationship between predictors and precipitation. The errors (residuals) are statistically independent,according to the Durbin–Watson test, and are normally distributed. A cross-validation was performed in orderto remove the artificial skill from each model (von Storch and Zwiers, 1999). In this procedure, a completewinter was omitted at each step and all the analyses were fully repeated for the remaining period. The valuesomitted were successively modelled and, at the end of this process, a synthetic time series was obtained foreach grid box.

The correlation coefficients between the synthetic and observed time series are: 0.83 (P1), 0.87 (P2) and0.80 (P3). All these coefficients are statistically significant at a confidence level of 99%. The skill scores ofthe regression models (Wilks, 1995) show that about two-thirds or more of the variance is explained by themodel at every grid box: 69% (P1), 75% (P2) and 64% (P3). The regression coefficients estimated for eachgrid box (models without intercept) represent a ‘typical daily precipitation amount’ of each weather regimeand are expressed in precipitation units (Table III). The C regime is the most significant in explaining theinterannual variability of the precipitation in the P1 and P2 boxes, whereas the NAO− regime is dominant inthe P3 box. The strong decrease in the relevance of the W regime and the increasing relevance of the NAO−

Table III. Regression coefficients estimated for the C, W and NAO−regimes and for each grid box of the gridded precipitation (P1, P2 and

P3)

Box Ccoef (mm) Wcoef (mm) NAO−coef (mm)

P1 14.7 8.8 6.7P2 10.6 4.0 7.3P3 7.8 — 8.6



regime towards the south are still noteworthy. In every grid box, the C and NAO− regimes are crucial inexplaining the interannual variability of precipitation.

6. SUMMARY AND CONCLUSIONS

Owing to the strong connection between winter rainfall over Portugal and the large-scale circulation patterns,a keying of the days into a set of non-overlapping large-scale rainfall regimes is particularly meaningful.Therefore, a K-means cluster analysis over the first three PCs of the daily MSLP weighted-anomalies wasperformed. An information measure was computed so that a maximum discrimination of the rainfall amountsamongst the weather regimes could be achieved. Five weather regimes (C, W, NAO−, NAO+ and E) wereisolated in this manner, and both their statistical and dynamical properties enabled a validation of the clusteringstrategy.

The C regime is characterized by low-pressure systems centred to the west of the British Isles. This featureleads to the prevalence of the westerlies and frontal systems over the country, which transport maritime airmasses (mT and mP) and establish rain-generating conditions. The NAO− regime is strongly linked to thenegative phase of the NAO; its main feature is an inverted dipolar structure, with a mean low-pressure centreover Iberia. The presence of local cyclones provides anomalously high levels of baroclinic instability andthey constitute the main precipitation source during this regime.

On the whole, the occurrence of both the C and NAO− regimes is largely dominant in the interannualvariability of winter rainfall over Portugal. The C regime is particularly meaningful over northern Portugal,whereas the NAO− regime plays a more relevant role over southern Portugal. The W regime is characterizedby the presence of low-pressure systems over the North Atlantic and northern Europe and by a weakenedAzores high. It allows an improvement in the description of the winter rainfall variability over northernand central Portugal. These three regimes jointly explain more than 60% of the total variance of winterprecipitation over the entire mainland of Portugal. It is still worth mentioning that the precipitation in boththe C and W regimes tends to be associated with remote cyclones (centred northward or northwestward ofthe Iberian Peninsula), whereas the precipitation in the NAO− regime tends to be linked to local cyclones.

The NAO+ regime is strongly coupled with the positive phase of the NAO; its principal feature is anenhanced Azores high, with a ridge that extends widely over Iberia, leading to generally settled and dryweather conditions. The E regime is clearly associated with a high-pressure area northward of Iberia. Duringthis regime, easterly advection of continental air masses (cT and cP) over Portugal usually yields dry weatherconditions.

This study also revealed that the frequencies of occurrence of the C, NAO− and W regimes are skilfulpredictors of winter rainfall over Portugal. In this manner, a significant prediction potential is available. In asubsequent study, under final development, the frequencies of occurrence of these weather regimes are usedto build up a downscaling strategy for the assessment of future rainfall scenarios in Portugal.

ACKNOWLEDGEMENTS

We would to like to thank Dr M. Hulme at the Climatic Research Unit, University of East Anglia,Norwich, UK, for supplying his monthly gridded precipitation dataset (‘gu23wld0098.dat’, Version 1.0).The construction of this dataset has been supported by the UK Department of the Environment, Transport andthe Regions (contract EPG 1/1/85). We are also grateful to the anonymous reviewers for their encouragingcomments.

REFERENCES

Barnston AG, Livezey RE. 1987. Classification seasonality and persistence of low-frequency atmospheric circulation patterns. MonthlyWeather Review 115: 1083–1126.

Corte-Real J, Wang X, Zhang X. 1995a. Modes of variability in the Northern Hemisphere’s mid tropospheric large-scale circulation.Theoretical and Applied Climatology 50: 133–146.



Corte-Real J, Zhang X, Wang X. 1995b. Downscaling GCM information to regional scales: a non-parametric multivariate regressionapproach. Climate Dynamics 11: 413–424.

Corte-Real J, Qian B, Xu H. 1998. Regional climate change in Portugal: precipitation variability associated with large-scale atmosphericcirculation. International Journal of Climatology 18: 619–635.

Corte-Real J, Qian B, Xu H. 1999. Circulation patterns, daily precipitation in Portugal and implications for climate change simulatedby the second Hadley Centre GCM. Climate Dynamics 15: 921–935.

Dunkeloh A, Jacobeit J. 2003. Circulation dynamics of Mediterranean precipitation variability 1948–98. International Journal ofClimatology 23: 1843–1866.

Esteban-Parra MJ, Rodrigo FS, Castro-Dıez Y. 1998. Spatial and temporal patterns of precipitation in Spain for the period 1880–1992.International Journal of Climatology 18: 1557–1574.

Fernandez J, Saenz J, Zorita E. 2003. Analysis of wintertime atmospheric moisture transport and its variability over southern Europein the NCEP reanalysis. Climate Research 23: 195–215.

Gonzalez-Rouco JF, Heyen H, Zorita E, Valero E. 2000. Agreement between observed rainfall trends and climate change simulationsin the southwest of Europe. Journal of Climate 13: 3057–3065.

Goodess CM, Jones PD. 2002. Links between circulation and changes in the characteristics of the Iberian rainfall. International Journalof Climatology 22: 1593–1615.

Hulme M. 1992. A 1951–80 global land precipitation climatology for the evaluation of general circulation models. Climate Dynamics7: 57–72.

Hulme M. 1994. Validation of large-scale precipitation fields in general circulation models. Global Precipitations and Climate Change,Desbois M, Desalmand F (eds). NATO ASI Series I, vol. 26. Springer-Verlag: Berlin; 387–406.

Hulme M, Osborn TJ, Johns TC. 1998. Precipitation sensitivity to global warming: comparison of observations with HadCM2simulations. Geophysical Research Letters 25: 3379–3382.

Hurrell JW. 1996. Influence of variations in extratropical wintertime teleconnections on Northern Hemisphere temperatures. GeophysicalResearch Letters 23: 665–668.

Hurrell JW, van Loon. 1997. Decadal variations in climate associated with the North Atlantic oscillation. Climate Change 36: 301–326.Hurrell JW, Kushnir Y, Visbeck M. 2001. The North Atlantic oscillation. Science 291: 603–605.Jones PD, Jonsson T, Wheeler D. 1997. Extension to the North Atlantic oscillation using early instrumental pressure observations from

Gibraltar and south-west Iceland. International Journal of Climatology 17: 1433–1450.Kalnay E, Kanamitsu M, Kistler R, Collins W, Deaven D, Gandin L, Iredell M, Saha S, White G, Woollen J, Zhu Y, Chelliah M,

Ebisuzaki W, Higgins W, Janowiak J, Mo KC, Ropelewski C, Wang J, Leetmaa A, Reynolds R, Jenne R, Joseph D. 1996. TheNCEP/NCAR 40-year reanalysis project. Bulletin of the American Meteorological Society 77: 437–471.

Klein Tank AMG, Wijngaard JB, Konnen GP, Bohm R, Demaree G, Gocheva A, Mileta M, Pashiardis S, Hejkrlik L, Kern-Hansen C,Heino R, Bessemoulin P, Muller-Westermeier G, Tzanakou M, Szalai S, Palsdottir T, Fitzgerald D, Rubin S, Capaldo M, Maugeri M,Leitass A, Bukantis A, Aberfeld R, Van Engelen AFV, Forland E, Mietus M, Coelho F, Mares C, Razuvaev V, Nieplova E,Cegnar T, Lopez JA, Dahlstrom B, Moberg A, Kirchhofer W, Ceylan A, Pachaliuk O, Alexander LV, Petrovic P. 2002. Daily datasetof 20th-century surface air temperature and precipitation series for the European Climate Assessment. International Journal ofClimatology 22: 1441–1453.

Knippertz P. 2004. A simple identification scheme for upper-level troughs and its application to winter precipitation variability innorthwest Africa. Journal of Climate 17: 1411–1418.

Knippertz P, Christoph M, Speth P. 2003. Long-term precipitation variability in Morocco and the link to the large-scale circulation inrecent and future climates. Meteorology and Atmospheric Physics 83: 67–88.

Leite SM. 1991. Estudo climatologico da bacia do Rio Douro: analise espectral e analise discriminante dos campos caracterısticos. PhDThesis UTAD, Vila Real, Portugal.

Michelangeli P, Vautard R, Legras B. 1995. Weather regimes: recurrence and quasi-stationarity. Journal of the Atmospheric Sciences52: 1237–1256.

Munoz-Dıaz D, Rodrigo FS. 2003. Effects of the North Atlantic oscillation on the probability for climatic categories of local monthlyrainfall in southern Spain. International Journal of Climatology 23: 381–397.

Murphy J. 1999. An evaluation of statistical and dynamical techniques for downscaling local climate. International Journal ofClimatology 12: 2256–2284.

Osborn TJ, Hulme M. 1997. Development of a relationship between station and grid-box rainday frequencies for climate modelevaluation. Journal of Climate 10: 1885–1908.

Osborn TJ, Briffa KR, Tett SFB, Jones PD, Trigo RM. 1999. Evaluation of the North Atlantic oscillation as simulated by a coupledclimate model. Climate Dynamics 15: 685–702.

Overland JE, Preisendorfer RW. 1982. A significance test for principal components applied to a cyclone climatology. Monthly WeatherReview 110: 1–4.

Pozo-Vazquez D, Esteban-Parra MJ, Rodrigo FS, Castro-Dıez Y. 2001. A study of NAO variability and its possible non-linear influenceson European surface temperature. Climate Dynamics 17: 701–715.

Prohaska JT. 1976. A technique for analyzing the linear relationship between two meteorological fields. Monthly Weather Review 11:1345–1353.

Qian BJ, Xu H, Corte-Real JM. 2000a. Is the North Atlantic oscillation the most important atmospheric pattern for precipitation inEurope? Journal of Geophysical Research 105: 901–910.

Qian BJ, Xu H, Corte-Real JM. 2000b. Spatial–temporal structures of the quasi-periodic oscillations in precipitation over Europe.International Journal of Climatology 20: 1583–1598.

Rocha A. 1999. Low-frequency variability of seasonal rainfall over Iberian Peninsula and ENSO. International Journal of Climatology19: 889–901.

Rodo X, Baert E, Comin FA. 1997. Variations in seasonal rainfall in southern Europe during the present century: relationships with theNorth Atlantic oscillation and the El Nino–southern oscillation. Climate Dynamics 13: 275–284.



Rodriguez-Puebla C, Encinas AH, Nieto S, Garmendia J. 1998. Spatial and temporal patterns of annual precipitation variability overthe Iberian Peninsula. International Journal of Climatology 18: 299–316.

Rogers JC. 1997. North Atlantic storm track variability and its association to the North Atlantic oscillation and climate variability ofnorthern Europe. Journal of Climate 10: 1635–1647.

Romero R, Sumner G, Ramis C, Genoves A. 1999. A classification of the atmospheric circulation in patterns producing significant dailyrainfall in the Spanish Mediterranean area. International Journal of Climatology 19: 765–785.

Solman SA, Menendez CG. 2003. Weather regimes in the South American sector and neighbouring oceans during winter. ClimateDynamics 21: 91–104.

Sumner GN, Romero R, Homar V, Ramis C, Alonso S, Zorita E. 2003. An estimate of the effects of climate change on the rainfall ofMediterranean Spain by the late twenty first century. Climate Dynamics 20: 789–805.

Trigo RM, DaCamara CC. 2000. Circulation weather types and their influence on the precipitation regime in Portugal. InternationalJournal of Climatology 20: 1559–1581.

Trigo RM, Osborn TJ, Corte-Real JM. 2002. The North Atlantic oscillation influence on Europe. Climate impacts and associated physicalmechanisms. Climate Research 20: 9–17.

Tullot IF. 1983. Climatologıa de Espana y Portugal. Instituto Nacional de Meteorologia: Madrid.Ulbrich U, Cristoph M, Pinto JG, Corte-Real J. 1999. Dependence of winter precipitation over Portugal on NAO and baroclinic wave

activity. International Journal of Climatology 19: 379–390.Von Storch H, Zwiers F. 1999. Statistical Analysis in Climate Research. Cambridge University Press: Cambridge.Wallace JM, Gutzler DS. 1981. Teleconnections in the geopotential height field during the Northern Hemisphere winter. Monthly Weather

Review 109: 784–812.Wibig J. 1999. Precipitation in Europe in relation to circulation patterns at the 500 hPa level. International Journal of Climatology 19:

253–269.Wilby RL, Wigley TML. 2000. Precipitation predictors for downscaling: observed and general circulation model relationships.

International Journal of Climatology 20: 641–661.Wilks DS. 1995. Statistical Methods in the Atmospheric Sciences: An Introduction. Academic Press.Zorita E, Kharin V, von Storch H. 1992. The atmospheric circulation and sea surface temperature in the North Atlantic area in winter:

their interaction and relevance for Iberian precipitation. Journal of Climate 5: 1097–1108.


weather regimes and their connection to the winter

Documents