INTRODUCTION
A
dense daily precipitation data-base, extending from 1964 to 1993,
was created for the Mediterranean regions of Spain. It is composed
of complete and homogeneous series at 410 rain gauge stations
(347 in the coastal fringe of peninsular Spain, and 63 in the
Balearic Islands, Fig. 1). The raw data consisted of the 3366
available daily precipitation records during some period between
1951 and 1995, provided by the Instituto Nacional de Meteorología
of Spain (INM). A first selection was made to consider only stations
with a minimum of 1000 data values (almost three years), which
yielded a set of 2842 rain gauges. The inventory of these stations
showed a great variety of record lengths, with only 5 stations
with no missing data. Then, a search was performed to choose the
longest sub-period of 1951-95 having the highest number of stations
with tolerable completion. The final decision was to keep the
410 stations with 90% of data available during 1964-93 (30 years).
An iterative method was followed to check the quality of the data,
as well as to fill the missing data by interpolation from the
surrounding stations. A detailed description of the method can
be found in Romero et al. (1998a).
The
region offers an interesting scenario for meso-climatological
studies on time and spatial rainfall variability. Geomorphologically,
it is characterized by important coastal relief units and complex
distribution of sea and land masses (Fig. 1), leading to different
exposures to the rain bearing maritime winds. Climatically, the
western Mediterranean is subject to strong seasonal variability,
since it is a transition zone between the mid-latitude low-pressure
belt and the subtropical highs as a result of its latitude (between
360 and 440 N). Another important climatic characteristic of the
region is the torrential aspect of rainfalls, mainly during the
autumn. Most of the coastal stations have registered daily rainfalls
greater than 200 mm (Font 1983).
The yearly mean precipitation (Fig. 2) shows high amounts about
Sierra de Ronda (in the form of an almost circular area with more
than 700 mm and peak values in its centre exceeding 1500 mm),
appreciable amounts also along Pyrenees, about Sierra de Aracena,
Sierra de Cazorla, Sierra de Aitana and the north of Mallorca,
contrasting with a general decrease toward inland areas, and very
low values in a vast area of the Southeast (south of Valencia,
Murcia and eastern Andalucía). In a general sense, the
spatial pattern of Fig. 2 reflects a general precipitation gradient
along the SE-NW direction. This feature is connected, apart from
the orography, to the degree of proximity to the Atlantic waters
from where the majority of storms arrive.
Fig.
1. The Spanish Mediterranean area. It includes a smoothed version
of its orography, the position of the stations of the daily rainfall
data base (410 in total), and the location of places mentioned
in the text.
Fig
2. Yearly mean precipitation for 1984-93
As
noted previously, Mediterranean Spain is frequently affected by
extreme daily rainfalls. Figure 3a shows that extreme rainfalls
tend to concentrate in coastal areas, and secondarily, in mountainous
interior lands. The maximum values are found in Sierra de Ronda,
where a station attains the 21-25 days category. Valencia and
Murcia define an extensive area where 100 mm occurrences (in general
in the 1-5 or 6-10 ranges) are given by most of the stations.
It is notable the high amount of 100 mm days given by the stations
located close to Sierra de Aitana. This zone has been classically
considered as the most torrential in Spain. Observe how the peninsular
arid Southeast participates actively in this case. Extreme rainfalls
are also frequent in the island of Mallorca along Sierra de Tramuntana,
and in its eastern part where a minor topographic ridge exists.
Ibiza also registered 100 mm rainfalls. The seasonal distribution
(not shown) demonstrates that most of the 100 mm rainfalls concentrate
in the autumn season. Winters occupy the second place, followed
by springs. Extreme rainfalls are rare during summer. Events of
200 mm are almost exclusive of autumn, although also tend to occur
during winter in Sierra de Ronda.
Fig
3a. Days with extreme rainfalls (> 100 mm) during 1984-93.
Fig
3b. Recurrence intervals of 100 mm events, for 1984-93.
A
direct demarcation of the torrential zones of Mediterranean Spain
can be done by visualizing the recurrence intervals. For 100 mm
daily rainfalls (Fig. 3b), recurrence intervals lesser than 5
years are obtained about Sierra de Aracena, in a wide zone around
Gibraltar Strait, in very localized zones of eastern Andalucía
close to the coast or at high altitudes, in Murcia and Valencia,
in the littoral of northern and southern Catalonia, zones of Pyrenees,
and north and east of Mallorca. A deeper analysis of the torrential
character of rainfalls is included in Romero et al. (1998a).
In
this study the main spatial patterns controlling torrential daily
rainfalls in Mediterranean Spain are derived. Furthermore, we
investigate the synoptic atmospheric circulations associated with
those patterns.
TORRENTIAL DAILY RAINFALL PATTERNS
From
the 30-years precipitation data-base, we define as torrential
days those days in which at least 2% of the stations registered
more than 50 mm, yielding a total of 449 events. Of the 449 selected
days, 35.2 % occurred in winter, 14.9 % in spring, 5.1 % in summer
and 44.8 % in autumn.
The
approach followed to derive the typical precipitation patterns
consists in subjecting the T-mode (day-by-day) correlation matrix
to principal components analysis (PCA), and carrying out cluster
analysis (CA) on the most important extracted components. That
is, days participating with similar loadings on the extracted
components are clustered together. This approach is aimed to join
days with similar precipitation distributions, irrespectively
of the precipitation amounts (Sumner et al. 1995).
For
deciding the number of PCs to retain, the simple scree test of
Cattell (1966) was adopted. We retained 15 PCs, which account
for 68.5 % of the total variance. For the cluster analysis, the
non-hierarchical k-means method (Anderberg 1973), as implemented
in the STATISTICA utility (1994), was used. The Euclidean distance
was taken as the similarity index. Hierarchical tree plots generated
by Ward's method (Ward 1963) were also considered as reference
for deciding how many clusters to create. A solution comprising
8 typical patterns was chosen. These pattern groups are presented
in Fig. 4.
The
obtained pattern groups are quite definite and clearly display
the dominant role exerted by the complex topography and its connection
with the main rain bearing flows. Plot of inter-seasonal variability
(Fig. 5) reveals a different incidence of torrential rainfall
patterns through the year. The western patterns, largely stimulated
by Atlantic flows, are similarly important in winter and autumn,
whereas the eastern patterns, which are strongly influenced by
the Mediterranean dynamics, dominate in autumn. A more detailed
analysis can be found in Romero et al. (1998b).
Fig
4. Daily rainfall composites for the 8 pattern groups of torrential
rainfall in Mediterranean Spain.
The number of days for 1964-93 included in each pattern group
is indicated in parentheses (total, 449)
Fig
5. Seasonal distribution for the 8 pattern groups of torrential
daily rainfall
ATMOSPHERIC CIRCULATION PATTERNS
This
section attempts to identify the main atmospheric circulation
settings associated with the torrential patterns presented in
the previous section. There are several studies in the literature
which have illustrated and interpreted statistical connections
between synoptic atmospheric patterns and certain surface weather
and/or environmental parameters. In his book, Yarnal (1993) offers
multiple examples of synoptic classifications, ranging from subjective
manual classifications to outputs from eigenvector-based techniques.
Bonell and Sumner (1992) establish, using S-mode PCA and CA, the
main daily precipitation affinity areas for Wales according to
surface wind direction. Sumner et al. (1995) associate the distribution
of significant rainfalls over the island of Mallorca with recognized
dominant surface circulation types.
Climatology-based
scenarios of rainfall distribution for defined synoptic types
have been conceptually recognized for a long time, and have become
apparent through numerous case studies of heavy precipitations
in eastern Spain (see Doswell et al. 1998 and references mentioned
therein).
The
meteorological data used to carry out the synoptic classification
are the European Centre for Medium Range Weather Forecast (ECMWF)
grid analyses of geopotential height at 925 and 500 hPa. The spatial
resolution is 0.750 in both latitude and longitude. Since the
ECMWF has only been operative since 1979, the meteorological data-base
has been restricted to the last decade of the 30 years of the
original rainfall data base (1984-93). During that decade, 1275
days were identified as significant rainfall days (5% of stations
registered more than 5 mm) in Mediterranean Spain. Of these, 165
were also classified as torrential (previously defined). The classification
of atmospheric circulation thus utilizes 1275 unique circulation
patterns, one per significant rainfall day.
Circulations
were classified using T-mode PCA and CA, as for the derivation
of the torrential patterns in last section. We have considered
data only within the rectangular geographical window between 33.750
and 45.750 N, and between 11.250 W and 6.000 E. It thus comprises
408 grid points. Classifications based on larger windows encompassing
much larger geographical areas were also tested, but associations
with rainfall patterns were very poor, because the classifications
were strongly influenced by circulation features of regions remote
from Mediterranean Spain.
Application
of the scree test suggests the retention of 6 PCs for the geopotential
field at 500 hPa (accounting for 96.7% of the total variance),
and 8 PCs for the geopotential field at 925 hPa (95.7% of the
total variance). Thus, 14 variables were used in the CA, performed
again with the k-means method. Solutions involving 4, 7, 11 and
19 clusters were clearly indicated. The former three solutions
produced too dilute an association with rainfall distributions,
and an appropriate compromise for a still relatively simple collection
of patterns, but with enough meaningful associations was found
in the 19 cluster solution. These clusters (referred to as atmospheric
patterns, APs) are presented in Fig. 6. Table 1 shows how the
1275 significant and the subset of 165 torrential rain days are
distributed amongst the derived 19 atmospheric patterns.
Table
1. Distribution of significant days and torrential days for
the 19 atmospheric circulation patterns.
Torrential days are also shown as a percentage of the total
number of significant days.
In
a general context, it is possible to argue that the synoptic
types obtained summarize four main more general scenarios. First,
a large scale disturbance lies to the west or north-west of
the Iberian Peninsula producing humid Atlantic flows which encourage
rainfall development in the bulk of western Andalucía
(APs 1, 2, 3, 4 and 5). The second occurs with the passage of
cold fronts over the Iberian Peninsula linked to higher latitude
low pressure systems (APs 7, 8 and 9). This scenario favors
the development of rainfalls in north-eastern Spain and in the
inland mountainous areas of Andalucía. In the third case,
relatively small lows at 500 hPa are found about the southern
part of Spain, and the associated low-level flux over the Mediterranean
is warm and humid from the east-southeast (APs 6, 10, 11, 12,
13, 14 and 15). This configuration leads to rainfalls over the
eastern flank of Spain, including the Balearic Islands. Finally,
for the fourth type, upper and low level disturbances are located
to the east of the Iberian Peninsula (APs 16, 17, 18 and 19),
inducing strong flows with a pronounced northerly component.
As
a further refinement of the analysis carried out, the days producing
torrential rainfall were extracted, and their association with
atmospheric circulation has been investigated. Figure 7 shows,
for each AP, how the torrential events distribute among the
8 torrential patterns (Fig. 4). Circulation patterns AP11 and
AP16 were not associated with any torrential event. By contrast,
AP13 is associated with the development of torrential rains
on almost 40 % of occasions (see Table 1). Also noteworthy are
AP3, AP6, AP12, AP14 and AP15, with the incidence of torrential
rain exceeding 20 %. As can be observed from Fig. 6, these more
torrential APs are characterized in the middle troposphere by
closed cyclonic circulations or very accentuated short-wave
troughs located in the south of the domain. At low levels, they
exhibit a significant level of warm advection towards some area
of Mediterranean Spain. These conditions have been observed
in most studies of severe rainfall events in this area (for
example, Doswell et al., 1998; Ramis et al., 1998).
Figure
7, although obtained from only 165 torrential events, is quite
informative. The distributions shown are physically consistent
with the dynamic processes contained in the corresponding APs.
The combination of important topographic enhancement of rainfall
for moist Atlantic or Mediterranean airflows, or both, and dynamic
factors associated with the upper level disturbance, in fact
render the physical interpretation of the relationship between
atmospheric and rainfall patterns quite straightforward. Focusing
the attention on the most torrential APs, AP3 favours torrential
patterns T1 and T2 in western Andalucía, AP6 favours
T4 and T6 and in a lesser extent T2 and T5, AP12 projects almost
exclusively on T5, AP13 and AP14 favour significantly torrential
rainfalls in Murcia, Valencia and the Balearic Islands through
T4, T5, T6 and T8, and AP15 produces T4, T6 and T8. See Romero
et al. (1998c) for details.
Fig.
7. Percentage frequency of the 8 torrential daily rainfall patterns
within
the 19 atmospheric circulation patterns for Mediterranean Spain.
CONCLUSIONS
Application
of T-mode PCA and CA on a 30-years data-base of daily precipitation
at 410 sites, has permitted to derive a simplified collection
of 8 spatial patterns governing torrential daily rainfall events
in Mediterranean Spain. A visual inspection of those patterns
confirms that the extent of the region and the exposure-sheltering
systems induced by the complex topography are sufficiently important
as to produce a clear regionalization of torrential rainfalls.
Distinct seasonal distributions have been observed: the western
Andalucía patterns occur with similar incidence in winter
and autumn, whereas the eastern patterns have a clear predilection
for the autumn season.
Using
a similar methodology, a total of 19 fundamental synoptic patterns
has been derived that explain the development of significant
daily rainfalls in the Spanish Mediterranean area. A study of
the torrentiality of the derived atmospheric patterns reveals
that some of them, although important for explaining significant
rainfalls, are irrelevant for the occurrence of torrential rainfalls
in the region. The situations characterized by disturbances
located about the south of the Iberian Peninsula, although not
very frequent, exhibit a high propensity towards the development
of torrential rainfalls focused on the eastern flank of the
Iberian Peninsula and the Balearic Islands.
The
results obtained should be most valuable in providing a more
detailed forecast of rainfall activity in Mediterranean Spain
within the context of synoptic numerical weather prediction,
and as a complement for meso-scale resolution models. The assessment
of the long-term rainfall variability in the region may also
benefit from these results.
ACKNOWLEDGEMENTS
Raw
precipitation data and ECMWF meteorological fields were provided
by the Instituto Nacional de Meteorologia of Spain. This work
has been sponsored by CICYT grant CLI95-1846.
REFERENCES
Anderberg,
M. R. 1973: Cluster analysis for applications. Academic Press,
359 pp.
Bonell,
M. and Sumner, G. N. 1992: Autumn-winter daily precipitation
areas in Wales, 1982/83 to 1986/87. Int. J. Climatol., 12, 77-102.
Cattell,
R. B. 1966: The scree test for the number of PCs. Mult. Behav.
Res., 1, 245-276.
Doswell
III, C. A., Ramis, C., Romero, R. and Alonso, S. 1998: A diagnostic
study of three heavy precipitation episodes in the western Mediterranean
region. Wea. Forecasting, 13, 102-124.
Font,
I. 1983: Climatología de España y Portugal. Instituto
Nacional de Meteorología, Apartado 285, 28071, Madrid,
296 pp.
Ramis,
C., Romero, R., Homar, V., Alonso, S. and Alarcón, M.
1998: Diagnosis and numerical simulation of a torrential precipitation
event in Catalonia (Spain). Meteorol. Atmos. Phys. (in press).
Romero,
R., Guijarro, J. A., Ramis, C. and Alonso, S. 1998a: A 30 year
(1964-1993) daily rainfall data base for the spanish mediterranean
regions: First exploratory study. Int. J. Climatol., 18, 541-560.
Romero,
R., Ramis, C. and Guijarro, J. A. 1998b: Daily rainfall patterns
in the Spanish Mediterranean area: An objective classification.
Int. J. Climatol., (in press).
Romero,
R., Sumner, G., Ramis, C. and Genovés, A. 1998c: A classification
of the atmospheric circulation patterns producing significant
daily rainfall in the Spanish Mediterranean area. Int. J. Climatol.,
(submitted).
STATISTICA,
1994: STATISTICA for Windows (Volume III): Statistics II. StatSoft
Inc., Tulsa OK.
Sumner,
G. N., Guijarro, J. A. and Ramis, C. 1995: The impact of surface
circulations on the daily rainfall over Mallorca. Int. J. Climatol.,
15, 673-696.
Ward,
J. H. 1963: Hierarchical grouping to optimize an objective function.
J. Amer. Stat. Assoc., 58, 236-244.
Yarnal,
B. 1993: Synoptic climatology in environmental analysis, Belhaven
Press, London, 195 pp.
