development of daily rainfall scenarios for southeast spain using a circulation-type approach to...

INTERNATIONAL JOURNAL OF CLIMATOLOGY

Int. J. Climatol. 10: 1051–1083 (1998)

DEVELOPMENT OF DAILY RAINFALL SCENARIOS FORSOUTHEAST SPAIN USING A CIRCULATION-TYPE APPROACH TO

DOWNSCALINGCLARE M. GOODESS* and JEAN P. PALUTIKOF

Climatic Research Unit, Uni6ersity of East Anglia, Norwich, NR4 7TJ, UK

Recei6ed 3 No6ember 1997Re6ised 27 February 1998Accepted 3 March 1998

ABSTRACT

A method for downscaling from the relatively coarse General Circulation Model (GCM) spatial scale to the finerspatial scale required for impact assessment has been developed and tested in the Guadalentin Basin, southeast Spain.The method uses a circulation-type approach and relates large-scale patterns of a predictor variable, gridded sea levelpressure, to local values of a surface climate variable (daily rainfall at six stations). The large-scale patterns aredefined using an automated version of the Lamb Weather Type classification scheme, originally developed for theBritish Isles. It is demonstrated that this scheme can be successfully transferred to another region, southeast Spain.The 14 basic circulation types are combined into eight groups. These provide a legitimate basis for downscalingbecause each has a characteristic pressure pattern which produces the expected type and direction of flow over thestudy region. Furthermore, a set of consistent and distinct relationships is identified between these circulation typesand daily rainfall in the Guadalentin Basin. The ability of the GCM to reproduce the observed circulation types isassessed before applying these relationships to control and perturbed-run GCM output using a statistical weathergenerator. The effects of the GCM’s failure to reproduce the observed frequency of the circulation types aredetectable in the weather generator output. The GCM changes in SLP and circulation-type frequency between thecontrol and perturbed-runs are generally small. Nonetheless the weather generator results indicate significant changesin the number of rain days in spring and summer. These scenarios are presented as illustrative results rather than asreliable predictions. It is concluded that the circulation-type based approach to downscaling offers great potential.© 1998 Royal Meteorological Society.

KEY WORDS: southeast Spain; climate modelling; downscaling; climate scenarios; circulation patterns; daily rainfall

1. INTRODUCTION

General circulation models (GCMs) are considered to provide the greatest potential for scenarioconstruction for enhanced greenhouse effect impact studies, but typically have a grid-resolution of a fewhundred kilometres. The need to downscale from this relatively coarse scale to the finer spatial scalerequired for impact assessment is now widely recognised. Downscaling methods must also provideinformation at the daily, or shorter, timescale. The growing demand for climate scenarios with a highspatial and temporal resolution has created a need for downscaling methods which are relatively simpleto apply, are parsimonious of computer time, do not require large amounts of observed data, and aretransferable between regions. Ideally, these methods should also reflect the underlying physical mecha-nisms, be easy to validate, and capable of producing self-consistent scenarios for a range of variables.

* Correspondence to: Climatic Research Unit, University of East Anglia, Norwich NR4 7TJ, UK; tel.: +44 1603 592875; fax: +441603 507784; e-mail: [email protected]

Contract grant sponsor: Commission of the European Union, MEDALUS II; Contract grant number: EV5V-CT92-0164Contract grant sponsor: MEDALUS III; Contract grant number: ENV4-CT95-C121

CCC 0899–8418/98/101051–33$17.50© 1998 Royal Meteorological Society

C.M. GOODESS AND J.P. PALUTIKOF1052

A number of different downscaling methods have been proposed and can be divided into twogeneral categories; (i) model-based and (ii) empirical (Cubasch et al., 1996; Hewitson and Crane, 1996;Schubert and Henderson-Sellers, 1997). The first approach includes nesting a finer-scale limited-areanumerical model within a GCM (Giorgi et al., 1990, 1992; Jones et al., 1995). This approach isconsidered to offer the greatest long-term potential (Hewitson and Crane, 1996), but is very computer-intensive and subject to a number of technical problems, related, in particular, to model boundaryconditions (Schubert and Henderson-Sellers, 1997). Other model-based approaches include variableresolution GCMs (Deque and Piedelievre, 1995) and high resolution time-slice experiments (Cubasch etal., 1995). Empirical approaches have the potential to provide a more immediate solution (Hewitsonand Crane, 1996) and are the focus of much ongoing research effort. Empirical downscaling requiresthe identification of relationships between the observed large-scale and regional climate, which arethen applied to large-scale GCM output. It encompasses methods based on multiple regression (Kimet al., 1984; Wigley et al., 1990; Palutikof et al., 1997; Winkler et al., 1997), and studies in whichcirculation classifications are used to describe the large-scale climate (Cubasch et al., 1996; Hewitsonand Crane, 1996; Schubert and Henderson-Sellers, 1997). In this paper the potential of the latterapproach is evaluated, which has many of the desirable features identified above.

The chosen methodology has been developed and tested in the Guadalentin Basin, southeast Spain(Figure 1). It relates large-scale patterns of a predictor variable, sea level pressure (SLP), to localvalues of a surface climate variable (daily rainfall at six stations). This approach is based on theexpectation that the predictive capacity of GCMs is greatest at the multiple, rather than the single,grid-point level (Grotch and MacCracken, 1991; von Storch et al., 1993). The large-scale patterns aredefined using circulation types. Provided that consistent and distinct relationships exist between thecirculation types and rainfall in the observed data, and making the major assumption that theserelationships will be unchanged in a future warmer climate regime, a perturbed-run GCM outputcan be used to investigate changes in the frequency and intensity of rain storms in response toglobal warming (von Storch et al., 1993; Hewitson and Crane, 1996; Schubert and Henderson-Sellers,1997).

Links between large-scale circulation and surface weather have been explored and applied to envi-ronmental problems in synoptic climatology for many years (Yarnal, 1993), and provide a soundtheoretical basis for this approach (Trenberth, 1995). These links have been used to explore thepotential for downscaling from GCM output (Hay et al., 1991; Bardossy and Plate, 1992; Wilsonet al., 1992; Bogardi et al., 1993; Hughes and Guttorp, 1994; Schubert, 1994; Wilby et al., 1995;Conway et al., 1996; O8 zelkan et al., 1996; Schubert and Henderson-Sellers, 1997) and have beenapplied to GCM output at the daily level in order to construct rainfall scenarios (Hay et al., 1992;

Figure 1. Grid used in the automated circulation-typing scheme. The hatched area indicates the location of the Guadalentin Basinstudy area shown in Figure 5

© 1998 Royal Meteorological Society Int. J. Climatol. 10: 1051–1083 (1998)

DAILY RAINFALL IN SOUTHEAST SPAIN 1053

Hughes et al., 1993; Frey-Buness et al., 1995; Zorita et al., 1995; Cubasch et al., 1996; Hewitson andCrane, 1996).

Two major decisions must be made when using the circulation-type approach to downscaling. First,how to model the circulation/surface climate relationship. Four general approaches have been widelyused: (i) Markov Chain models (Bardossy and Plate, 1992; Hay et al., 1992; Wilson et al., 1992;Bogardi et al., 1993; Hughes and Guttorp, 1994; Schubert, 1994; Wilby et al., 1994); (ii) linearregression (Conway et al., 1996; O8 zelkan et al., 1996; Schubert and Henderson-Sellers, 1997); (iii)canonical correlation analysis (von Storch et al., 1993; Gyalistras et al., 1994; Noguer, 1994; Heyen etal., 1996); and, (iv) sampling from present-day instrumental analogue data (Zorita et al., 1995; Con-way et al., 1996; Cubasch et al., 1996). Most recently, fuzzy-rule based methods (Bardossy, 1994;O8 zelkan et al., 1996) and neural nets (Bardossy et al., 1994; Hewitson and Crane, 1994) have beenused. Here, the first approach is adopted. A statistical weather generator is used in which rainfalloccurrence is conditional upon the circulation type of each day, and the transition from one circula-tion type to another is modelled as a Markov Chain process.

The second decision which must be made is the choice of an appropriate automated circulationclassification scheme. The major distinction is between schemes based on existing circulation-typecatalogues, such as Lamb Weather Types (LWTs) (Lamb, 1972, used by Wilby et al. 1994, 1995), andschemes based on statistical grouping methods (Huth, 1996). The statistical methods most frequentlyused in downscaling studies include principal components analysis (Bogardi et al., 1993; Hughes andGuttorp, 1994; Cubasch et al., 1996; Hewitson and Crane, 1996; Schubert and Henderson-Sellers,1997; Zhang et al., 1997), canonical correlation analysis (Zorita et al., 1992; von Storch et al., 1993;Heyen et al., 1996), and cluster analysis (Wilson et al., 1992; Bogardi et al., 1993; Zhang et al., 1997).Neural nets and fuzzy-rule based methods are also being developed (Bardossy, 1994; Bardossy et al.,1994; Hewitson and Crane, 1994; Verdecchia et al., 1996).

Here an automated method based on LWTs is used (Jenkinson and Collinson, 1977; Jones et al.,1993) which has a number of advantages for downscaling. It is computationally simple and is basedon a single, widely-available, free atmosphere variable: daily gridded SLP. Free atmosphere variablesare reasonably well simulated by GCMs, at least over some regions (Crane and Barry, 1988; Hansenand Sutera, 1990; Hewitson and Crane, 1992) and are therefore considered to have advantages oversurface variables as predictor variables (Karl et al., 1990; Palutikof et al., 1997). Observed griddedSLP data sets have a resolution comparable to that of GCMs. The automated LWT scheme is easy tointerpret and has a sound physical basis in synoptic climatology.

The LWT-catalogue was initially developed for the British Isles using subjective methods (Lamb,1972) and has been used to predict present-day daily rainfall in the British Isles (Wilby et al., 1994,1995). The automated version categorises surface flow by direction (with a resolution of 45°) and type(cyclonic/anticyclonic, light/hybrid flow) (Jenkinson and Collinson, 1977; Jones et al., 1993). It hasbeen applied to control-run GCM output for the British Isles (Hulme et al., 1993) and the underlyingair flow indices used to predict present-day rainfall at two British sites (Conway et al., 1996). Intheory, it can be applied anywhere in Northern Hemisphere mid-latitudes. Here, the authors furtherdevelop and test its potential in a very different (Mediterranean) climate regime, using output from theUK Meteorological Office high-resolution transient GCM experiment (Murphy, 1995; Murphy andMitchell, 1995).

Because the automated scheme is transferred to a new region, it must first be demonstrated that itis physically valid for the Iberian Peninsula (Section 2). It is then shown that there are consistent anddistinct relationships between the circulation types and rainfall in the Guadalentin Basin (Section 3).The next step is to assess the accuracy of the GCM simulation of the observed circulation types(Section 4). The simulated perturbed-control run changes in circulation-type frequency are then investi-gated (Section 5), and translated into changes in rain-day frequency using a statistical weather genera-tor (Section 6).



Table I. The fourteen basic circulation types

Values of Z and F Description Type

ZBF Directional (resolution of 45°) N, NE ,E ,SE ,S ,SW ,W, NWZ\2F Z\0 Cyclonic C

ZB0 Anticyclonic AZB6 and FB6 Z\0 Unclassified cyclonic UC

ZB0 Unclassified anticyclonic UAFBZB2F Z\0 Hybrid cyclonic HYC

ZB0 Hybrid anticyclonic HYA

2. APPLICATION AND VALIDATION OF THE AUTOMATED CIRCULATION-TYPINGSCHEME

2.1. Application of the scheme

The automated LWT classification scheme has been used to classify SLP data for 1956–1989 from theNational Meteorological Center (NMC) CD-ROM data set. Data for the years (1961–64 and 1977) inwhich there are concerns about homogeneity (Trenberth and Paolino, 1980; Trenberth and Olson, 1988)are not used. The data have been interpolated to a 32-point 2.5° latitude by 3.75° longitude grid over thearea 36.25°N–46.25°N and 16.88°W–9.38°E (Figure 1). The grid spacing is the same as that used in theUK Meteorological Office high-resolution transient GCM experiment (Murphy, 1995). Fourteen basiccirculation types are defined using the calculated values of the resultant flow (F) and total shear vorticity(Z) (Jenkinson and Collinson, 1977; Jones et al., 1993), as shown in Table I.

The F and Z threshold values used to define the unclassified or indeterminate/light-flow circulationtypes (UC and UA), and the hybrid circulation types (HYC and HYA), were initially chosen to defineLWTs in the British Isles (Jenkinson and Collinson, 1977; Jones et al., 1993), and may not be the mostappropriate for use in other regions. Values of F and Z calculated from observed SLP over the IberianPeninsula have been plotted. These scatterplots do not show any clustering or grouping, indicating thatin this region, as elsewhere (Conway et al., 1996), the threshold values represent arbitrary values imposedon a smooth distribution. The identification of a more appropriate cut-off point within this smoothdistribution is not straightforward and hence a value of 6 has been retained.

2.2. The eight circulation-type groups

Some of the 14 circulation types are relatively infrequent over the study area in at least some seasons.This makes it difficult to establish reliable statistics for the precipitation regime associated with each type,particularly given the low number of rain days in the Guadalentin Basin (Section 3.1). For this reason,the 14 basic types were regrouped into eight types, including three directional groups (Table II).

Table II. The eight circulation-type groups

Type Description

C CyclonicHYC Hybrid-cyclonicUC Unclassified/light flow-cyclonic

Anticyclonic/hybrid-anticyclonicA/HYAUA Unclassified/light flow-anticyclonicW/NW/SW/N Westerly/northwesterly/southwesterly/northerly directional types (202.5–22.5°)E/NE Easterly/northeasterly directional types (22.5–112.5°)S/SE Southerly/southeasterly directional types (112.5–202.5°)



Table III. Mean annual (calendar year) and seasonal frequencies (days) of the eightcirculation-type groups calculated from the observed data, 1956–1989

Type Annual Winter Spring Summer Autumn

C 19.1 2.0 6.3 8.2 2.8HYC 19.4 3.2 5.4 7.1 3.4UC 90.8 5.9 20.9 42.4 21.1A/HYA 41.9 21.8 7.0 1.8 11.0UA 79.0 17.2 19.6 18.4 25.0W/NW/SW/N 68.6 26.7 19.5 5.0 16.7E/NE 33.1 8.3 9.3 7.9 7.8S/SE 12.3 4.1 4.0 1.2 3.2

These groupings were made after a preliminary examination of the rainfall data and SLP maps for somerandomly-selected days. It is only legitimate to combine types which share precipitation regimes andunderlying surface pressure patterns (Huth, 1996). The W/NW/SW/N directional types can be combinedbecause their associated air-masses must all cross the Iberian Peninsula before reaching the Guadalentin,and are not therefore major rain-producers. It would, however, be inappropriate to combine these typesif the study area were located on the western or northern Spanish coast where much of the rainfall isassociated with Atlantic weather systems (Lines Escardo, 1970). The A and HYA-types can be combinedbecause they must both by definition be characterised by high-pressure conditions and were immediatelyidentifiable as low-rainfall types. It was decided not to combine the C and HYC-types at this stagebecause of uncertainty over the direction of flow associated with them.

The mean annual and seasonal frequencies of the eight circulation-type groups calculated fromobserved SLP data for the period 1956–1989 are given in Table III. The mean seasonal cycles are shownin Figure 2. Mean frequencies simulated by the GCM are also shown and are discussed in Section 4.2.

Over the year as a whole, the most frequently observed circulation types are the two unclassified orindeterminate/light-flow types (UC and UA). The UC-type has a very strong seasonal cycle, with a winterminimum and a summer maximum. The UA-type is most frequent in autumn, but does not have a strongseasonal cycle. In contrast, the A/HYA-group has a very strong seasonal cycle, with a pronounced wintermaximum. The C and HYC-types have a late spring/summer maximum and a less pronounced seasonalcycle. The W/NW/SW/N-group has a strong seasonal cycle which peaks in late autumn/winter. TheE/NE-group is one of the least frequent types and does not have a seasonal cycle. The least frequentcategory is the S/SE-group and is particularly infrequent from May–September.

2.3. Validation of the typing scheme for southeast Spain

2.3.1. Construction of SLP composites from the obser6ations. For the typing scheme to be valid forsoutheast Spain, each of the eight circulation-type groups should have a characteristic underlying synopticpattern, which is physically distinct and produces the expected type and direction of surface flow (Huth,1996). To test this, composite SLP maps showing the mean pattern and the anomaly pattern wereconstructed, for each circulation type and each season, from the NMC CD-ROM data. Mean pressureand anomaly maps for winter and summer are shown in Figures 3 and 4, respectively.

The anomaly maps (Figure 4) are constructed by subtracting the long-term day mean from the meanof all days of a particular type. For example, to calculate the anomaly for the E/NE-type in winter, themean SLP pattern for all days in December, January and February for the period 1956–1989 classifiedas E/NE is calculated. The long-term day mean is calculated as the mean of all other days with the samedate.

2.3.2. Interpretation of the SLP composites. All the mean and anomaly patterns produce the expectedtype and direction of flow over southeast Spain (Figures 3 and 4). Most are consistent from season-to-sea-son (only winter and summer maps are shown), although the magnitude of the anomalies and the detail



of the patterns show some variability. Larger anomalies occur in winter and autumn than in summer, forexample.

The cyclonic-types (C and HYC) are characterised by a weaker Icelandic Low, a weaker Azores Highand a low-pressure anomaly centred over the Iberian Peninsula. The A/HYA-group is associated with astronger and more extensive Azores High, and with a deeper Icelandic Low. The cyclonic and anticyclonic

Figure 2. Observed and simulated monthly frequencies of the eight circulation types. The solid line shows the mean frequency (no.of days per month) calculated from observed SLP data (1956–1989). The shaded area shows the maximum and minimum frequencyrange observed over any ten-year period. The dashed line shows the mean frequency calculated from UKTR control-run model

output



patterns are reverse images of each other, and appear to represent different phases of the North AtlanticOscillation (NAO). The cyclonic pattern is very similar to the Greenland Above mode of the NAO, whilethe anticyclonic pattern resembles the Greenland Below mode (van Loon and Rogers, 1978).

The HYC-type anomaly patterns are very similar to the C-type patterns, although the HYC-typeanomalies are somewhat smaller in magnitude. The similarity of these patterns suggests that it may be

Figure 3. Seasonal mean SLP (hPa) for the eight circulation types calculated from observed data, 1956–1989, (a) winter; and, (b)summer



Figure 3 (Continued)

legitimate to combine the two types. However, this cannot be done without first demonstrating that theyalso share a similar rainfall regime (see Section 3.2). The UC-type anomalies are considerably smaller inmagnitude than both the C and HYC-type anomalies in every season, and the pattern of these anomaliesis different, particularly in winter and summer. The A/HYA and UA-type patterns are broadly similar,with smaller anomalies in the latter case.



In the case of the W/NW/SW/N-group, a low pressure anomaly is located over southern UK/north-western France (Figure 4), suggesting that the North Atlantic storm tracks are shifted south of theirnormal position. The Azores High maintains its normal position (Figure 3).

With E/NE-type flow, a high-pressure centre appears to the northwest of France/southeast of Ireland(Figure 3). Another characteristic feature of this circulation type is a small area of negative anomalies

Figure 4. Seasonal SLP anomalies (hPa) for the eight circulation types calculated from observed data, 1956–1989. Negativeanomalies are indicated by dashed lines and positive anomalies by solid lines. (a) winter; and, (b) summer




centred over the southern/central Mediterranean Sea (Figure 4). These anomalies may be related tointensification of action in the areas of cyclogenesis in the Gulf of Genoa and Northwest Africa, andpossibly the area of Alpine lee cyclogenesis in the Gulf of Lions (Lines Escardo, 1970; Dalu and deGregorio, 1987; Alpert et al., 1990; Prezerakos et al., 1990). The boundary between the low-pressureanomaly and the more extensive, high-pressure anomaly to the north lies across southeast Spain,channelling flow along the coast.



Figure 5. Mean annual rainfall (mm) for 20 stations in the Guadalentin Basin, 1958–1987 and the location of the six stations usedhere: AG—Aguilas; AL—Alcantarilla; AM—Alhama de Murcia; FA—Fuente Alamo; LO—Lorca; and, TO—Totana

Except in summer, S/SE-type flow is associated with higher pressure throughout central and northwestEurope, and with an area of lower pressure to the southwest of the Iberian Peninsula (Figure 4). TheAzores High is almost unrecognisable as a high pressure feature (Figure 3). It is possible that the highpressure over central and northwest Europe is related to the westward extension of the Siberiananticyclone. The boundary between the high-pressure and low-pressure anomalies lies just to the east ofthe Guadalentin Basin (Figure 4), giving strong southerly/south-easterly flow over the region. Theanomaly pattern is somewhat different in summer with a strengthening of the Azores High, although thistype only occurs, on average, once each summer.

On the basis of the SLP composite maps, it is concluded that each of the eight circulation-type groupshas a characteristic underlying pressure pattern, which is physically distinct and produces the expectedtype and direction of surface flow over southeast Spain and the Guadalentin Basin. (It is noted, however,that the C and HYC-types have similar underlying pressure patterns and that it may be legitimate tocombine them.) It is therefore valid to apply this classification scheme in southeast Spain. It provides anappropriate basis for the next stage, the identification of relationships between the circulation types anddaily rainfall.

Table IV. Details of the six Guadalentin stations. Means are for the period 1958–1987

Altitude Annual rainfall: Rain-days:Mean annualLatitude Longitude Mean no. ofS.D.rain daysS.D.rainfall (mm)(m)

5 178 74 29Aguilas 37.4 11−1.6289 113 48Alcantarilla 38.0 −1.2 75 12

44145418 13760Alhama de −1.537.9Murcia

200 272 107 34Fuente Alamo 37.7 8−1.283 38 11Lorca 37.7 −1.7 335 234

200 293 121 28Totana 37.8 7−1.5



Figure 6. Seasonal cycle of mean monthly rainfall (mm) for the wettest (solid line, Alhama de Murcia) and driest (dashed line,Aguilas) of the six Guadalentin Basin stations

3. OBSERVED RELATIONSHIPS BETWEEN THE CIRCULATION TYPES AND THERAINFALL REGIME

3.1. Rainfall in the Guadalentin Basin

Before implementing the downscaling scheme to generate rainfall scenarios, it must be shown thatmeaningful relationships exist in the observations between the eight circulation-type groups and precipita-tion. The five stations within the Guadalentin Basin selected for analysis are: Alcantarilla, Alhama deMurcia, Fuente Alamo, Lorca and Totana. The sixth station, Aguilas, is located on the coast just outsidethe Basin itself. Station details are given in Table IV. Their varying precipitation regimes reflect the effectsof complex topography in this region (Figure 5). Mean annual rainfall ranges from 178 mm at Aguilas to

Figure 7. The mean annual frequency (shown as the percentage of all days) of the eight circulation types (solid bars); and theircontribution to annual rainfall (shown as the percentage of total annual rainfall) for six stations in the Guadalentin Basin (open

bars)



Figure 8. Standardised ratios PROPct/PROPtot for six stations in the Guadalentin Basin. Circulation type ‘W’=W/NW/SW/N. (a)winter; and, (b) summer

418 mm at Alhama de Murcia. The mean annual number of rain days ranges from 28 days at Totana to48 days at Alcantarilla. The seasonal rainfall regime is typically Mediterranean in that winter rainfall isat least three times the summer rainfall (Koppen, 1936), but differs from that across much of theMediterranean Basin in having two peaks: a major peak in October and a slightly lower peak in April(Figure 6).

3.2. Relationships between the circulation types and daily rainfall

It is possible to identify those of the eight circulation-type groups for which the contribution to annualrainfall is significantly greater (or less) than average (Figure 7). The percentage contribution to annualrainfall from the UC-type is the same as the percentage of days which are of this type (about 25%). Incontrast, the E/NE-type also contributes about 25% of annual rainfall, but occurs on less than 10% ofdays. The percentage contributions to annual rainfall of the C, HYC and S/SE-types are also greater than



the percentage of days on which they occur. The rainfall contribution from the A/HYA, UA andW/NW/SW/N-types is considerably lower than their percentage occurrence.

The high percentage rainfall contribution from the C, HYC, E/NE and S/SE-types may be because ahigh proportion of days of a particular type are wet, or because there are only a few wet days, with a largeamount of rainfall on each. It is important to distinguish between the two, because the potential impactsare quite different. For example, runoff and erosion may be limited when rainfall is spread over a largenumber of low-intensity days but may be a serious problem where there are a few high-intensity rain days.

The first possibility, that a high proportion of type days are wet, is indicated by the ratio PROPct/PROPtot, where PROPct is the proportion of type days which are wet and PROPtot is the proportion ofall days which are wet. The second possibility, that a large amount of rain falls on each type day, isindicated by the ratio PRECct/PRECtot, where PRECct is the mean amount of rain which falls on a wettype day and PRECtot is the mean amount of rain which falls on any wet day. Annual and seasonal ratioshave been calculated for each station. Ratios for winter and summer are plotted in Figures 8 and 9. Fora circulation type with a ratio greater than 1.0, the likelihood of rain or the amount of rain per rain day

Figure 9. Standardised ratios PRECct/PRECtot for six stations in the Guadalentin Basin. Circulation type ‘W’=W/NW/SW/N. (a)winter; and, (b) summer



Table V. High-rainfall circulation types with a higher than average proportion of wetdays at every station (+) and a higher than average amount of rain per rain day at

every station (*)

C HYC UC A/HYA UA W/NW/SW/N E/NE S/SE

High rainfallAnnual + + +* +Winter + + + +* +Spring + +*1 + +*Summer + +* +Autumn + + * +* +

Low rainfallAnnual − − −xWinter − −xSpring − −x −xSummer x x −x − −Autumn x x − − −x

1 Rain per rain day is higher than average at every station except Aguilas.Low-rainfall circulation types with a lower than average proportion of wet days at every station (–)and a lower than average amount of rain per rain day at every station (x).

is greater than the station mean, and 6ice 6ersa. The seasonal ratios indicate considerable between-seasonvariability, but a number of consistent relationships can be identified. While the C, HYC, and (in winter)the UC-types tend to have a high proportion of wet days of average intensity, the E/NE and S/SE-typesalso tend to have a high proportion of wet days, but of high intensity. The amount of rain per rain daytends to show greater variability from station to station than the proportion of wet days, particularly insummer and autumn.

The similarity of the pressure patterns underlying the C and HYC-types was noted in Section 2.3.2 andit was concluded that it would be legitimate to combine them if they share a similar rainfall regime. Atthe annual level, the rainfall regimes of the two types are similar. Both have a higher than averageproportion of wet days, for example. The rainfall regimes are broadly similar in winter and spring, but aredifferent in summer and autumn when both the probability of rain and the amount of rain per rain dayare consistently higher for the HYC-type. For these reasons, it was decided not to combine these twotypes.

3.3. E6aluation of the circulation type/rainfall relationships

A number of consistent relationships can be identified between the circulation types and the rainfallregime and are summarised in Table V. The important high-rainfall circulation types are the C, HYC,UC, E/NE and S/SE-types whereas the A/HYA, UA and W/NW/SW/N-types are the most consistentlow-rainfall types. The extent to which these relationships can be explained by the synoptic situationunderlying each type is discussed here.

The cyclonic types (C, HYC and UC) have a higher than average proportion of rain days, in at leastsome seasons (Table V), but, except in spring for the HYC-type and in autumn for the UC-type, do nothave a higher than average amount of rain per rain-day. This may be because the rain is associated withfrontal events rather than more ‘explosive’ convective events, and/or because of the short sea track acrossthe Mediterranean (Figure 3). In Section 2.3.2, it is noted that the C and HYC-type anomaly patterns(Figure 4) resemble the Greenland Above mode of the NAO (van Loon and Rogers, 1978). Thiscirculation mode produces high-pressure blocking in the Northeast Atlantic, and a more meridionalcirculation (Jacobeit, 1987; Moses et al., 1987; Maheras, 1988). Upper-air troughs and incursions of polarair over the Mediterranean are more frequent, and the Atlantic storm tracks are displaced south. All thesefactors are conducive to wetter conditions in the western Mediterranean (Jacobeit, 1987; Moses et al.,1987; Maheras, 1988; Kutiel et al., 1996).



The A/HYA anomaly pattern (Figure 4) resembles that associated with the Greenland Below mode ofthe NAO (van Loon and Rogers, 1978) and with high positive values of the NAO pressure index (Hurrell,1995). During times of high NAO index, moisture transport across the North Atlantic has a moresouthwest-to-northeast orientation and extends further into northern Europe and Scandinavia. Incontrast, moisture transport, and hence rainfall, is reduced over southern Europe and the Mediterranean(Hurrell, 1995; Moulin et al., 1997). The southwest-to-northeast orientation of prevailing flow is evidentin the composite SLP maps for the A/HYA-type (Figures 3 and 4). The high-pressure centre over theIberian Peninsula, together with the relatively short sea track of the prevailing circulation (Figure 3), isunlikely to cause precipitation over southeast Spain. The synoptic conditions associated with theW/NW/SW/N group (Figure 3) also result in zonal flow which, in this case, will bring rain to the Atlanticcoast of the Iberian Peninsula, but not to the sheltered southeast Mediterranean coast (Lines Escardo,1970). A number of studies confirm that rainfall tends to be below average across the westernMediterranean when zonal circulation dominates (Jacobeit, 1987; Maheras, 1988; Bardossy and Caspary,1990; Kutiel et al., 1996).

High-pressure blocking in the Northeast Atlantic occurs in the Greenland Above mode of the NAO(van Loon and Rogers, 1978) and is also seen in the cyclonic and E/NE-type anomaly patterns (Figure4). In comparison with the cyclonic types, the E/NE-type positive anomalies are associated with a muchstronger and more clearly-defined high-pressure centre (Figure 3). Moreover, the area of below averagepressure is much smaller, and is located over the central–southern Mediterranean rather than over theIberian Peninsula (Figure 4). This configuration gives a longer sea track across the Mediterranean (Figure3), allowing surface and near-surface air masses to pick up more moisture and heat. The prevailing windsare onshore along the southeast Spanish coast and over the Guadalentin Basin. Rainfall, often convectivein nature, is likely to occur when warm, moist air masses meet the coastal mountains (Lines Escardo,1970; Dalu and de Gregorio, 1987; Fernandez Mills, 1995).

Onshore flow over the Guadalentin Basin is also produced by the S/SE-type although the direction ofapproach is somewhat different, with a shorter track over the Mediterranean (Figure 3). Flow from thisdirection is sometimes associated with Atlantic depressions, approaching from the southwest andfunnelled through the Straits of Gibraltar (Tout, 1991), or with depressions formed in the Gulf of Cadizand moving eastwards (Lines Escardo, 1970). These systems can bring high rainfall to the Andaluciaregion, but are less likely to be associated with high-intensity rainfall in the Guadalentin.

The synoptic conditions underlying the E/NE-type resemble those associated with the most destructivehigh-rainfall events in this region of Spain (Lines Escardo, 1970; Lawson, 1989; Tout and Wheeler, 1990;Wheeler, 1990; Tout, 1991). These storm events are most frequent in September and October and it isnoted that severe storms were reported across southeast Spain and the Guadalentin Basin in October 1973and 1982 (Tout and Wheeler, 1990) on days classified as E/NE. But severe storms also occur on othertype days and not all E/NE days are associated with storms. The October 1988 storms which affectedmost of southeast Spain (Lawson, 1989), for example, occurred on days classified as HYC.

4. VALIDATION OF GCM CIRCULATION TYPES

4.1. The UKTR model

Before the circulation type/rainfall relationships identified from the observations can be used tointerpret GCM output, it is necessary to validate SLP output from the GCM control run. The model usedin this study is the UK Meteorological Office high resolution GCM (UKTR) run in transient mode(Murphy, 1995; Murphy and Mitchell, 1995). Daily output is available from the years 66–75 of the75-year long control and perturbed simulations. The grid spacing is 2.5° latitude by 3.75° longitude, asshown in Figure 1. In the perturbed simulation, CO2 forcing is increased by 1% per annum compound anddoubles in year 70.

Inspection of the mean SLP fields from the control run for winter and summer indicates that the modelis reasonably successful in reproducing the main features of the general circulation system although some



systematic errors can be identified (Figure 25 from Murphy, 1995). In winter, for example, the Siberiananticyclone is too weak, and the Icelandic Low extends too far eastwards into central Europe. Meanseasonal SLP maps for the Northeast Atlantic and Europe were constructed from UKTR control-rundaily data, as shown in Figure 10. Control-run minus observed differences are also shown and indicatethat the simulated Icelandic Low is too intense in every season. The position and intensity of the AzoresHigh is well simulated in winter and spring. In summer, it does not extend far enough eastwards and inautumn it extends too far eastwards.

Figure 10. (a) Seasonal mean SLP (hPa) calculated from UKTR control-run model output; (b) UKTR control-run−observeddifferences in seasonal mean SLP (hPa). Negative differences are indicated by dashed lines and positive differences by solid lines



A more detailed study of UKTR storm tracks and cyclones during winter confirms that there are‘substantial errors in the position of the mid-latitude storm tracks’ particularly in the eastern NorthAtlantic (Carnell et al., 1996). Atlantic storm tracks extend too far into Europe and too far south,reaching a maximum over Mediterranean latitudes. The model tends to over-predict the frequency ofweak Atlantic depressions and to under-predict the frequency of very deep depressions (Carnell et al.,1996).

These studies (Murphy, 1995; Carnell et al., 1996) indicate that, like other GCMs (Liang et al., 1996;Risbey and Stone, 1996), the UKTR model is able to reproduce the main features of the large-scalegeneral circulation but fails to reproduce its finer details. However, a validation study more appropriatefor these purposes is provided by determining how well the model simulates circulation types over theIberian Peninsula.

4.2. Simulated circulation-type frequencies

Seasonal and monthly frequencies of the eight circulation-type groups have been calculated from the10 years of control-run output of the UKTR model. These are compared with the observations inFigure 2. The minimum and maximum observed frequencies in any decade during the period 1956–1989 are indicated by the hatched area. Whenever the UKTR values fall within this area the model isconsidered to be performing well. Thus the model reproduces the seasonal cycles of the HYC, UC andS/SE-types reasonably well.

The differences (in no. of days and S.D. units) between the mean seasonal circulation-type frequen-cies calculated from model output and from the observations are given in Table VI. Significantdifferences are identified using the Mann Whitney/Wilcoxon rank sum test. The following discussionfocuses on the circulation types identified as high-rainfall types in Section 3 (the C, HYC, UC, E/NEand S/SE-types).

Fifty per cent of all differences are significant. Some of the errors are quite large, greater than 1S.D. unit in nearly 40% of cases, and greater than 2 S.D. units in 10% of cases. More significanterrors occur in winter and summer than in spring and autumn.

The UC-type is underestimated in every season (significantly so in every season except autumn), andthe A/HYA-type overestimated in every season (significantly so in every season except winter). How-ever, most of the differences between the simulated and observed frequencies are not consistent fromseason-to-season. The model underestimates the frequency of the C, HYC and UC-types in spring andsummer, and over the year as a whole. These circulation types are all associated with a weaker thannormal Icelandic Low (Figure 4) so this error is probably related to the overestimation of the intensityof the Icelandic Low (Figure 10). The frequency of the E/NE-type is underestimated by the model inevery season except summer, when it is overestimated. This gives a seasonal cycle which is not seen inthe observations. The frequency of the S/SE-types is simulated fairly well, except in winter (underesti-mated) and autumn (overestimated).

The errors associated with the simulation of the circulation-type frequencies identified here suggestthat there may also be problems with the day-to-day sequence and persistence of the simulatedcirculation types. However, it is considered that the performance of the UKTR model is adequate forthe major purpose of this paper, which is to evaluate the potential of the circulation-type basedapproach for regional scenario construction and to provide some illustrative results.

4.3. Simulated relationships between the circulation types and synoptic features

A further validation test is provided by comparing SLP composite maps constructed from control-run model output with maps constructed from observed data (Figure 4). The pressure patterns overthe 32 grid-points used to define each type (Figure 1) must be similar because the same classificationsystem is used for observed and model data. However, the composites cover a wider geographicalarea, and most interesting is the ability of the model to reproduce the variations in large-scalecirculation with circulation type which were identified in Section 2.3.2. Comparisons have been made


DA

ILY

RA

INF

AL

LIN

SOU

TH

EA

STSP

AIN

1069

©1998

Royal

Meteorological

SocietyInt.

J.C

limatol.

10:1051

–1083

(1998)

Table VI. Actual differences (modelled−observed (days)) and standardised differences (difference/observed S.D. (s units)) between observed and simulatedcirculation-type frequencies

Winter SummerSpring SpringSummer WinterAutumn AnnualAnnualType Autumns unitsdaysdays s unitss unitsdays s unitsdays s unitsdays

0 −1.0 +1.3 −0.7 −1.0−3 0.0−5**+2**−6**CHYC −1 B1 −0.2 +0.4 −0.3 −0.3 +0.1−1 +1 −1

−1 −1.8 −0.9 −1.7 −1.7 −0.1UC −28** −2** −9** −15**+6** +2.6 +0.4 +3.5 +2.6+8** +1.3+12**+3+28**A/HYA+2 +0.1 −1.3 +0.3 +0.6 +0.3UA +2 −9** +2 +6**

−10** +0.6 +1.8 +0.7 −0.1B1 −1.4+4+14**+8W/NW/SW/N−6** +8** −1 −0.6 −1.2 −2.0 +2.3 −0.4E/NE −5 −6**

+4** +0.4 −1.0 +0.2 −0.1−1 +1.3−2**S/SE +2 +1

** indicates changes which are significant at the 5% level.


for each season; anomaly maps for the UKTR model, for winter and summer only, are shown in Figure11.

The position and sign of the anomalies are very similar in the observed (Figure 4) and simulated (Figure11) composites. However, the simulated anomalies tend to be smaller in magnitude than the observed.This is most evident in the case of the negative anomalies associated with the W/NW/SW/N-type in

Figure 11. Seasonal SLP anomalies (hPa) for the eight circulation types calculated from UKTR control-run model output. Negativeanomalies are indicated by dashed lines and positive anomalies by solid lines. (a) winter; and, (b) summer.



winter. The magnitudes of the positive anomalies associated with the E/NE-type are underestimated inevery season except winter. The tendency to underestimate the strength of the anomalies may arisebecause only 10 years of model data are available, but is most likely to be related to low year-to-yearvariability in the model.




Figure 12. UKTR perturbed-run minus control-run differences in seasonal mean SLP (hPa). Negative differences are indicated bydashed lines and positive differences by solid lines

5. FUTURE CHANGES IN CIRCULATION-TYPE FREQUENCY

5.1. Mean sea le6el pressure changes

The mean winter and summer SLP fields simulated by the UKTR GCM show some coherent changesover the North Atlantic/Europe region from the control to the perturbed simulation (Figure 24 inMurphy and Mitchell, 1995). In winter, the strength of Atlantic storm tracks (Carnell et al., 1996), andthe temperature gradient between 30 and 60°N (Murphy and Mitchell, 1995), increase in the perturbedsimulation. Enhanced westerly flow is simulated across northwest Europe in winter and the frequency ofdeep depressions increases (Carnell et al., 1996). These changes are related to a reduction in SLP over theArctic Ocean and over the mid-latitude continents, and to an increase over the mid-latitude oceans. TheSLP changes are similar in winter and summer. Summer changes, although smaller in magnitude, have agreater statistical significance (Murphy and Mitchell, 1995). These patterns of change are apparent afterthe first few decades of the experiment, but, even by the final decade, their amplitude is ‘small whencompared with unforced inter-annual variations’ (Murphy and Mitchell, 1995). The storm-track changesare ‘unlikely to be statistically detectable’ (Carnell et al., 1996). Moreover, the ‘largest changes occur inthe region of greatest model systematic error and must therefore be treated with caution’ (Carnell et al.,1996).

The change (perturbed-run minus control-run) in mean seasonal SLP for the study area is shown inFigure 12. The winter map shows the increase in westerly flow across northwest Europe indicated byMurphy and Mitchell (1995), with lower pressure to the north and higher pressure over the British Isles,western and central Europe, and the Mediterranean. The mean SLP increase over the Iberian Peninsulais in the range 1–2 hPa. In every other season, SLP decreases over the Peninsula, by 0–0.5 hPa in springand by 0.5–1 hPa in summer and autumn. How do these ‘small’ changes in SLP translate into changesin circulation-type frequency?



5.2. Seasonal changes in circulation-type frequency

Seasonal and monthly frequencies of the eight circulation-type groups have been calculated using SLPdata from the perturbed run. The mean seasonal changes (perturbed-run minus control-run) are shown inTable VII. Significant changes are identified using the Mann Whitney/Wilcoxon rank sum test.

The largest changes occur in summer. Of the high-rainfall types, there is a significant increase in thefrequency of the C and HYC-types, reflecting the lower mean SLP over the Iberian Peninsula (Figure 12).The frequency of both the E/NE and S/SE-types increases in summer, but the changes are not significant.Of the low-rainfall types, the A/HYA and UA-types show a significant decrease in frequency in summerand the W/NW/SW/N group shows a non-significant decrease. The effect of these changes is expected tobe an increase in the number of rain days in summer. This is in contrast to the rainfall output for the gridsquare closest to the Guadalentin study area which indicates a 60% reduction in the number of rain daysin summer.

Few significant or consistent changes in circulation-type frequency are predicted in other seasons. Inspring, however, decreases in the high-rainfall C, HYC, E/NE and S/SE-types, together with increases inthe low-rainfall UA and W/NW/SW/N-types suggest a reduction in the number of rain days. This is inagreement with the direction of change in spring indicated by the nearest model grid-point.

SLP composite maps were constructed using perturbed-run model output. The anomaly patterns arevery similar to the control-run patterns (Figure 11), so are not shown here. This similarity indicates thatthe relatively small changes in mean SLP are not associated with changes in the synoptic situationunderlying each circulation type but rather with changes in their frequency. In order to investigaterigorously how these frequency changes translate into changes in the number of rain days, a stochasticweather generator is used, as described in the next section.

6. USE OF A WEATHER GENERATOR TO PREDICT RAIN-DAY CHANGES

6.1. The conditional weather generator

A number of studies have used statistical models in which rainfall occurrence is conditional upon thecirculation pattern of each day, and in which the transition from one circulation type to another ismodelled as a Markov Chain process (Hay et al., 1991; Bardossy and Plate, 1992; Wilson et al., 1992;Hughes et al., 1993; Hughes and Guttorp, 1994; Wilby et al., 1994). This type of a model is referred toas a conditional weather generator (CWG).

The CWG used here follows the approach of Hay et al. (1991) and Wilby et al. (1994). Rainfalloccurrence is defined by two parameters calculated for each of the eight circulation types and for eachseason: the probability (PROBct1–8

) (expressed as a transition matrix) of the next day being circulation-type 1(C)–8(S/SE), and the mean probability of rain (PROBPRECct) related to each type. The PROBct1–8

Table VII. Mean seasonal changes (perturbed−control run) in the frequency (no. ofdays) of the eight circulation-type groups

Summer AutumnSpringType Winter

−0.8+4.4**−0.7−0.5C−1.5+0.4 +3.1*HYC −0.4

+2.4−0.8 +0.3UC +3.6−2.70.0 −0.8A/HYA −4.3**−2.7−7.3**+2.3+3.5UA+3.5*−4.5 +3.1 −1.0W/NW/SW/N

+1.1−1.4*−0.2E/NE −0.4S/SE +1.2+0.4−1.1+2.1*

** Indicates changes which are significant at the 5% level.* Indicates changes which are significant at the 10% level.



Table VIII. The circulation-type transition matrix (PROBct1–8) for Alcantarilla in winter and the probability

of rainfall (PROBPRECct) calculated from the observations

Next day S/SEC HYC UC A/HYA UA W/NW/SW/N E/NECurrent day

C 0.1484 0.3704 0.5000 0.5370 0.6481 0.8519 0.9815 1.000HYC 0.1236 0.2921 0.4045 0.4607 0.4944 0.7416 0.9551 1.000UC 0.0500 0.1250 0.2938 0.4063 0.6563 0.8313 0.9125 1.000A/HYA 0.0066 0.0116 0.0314 0.4645 0.6430 0.9273 0.9570 1.000UA 0.0108 0.0280 0.1312 0.3634 0.7720 0.8860 0.9462 1.000W/NW/SW/N 0.0187 0.0576 0.1058 0.3092 0.3681 0.9304 0.9866 1.000E/NE 0.0045 0.0090 0.0769 0.2353 0.4796 0.5701 0.9819 1.000S/SE 0.0459 0.1284 0.2110 0.3761 0.5780 0.7248 0.7431 1.000PROBPRECct 0.40 0.38 0.29 0.06 0.15 0.12 0.22 0.29

transition matrix and PROBPRECct values for Alcantarilla in winter calculated from the observations areshown in Table VIII. On each day, a random number is selected and used to determine the next day’scirculation type from the transition matrix. A second random number is selected, and used to determinewhether the day is wet or dry. (For the purposes of the sensitivity experiments described here, only thenumber of rain-days (NRD) is of interest. The amount of rain can, however, be simulated by samplingfrom an appropriate distribution.)

The CWG is used to produced climate-change scenarios based on the assumption that changes incirculation-type frequency will be propagated through to changes in rainfall frequency and amount. Thisassumption cannot be tested for the future but can be tested, in part, for the past by looking at observedcirculation-type/rainfall relationships. The proportion of circulation-type days which are wet at Alcantar-illa is shown in Figure 13 for each season and for overlapping decades during the period 1958–1987. Thegreatest variability occurs in cases where there are very few (less than 3) type-days (C in winter andautumn, and S/SE in summer). Figure 13 demonstrates that there are coherent and stable relationshipsbetween the circulation-type frequencies and rainfall occurrence. Thus, it is reasonable to assume thatchanges in circulation-type frequency will be reflected as changes in rainfall occurrence.

Three sets of 100 30-year long simulations have been performed for the six Guadalentin Basin stations.The sequence of circulation types in each 30-year simulation is dependent on the transition matrix and therandom number generator. Thus the 100 sequences making up each simulation set should be differentwhereas, if the circulation types were taken directly from the UKTR model, all 100 sequences would bethe same. The method adopted here therefore introduces a greater ‘Monte Carlo’ and probabilisticelement to scenario construction.

In the first simulation set (Gen; Section 6.2), the CWG parameters (PROBct1–8and PROBPRECct) are

calculated from the observed data. All available data for 1958–1987 are used in order to maximise thesample sizes. The output from these simulations cannot, therefore, be used for independent validation ofmean values, but can be used to explore the effect of the CWG on second-order statistics such as varianceand persistence.

In the second set of simulations (Cont.; Section 6.3), the PROBct1–8parameters are calculated from

control-run output of the UKTR GCM and PROBPRECct is calculated from the observations. Thesesimulations allow further investigation of the GCM’s ability to simulate the frequency of circulation types.In the final set of simulations (Pert.; Section 6.4), the PROBct1–8

probabilities are calculated fromperturbed-run GCM output and PROBPRECct is again calculated from the observations. The differencesbetween the Pert. and Cont. simulations provide the rain-day climate change scenarios (Table XI).

The performance of the CWG is similar at all stations. For reasons of space, therefore, results for onestation, Alcantarilla, are summarised in Table IX (circulation types: CT) and Table X (rain days: NRD).The mean (M) and S.D. (s) are calculated for every 30-year time series from the Gen. (g), Cont. (c), andPert. (p) simulation sets. Thus, for each season, the vectors indicated by notation such as Mg(CT) or


DA

ILY

RA

INF

AL

LIN

SOU

TH

EA

STSP

AIN

1075

©1998

Royal

Meteorological

SocietyInt.

J.C

limatol.

10:1051

–1083

(1998)

Table IX. Summary of CWG circulation-type results for Alcantarilla in winter and summer. See text for full explanation

Number of circulation-type days S.D.

UC A/HYA UA W/NW/SW/N E/NE S/SE C HYC UC A/HYA UA W/NW/SW/N E/NE S/SEC HYC

Winter16.8 28.0 8.7Obs(CT)/sobs(CT) 3.9 1.3 1.7 2.2 6.5 6.4 7.1 4.8 1.92.1 3.1 5.7 21.6

8.5 41.3 1.8 1.9 2.0 2.5 1.5 6.824.6 3.4UKTR(CT)/suktr(CT) 7.0 1.2 1.73.84.14.021.6Mean of Mg(CT)/sg(CT) 16.5 28.3 7.9 4.1 1.6 2.2 2.6 5.1 4.8 6.4 3.6 2.42.1 3.5 6.1

2 1Sig+ 1 3 2 32 1 16 3 46 2 77Sig− 25

24.3Mean of Mc(CT)/sc(CT) 8.5 41.3 1.7 2.0 2.2 1.9 2.0 4.5 3.4 6.0 1.5 1.84.1 4.2 3.9100Sig+ 92 11 34

100 100 100 4 12Sig− 10024.9Mean of Mp(CT)/sp(CT) 12.0 36.9 1.6 4.0 2.2 2.0 1.6 5.1 3.9 5.8 1.4 2.33.5 4.3 3.0

− + + − + + + − + + +−Sig +

Summer19.1 4.9 7.8 1.1 5.3 4.2Obs(CT)/sobs(CT) 9.47.9 3.1 9.1 2.6 3.3 1.16.9 42.3 2.0

27.4 9.8 24.2 4.6 16.2 1.1 1.5 2.5 4.4 2.7 3.4 2.9 5.3 1.22.8 5.9UKTR(CT)/suktr(CT)

16.9 4.8 7.9 1.3 3.8 3.0 6.1Mean of Mg(CT)/sg(CT) 1.3 4.8 2.4 3.4 1.28.3 7.5 43.9 1.41 10Sig+ 1 1

9Sig− 5 23 1 3 12 54 41 3 2Mean of Mc(CT)/sc(CT) 9.3 23.8 4.8 16.5 1.1 1.6 2.6 5.6 3.7 5.6 2.4 4.9 1.32.8 6.1 27.6

100 100 21100 1 3Sig+Sig− 11 53 15 2 26 2100 66 100

16.8Mean of Mp(CT)/sp(CT) 3.5 17.6 1.6 4.0 3.7 6.1 3.0 4.9 2.0 4.8 1.37.5 9.3 30.6 5.1− − + + + + + − −− −Sig + ++

C.M

.G

OO

DE

SSA

ND

J.P.

PA

LU

TIK

OF

1076

©1998

Royal

Meteorological

SocietyInt.

J.C

limatol.

10:1051

–1083

(1998)

Table X. Summary of CWG rain-day results for Alcantarilla

LW LDNo. of rain days S.D.

Mc(NRD) Mp(NRD) sobs(NRD) sg(NRD) sc(NRD) sp(NRD) Lwobs LWg LWc LWp LDobs LDg LDc LDpObs(NRD) Mg(NRD)

Winter6.4Mean 3.5 3.3 3.4 8 3 3 3 58 35 37 3914.8 14.1 12.9 13.2

4.9 4.2 4.515.0Max 14.215.811.8Min 2.4 2.3 2.312.3 11.2

Sig+ ****73 741Sig− 26

Spring11.2 10.5 6.0 3.6 3.3 3.1 8 4 3 3 58 34 43 4614.7 14.9Mean

4.7 4.8 4.4Max 16.8 12.5 12.12.6 2.2 2.39.2Min 9.713.5

Sig+97 ** 96 94 **Sig−

Summer2.1 2.2 2.1 2.2 3 2Mean 15.2 2 73 68 68 645.4 5.0 5.7

6.1 7.0 3.3 2.8 3.46.6Max1.5 1.5 1.3Min 4.3 4.1 4.7

****Sig+1Sig− 15 22

Autumn13.2 13.5 6.3 3.4 3.4 3.6 6 3 3 3 58 39 39 3813.3 13.2Mean

4.6 4.7 4.615.314.7 14.9Max11.9Min 2.2 2.5 2.411.5 11.6

Sig+ ****100 100Sig−

See text for full explanation.


Figure 13. Proportion of wet days at Alcantarilla for each circulation type and each season calculated for overlapping decades fromthe observations. Solid line, median; box, interquartile range; whiskers, spread of points within 1.5× interquartile range;�, outliers.

Circulation type ‘W’, W/NW/SW/N



sp(NRD) contain 100 values. For winter and summer only, means of the CT vectors are shown in Table IX.Maximum and minimum values of Mc(CT) and sc(CT) were also calculated but are not shown. Meanfrequencies and S.D.s for the observed series (Obs(CT) and sobs(CT)) and for UKTR control-run output(UKTR(CT) and suktr(CT)) are shown in the first two rows.

In Table X the following values are shown for the observed rainfall series: mean NRjD (Obs(NRD)) andS.D. (sobs(NRD)), and length in days of the longest wet (LWobs) and dry (LDobs) periods. The LW and LDparameters provide an indication of rainfall persistence. The means of the Mg(NRD), Mc(NRD), Mp(NRD) andsg(NRD), sc(NRD), sp(NRD) vectors are also shown, together with the maximum and minimum vector values.The right-hand columns show vector means for the LW and LD parameters.

The 100 individual time-series from the Gen. and Cont. simulations are each compared with theobserved series using the Mann Whitney/Wilcoxon rank sum test. The variance of these individual seriesis compared using the Siegel-Tukey rank sum dispersion test (Kanji, 1993). Tables IX and X show thenumber of times out of 100 simulations that the mean or S.D. of the simulated series is significantlygreater (Sig+ ) or smaller (Sig− ) than that of the observed series. The Mc and Mp vectors are comparedusing the Mann Whitney-Wilcoxon rank sum test. A ‘+ ’ (− ) in the Sig row of Table IX indicates thatthe mean of the Mp vector is significantly greater(smaller) than the mean of the Mc vector. The equivalentfor Table X is the symbol ‘**’ in either the Sig+ or Sig− row. A 5% significance level is used for alltests.

6.2. E6aluation of the Gen. simulations

The PROBct1-8parameters used in the Gen. simulations are calculated from the observations so the

simulated circulation-type frequencies (Mg (CT)) are expected to agree well with the observations (Obs(CT)).

Table IX confirms that agreement is good. The greatest differences occur in winter for the C-type: 25simulations have means significantly below observed. This discrepancy may be related to the lowprobability of the C-type in winter (on average only 2 days per winter (Table III)). The most infrequentcirculation type is the S/SE-type in summer (1.2 days) and this type is also underestimated by the CWG(23 simulations are significantly lower than observed). However, in all seasons and for every circulationtype, the simulated maximum and minimum values (not shown) fall either side of the observed mean, i.e.the observed means are within the simulated range.

Circulation-type S.D.s (sg(CT)) are generally underestimated by the CWG (Table IX). The greatestnumber of significant differences (54 underestimates) occurs in summer for the UA-type. For mostcirculation types the simulated maximum and minimum values (not shown) fall either side of the observedS.D., but in summer the simulated maximum is smaller than observed for five circulation types. Thus theCWG has a tendency to underestimate the observed variance but, except in summer, the effect is notgreat.

The mean NRD (Mg(NRD)) is very well simulated, with a maximum of five out of 100 significantunderestimates in summer and no significant differences in spring and autumn (Table X). The effect of theCWG on rain-day variance is much greater. With the exception of summer the simulated maximum S.D.(sg(NRD)) is always smaller than observed (sobs(NRD)). The CWG also underestimates the LW and LDparameters. The CWG simulates a wet or dry day only on the basis of the circulation type for that day,with no memory of previous rainfall occurrence. Thus the random number generator (i.e. chance) willhave a major control on the simulated LW and LD values. A possible way to improve performance wouldbe to include a persistence parameter (Hay et al., 1991; Wilby et al., 1994). A number of sensitivityexperiments have been performed using a range of arbitrary persistence parameters. Although it ispossible to increase the persistence of wet and dry spells, because the NRD is well simulated to start with,new errors are introduced. This approach was not, therefore, pursued further.

6.3. E6aluation of the Cont. simulations

The PROBct1–8parameters used in the Cont. simulations are calculated from UKTR control-run

output. Hence good agreement is expected, and found, between the Cont. (Mc(CT)) and UKTR



Table XI. Mean change (Pert.−Cont.) from the 100 simulations, of the mean (mean ofMp(NRD)−mean of Mc(NRD)) and S.D. (mean of sp(NRD)−mean of sc(NRD)) of thenumber of rain days simulated by the CWG for six stations in the Guadalentin Basin

Winter Spring Summer Autumn

MeanAguilas +0.3 −0.7 +0.3 0.0Alcantarilla +0.3 −0.7 +0.7 +0.3Alhama de Murcia +0.3 −1.0 +0.6 +0.1Fuente Alamo +0.2 −0.5 +0.6 +0.1Lorca +0.3 −0.8 +0.6 +0.1Totana +0.3 −0.7 +0.4 0.0

S.D.Aguilas +0.1 −0.2 0.0 0.0Alcantarilla +0.1 −0.2 +0.1 +0.2Alhama de Murcia 0.0 −0.2 +0.2 0.0Fuente Alamo 0.0 −0.1 +0.2 0.0Lorca 0.0 −0.1 +0.2 +0.1Totana +0.2 −0.2 +0.1 0.0

Significant changes are shown in bold.

(UKTR(CT)) circulation-type frequencies (Table IX). For the same reason, the Cont./Obs. differences inTable IX (Mc(CT)/Obs(CT)) closely follow those identified in Section 4.2. All circulation types with morethan 90 out of 100 significant differences in Table IX, with the exception of the C-type in spring, are alsoshown as significantly different in Table VI. The CWG performance is worst in winter, when none of theeight circulation-type mean frequencies fall within the simulated range (not shown), and best in autumnwhen five are within range.

The Cont. circulation-type S.D.s (sc(CT)) tend to be lower than the UKTR S.D.s (suktr(CT)), except insummer. They also tend to be somewhat lower than the Gen. S.D.s (sg(CT)). In summer the observed S.D.s(sobs(CT)) of only three circulation types are within the simulated range while seven circulation types arewithin range in winter.

In spring, almost all of the 100 simulated means (Mc(NRD)) are significantly lower than observed and theoverall-mean NRD is 24% lower than observed (Table X). About a quarter of the simulated means aresignificantly lower than observed in winter with a percentage error in the overall-mean NRD of 13%.

The mean NRD is reproduced well in the Gen. simulations (Mg(NRD)). Thus the discrepancies betweenthe Cont. and Obs. means (Mc(NRD) and Obs(NRD)) must be due to errors in the GCM simulation ofcirculation-types. The largest errors in the Cont. time-series occur in winter and spring. The frequency ofthe UC and E/NE-types (both high rainfall types) is underestimated in these seasons (Tables VI and IX).The low-rainfall A/HYA and W/NW/SW/N-types are overestimated.

The smallest circulation-type errors occur in autumn and tend to balance each other out. For example,underestimation of the dry W/NW/SW/N-type is offset by overestimation of the dry A/HYA-type. Hence,the smallest NRD errors also occur in autumn.

For NRD, the Cont. S.D.s, and the LW and LD parameters, are underestimated in all seasons and aresimilar to the Gen. values. This indicates that loss of variance and persistence is an inherent feature of theCWG and that the relatively small underestimation of circulation-type variance in the GCM has littleeffect on the rain-day series. It is concluded that, for NRD, while problems with the simulated varianceare related primarily to the effects of the CWG, problems with the simulated means are related primarilyto errors in the underlying GCM output.

6.4. The climate-change scenarios

The change in the mean NRD (calculated as the mean of vector Mp(NRD) minus the mean of vectorMc(NRD)) and in the NRD S.D. (calculated as the mean of vector sp(NRD) minus the mean of vector



sc(NRD)) for each season and each station is shown in Table XI. The pattern of change is consistent forall stations. These scenarios indicate that the average NRD in the Guadalentin Basin could increase by10–18% in summer in a future warmer world, and decrease by 5–9% in spring. A very small increase(2–4%) is indicated in winter, and little change in autumn (0–2%).

The scenarios in Table XI are calculated as the difference between the means of 100 simulated series.Alternatively, differences between pairs of simulations could be used to produce a range of scenarios. Forthe NRD, the maximum-possible range of these scenarios is provided by the maximum Mp(NRD) minusminimum Mc(NRD) value given in Table X (upper limit) and by the minimum Mp(NRD) minus maximumMc(NRD) value (lower limit). For the example of summer rain-day changes at Alcantarilla, this range is+2.9 to −1.4 days. The upper limit of 2.9 days is greater than the observed summer S.D. (sobs(NRD)). Thelower limit of −1.4 days is slightly smaller than the observed S.D. This pattern occurs at all stations insummer.

The rain-day scenarios presented here are intended as illustrative results rather than as reliablepredictions. Nonetheless, they confirm that downscaling methods can produce changes of the oppositesign to the ‘raw’ GCM changes (von Storch et al., 1993). Grid-point rain-day data from the UKTR modelfor the square closest to the Guadalentin Basin indicate little change or an insignificant decrease in thenumber of rain days in winter and autumn, a greater decrease in spring (in agreement with ourdownscaled scenarios), and the largest percentage decrease (about 60%) in summer, whereas thesescenarios indicate an increase in summer.

7. CONCLUSIONS

It has been demonstrated that the automated LWT classification scheme (Jenkinson and Collinson, 1977;Jones et al., 1993) can be successfully transferred to another region, southeast Spain. The 14 basiccirculation types were combined into eight groups to facilitate the analysis. These groups provide alegitimate basis for downscaling because each is shown to have a characteristic pressure pattern whichproduces the expected type and direction of flow over the study region (Section 2). Furthermore, a set ofconsistent and distinct relationships has been identified between these circulation types and daily rainfallin the Guadalentin Basin (Section 3).

Most of these relationships are physically realistic in terms of the underlying synoptic situation. Theinfluence of large-scale (North Atlantic–European) circulation is very evident. More regional (Mediter-ranean) influences can also be detected, particularly in the case of the E/NE and S/SE circulation types.The circulation type/rainfall relationships do, however, vary between season and between station. Part ofthis variability may be due to differences in topography, and part due to the effects of convective activity.The extent to which the typing scheme is capturing convective events is not clear and requires furtherinvestigation. The fact that summer rainfall is well simulated in the Gen. runs of the CWG (Section 6),and is better simulated than winter rainfall in the Cont. runs, suggests that some account is taken ofconvective activity. In part, however, the latter finding reflects the more reliable GCM simulation of theobserved circulation types in summer compared with winter.

Thus, it is essential to assess the ability of the GCM to reproduce the observed circulation types beforethe observed circulation type/rainfall relationships can be applied to GCM output. In the case of theUKTR model used here, the major problems are associated with the frequency of the circulation typesrather than with the underlying synoptic situation (Section 4). The effect of these problems on thesimulated means is clearly traceable in the CWG results, while the underestimation of variance andpersistence is a direct effect of the CWG itself (Section 6).

The simulated rainfall series based on UKTR control and perturbed-run output (Cont. and Pert.)contain errors because of the GCM’s failure to reproduce the observed circulation types. The illustrativeclimate-change scenarios presented here (Table IX) are, therefore, calculated as the difference between theCont. and Pert. means to ensure that the errors are consistent. The GCM changes in SLP andcirculation-type frequency between the control and perturbed-runs are generally small and, except in



summer, not significant (Section 5). Nonetheless the CWG results indicate changes in the number of raindays which are significant (in terms of model variance but not observed variance) in spring and summer.The pattern of change is unlikely to be beneficial for the Guadalentin Basin. Fewer rain days are indicatedduring spring which is an important season for agriculture and groundwater recharge. An increase in thenumber of rain days is indicated during the non-growing season and the period of highest evaporation(summer) when the soil surface is most vulnerable to erosion.

There are a number of ways in which the downscaling methodology described here might be adaptedor improved. In particular, a more sophisticated weather generator could be used to describe thecirculation type/rainfall relationships. It might be appropriate, for example, to incorporate some index ofupper-air trough and cut-off low activity in order to improve the prediction of storm events (Section 3.3).Techniques for increasing variance and persistence in the simulated series also require investigation.

Any modification to the methodology described here must, however, strike a careful balance betweenthe criteria outlined in Section 1. The incorporation of additional physical processes may improve thesimulation of present-day rainfall, but only at the expense of computational simplicity and ease oftransferability between regions, which are seen as major advantages of the proposed methodology. It mustalso be recognised that downscaled scenarios can never be more reliable than the simulation by the GCMof the underlying variable (here SLP).

The assumption that the observed relationships between the circulation types and the daily rainfallregime (or any other aspect of the climate) will be unchanged in a future warmer world is fundamentalto this approach to downscaling. Clearly this assumption cannot be fully tested. However, a similarassumption must also be made when using any of the empirical approaches to downscaling (Section 1),and in many other areas of climate research (Heyen et al., 1996).

Despite the issues identified above, it is considered that the circulation-type based approach todownscaling presented here offers great potential. It provides information about regional climate changein a form which is readily understood and can be related to underlying physical mechanisms. It does notrequire large computing or data resources. The method was developed and tested in a particularlychallenging area. Daily rainfall in the Guadalentin Basin is subject to large-scale, regional and localinfluences. The highly seasonal nature of this Mediterranean climate regime creates additional difficulties.Nonetheless, the authors are encouraged by the performance of the method and, as a further test oftransferability, it is being applied to study regions in the central and eastern Mediterranean using GCMoutput from the HadCM2 set of simulations (Johns et al., 1997).

ACKNOWLEDGEMENTS

This work was funded by the Commission of the European Union under the MEDALUS II (ContractEV5V-CT92-0164) and MEDALUS III (Contract ENV4-CT95-C121) projects. The UKTR SLP data hasbeen supplied by the Climate Impacts LINK Project (Department of the Environment Contract EPG1/1/16) on behalf of the Hadley Centre and UK Meteorological Office. The authors would like to thankthe two anonymous reviewers for their very helpful and useful comments and suggestions.

REFERENCES

Alpert, P., Neeman, B.U. and Shay-El, Y. 1990. ‘Climatological analysis of Mediterranean cyclones using ECMWF data’, Tellus,42(A), 65–77.

Bardossy, A. 1994. ‘Downscaling from GCMs to local climate through stochastic linkages’, in Paoli, G. (ed.), Climate Change,Uncertainty and Decision Making, Institute for Risk Research, Waterloo, pp. 33–46.

Bardossy, A. and Caspary, H.J. 1990. ‘Detection of climate change in Europe by analyzing European atmospheric circulationpatterns from 1881 to 1989’, Theor. Appl. Climatol., 42, 155–167.

Bardossy, A. and Plate, E.J. 1992. ‘Space-time model for daily rainfall using atmospheric circulation patterns’, Wat. Resour. Res.,28, 1247–1259.

Bardossy, A., Muster, H., Duckstein, L. and Bogardi, I. 1994. ‘Knowledge based classification of circulation patterns for stochasticprecipitation modelling’, in Hipel et al., (eds), Time Series Analysis in Hydrology and En6ironmental Engineering, 3, KluwerAcademic Publishers, pp. 19–32.



Bogardi, I., Matyasovszky, I., Bardossy, A. and Duckstein, L. 1993. ‘Application of a space-time stochastic model for dailyprecipitation using atmospheric circulation patterns’, J. Geophys. Res., 98, 16653–16667.

Carnell, R.E., Senior, C.A. and Mitchell, J.F.B. 1996. ‘An assessment of measures of storminess: simulated changes in northernhemisphere winter due to increasing CO2’, Clim. Dyn., 12, 467–476.

Conway, D., Wilby, R.L. and Jones, P.D. 1996. ‘Precipitation and air flow indices over the British Isles’, Clim. Res., 7, 169–183.Crane, R.G. and Barry, R.G. 1988. ‘Comparison of the MSL synoptic pressure patterns of the Arctic as observed and simulated by

the GISS general circulation model’, Meteorol. Atmos. Phys., 39, 169–183.Cubasch, U., Waszkewitz, J., Hegerl, G. and Perlwitz, J. 1995. ‘Regional climate changes as simulated in time-slice experiments’,

Clim. Change, 31, 273–304.Cubasch, U., von Storch, H., Waszkewitz, J. and Zorita, E. 1996. ‘Estimates of climate change in southern Europe derived from

dynamical climate model output’, Clim. Res., 7, 129–149.Dalu, G.A. and de Gregorio, S. 1987. ‘Some simple solutions for the low-pressure anomaly observed over the Mediterranean Sea’,

Ann. Geophys., 5B, 461–467.Deque, M. and Piedelievre, J.P. 1995. ‘High resolution climate simulation over Europe’, Clim. Dyn., 11, 321–339.Fernandez Mills, G. 1995. ‘Principal component analysis of precipitation and rainfall regionalization in Spain’, Theor. Appl.

Climatol., 50, 169–183.Frey-Buness, F., Heimann, D. and Sausen, R. 1995. ‘A statistical-dynamical downscaling procedure for global climate simulations’,

Theor. Appl. Climatol., 50, 117–131.Giorgi, F., Marinucci, M.R. and Visconti, G. 1990. ‘Use of a limited-area model nested in a general circulation model for regional

climate simulation over Europe’, J. Geophys. Res., 95, 18413–18431.Giorgi, F., Marinucci, M.R. and Visconti, G. 1992. ‘A 2×CO2 climate change scenario over Europe generated using a limited area

model nested in a general circulation model. 2. Climate change scenario’, J. Geophys. Res., 97, 10011–10028.Grotch, S.L. and MacCracken, M.C. 1991. ‘The use of general circulation models to predict regional climatic change’, J. Climate,

4, 286–303.Gyalistras, D., von Storch, H., Fischlin, A. and Beniston, M. 1994. ‘Linking GCM-simulated climatic changes to ecosystem models:

case studies of statistical downscaling in the Alps’, Clim. Res., 4, 167–189.Hansen, A.R. and Sutera, A. 1990. ‘Weather regimes in a general circulation model’, J. Atmos. Sci., 47, 380–391.Hay, L.E., McCabe, G.J., Wolock, D.M. and Ayers, M.A. 1991. ‘Simulation of precipitation by weather type analysis’, Wat. Resour.

Res., 27, 493–501.Hay, L.E., McCabe, G.J., Wolock, D.M. and Ayers, M.A. 1992. ‘Use of weather types to disaggregate general circulation model

predictions’, J. Geophys. Res., 97, 2781–2790.Hewitson, B. and Crane, R.G. 1992. ‘Regional climates in the GISS global circulation model: synoptic-scale circulation’, J. Climate,

5, 1002–1011.Hewitson, B.C. and Crane, R.G. 1994. ‘Precipitation controls in southern Mexico’, in Hewitson, B.C. and Crane, R.G. (eds). Neural

Nets: Applications in Geography, Kluwer Academic Publishers, p. 121–143.Hewitson, B.C. and Crane, R.G. 1996. ‘Climate downscaling: techniques and application’, Clim. Res., 7, 85–95.Heyen, H., Zorita, E. and von Storch, H. 1996. ‘Statistical downscaling of monthly mean North Atlantic air-pressure to sea level

anomalies in the Baltic Sea’, Tellus, 48(A), 312–323.Hughes, J.P. and Guttorp, P. 1994. ‘A class of stochastic models for relating synoptic atmospheric patterns to regional hydrologic

phenomena’, Wat. Resour. Res., 30, 1535–1546.Hughes, J.P., Lettenmaier, D.P. and Guttorp, P. 1993. ‘A stochastic approach for assessing the effect of changes in synoptic

circulation patterns on gauge precipitation’, Wat. Resour. Res., 29, 3303–3315.Hulme, M., Briffa, K.R., Jones, P.D. and Senior, C.A. 1993. ‘Validation of GCM control simulations using indices of daily airflow

types over the British Isles’, Clim. Dyn., 9, 95–105.Hurrell, J.W. 1995. ‘Decadal trends in the North Atlantic Oscillation: regional temperatures and precipitation’, Science, 269,

676–679.Huth, R. 1996. ‘An intercomparison of computer-assisted circulation classification methods’, Int. J. Climatol., 16, 893–922.Jacobeit, J. 1987. ‘Variations of trough positions and precipitation patterns in the Mediterranean area’, J. Climatol., 7, 453–476.Jenkinson, A.F. and Collinson, P. 1977. An Initial Climatology of Gales o6er the North Sea, Synoptic Climatology Branch

Memorandum No. 62, Meteorological Office, London, 18 pp.Johns, T.C., Carnell, R.E., Crossley, J.F., Gregory, J.M., Mitchell, J.F.B., Senior, C.A., Tett, S.F.B. and Wood, R.A. 1997. ‘The

second Hadley Centre coupled ocean-atmosphere GCM: model description, spinup and validation’, Clim. Dyn., 13, 103–134.Jones, P.D., Hulme, M. and Briffa, K.R. 1993. ‘A comparison of Lamb circulation types with an objective classification scheme’,

Int. J. Climatol., 13, 655–663.Jones, R.G., Murphy, J.M. and Noguer, M. 1995. ‘Simulation of climate change over Europe using a nested regional-climate model.

Part 1: Assessment of control climate, including sensitivity to location of lateral boundaries’, Q. J. R. Meteorol. Soc., 121,1413–1449.

Kanji, G.K. 1993. 100 Statistical Tests, SAGE Publications, London, 216 pp.Karl, T.R., Wang, W.-C., Schlesinger, M.E., Knight, R.W. and Portman, D. 1990. ‘A method of relating general circulation model

simulated climate to the observed local climate. Part I: seasonal statistics’, J. Clim., 3, 1053–1079.Kim, J-W., Chang, J-T., Baker, N.L., Wilks, D.S. and Gates, W.L. 1984, ‘The statistical problem of climate inversion: determination

of the relationship between local and large-scale climate’, Mon. Wea. Re6., 112, 2069–2077.Koppen, W. 1936. ‘Das geographische system der klimate’, Handbuch der Klimatologie 3, in Koppen, W. and Geiger, R. (eds),

Gebruder Borntraeger, Berlin, 46 pp.Kutiel, H., Maheras, P. and Guika, S. 1996. ‘Circulation indices over the Mediterranean and Europe and their relationship with

rainfall conditions across the Mediterranean’, Theor. Appl. Climatol., 54, 125–138.Lamb, H.H. 1972. British Isles Weather Types and a Register of Daily Sequence of Circulation Patterns, 1861–1971, Geophysical

Memoir, 116, HMSO, London, 85 pp.



Lawson, J.S. 1989. ‘The wet spell in southern Spain, mid-October 1988’, Weather, 44, 475–478.Liang, X-Z., Wang, W-C. and Dudek, M.P. 1996. ‘Northern hemispheric interannual teleconnection patterns and their changes due

to the greenhouse effect’, J. Clim., 9, 465–479.Lines Escardo, A. 1970. ‘The climate of the Iberian Peninsula’, in Wallen, C.C. (ed.), Climates of Northern and Western Europe,

World Survey of Climatology, Volume 5, Elsevier Publishing, Amsterdam-London-New York, pp. 195–239.Maheras, P. 1988. ‘Changes in precipitation conditions in the western Mediterranean over the last century’, J. Climatol., 8, 179–189.Moses, T., Kiladis, G.N., Diaz, H.F. and Barry, R.G. 1987, ‘Characteristics and frequency of reversals in mean sea level pressure

in the North Atlantic sector and their relationship to long-term temperature trends’, J. Climatol., 7, 13–30.Moulin, C., Lambert, C.E., Dulac, F. and Dayan, U. 1997. ‘Control of atmospheric export of dust from North Africa by the North

Atlantic Oscillation’, Nature, 387, 691–694.Murphy, J.M. 1995. ‘Transient response of the Hadley Centre coupled ocean-atmosphere model to increasing carbon dioxide. Part

I: control climate and flux adjustment’, J. Clim., 8, 36–56.Murphy, J.M. and Mitchell, J.F.B. 1995. ‘Transient response of the Hadley Centre coupled ocean-atmosphere model to increasing

carbon dioxide. Part II: spatial and temporal structure of response’, J. Clim., 8, 57–80.Noguer, M. 1994. ‘Using statistical techniques to deduce local climate distributions. An application for model validation’, Met.

Apps., 1, 277–287.O8 zelkan, E.C., Ni, F. and Duckstein, L. 1996. ‘Relationship between monthly atmospheric circulation patterns and precipitation:

fuzzy logic and regression approaches’, Water Resour. Res., 32, 2097–2103.Palutikof, J.P., Winkler, J.A., Goodess, C.M. and Andresen, J.A. 1997. ‘The simulation of daily temperature time series from GCM

output. Part I: comparison of model data with observations’, J. Climate, 10, 2497–2513.Prezerakos, N.G., Michaelides, S.C. and Vlassi, A.S. 1990. ‘Atmospheric synoptic conditions associated with the initiation of

north–west African depressions’, Int. J. Climatol., 10, 711–729.Risbey, J.S. and Stone, P.H. 1996. ‘A case study of the adequacy of GCM simulations for input to regional climate change

assessments’, J. Climate, 9, 1441–1467.Schubert, S. 1994. ‘A weather generator based on the European Grosswetterlagen’, Clim. Res., 4, 191–202.Schubert, S. and Henderson-Sellers, A. 1997. ‘A statistical model to downscale local daily temperature extremes from synoptic-scale

atmospheric circulation patterns in the Australian region’, Clim. Dyn., 13, 223–234.Tout, D.G. 1991. ‘A very wet autumn-early winter in southern Spain in 1989’, Weather, 46, 11–15.Tout, D.G. and Wheeler, D.A. 1990. ‘The early autumn storms of 1989 in eastern Spain’, J. Meteorol., 15, 238–248.Trenberth, K.E. 1995. ‘Atmospheric circulation climate changes’, Clim. Change, 31, 427–453.Trenberth, K.E. and Paolino, D.A. 1980. ‘The Northern Hemisphere sea-level pressure data set: trends, errors and discontinuities’,

Mon. Wea. Re6., 108, 855–872.Trenberth, K.E. and Olson, J.G. 1988. ‘An evaluation and intercomparison of global analysis from NMC and ECMWF’, Bull. Am.

Meteorol. Soc., 69, 1047–1057.van Loon, H. and Rogers, J.C. 1978. ‘The seesaw in winter temperatures between Greenland and Northern Europe. Part 1: general

description’, Mon. Wea. Re6., 106, 296–310.Verdecchia, M., Visconti, G., D’Andrea, F. and Tibaldi, S. 1996. ‘A neural network approach for blocking recognition’, Geophys.

Res. Lett., 23, 2081–2084.von Storch, H., Zorita, E. and Cubasch, U. 1993. ‘Downscaling of global climate change estimates to regional scales: an application

to Iberian rainfall in wintertime’, J. Climate, 6, 1161–1171.Wheeler, D. 1990. ‘Majorca’s severe storms of September 1989: a reminder of Mediterranean uncertainty’, Weather, 45, 21–26.Wigley, T.M.L., Jones, P.D., Briffa, K.R. and Smith, G. 1990. ‘Obtaining sub-grid-scale information from coarse-resolution general

circulation model output’, J. Geophys. Res., 95, 1943–1953.Wilby, R.L., Greenfield, B. and Glenny, C. 1994. ‘A coupled synoptic-hydrological model for climate change impact assessment’,

J. Hydrol., 153, 265–290.Wilby, R.L., Barnsley, N. and O’Hare, G. 1995. ‘Rainfall variability associated with Lamb Weather Types: the case for

incorporating weather fronts’, Int. J. Climatol., 15, 1241–1252.Wilson, L.L., Lettenmaier, D.P. and Skyllingstad, E. 1992. ‘A hierarchical stochastic model of large-scale atmospheric circulation

patterns and multiple station daily precipitation’, J. Geophys. Res., 97, 2791–2809.Winkler, J.A., Palutikof, J.P., Andresen, J.A. and Goodess, C.M. 1997. ‘The simulation of daily temperature time series from GCM

output. Part II: Sensitivity analysis of an empirical transfer function methodology’, J. Climate, 10, 2514–2532.Yarnal, B. 1993. Synoptic Climatology in En6ironmental Analysis. A Primer, Studies in Climatology Series, Belhaven Press, London,

195 pp.Zhang, X., Wang, X.L. and Corte-Real, J. 1997. ‘On the relationships between daily circulation patterns and precipitation in

Portugal’, J. Geophys. Res., 102, 13495–13507.Zorita, E., Kharin, V. and 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’, J. Climate, 5, 1097–1108.Zorita, E., Hughes, J.P., Lettenmaier, D.P. and von Storch, H. 1995. ‘Stochastic characterization of regional circulation patterns for

climate model diagnosis and estimation of local precipitation’, J. Climate, 8, 1023–1042.


development of daily rainfall scenarios for southeast spain using a circulation-type approach to...

Documents