You are on page 1of 14

Journal of Hydrology 529 (2015) 106–119

Contents lists available at ScienceDirect

Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol

Delimiting homogeneous regions using the multifractal properties


of validated rainfall data series
A.P. García-Marín ⇑, J. Estévez, M.T. Medina-Cobo, J.L. Ayuso-Muñoz
Dept. of Rural Engineering, University of Córdoba, P.O. Box 3048, 14080 Córdoba, Spain

a r t i c l e i n f o s u m m a r y

Article history: In this work, a regional frequency analysis of extreme annual rainfall data in Malaga (Southern Spain) has
Received 13 March 2013 been performed. Rainfall records have been validated, applying various quality control tests as a
Received in revised form 10 April 2015 pre-requisite before their use, ensuring their reliability and discarding anomalous data. For grouping
Accepted 16 July 2015
the stations into potential homogeneous regions, the multifractal properties of daily rainfall data series
Available online 22 July 2015
This manuscript was handled by Geoff
recorded at 72 locations have been studied. The scaling of the rainfall moments has been analyzed and
Syme, Editor-in-Chief, with the assistance of the empirical moments scaling exponent functions have been obtained. The corresponding multifractal
Alin Andrei Carsteanu, Associate Editor values have been used to group stations into regions, resulting some of them homogeneous.
Ó 2015 Elsevier B.V. All rights reserved.
Keywords:
Regional frequency analysis
Meteorological data validation
Homogeneous regions
Multifractal properties

1. Introduction Different works back up regionalization as being the technique


that improves the estimations of the quantiles when working with
One of the most common problems when considering a hydro- rain or flows (Greis and Wood, 1981; Hosking et al., 1985; Saenz de
logical study in a particular area is the availability of rainfall data Ormijana et al., 1991; Hosking and Wallis, 1997; Parida et al.,
and the resolution and the length of the available series. 1998; Yun and Chen, 1998; Ferrer and Mateos, 1999; Chiang
Obtaining an accurate estimate of the relationship between the et al., 2002a,b). The improvement of the RFA over the local one
extreme values of precipitation and their associated return periods depends on the regional homogeneity, always considering that a
becomes more complicated if the length of the records available at slight heterogeneity does not override the advantages (e.g.
the place of interest is shorter than the return period considered. Cunnane, 1988). In cases of extreme regional heterogeneity, local
Another difficulty arises when there are no records available at estimations could be better than those based on RFA
the site of interest. (Lettenmaier and Potter, 1985).
In order to compensate the insufficient length of the available The lineal moments were introduced by Hosking (1990, 1992)
data series a link can be made between the spatial and temporal and they are used in all the steps of RFA (Hosking and Wallis,
characterization of the extreme values of the precipitation by using 1993, 1995, 1997; Rao and Hamed, 2000). Since then, they have
regional frequency analysis (RFA). In addition, a quality control is been widely applied on regional analysis of extreme hydrologic
required as a major prerequisite for using rainfall data. High qual- events like floods (Burn, 1990; Farquharson et al., 1992; Zrinji
ity rainfall data sources are vital to ensure their reliability and for and Burn, 1994; Adamowski, 2000; Ouarda et al., 2001, 2008;
detecting erroneous data using validation procedures. Kjeldsen et al., 2002; Zhang and Hall, 2004; Kumar and
The RFA increases the data at the site of interest considering Chatterjee, 2005; Hussain and Pasha, 2009; Bhuyan et al., 2010),
data from other places that share the same probability distribution low flow rates (Chen et al., 2006; Arbeláez and Castro, 2007;
functions. Although some problems can arise, the RFA leads to Modarres, 2008; Tsakiris et al., 2011), rainfall (Schaefer, 1990;
more accurate quantile estimations than those from local fre- Adamowski and Alila, 1991; Lee and Maeng, 2003; Fowler and
quency analysis (Hosking et al., 1985; Lettenmaier and Potter, Kilsby, 2003; Di Baldassarre et al., 2006; Norbiato et al., 2007;
1985; Wallis and Wood, 1985; Hosking and Wallis, 1988, 1997). Wallis et al., 2007; Castellarin et al., 2009; Ngongondo et al.,
2011; García-Marín et al., 2011; Satyanarayana and Srinivas,
⇑ Corresponding author. Tel.: +34 957212655. 2011) and droughts (Clausen and Pearson, 1995; Lee and Maeng,
E-mail address: amanda.garcia@uco.es (A.P. García-Marín).

http://dx.doi.org/10.1016/j.jhydrol.2015.07.021
0022-1694/Ó 2015 Elsevier B.V. All rights reserved.
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 107

2005; Modarres, 2010; Núñez et al., 2011; Santos et al., 2011; Valid data are required to make climate assessments and to make
Eslamian et al., 2011). climate-related decisions. These procedures are part of the quality
Within regionalization, the determination of homogeneous control systems and consist of methods, tests or rules against
regions is the most complex step and the continuity of the analysis which data are tested. A quality control system is not only based
depends on its result. Authors such as Greis and Wood (1981), on the validation of meteorological records but on a set of activities
Hosking et al. (1985), and Lettenmaier et al. (1987), showed the and decisions that ensure the reliability of the records obtained
importance of regional homogeneity. Although is obvious that cli- from meteorological stations (WMO, 1993). The purpose of the val-
matologically, geological and geomorphological homogeneous idation process is to identify erroneous data from sensor measure-
region can let the transfer of information within it, a mathematical ments in order to make optimal use of them (Estévez et al., 2011a).
criteria is advisable. Different methodologies can be applied to In the validation process, data of a doubtful quality must be
delimit homogeneous regions: cluster analysis (e.g. Periago et al., detected and appropriately flagged. These quality control flags,
1991; Bonell and Sumner, 1992), principal component analysis which supplement but do not alter the data, are employed to
(García-Marín et al., 2011), Fuzzy C-means clustering combined describe which test the data failed. All archived meteorological
with artificial neuronal networks (Hall and Minns, 1999; Jingyi data must be coupled with flags (‘‘good’’, ‘‘suspect’’, ‘‘warning’’ or
and Hall, 2004; Srinivas et al., 2008; Satyanarayana and Srinivas, ‘‘failure’’) that indicate the level of confidence that network man-
2011), and visual inspection of L-Moments diagrams (Yürekli and agers place upon their observations (Fiebrich and Crawford, 2001).
Modarres, 2007), among others. The quality control methods applied in this work are: range/li-
The characteristics that are usually considered for regional fre- mit (fixed or dynamic) test and persistence test. Meteorological
quency analysis of rainfall data include, statistical values like data that do not pass the fixed range test should be flagged as erro-
extreme daily annual rainfall (e.g. García-Marín et al., 2011), neous and shall not be used in the subsequent procedures.
parameters from the probability distribution functions of the data Dynamic and persistence tests are based on statistical decisions
(e.g. Easterling, 1989), location attributes like latitude, longitude and were developed by Hubbard et al. (2005). Data rejected by
and altitude (e.g. Guttman, 1993) and distance to the sea (e.g. these last tests should be flagged as suspect, and they should be
Moreno and Roldán, 1999), among others. Recent works deal with validated by manual inspection in order to decide their rejection
atmospheric variables (e.g. Satyanarayana and Srinivas, 2011). This or acceptance.
background shows that other rainfall characteristics can be used to
form homogeneous regions in a certain area. The multifractal char- 2.2.1. Range test
acterization of rainfall data series can then be considered with this This test is based upon a combination of performance specifica-
aim, being the application of this theory justified because of the tions for each sensor and physical/climate extremes for each loca-
high variability of rainfall in a wide range of temporal scales. tion and variable (Estévez et al., 2011b). Limits of different
During the last decades rainfall process has been widely analyzed meteorological variables may depend on climate conditions at
from a multifractal point of view (e.g. Schertzer and Lovejoy, the station’s site and on a season or a month. Any observation that
1987; Ladoy et al., 1993; Fraedrich and Larnder, 1993; Over and occurs outside the maximum or minimum allowable value is prop-
Gupta, 1994; Svensson et al., 1996; Tessier et al., 1993, 1996; De erly flagged. Many authors have used this upper and lower thresh-
Lima and Grasman, 1999; Kiely and Ivanova, 1999; Sivakumar, old in order to consider valid precipitation data or not (Reek et al.,
2001; Veneziano and Furcolo, 2002; Labat et al., 2002; Olsson 1992; Meek and Hatfield, 1994; Shafer et al., 2000; Feng et al.,
and Burlando, 2002; Kantelhardt et al., 2006; García-Marín et al., 2004). There are two types of ranges: fixed (physics and instru-
2008a,b; García-Marín et al., 2013). The main advantages of multi- mental) and dynamic ranges. The fixed range test used in this work
fractal analysis are that multifractal parameters are independent of compares each precipitation value (P) with established extreme
the available data for the different scales and that no probability values, one based on physical limit (negative precipitation values
distribution function has to be assumed for the data set. are not possible) and the upper limit proposed by Shafer et al.
The main objective of the present work is to perform a regional (2000), according to:
frequency analysis of historical daily rainfall data series in the pro-
0 6 P 6 508 mm ð1Þ
vince of Malaga using their multifractal characterization to delimit
the homogeneous regions. Any observation that occurs outside the acceptable range is
flagged as erroneous and is not validated by the next tests.
Dynamic bounds for each meteorological variable are based on
2. Materials and methods
extreme values measured for each location or on the theoretical
possible extreme for each site and time period (Feng et al., 2004).
2.1. Source of data
Variable high/low extreme values recorded for precipitation used
in this test were reported by the Spanish Meteorological Agency
Daily precipitation data from 72 stations located in the province
(AEMET, 2012) and they were obtained from the long-term mete-
of Malaga and supplied by the Hydrographic Confederation of the
orological observatory of Malaga airport (Table 2). Data rejected by
Southern Basin were used to carry out this work. Malaga is located
this test are flagged as suspect and they should be verified by man-
at the western end of the Mediterranean Sea and in the south of the
ual inspection.
Iberian Peninsula. The geographical distribution of the stations
throughout the province of Malaga is shown in Fig. 1. Site eleva-
2.2.2. Tests based on statistical decisions
tions range from 5 to 1190 m above mean sea level, longitude, from
This type of tests is a quality control procedure that will either
3°510 4600 to 5°240 3100 W and latitude, from 36°260 3500 to 37°110 0100 N
accept the precipitation datum as being true, or reject the datum
(Table 1).
and label it as an outlier. The number of outliers for precipitation
dataset is related to a dynamic range used on each test. If the
2.2. Validation procedures and precipitation data quality control datum is valid and is accepted as such or the datum is invalid
and rejected, the test is working appropriately. When the datum
In recent years various methods have been developed as valida- is valid and is rejected by the test, a Type I error is committed. If
tion procedures of meteorological data (Feng et al., 2004; the datum is not valid but is accepted by the test, a Type II error
Zahumensky, 2004; Kunkel et al., 2005; Estévez et al., 2011b). is committed (Hubbard et al., 2005). Two procedures were carried
108 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

Fig. 1. Spatial distribution of stations located in Malaga province (Andalusia, Southern Spain).

out based on climatic conditions: seasonal thresholds and seasonal 2.3. Regional frequency analysis
persistence. These tests are related to station climatology at the
monthly level, using dynamic limits. The main objective of the regional frequency analysis (RFA) is
The threshold test is a quality control approach that checks the estimation of extreme events of different return periods by
whether the variable x falls in a specific range for the month in using probability distribution functions. Regionalization tech-
question. The equation for precipitation (x) is: niques are usually applied in hydrology to help the extrapolation
of information from gauged to ungauged sites. RFA lets the estima-
x  f rX 6 x 6 x þ f rX ð2Þ tion of quantiles in a certain place considering data from other
locations.
Lineal moments (L-Moments) are lineal functions of the proba-
where x is the mean value for each month and rX is the standard bility weighted moments (Greenwood et al., 1979) and were intro-
deviation of the daily values for the month in question. This rela- duced by Hosking (1990, 1992). They are used on all the steps of
tionship indicates that with larger values of f, the number of poten- RFA (Hosking and Wallis, 1993, 1997; Rao and Hamed, 2000): data
tial outliers decreases. screening and homogeneous regions checking, regional distribu-
The persistence test checks the variability of the measurements. tion function identification, and development of the regional
When the variability is too high or too low, the data should be growth curve.
flagged for further checking. If the sensor fails it will often report
a constant value and the standard deviation (r) will become smal- (a) Data screening. L-Moments and their corresponding
ler. When the sensor is out for an entire period, r will be zero. If the L-Moments ratios (L-CV, L-Skewness and L-Kurtosis) have
instrument works intermittently and produces reasonable values to be obtained for all the data series. These values are later
interspersed with zero values, thereby greatly increasing the vari- used to obtain the discordancy measure (Di), given by:
ability for the period. This test compares the standard deviation for
the time period being tested to the limits expected as follows: 1
Di ¼  ÞT A1 ðui  u
Nðui  u Þ ð4Þ
3
r j  f rj;r 6 r 6 r j þ f rj;r ð3Þ PN PN
being, A¼ i¼1 ðui
 Þðui  u
u  Þ,  ¼ N 1
u i¼1 ui , ui ¼
½LC v ; LC is ; LC ik 
i
and N = the number of stations.
where r is the standard deviation from daily values for each month Regard the sample L-Moments ratios of a site as a point in a
(j) and year, rj;r is the standard deviation of r for the month in three-dimensional space. A group of sites will then yield a
question and r j is the mean of these standard deviations for each cloud of such points. Flag as discordant any point that is
month. far from de center of the cloud. Declare site i to be discordant
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 109

Table 1
Name of the location, data time-period, coordinates and elevation of the weather stations used in this study (Málaga province).

No. Name Time-period Latitude (N°) Longitude (W°) Elevation (m)


0 00 0 00
1 Agujero 1952–1999 36°46 25 4°25 57 100
2 Alcaucín Cortijo 1947–1999 36°510 5000 4°090 3100 220
3 Alcaucín Forestal 1946–1999 36°540 0400 4°070 4600 508
4 Alfarnate 1941–1999 36°590 4000 4°150 3600 880
5 Alhaurín El Grande 1964–1999 36°370 4000 4°410 4600 400
6 Aljaima 1946–1999 36°440 0400 4°400 2100 50
7 Almargen 1945–1999 37°000 1000 5°010 1600 510
8 Almogia Los Llanes 1949–1999 36°490 3400 4°320 2600 363
9 Alora 1946–1999 36°490 0900 4°410 5600 110
10 Alozaina 1944–1999 36°430 4000 4°510 2600 386
11 Alpandeire 1940–1999 36°380 0000 5°120 600 695
12 Antequera El Águila 1943–1999 37°010 4500 4°330 3600 460
13 Archidona 1936–1999 37°050 4000 4°230 0700 598
14 Arriate 1947–1999 36°480 0400 5°080 2600 7
15 Benahavis 1952–1999 36°310 2500 5°020 4600 190
16 Benalmádena 1964–1999 36°350 4200 4°340 2400 240
17 Benamargosa 1942–1999 36°500 0400 4°110 3100 96
18 Benamocarra 1942–1999 36°470 2500 4°090 3600 179
19 Benaoján CP 1959–1999 36°410 1500 5°160 5100 920
20 Bobadilla Estación 1944–1999 37°020 0900 4°430 3600 384
21 Borregos 1949–1999 37°040 5500 4°370 0700 410
22 Buitreras CE 1941–1999 36°320 2000 5°220 4600 212
23 Buitreras Presa 1943–1999 36°340 4500 5°200 0600 334
24 Campillos 1945–1999 37°020 5000 4°510 4600 460
25 Canillas Aceituno 1942–1999 36°520 2600 4°040 5700 645
26 Cartajima 1943–1999 36°380 4700 5°090 1500 846
27 Cartama Estación 1948–1999 36°440 0400 4°360 2100 35
28 Casabermeja 1945–1999 36°530 2900 4°250 4600 547
29 Casabermeja V. Pineda 1947–1999 36°520 2500 4°260 6000 693
30 Casapalma 1959–1999 36°560 5000 4°200 3900 880
31 Casarabonela Forestal 1946–1999 36°470 0000 4°500 2600 494
32 Casares 1945–1999 36°260 3500 5°160 2600 415
33 Chorro Estación 1948–1999 36°540 3000 4°450 3600 225
34 Coín 1943–1999 36°390 4000 4°450 3100 209
35 Colmenar 1943–1999 36°540 1000 4°200 600 680
36 Cómpeta 1942–1999 36°500 0000 3°580 3100 636
37 Conde de Guadalhorce 1943–1999 36°560 0000 4°480 0600 325
38 Contaderas Forestal 1956–1999 36°500 4000 4°230 1600 630
39 Corchado Central 1942–1999 36°290 4500 5°240 3100 98
40 Cuevas Becerro 1945–1999 36°520 4000 5°020 4600 735
41 El Burgo 1943–1999 36°470 2000 4°560 4600 591
42 Fuente Piedra 1949–1999 37°080 0500 4°430 4700 440
43 Fuente Piedra Herriza 1964–1999 37°070 0000 4°430 4600 425
44 Gobantes Vivero 1945–1999 36°560 2000 4°470 3600 360
45 Humilladero 1946–1999 37°060 5000 4°510 1600 448
46 Hundidero Pto. Sapo 1954–1999 36°500 5000 4°470 4600 870
47 Istán 1942–1999 36°350 0000 4°560 5600 310
48 Jimena Libar CE 1946–1999 36°390 4000 5°160 5100 429
49 Las Mellizas 1946–1999 36°520 0000 4°430 1600 140
50 Málaga Oficina 1961–1999 36°430 2000 4°240 1700 31
51 Marbella Inst. Laboral 1946–1999 36°300 2900 4°530 0600 20
52 Mijas Faro Cal 1942–1999 36°300 2900 4°370 6000 20
53 Moclinejo 1940–1999 36°460 1500 4°150 1600 451
54 Montejaque CE 1942–1999 36°450 1000 5°140 2600 530
55 Nerja 1947–1999 36°440 5000 3°520 3600 21
56 Ojen 1948–1999 36°330 5000 4°510 2100 335
57 Parauta Taramal 1947–1999 36°420 3500 5°010 3600 1190
58 Parchite 1946–1999 36°480 4500 5°060 3600 755
59 Peña Enamorados 1943–1999 37°030 2500 4°300 0100 470
60 Periana 1939–1999 36°550 4000 4°110 3100 547
61 Pizarra 1946–1999 36°460 0000 4°420 2600 85
62 Rincón de la Victoria 1964–1999 36°430 0000 4°170 2100 5
63 Riogordo 1945–1999 36°540 5400 4°170 3600 400
64 Ronda CE 1940–1999 36°440 3000 5°100 1600 600
65 SP Alcántara 1946–1999 36°280 3900 4°590 3600 25
66 Tolox Millanas 1964–1999 36°410 2500 4°510 3100 200
67 Torrox 1944–1999 36°450 2900 3°570 0600 145
68 Vegueta Grama 1943–1999 36°490 1000 3°510 4600 410
69 Vélez Málaga 1940–1999 36°470 00°00 4°060 1600 165
70 Villanueva de Tapias 1950–1999 37°110 0100 4°200 0600 670
71 Viñuela 1942–1999 36°510 0500 4°080 0600 130
72 La Yedra 1946–1999 36°590 0000 4°270 4100 780
110 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

Table 2 canonical correlation analysis, artificial neuronal network


Precipitation extreme values at Málaga airport observatory (AEMET, 2012). and fuzzy logic) being cluster analysis of site characteristics
Month Precipitation (mm) Date the most practical one (Hosking and Wallis, 1997). Cluster
January 121.5 18/01/1979 analysis (e.g. Periago et al., 1991; Bonell and Sumner,
February 151 22/02/1969 1992) is a standard method of statistical multivariate analy-
March 102.1 27/03/2004 sis used to investigate, interpret and group, if possible, data
April 47.4 9/04/1994 (Rao and Srinivas, 2008). This technique has been widely
May 44 4/05/1946
June 72.5 14/06/1974
used in hydrology (e.g. Burn, 1989; Hall and Minns, 1999;
July 22.2 19/07/1942 Lecce, 2000; Jingyi and Hall, 2004; Kyselý et al., 2007;
August 70.3 28/08/1987 Srinivas et al., 2008; Meshgi and Khalili, 2009;
September 313 27/09/1957 Satyanarayana and Srinivas, 2011) and can identify regions
October 109.4 31/10/2008
that are not necessary geographically contiguous.
November 164 28/11/1942
December 132.7 18/12/2010 Once a set of regions has been identified it has to be tested if
the proposed regions may be accepted as being homogeneous.
The heterogeneity test of Hosking and Wallis (1997) can be
used. The heterogeneity measure compares the between-site
if Di exceeds a critical value that depends on the number of variations in sample L-Moments for the group of sites with
sites of the group (Hosking and Wallis, 1997). what would be expected for a homogeneous region.
(b) Homogeneous regions checking. Of all the stages in a RFA, According to Hosking and Wallis (1997) the heterogeneity
the identification of homogeneous regions is usually the measure, H, is obtained by:
most difficult. Authors such as Greis and Wood (1981),
Hosking et al. (1985) and Lettenmaier et al. (1987), among ðV  lV Þ
H¼ ð5Þ
others, showed the importance of the regional homogeneity. rV
The aim is to form groups of sites with identical frequency P 2 P 1=2 P
distributions apart from a site-specific scale factor. Several being V ¼ f Ni¼1 ni ðt ðiÞ  t R Þ = Ni¼1 ni g , tR ¼ Ni¼1 ni t ðiÞ =
PN
methods exist for forming groups (visual inspection, cluster i¼1 ni , N the number of sites, ni the record length of site i,
analysis, principal component analysis, residuals method, and t ðiÞ the sample L-Moment ratios. These authors suggest

Fig. 2. Quality control procedure flowchart.


A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 111

Table 3 homogenous region. It is supposed that all the stations fol-


Number of records flagged by AEMET limits check (fixed range test). low the same distribution except than a scale factor repre-
Station Records Station Records sented by the mean of the frequency distribution at each
flagged flagged site and estimated by the sample mean of the at-site data.
Agujero 7 Conde de 5 This method is known as Flood index (Dalrymple, 1960;
Guadalhorce Hosking and Wallis, 1997). To measure the goodness of fit
Alcaucín Cortijo 8 Contaderas Forestal 9 to a distribution of three parameters, Hosking and Wallis
Alcaucín Forestal 42 Corchado Central 13
Alfarnate 19 Cuevas Becerro 17
(1993, 1997) developed the statistic ZDIST given by:
Alhaurín el Grande 6 El Burgo 7
Aljaima 9 Fuente Piedra 3
Z DIST ¼ ðsDIST
4  tR4 þ B4 Þ=r4 ð6Þ
Almargen 5 Fuente Piedra 4
Herriza
Almogia los Llanes 8 Gobantes Vivero 9 Being sDIST
4 the L-Kurtosis of the fitted distribution; t R4 the
Álora 7 Humilladero 3 L-Kurtosis coefficient for the region; r4 the standard devia-
Alozaina 13 Hundidero Pto. Sapo 14
tion of tR4 , and B4 the bias of tR4 . The fit of a specific distribu-
Alpandeire 27 Istán 29
Antequera Águila 2 Jimena Libar CE 19 tion is considered to be adequate if the statistic ZDIST is
Archidona 5 Las Mellizas 9 sufficiently close to zero, a reasonable criterion being
Arriate 4 Málaga Oficina 4 |ZDIST| 6 1.64 (Hosking and Wallis, 1997).
Benahavis 14 Marbella Inst. 11 (d) Development of the regional growth curve. Considering the
Laboral
index-flood procedure for a homogeneous region, the quan-
Benalmádena 5 Mijas Faro Cal 9
Benamargosa 10 Moclinejo 9 tile function Q i ðFÞ may be written
Benamocarra 12 Montejaque CE 7
Benaoján CP 54 Nerja 5
Bobadilla Estación 7 Ojén 19
Q i ðFÞ ¼ li qðFÞ; i ¼ 1; . . . N ð7Þ
Borregos 5 Parauta Taramal 13
Buitreras CE 21 Parchite 9 where li is the index flood, and qðFÞ is the regional growth
Buitreras Presa 21 Peña Enamorados 5 curve calculated with the regional L-Moment algorithm
Campillos 8 Periana 11 (Hosking and Wallis, 1997).
Canillas Aceituno 15 Pizarra 12
Cartajima 37 Rincón de la Victoria 4
2.4. Multifractality of rainfall
Cártama Estación 6 Riogordo 8
Casabermeja 6 Ronda CE 4 The multifractal description has been proposed as an efficient
Casabermeja VP 8 SP Alcántara 14 tool to study the time structure of some phenomena
Casapalma 7 Tolox Millanas 5
(Jiménez-Hornero et al., 2011). In order to identify the multifractal-
Casarabonela 27 Torrox 7
Forestal ity of any process different methodologies can be applied. One of
Casares 25 Vegueta Grama 12 the most widely used in hydrology is the turbulence formalism
Chorro Estación 12 Vélez Málaga 7 developed by Schertzer and Lovejoy (1987) (e.g. Schertzer and
Coín 20 Villanueva de Tapias 5 Lovejoy, 1988; De Lima and Grasman, 1999; De Lima and de
Colmenar 12 Viñuela 10
Cómpeta 20 Yedra 8
Lima, 2009; García-Marín et al., 2012). This approach assumes that
the variability of the process could be directly modeled as a
stochastic (or random) turbulent cascade process (Schertzer and
that the region can be regarded as ‘‘acceptably homoge- Lovejoy, 1987; Gupta and Waymire, 1993; Over and Gupta, 1994;
neous’’ if H < 1, ‘‘possibly heterogeneous’’ if 1 6 H < 2, and Lovejoy and Schertzer, 1995). According to this methodology, the
‘‘definitely heterogeneous’’ if H P 2. multifractal temporal structure of a process can be investigated
(c) Regional distribution function identification. In RFA a single by studying the (multiple) scaling of its statistical moments.
frequency distribution is fitted to data from several sites, The scaling of the moments can be described by the exponent
once it has been confirmed that they compound a function K(q), that satisfies (Schertzer and Lovejoy, 1987):

Fig. 3. Results of the application of dynamic range test, showing the fraction of precipitation data flagged from ‘‘Agujero, Almargen. Pizarra and Rincón de la Victoria’’ stations
for the values of f shown.
112 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

heqk i  kKðqÞ ð8Þ AEMET extreme values are reported in Table 3. It can be observed
that the station with the highest number of outliers was ‘‘Benaoján
where heqk i is the average qth moment of the intensity of the process CP’’ (54) and the lowest number of outliers detected was given in
at a scale k (the ratio between the length of the data set and any Antequera, with only 2 records flagged. The weakness of this check
time interval) and K(q) is the so-called moments scaling exponent is that there are no long-term extreme values from each station,
function, that can be obtained using the Trace Moment method only from Malaga airport observatory (Table 2).
(Schertzer and Lovejoy, 1987). Subsequently, dynamic range and persistence tests (based on
K(q) can be regarded as a characteristic function of scaling statistical decisions) were applied, establishing a relationship
behavior (e.g. Svensson et al., 1996). For simple scaled or (mono) between the number of potentially erroneous data and the factor
fractal processes (e.g. Mandelbrot, 1972) the plot of K(q) versus q f used. Figs. 3 and 4 show an example of the application of each
is a straight line but does not pass through the origin, showing only test, at ‘‘Agujero’’ and ‘‘Alozaina’’ stations. The values of f selected
an intensity level for the process. If K(q) is linear through the origin, for both tests have been those that mark the 1.5% of the original
the measure is self-similar. However, if the moment scaling func- precipitation dataset, which is a reasonable percentage of data
tion is nonlinear, the measure is multifractal (Veneziano et al., flagged as potentially erroneous records (Estévez et al., 2011a).
2006). Only flagged data as outliers by both tests were initially dismissed
Some important multifractal parameters can be obtained from at this early stage of the validation process. The next verification is
the K(q) function: the orders of singularities (c) and the moment for checking if these records have successfully passed the fixed
qcrit . The singularities values c are upper limited by cmax. This range test defined by the extreme values of the AEMET, only for
implies that the K(q) function is linear for q > qcrit . Such a disconti- those stations that are part of the homogeneous region reported
nuity in the first or second derivative of K(q) arises either because by García-Marín et al. (2011), corresponding to Malaga airport.
of divergence of moments at qcrit (called qD in the first order case) Thus data flagged by the three tests will be discarded for the sta-
or due to the inadequate sample size (at qs in the second order tions mentioned above and data flagged by the dynamic range test
case). cmax is the corresponding largest singularity present in the and persistence test for all other stations.
sample data (e.g. Schertzer and Lovejoy, 1987; Tessier et al., Table 4 shows the number of precipitation records from the
1993; Lovejoy and Schertzer, 1995; De Lima and de Lima, 2009) original dataset, the number of data discarded and flagged as out-
and can be determined from, liers and the number of data for each station that have been used in
cmax ¼ maxðK 0 ðqÞÞ ð9Þ this work. ‘‘Almargen’’ and ‘‘Cartajima’’ were the stations with
more data flagged (24 records) and the station with the lowest
In terms of K(q), the low rainfall intensities are characterized by number of data flagged was ‘‘Málafa oficina’’ (1 record).
Kð0Þ, being the corresponding fractal dimension of the set over
which the measure is carried out, D ¼ 1 þ Kð0Þ. 3.2. Regional frequency analysis

3. Results and discussion Extreme daily annual rainfall data series were obtained by using
the validated data series from the 72 available stations on the pro-
3.1. Quality control vince of Malaga (Table 1, Fig. 1). Each data series were used to
characterize each site with the L-Moments values and ratios
The making decision process based on quality control tests is (L-Cv, L-Cs, L-Ck) (Table 5). These L-Moments ratios have been used
described in Fig. 2. The first test applied was the fixed range one, on all the steps needed for the subsequent development of the RFA.
which consists of two checks. In the first of them is verified that A first region named Malaga, composed by the whole of 72 sta-
there are no values below 0 mm or above 508 mm of rainfall in tion, is firstly tested in order to verify its possible homogeneity. For
the initial data set. In the second one, the extreme values obtained regions with more than 15 sites, the critical value of Di is 3.00
from AEMET are used in order to select those data that exceed (Hosking and Wallis, 1997). Four stations overcame this value
these long-term monthly maximum precipitation values in the and had to be taken out from the Malaga Region: Alozaina (3.55),
province of Malaga. No data were flagged as erroneous by the first Alpandeire (3.02), Rincón (5.66) y Yedra (3.06). Considering then
check of the fixed range test. Data flagged by the test based on a region with the rest 68 sites, a new discordancy test were done

Fig. 4. Results of the application of persistence test, showing the fraction of precipitation data flagged from ‘‘Agujero, Almargen. Pizarra and Rincón de la Victoria’’ stations for
the values of f shown.
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 113

Table 4 Table 5
Number of precipitation records from the original dataset and number of data flagged L-Moment ratios for the 72 stations.
by validation procedures for each station.
No. L-Cv L-Cs L-Ck No. L-Cv L-Cs L-Ck
No. Station Data from the original Data
1 0.2170 0.2343 0.1806 37 0.1983 0.1923 0.1131
dataset flagged
2 0.1477 0.1360 0.0912 38 0.1956 0.2047 0.1471
1 Agujero 17.532 2 3 0.1970 0.0028 0.1403 39 0.1993 0.0670 0.1356
2 Alcaucín Cortijo 19.358 16 4 0.1518 0.0308 0.0231 40 0.1963 0.2274 0.1982
3 Alcaucín Forestal 19.723 11 5 0.1781 0.2340 0.1230 41 0.2158 0.2903 0.1633
4 Alfarnate 21.549 12 6 0.2008 0.1135 0.0994 42 0.1847 0.1686 0.1760
5 Alhaurín el Grande 13.149 11 7 0.1737 0.1643 0.1710 43 0.1640 0.2476 0.2639
6 Aljaima 19.723 7 8 0.2492 0.2071 0.1023 44 0.2187 0.2456 0.1085
7 Almargen 20.088 24 9 0.2073 0.0794 0.0087 45 0.1616 0.1699 0.2070
8 Almogia los Llanes 18.627 4 10 0.2801 0.3198 0.1661 46 0.2150 0.1269 0.1521
9 Alora 19.723 4 11 0.1532 0.0104 0.1663 47 0.1926 0.2253 0.1286
10 Alozaina 20.454 21 12 0.1567 0.1138 0.0692 48 0.1671 0.1648 0.1871
11 Alpandeire 21.915 11 13 0.1710 0.1643 0.0877 49 0.2164 0.1417 0.1201
12 Antequera el Águila 20.819 17 14 0.1991 0.3065 0.1800 50 0.2135 0.2581 0.2271
13 Archidona 23.376 19 15 0.2193 0.2015 0.1562 51 0.2614 0.2168 0.1094
14 Arriate 19.358 10 16 0.1902 0.2036 0.1544 52 0.2176 0.2641 0.2423
15 Benahavis 17.532 3 17 0.1765 0.1766 0.1990 53 0.1855 0.0484 0.1074
16 Benalmádena 13.149 2 18 0.1946 0.1704 0.1195 54 0.1951 0.1702 0.1131
17 Benamargosa 21.184 21 19 0.2100 0.1951 0.1734 55 0.1893 0.1295 0.1130
18 Benamocarra 21.184 11 20 0.2137 0.2129 0.1275 56 0.2229 0.1135 0.0853
19 Benaoján CP 14.975 7 21 0.1927 0.1414 0.0770 57 0.1683 0.1092 0.0250
20 Bobadilla Estación 20.454 17 22 0.1972 0.1284 0.1254 58 0.2040 0.3012 0.1664
21 Borregos 18.627 8 23 0.1741 0.1106 0.1269 59 0.2029 0.1880 0.1572
22 Buitreras CE 21.549 4 24 0.1712 0.2551 0.2320 60 0.1560 0.1412 0.1291
23 Buitreras Presa 20.819 21 25 0.1559 0.1461 0.0677 61 0.2389 0.1658 0.1294
24 Campillos 20.088 13 26 0.1917 0.1026 0.0557 62 0.2328 0.3228 0.3796
25 Canillas Aceituno 21.184 8 27 0.2127 0.1246 0.1175 63 0.2049 0.0468 0.1259
26 Cartajima 20.819 24 28 0.2012 0.1835 0.1001 64 0.1664 0.0598 0.0700
27 Cartama Estación 18.993 3 29 0.2088 0.0942 0.0241 65 0.2022 0.2343 0.1237
28 Casabermeja 20.088 8 30 0.1981 0.0714 0.0555 66 0.2132 0.1659 0.1682
29 Casabermeja VP 19.358 13 31 0.2199 0.0530 0.0940 67 0.1851 0.0492 0.0637
30 Casapalma 14.975 5 32 0.2026 0.1282 0.0965 68 0.1979 0.1336 0.0773
31 CasarabonelaForestal 19.723 18 33 0.2174 0.2401 0.1670 69 0.1651 0.2046 0.2851
32 Casares 20.088 20 34 0.2343 0.1997 0.0751 70 0.1477 0.1711 0.1969
33 Chorro Estación 18.993 16 35 0.1644 0.0614 0.0182 71 0.1782 0.2465 0.0975
34 Coín 20.819 6 36 0.1551 0.0805 0.1136 72 0.1631 0.0683 0.0900
35 Colmenar 20.819 17
36 Cómpeta 21.184 11 No.: number of the station; L-Cv: Lineal coefficient of variation; L-Cs: Lineal
37 Conde Guadalhorce 20.819 18 skewness coefficient; L-Ck: Lineal kurtosis coefficient.
38 Contaderas Forestal 16.071 15
39 Corchado Central 21.184 4
and two stations resulted discordant, Alcaucin Forestal and Velez,
40 Cuevas Becerro 20.088 13
41 El Burgo 20.819 22
with Di values of 3.32 and 3.25, respectively. The new region of
42 Fuente Piedra 18.627 13 Malaga was then composed by 66 stations. No stations resulted
43 Fuente Piedra Herriza 13.149 12 discordant and so, the value of H (3.75) was then obtained showing
Herriza that the region was heterogeneous.
44 Gobantes Vivero 20.088 15
The main area had to be then divided into sub-regions. With
45 Humilladero 19.723 15
46 Hundidero Pto. Sapo 16.801 16 this purpose, the multifractal characterization of daily rainfall val-
47 Istán 21.184 19 idated data series was carried out for all the available stations. To
48 Jimena Líbar CE 19.723 9 analyse the multifractal behavior of daily rainfall time series, the
49 Las Mellizas 19.723 11
scaling of the moments q was determined. For a example selection
50 Málaga Oficina 14.244 1
51 Marbella Inst Laboral 19.723 4
of four stations (Agujero, Almargen, Pizarra and Rincón de la
52 Mijas Faro Cal 21.184 3 Victoria), Fig. 5 shows the log–log plot of the average qth moments
53 Moclinejo 21.915 11 of the rainfall intensity ek against the scale ratio k. For each place,
54 Montejaque CE 21.184 19 the top plot (a) shows moments larger than 1 and the bottom one
55 Nerja 19.358 7
(b) shows those moments smaller than one. In all the cases, only
56 Ojén 18.993 5
57 Parauta Taramal 19.358 5 some of the q moments calculated are plotted in order to clarify
58 Parchite 19.723 13 the figure and the results. As it can be seen, the scaled behavior
59 Peña Enamorados 20.819 15 was detected from 1 day to 16 or 32, depending on the place.
60 Periana 22.280 12 Table 6 summarizes the limits of this scaled behavior for all the
61 Pizarra 19.723 4
62 Rincón de la Victoria 13.149 4
places. Other authors have found similar values of these scaling
63 Riogordo 20.088 14 break all over the world: from 16 days (e.g. Ladoy et al., 1991,
64 Ronda CE 21.915 16 1993; Tessier et al., 1996), 1 month (e.g. Fraedrich and Larnder,
65 SP Alcántara 19.723 5 1993; Svensson et al., 1996; Labat et al., 2002), up to 42 or even
66 Tolox Millanas 13.149 3
64 days (e.g. García-Marín et al., 2008a,b). This scaling break is
67 Torrox 20.454 8
68 Vegueta Grama 20.819 13 usually explained in terms of a synoptic maximum, which corre-
69 Vélez Málaga 21.914 12 sponds to temporal scales associated with planetary size structures
70 Villanueva Tapia 18.262 13 (e.g. Kolesnikov and Monin, 1965) and a natural scale for separat-
71 Viñuela 21.184 12 ing meteorological from climatological regimes (e.g. Pandey et al.,
72 Yedra (La) 19.723 14
1998).
114 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

Fig. 5. Log–log plot of the qth moments of daily rainfall intensity ek versus the scale ratio k at the locations of ‘‘Agujero, Almargen. Pizarra and Rincón de la Victoria’’. (a) For
moments higher than 1; (b) for moments lower than 1.

Table 6
Upper limit (days) for the scaled behavior for the 72 stations. The lower limit is 1 day for all of them.

No. Upper limit No. Upper limit No. Upper limit No. Upper limit No. Upper limit No. Upper limit
1 32 13 32 25 32 37 32 49 16 61 16
2 32 14 32 26 32 38 16 50 32 62 16
3 32 15 32 27 32 39 8 51 16 63 32
4 32 16 16 28 16 40 16 52 16 64 32
5 32 17 16 29 16 41 32 53 32 65 32
6 32 18 32 30 16 42 16 54 16 66 16
7 32 19 32 31 32 43 16 55 16 67 32
8 16 20 32 32 16 44 16 56 32 68 16
9 16 21 32 33 16 45 16 57 16 69 16
10 32 22 16 34 32 46 32 58 16 70 16
11 32 23 32 35 32 47 32 59 32 71 16
12 16 24 32 36 32 48 16 60 16 72 32

As an example, Fig. 6 shows the K(q) functions obtained for Rincon de la Victoria. This variation occurs for all the sites
scales from 1 to 16 days in Pizarra and Rincón, and from 1 to analyzed.
32 days in Agujero and Almargen. As it can be checked, the behav- The values of the critical moment qcrit are also different, being
ior of the K(q) functions around the mean (q = 1) is similar at all the 1.05, 1.60, 2.85 and 3.50 for the four stations showed in Fig. 6
locations, being Kð1Þ  0, showing the conservation condition (Agujero, Pizarra, Almargen and Rincon de la Victoria, respectively)
hek i ¼ 1 (De Lima, 1998). For the not showed stations, similar and can be also related to the degree of multifractality of a data set.
behavior was found. The values of Kð0Þ are related to the ‘zeros’ This critical moment has been traditionally related to the algebraic
(De Lima, 1998) in the data sets and with the fractal dimension decay of the probability distribution of the extreme events size
D, being close to 1 for data series with very few zero data values (e.g. Schertzer et al., 1993; Schertzer and Lovejoy, 1994; Tessier
(e.g. De Lima and de Lima, 2009; García-Marín et al., 2013). For et al., 1996; García-Marín et al., 2013) and suggests that values
all the represented sites (and also for non-represented ones) in of qcrit higher than 2 are present in multifractal data sets. Table 7
Fig. 6 the values of Kð0Þ led to fractal dimension values D of around shows the values for the critical moment qcrit at each station.
0.5, showing an important presence of zero values in the rainfall According to the values of qcrit shown in Table 7, several groups
data sets analyzed. of stations could be formed. Considering the limit between mono
Despite this similar behavior of the K(q) functions at the values and multifractal behavior (qcrit = 2), two initial groups were
of q = 0 and q = 1 for all the places, some differences appear, being formed. For each group of stations, a RFA of extreme daily rainfall
the basis for grouping stations into regions for the later RFA. was performed being both regions heterogeneous according to the
According to the linear or convex shape of the K(q) function (e.g. H values obtained.
Parisi and Frisch, 1985), Agujero and Pizarra seem to have a After these results, subgroups of stations were made below and
monofractal behavior. Multifractality appears for Almargen and above the qcrit values of 2, and the RFA was again tested at each
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 115

subgroup. Different combinations were tested and stations were regions can be considered possibly heterogeneous. Following the
eliminated and or added to the groups until the H values obtained same authors, values of H < 2 represent an acceptable value for
were satisfactory (i.e. corresponded to homogenous or possible considering a region as homogeneous. The values obtained for
heterogeneous). The final results are described below and summa- Regions 3 and 4 are besides closer to 1 than to 2, so it was decided
rized in Table 8, where the information related to each region to consider both regions as homogeneous.
(Regions 1–4) can be found. Table 8 shows the number and names As it can be seen in Table 8, there is an overlap between Region
of the sites at each region, the values of qcrit used to delimit the 1 and Region 4, being all the stations included in the former also
groups, and value of H once the RFA is done. included in the latter. Thus, if any extreme rainfall quantile has
Region 1 is composed by 12 stations with values of qcrit between to be calculated for any common station, regional data from
1.05 and 1.75. No stations resulted discordant and the value of H Region 1 should be used. For no common sites, Region 4 will be
was 0.06, being then this region homogeneous. Region 2 was orig- used for quantile estimations.
inally composed by 29 stations (with qcrit > 2.60) but two of them Stations with qcrit values between 1.85 and 2.50 are not
(Rincon de la Victoria and Yedra) resulted discordant attending included in any homogeneous or possibly heterogeneous region.
to the value of Di. Once these stations were removed, no sites were Thus, with these sites and the discordant ones (Rincón de la
discordant and the value of H was 0.59, being Region 2 also homo- Victoria and Yedra), new groups should be made considering more
geneous. Regions 3 and 4 are composed by 4 and 16 stations, characteristic parameters. The new bunch is composed by 25 sta-
respectively. The values of qcrit are lower than 0.40 for Region 3, tions that will be now characterized by the values of qcrit and cmax
while for Region 4 vary from 0.80 to 1.75. No discordant station from the K(q) function, and the 100-year return period extreme
was found, and the values of H were 1.21 for Region 3 and 1.28 daily annual rainfall (R100) (Table 9). This last value was obtained
for Region 4. Following Hosking and Wallis (1997) criterion, both by applying a local frequency analysis to each site’s extreme daily

Fig. 6. Empirical moments scaling exponent functions K(q) for the range of scales detected at the stations of ‘‘Agujero, Almargen. Pizarra and Rincón de la Victoria’’ and
considering the qth moments of the daily rainfall intensities.
116 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

Table 7 Table 9
Values of the critical moments obtained (qcrit) from the K(q) function at each station. Characteristics (qcrit, cmax, R100) of the 25 stations used for the K-means cluster
application.
No. qcrit No. qcrit No. qcrit
No. Stations R100 (mm) qcrit cmax
1 1.05 25 2.50 49 3.00
2 2.40 26 0.00 50 2.50 2 Alcaucín Cortijo 109.32 2.40 0.571
3 2.00 27 1.50 51 1.00 3 Alcaucín Forestal 199.02 2.00 0.564
4 1.85 28 2.25 52 1.15 4 Alfarnate 142.81 1.85 0.541
5 2.15 29 1.75 53 1.85 5 Alhaurín 167.42 2.15 0.744
6 2.15 30 2.85 54 2.85 6 Aljaima 119.49 2.15 0.664
7 2.85 31 1.85 55 1.40 8 Almogia 169.51 2.25 0.771
8 2.25 32 1.85 56 2.50 10 Alozaina 194.07 2.00 0.771
9 3.5 33 3.00 57 3.50 11 Alpandeire 158.01 1.85 0.469
10 2.00 34 1.50 58 2.60 12 Antequera 84.15 1.85 0.623
11 1.85 35 2.25 59 0.80 13 Archidona 83.54 2.15 0.597
12 1.85 36 2.75 60 0.30 15 Benahavis 172.25 1.85 0.550
13 2.15 37 2.50 61 1.60 20 Bobadilla 97.24 1.85 0.644
14 2.75 38 2.85 62 3.50 25 Canillas 114.25 2.50 0.615
15 1.85 39 1.25 63 3.00 28 Casabermeja 121.21 2.25 0.674
16 2.75 40 2.85 64 2.85 31 Casarabonela 192.06 1.85 0.652
17 0.80 41 3.00 65 2.75 32 Casares 176.83 1.85 0.464
18 1.75 42 1.75 66 0.40 35 Colmenar 124.72 2.25 0.504
19 2.60 43 3.00 67 1.60 37 Conde 98.97 2.50 0.667
20 1.85 44 2.75 68 0.80 46 Hundidero 162.92 2.15 0.647
21 1.25 45 3.00 69 3.50 48 Jimena Libar CE 160.23 1.85 0.437
22 2.85 46 2.15 70 3.00 50 Málaga Oficina 147.22 2.50 0.746
23 0.00 47 2.85 71 3.00 53 Moclinejo 112.89 1.85 0.510
24 3.00 48 1.85 72 3.50 56 Ojen 205.85 2.50 0.551
62 Rincón Vª 134.4 3.50 0.961
72 Yedra 108.1 3.50 0.514

annual rainfall validated data series by using the two-parameter


probability distribution function Gumbel. K-means cluster analysis
has been performed and two new groups arose, Region 5 and and for all its sites, the best choice is a local frequency analysis if
Region 6, composed by 12 and 13 stations, respectively any extreme rainfall value is needed.
(Table 10). No discordant stations appeared in any group, and the The results described above show a bunch of homogeneous
values for the parameter H were 0.77 for Region 5, and 7.24 for regions (or possible heterogeneous) obtained by combining multi-
Region 6. According to these results, a new homogeneous region fractal properties of rainfall and cluster analysis. If other grouping
arose (Region 5). Region 6 seemed to be completely heterogeneous methods and rainfall characteristics for the same area were used,
the obtained regions could be different (e.g. García-Marín et al.,
2011).
Once the regions have been compounded and their homogene-
Table 8 ity has been tested, a single frequency distribution can be fitted to
Stations (name), number of stations for each region, qcrit and H values for Regions 1, 2, the data from each region and the regional growth curve can be
3 and 4.
then obtained. The value of the statistic ZDIST has been calculated
Region qcrit value Number Stations H for all the regions considering some three-parameter probability
of value distribution functions: Generalized Normal (GEN-NOR),
stations
Generalized Pareto (GEN-PAR), Generalized Extreme Value (GEV),
1 1.05 < qcrit < 1.75 12 Agujero, Benamocarra, 0.06 Generalized Logistic (GEN-LOG) and Pearson Type III (PT-III) distri-
Borregos, Cártama,
butions. All the results obtained are shown in Table 11. Following
Casabermeja Venta Pineda,
Coin, Corchado, Fuente the criterion of |ZDIST| 6 1.64 for an adequate fit (Hosking and
Piedra, Mijas, Nerja, Pizarra Wallis, 1997), the Generalized Normal and Pearson Type III proba-
2 2.60 < qcrit < 3.50 27 Almargen, Alora, Arriate, 0.59 bility distribution functions could be a good choice for all the
Benalmádena, Benaoján CP, regions. The GEV function is also adequate for Regions 2, 3 and 5.
Buitreras CE, Campillos,
Considering the proper function the one with the closest value to
Casapalma, Chorro,
Cómpeta, Contaderas F., zero of ZDIST, the PT-III distribution is the one that can be used in
Cuevas Becerro, El Burgo, all the regions, except in Region 2 where the lower value belongs
Fuente Piedra Herriza, to the Generalized Normal distribution. Considering these
Gobantes Vivero,
Humilladero, Istán, Las
Mellizas, Montejaque,
Parauta T., Parchite, Table 10
Riogordo, Ronda, SP Stations (name), number of stations for each region and H values for Regions 5 and 6,
Alcántara, Vélez Málaga, obtained after applying K-means cluster analysis.
Villanueva Tapia, Viñuela
Region Number of Stations H
3 0.00 < qcrit < 0.40 4 Buitreras P, Cartajima, 1.21
stations value
Periana y Tolox
4 0.80 < qcrit < 1.75 16 Agujero, Benamargosa, 1.28 5 12 Casabermeja, Casarabonela, Casares, 0.77
Benamocarra, Borregos, Colmenar, Conde de Guadalhorce, Hundidero,
Cártama, Casabermeja Venta Jimena Líbar CE, Málaga, Moclinejo, Ojén,
Pineda, Coin, Corchado Rincón and Yedra
Central, Fuente Piedra, 6 13 Alcaucín Cortijo, Alcaucín Forestal, Alfarnate, 7.24
Marbella, Mijas, Nerja, Peña Alhaurín el Grande, Aljaima, Almogia los
Enamorados, Pizarra, Torrox, Llanes, Alozaina, Alpandeire, Antequera,
Vegueta Grama Archidona, Benahavis, Bobadilla and Canillas
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 117

Table 11 basis of this subdivision. The moments scaling exponent functions


ZDIST values for the five three-parameter probability distribution functions (pdd) of the validated data sets were obtained, and two multifractal
considered for each region: Generalized Logistic (GEN-LOG), Generalized Extreme
Values (GEV), Generalized Normal (GEN-NOR), Pearson Type III (PT-III) and
parameters (the critical moment, qcrit, and the largest singularity,
Generalized Pareto (GEN-PAR). cmax) were derived for each station.
The critical moment value was firstly used for grouping and
Region GEN-LOG (GEV) GEN-NOR PT-III GEN-PAR
four regions were obtained with a total of 47 stations. Two of these
1 4.75 1.65 1.50* 0.74* 4.94 regions resulted homogeneous when the regional frequency anal-
2 4.62 0.93* 0.29* 1.21* 7.33
3 3.04 0.94* 0.98* 0.63* 3.38
ysis was again performed. The two remaining regions were possi-
4 5.24 1.74 1.48* 0.52* 5.78 bly heterogeneous. Nevertheless, they can be considered as
5 3.49 0.32* 0.35* 0.21* 6.23 homogeneous following the criteria of Hosking and Wallis (1997).
Two new groups of stations were made by performing a
ZDIST values marked in bold for the selected pdd.
*
denotes all the adequate fits. K-means cluster analysis. The stations were previously character-
ized by the values of qcrit, cmax and the 100-year return period
extreme daily annual rainfall (R100). The first group was composed
Table 12
Regional growth curves for all the regions and for various return periods (2, 5, 10, 25, by 12 sites and behaved as homogenous after testing the regional
50 and 100 years). frequency analysis of extreme daily annual rainfall data. The last
group of stations was heterogeneous and only a local frequency
Region 2 5 10 25 50 100
analysis of extreme daily annual rainfall is possible for its 13 sites.
1 0.94 1.29 1.51 1.77 1.95 2.13 According to these results, the multifractal characterization of
2 0.94 1.25 1.46 1.74 1.94 2.14
3 0.96 1.26 1.44 1.66 1.81 1.96
rainfall data can be directly used to identify homogeneous regions
4 0.94 1.29 1.51 1.78 1.97 2.15 in a regional frequency analysis. Multifractal parameters can be
5 0.95 1.28 1.48 1.73 1.89 2.06 also the basis of statistical multivariate analysis to group data.

Acknowledgment
Table 13
Extreme daily annual rainfall values (mm) for various return periods (2, 5, 10, 25, 50 We applied the ‘‘sequence-determines-credit’’ (SDC) approach
and 100 years) and for five stations (one per region as an example). for the sequence of authors.
Region Stations Rainfall quantile (mm)
References
2 5 10 25 50 100
1 Agujero 59.05 80.96 94.66 111.08 122.71 133.86 Adamowski, K., 2000. Regional analysis of annual maximum and partial duration
2 Almargen 40.62 54.39 63.61 75.36 84.17 93.05 flood data by nonparametric and L-moment methods. J. Hydrol. 229, 219–231.
3 Periana 55.75 73.38 84.17 96.91 105.84 114.35 Adamowski, K., Alila, Y., 1991. Regional analysis of annual maxima precipitation
4 Torrox 55.39 76.13 89.23 104.98 116.18 126.94 using L-moments. Atmos. Res. 27, 81–92.
5 Ojen 86.61 116.79 135.34 157.31 172.75 187.49 AEMET, 2012. Valores extremos – Valores extremos absolutos – Mapa de España –
Agencia Estatal de Meteorología – AEMET. Gobierno de España. <http://www.
aemet.es/es/serviciosclimaticos/datosclimatologicos/efemerides_extremos>
(accessed 05.07.12) (in Spanish).
probability distribution functions, the regional growth curves were Arbeláez, Castro, 2007. Low flow discharges regional analysis using wakeby
distribution in an ungauged basin in Colombia. Hydrol. Days, 198–208.
obtained for various return periods (Table 12). Bhuyan, A., Bora, M., Kumar, R., 2010. Regional flood frequency analysis of North-
According to the Flood Index method and applying Eq. (7) the Bank of the River Brahmaputra by using LH-moments. Water Resour. Manage.
values of the extreme daily annual rainfall data with a certain 24, 1779–1790.
Bonell, M., Sumner, G., 1992. Autumn and winter daily precipitation areas in Wales,
return period can be obtained. As an example, these values are 1982–1983 to 1986–1987. Int. J. Climatol. 12, 77–102.
shown in Table 13 for five stations (each one belonging to a differ- Burn, D.H., 1989. Cluster analysis as applied to regional flood frequency. J. Water
ent region). Resour. Plan. Manage. 115, 567–582.
Burn, D.H., 1990. Evaluation of regional flood frequency analysis with a region of
influence approach. Water Resour. Res. 26 (10), 2257–2265.
Castellarin, A., Merz, R., Blöschl, G., 2009. Probabilistic envelope curves for extreme
4. Conclusions rainfall events. J. Hydrol. 378, 263–271.
Chen, Y.D., Huang, G., Shao, Q., Xu, C.Y., 2006. Regional analysis of low flow using L-
The results of this study related to the quality control phase moments for Dongjiang basin, South China. Hydrol. Sci. J. 51 (6), 1051–1064.
Chiang, S.-M., Tsay, T.-K., Nix, Y.S., 2002a. Hydrologic regionalization of watersheds.
show that the application of the validation tests to precipitation I: methodology development. J. Water Resour. Plan. Manage. 128 (1), 3–11.
data reveals the existence of anomalous records in various weather Chiang, S.-M., Tsay, T.-K., Nix, Y.S., 2002b. Hydrologic regionalization of watersheds.
stations in Malaga province. In this regard it is noteworthy that the II: applications. J. Water Resour. Plan. Manage. 128 (1), 3–11.
Clausen, B., Pearson, C.P., 1995. Regional frequency analysis of annual máximum
quality control applied to the original precipitation datasets guar- streamflow drought. J. Hydrol. 173, 111–130.
antees the reliability of the subsequent studies or analysis using Cunnane, C., 1988. Methods and merits of regional flood frequency analysis. J.
these meteorological data. The number of discarded data ranged Hydrol. 100, 269–290.
Dalrymple, T., 1960. Flood Frequency Analysis. U.S. Geological Survey Water Supply
from a minimum of 1 (Malaga Office station) to 24 (Cartajima Paper 1543-A, Reston, VA.
and Almargen stations), depending on the geographical area to De Lima, M.I.P., de Lima, J.L.M.P., 2009. Investigating the multifractality of point
which the station in question belongs. Furthermore, the use of a precipitation in the Madeira archipelago. Nonlinear Process. Geophys. 16, 311.
De Lima, M.I.P., Grasman, J., 1999. Multifractal analysis of 15-min and daily rainfall
specific value of factor f for each station has allowed prefixing an from a semi-arid region in Portugal. J. Hydrol. 220, 1–11.
equal potential error rate equal for all stations (1.5%). The results De Lima, M.I.P., 1998. Multifractals and the Temporal Structure of Rainfall. Ph.D.
show that the efficiency of the tests based on statistical decisions Thesis, Wageningen Agricultural Univ., 229 pp.
Di Baldassarre, G., Castellarin, A., Brath, A., 2006. Relationships between statistics of
increases with the magnitude of the error.
rainfall extremes and mean annual precipitation: an application for design-
The regional frequency analysis of extreme daily annual rainfall storm estimation in northern central Italy. Hydrol. Earth Syst. Sci. 10, 589–601.
data in the province of Malaga was not satisfactory when the 72 Easterling, D.R., 1989. Regionalization of thunderstorm rainfall in the contiguous
available stations were grouped into a region. A division into United States. Int. J. Climatol. 9, 567–579.
Eslamian, S., Hassanzadeh, H., Abedi-Koupai, J., Gheysari, M., 2011. Application of L-
potential homogeneous regions had to be done, and as a novelty, moments for regional frequency analysis of monthly drought indices. J. Hydrol.
the multifractal characterization of rainfall data was used as the Eng. http://dx.doi.org/10.1061/(ASCE)HE.19435584.0000396.
118 A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119

Estévez, J., Gavilán, P., García-Marín, A.P., 2011a. Data validation procedures in Kunkel, K.E., Easterling, D.R., Hubbard, K., Redmond, K., Andsager, K., Kruc, M.C.,
agricultural meteorology. A prerequisite for their use. Adv. Sci. Res. 6, 141–146. Spinar, M.L., 2005. Quality control of pre-1948 cooperative network observer
Estévez, J., Gavilán, P., Giráldez, J.V., 2011b. Guidelines on validation procedures for data. J. Atmos. Ocean. Technol. 22, 1691–1705.
meteorological data from automatic weather stations. J. Hydrol. 402, 144–154. Kyselý, J., Picek, J., Huth, R., 2007. Formation of homogeneous regions for regional
Farquharson, F.A.K., Meigh, J.R., Sutcliffe, J.V., 1992. Regional flood frequency frequency analysis of extreme precipitation events in the Czech Republic. Stud.
analysis in arid and semi-arid areas. J. Hydrol. 138, 487–501. Geophys. Geod. 51, 327–344.
Feng, S., Hu, Q., Qian, Q., 2004. Quality control of daily meteorological data in China, Labat, D., Mangin, A., Ababou, R., 2002. Rainfall-runoffs relations for karstic springs:
1951–2000: a new dataset. Int. J. Climatol. 24, 853–870. multifractal analyses. J. Hydrol. 256, 176–195.
Ferrer, F.J., Mateos, C., 1999. Analisis de maximas lluvias diarias. Un nuevo método Ladoy, P., Lovejoy, S., Schertzer, D., 1991. Extreme variability of climatological data:
regional de estimacion de parametros de la función de distribución SQRT-ET scaling and intermittency. In: Schertzer, D., Lovejoy, S. (Eds.), Non-linear
ḿax. Ingen. Civil 115, 109–118. Variability in Geophysics. Kluwer Acad., Norwell, Mass., pp. 241–250.
Fiebrich, C.A., Crawford, K.C., 2001. The impact of unique meteorological Ladoy, P., Schmitt, F., Schertzer, D., Lovejoy, S., 1993. The multifractal temporal
phenomena detected by the Oklahoma Mesonet and ARS Micronet on variability of Nimes rainfall data. C.R. Acad. Sci. Ser. II 317 (6), 775–782.
automated quality control. Bull. Am. Meteorol. Soc. 82, 2173–2187. Lecce, S.A., 2000. Spatial variations in timing of annual floods in the southeastern
Fowler, H.J., Kilsby, C.G., 2003. A regional frequency analysis of United Kingdom United States. J. Hydrol. 235, 151–169.
extreme rainfall from 1961 to 2000. Int. J. Climatol. 23, 1313–1334. Lee, S.H., Maeng, S.J., 2003. Frequency analysis of extreme rainfall using L-moment.
Fraedrich, K., Larnder, C., 1993. Scaling regimes of composite rainfall time series. Irrigat. Drain. 52, 219–230.
Tellus Ser. A – Dynam. Meteorol. Oceanogr. 45A, 289–298 (Meteorologische Lee, S.H., Maeng, S.J., 2005. Estimation of drought rainfall using L-moment. Irrigat.
Institut, Universität Hambur, D-2000 Hamburg 13, Germany) (la numeración la Drain. 54, 279–2940.
misma). Lettenmaier, D.P., Wallis, J.R., Wood, E.F., 1987. Effect of regional heterogeneity on
García Marín, A., Ayuso Muñoz, J.L., Taguas Ruiz, E.V., Estévez, J., 2011. Regional flood frequency estimation. Water Resour. Res. 23, 313–323.
analysis of the annual maximum daily rainfall in the province of Malaga Lettenmaier, D.P., Potter, K.W., 1985. Testing flood frequency estimation methods
(southern Spain) using the principal component analysis. Water Environ. J. 25 using a regional flood generating model. Water Resour. Res. 21 (12), 190–1914.
(4), 522–531. Lovejoy, S., Schertzer, D., 1995. Multifractals and rain. In: Kundzewicz, Z.W. (Ed.),
García-Marín, A.P., Jiménez-Hornero, F.J., Ayuso-Muñoz, J.L., 2008a. Applying New Uncertainty Concepts in Hydrology and Water Resources. Cambridge
multifractality and the self-organized criticality theory to describe the University Press – UNESCO International Hydrology Series, New York, pp. 61–
temporal rainfall regimes in Andalucía (Southern Spain). Hydrol. Process. 22, 103.
295–308. Mandelbrot, B., 1972. Possible refinement of the lognormal hypothesis concerning
García-Marín, A.P., Jiménez-Hornero, F.J., Ayuso-Muñoz, J.L., 2008b. Multifractal the distribution of energy dissipation in intermittent turbulence. In: Statistical
analysis as a tool for validating a rainfall model. Hydrol. Process. 22, 2672– Models and Turbulence. Lect. Notes Phys., vol. 12, pp. 333–351.
2688. Meek, D.W., Hatfield, J.L., 1994. Data quality checking for single station
García-Marín, A.P., Ayuso-Muñoz, J.L., Jiménez-Hornero, F.J., Estévez, J., 2012. meteorological databases. Agric. For. Meteorol. 69, 85–109.
Selecting the best IDF model by using the multifractal approach. Hydrol. Meshgi, A., Khalili, D., 2009. Comprehensive evaluation of regional flood frequency
Process. http://dx.doi.org/10.1002/hyp.9272. analysis by L- and LH-moments. I. A re-visit to regional homogeneity. Stoch.
García-Marín, A.P., Estévez, J., Jiménez-Hornero, F.J., Ayuso-Muñoz, J.L., 2013. Environ. Res. Risk Assess. 23 (1), 119–135.
Multifractal analysis of validated wind speed time series. Chaos 23, 013133. Modarres, R., 2008. Regional frequency distribution type of low flow in North of Iran
http://dx.doi.org/10.1063/1.4793781. by L-moments. Water Resour. Manage. 22, 823–841.
Greenwood, J., Landwehr, J., Matalas, N., Wallis, J., 1979. Probability weighted Modarres, R., 2010. Regional dry spells frequency analysis by L-moment and
moments: definition and relation to parameters of several distributions multivariate analysis. Water Resour. Manage. 24, 2365–2380.
expressed in inverse form. Water Resour. Res. 15 (6), 1049–1054. Moreno, F., Roldán, J., 1999. Regionalization of daily precipitation stochastic model
Greis, N.P., Wood, E.F., 1981. Regional flood frequency estimation and network parameters, application to the Guadalquivir valley in Southern Spain. Phys.
design. Water Resour. Res. 17 (4). http://dx.doi.org/10.1029/ Chem. Earth 24, 65–71.
WR017i004p01167. Ngongondo, C.S., Xu, Ch.Y., Tallaksen, L.M., Alemaw, B., Chirwa, T., 2011. Regional
Gupta, V.K., Waymire, E.C., 1993. A statistical analysis of mesoscale rainfall as a frequency analysis of rainfall extremes in Southern Malawi using the index
random cascade. J. Appl. Meteorol. 32, 251–267. rainfall and L-moments approaches. Stoch. Environ. Res. Risk Assess. 25, 939–955.
Guttman, N.B., 1993. The use of L-moments in the determination of regional Norbiato, D., Borga, M., Sangati, M., Zanon, F., 2007. Regional frequency analysis of
precipitation climates. J. Clim. 6, 2309–2325. extreme precipitation in the eastern Italian Alps and the August 29, 2003 flash
Hall, M.J., Minns, A.W., 1999. The classification of hydrological homogeneous flood. J. Hydrol. 345, 149–166.
regions. J. Hydrol. Sci. 44 (5), 693–704. Núñez, J.H., Verbist, K., Wallis, J.R., Schaefer, M.G., Morales, L., Cornelis, W.M., 2011.
Hosking, J.R.M., Wallis, J.R., 1993. Some statistics useful in regional frequency Regional frequency analysis for mapping drought events in north-central Chile.
analysis. Water Resour. Res. 29 (1), 271–281. J. Hydrol. 405, 352–366.
Hosking, J.R.M., Wallis, J.R., 1995. Correction to ‘‘some statistics useful in regional Olsson, J., Burlando, P., 2002. Reproduction of temporal scaling by rectangular
frequency analysis’’. Water Resour. Res. 31 (1), 251. pulses rainfall model. Hydrol. Process. 16, 611–630.
Hosking, J.R.M., Wallis, J.R., 1997. Regional Frequency Analysis—An Approach Based Ouarda, T.B.M.J., Girard, C., Cavadias, G.S., Bobeé, B., 2001. Regional flood frequency
on L-moments. Cambridge University Press, Cambridge. estimation with canonical correlation analysis. J. Hydrol. 254 (1–4), 157–173.
Hosking, J.R.M., 1990. L-moments: analysis and estimation of distributions using Ouarda, T.B.M.J., Bâ, K.M., Diaz-Delgado, C., Cârsteanu, A., Chokmani, K., Gingras, H.,
linear combinations of order statistics. J. R. Stat. Soc. B 52, 105–124. Quentin, E., Trujillo, E., Bobée, B., 2008. Intercomparison of regional flood
Hosking, J.R.M., 1992. Moments or L-moments? An example comparing two frequency estimation methods at ungauged sites for a Mexican case study. J.
measures of distributional shape. Am. Stat. 46 (3), 186–189. Hydrol. 348, 40–58.
Hosking, J.R.M., Wallis, J.R., 1988. The effect of intersite dependence on regional Over, T.M., Gupta, V.K., 1994. Statistical analysis of mesoscale rainfall: dependence
flood frequency analysis. Water Resour. Res. 24, 588–600. of a random cascade generator on large scaling forcing. J. Appl. Meteorol. 33,
Hosking, J.R.M., Wallis, J.R., Wood, E.F., 1985. An appraisal of the regional flood 1526–1543.
frequency procedure in the UK Flood Studies Report. Hydrol. Sci. J. 30, 85–109. Pandey, G., Lovejoy, S., Schertzer, D., 1998. Multifractal analysis of daily river flows
Hubbard, K.G., Goddard, S., Sorensen, W.D., Wells, N., Osugi, T.T., 2005. Performance including extremes for basins of five to two million square kilometres, one day
of quality assurance procedures for an applied climate information system. J. to 75 years. J. Hydrol. 208, 62–81.
Atmos. Ocean. Technol. 22, 105–112. Parida, B.P., Kachroo, R.K., Shrestha, D.B., 1998. Regional flood analysis of Mahi-
Hussain, Z., Pasha, G.R., 2009. Regional flood frequency analysis of the seven sites of Sabarmati Basin (subzone 3-1) using index flood procedure with L-moments.
Punjab, Pakistan, using L-moments. Water Resour. Manage. 23, 1917–1933. Water Resour. Manage. 12, 1–12.
Jiménez-Hornero, F.J., Pavón-Domínguez, P., Gutiérrez de Ravé, E., Ariza-Villaverde, Parisi, G., Frisch, U., 1985. Fully developed turbulence and intermittency. In: Ghil,
A.B., 2011. Joint multifractal description of the relationship between wind M., Benzi, R., Parisi, G. (Eds.), Proceedings International School of Physics Enrico
patterns and land surface temperature. Atmos. Res. 99, 366–376. Fermi, Course LXXXVIII, Italian Physical Society. Turbulence and Predicability in
Jingyi, Z., Hall, M.J., 2004. Regional flood frequency analysis for Gan-Ming river Geophysical Fluid Dynamics and Climate Dynamics. North-Holland,
basin in China. J. Hydrol. 296, 98–117. Amsterdam.
Kantelhardt, J.W., Koscielny-Bunde, E., Rybski, D., Braun, P., Bunde, A., Havlin, S., Periago, M.C., Lana, X., Serra, C., Mills, G.F., 1991. Precipitation regionalization: an
2006. Long-term persistence and multifractality of precipitation and river application using a meteorological network in Catalonia (NE Spain). Int. J.
runoff records. J. Geophys. Res. – Atmos. 111 (D1). Art. No. D01106. Climatol. 11, 529–543.
Kiely, G., Ivanova, K., 1999. Multifractal analysis of hourly precipitation. Phys. Chem. Rao, A.R., Hamed, K.H., 2000. Flood Frequency Analysis. CRC Press, Boca Raton.
Earth B – Hydrol. Oceans Atmos. 24, 781–786. Rao, A.R., Srinivas, V.V., 2008. Regionalization of Watersheds – An Approach Based
Kjeldsen, T.R., Smithers, J.C., Schulze, R.E., 2002. Regional flood frequency analysis in on Cluster Analysis. Springer Publishers, 241 p.
the KwaZulu-Natal province, South Africa, using the index-flood method. J. Reek, T., Doty, S.R., Owen, T.W., 1992. A deterministic approach to the validation of
Hydrol. 255, 194–211. historical daily temperature and precipitation data from the Cooperative
Kolesnikov, V.N., Monin, A.S., 1965. Spectra of meteorological field fluctuations. Network. Bull. Am. Meteorol. Soc. 73, 753–762.
Izvestiya Atmos. Ocean. Phys. 1, 653–669. Saenz De Ormijana, F., Hidalgo, F.J., Santa, A., 1991. Estimacion de precipitaciones
Kumar, R., Chatterjee, C., 2005. Regional flood frequency analysis using L-moments maximas mediante el metodo regional del índice de avenida. Rev. Obras Públ.
for North Brahmaputra Region of India. J. Hydrol. Eng. 4 (3), 240–244. 138, 9–22.
A.P. García-Marín et al. / Journal of Hydrology 529 (2015) 106–119 119

Santos, J.F., Portela, M.M., Pulido, I., 2011. Regional frequency analysis of droughts in Tessier, Y., Lovejoy, S., Hubert, P., Schertzer, D., Pecknold, S., 1996. Multifractal
Portugal. Water Resour. Manage. 25, 3537–3558. analysis and modelling of rainfall and river flows and scaling, causal transfer
Satyanarayana, P., Srinivas, V.V., 2011. Regionalization of precipitation in data functions. J. Geophys. Res. – Atmos. 101, 26427–26440.
sparse areas using large scale atmospheric variables – a fuzzy clustering Tessier, Y., Lovejoy, S., Schertzer, D., 1993. Universal multifractals in rain and
approach. J. Hydrol. 405, 462–473. clouds: theory and observations. J. Appl. Meteorol. 32, 223–250.
Schaefer, M.G., 1990. Regional analyses of precipitation annual maxima in Tsakiris, G., Nalbantis, I., Cavadias, G., 2011. Regionalization of low flows based on
Washington state. Water Resour. Res. 26 (1), 119–131. Canonical Correlation Analysis. Adv. Water Resour. 34, 865–872.
Schertzer, D., Lovejoy, S., Lavallée, D., 1993. Generic multifractal phase transitions Veneziano, D., Furcolo, P., 2002. Multifractality of rainfall and scaling of intensity-
and self-organized criticality. In: Perdang, J.M., Lejeune, A. (Eds.), Cellular duration-frequency curves. Water Resour. Res. 38 (12). http://dx.doi.org/
Automata: Prospects in Astrophysical Applications. World Scientific, pp. 216– 10.1029/2001WR000372.
227. Veneziano, D., Langousis, A., Furcolo, P., 2006. Multifractality and rainfall extremes:
Schertzer, D., Lovejoy, S., 1987. Physical modelling and analysis of rain and clouds a review. Water Resour. Res. 4.
by anisotropic scaling multiplicative processes. J. Geophys. Res. – Atmos. 92, Wallis, J.R., Schaefer, M.G., Barker, B.L., Taylor, G.H., 2007. Regional precipitation-
9693–9714. frequency analysis and spatial mapping for 24-hour and 2-hour durations for
Schertzer, D., Lovejoy, S., 1988. Multifractal simulations and analysis of rain and Washington State. J. Hydrol. Earth Syst. Sci. 11 (1), 415–442.
clouds by anisotropic scaling multiplicative processes. Atmos. Res. 21, 337–361. Wallis, Wood, 1985. Grouping basins for regional flood frequency analysis. Hydrol.
Schertzer, D., Lovejoy, S., 1994. Multifractal generation of self-organized criticality. Sci. J. 30, 151–159.
In: Novak, M.M. (Ed.), Fractals in the Natural and Applied Sciences. Elsevier, World Meteorological Organization, 1993. Guide on the Global Data-Processing
North-Holland, pp. 325–339. System. WMO-No. 305, Geneva, Switzerland.
Shafer, M.A., Fiebrich, C.A., Arndt, D.S., Fredrickson, S.E., Hughes, T.W., 2000. Quality Yun, P., Chen, C., 1998. Incorporating uncertainty analysis into a regional IDF
assurance procedures in the Oklahoma Mesonet. J. Atmos. Ocean. Technol. 17, formula. Hydrol. Process. 12, 713–726.
474–494. Yürekli, K., Modarres, R., 2007. Regionalization of máximum daily rainfall data over
Sivakumar, B., 2001. Is a chaotic multi-fractal approach for rainfall possible? Hydrol. Tokat province, Turkey. Int. J. Nat. Eng. Sci. 1 (2), 1–7.
Process. 15, 943–955 (with a minor correction in Hydrol. Process. 15, 2381– Zahumensky, I., 2004. Guidelines on Quality Control Procedures for Data from
2382). Automatic Weather Stations. WMO-No. 955, Geneva, Switzerland.
Srinivas, V.V., Tripathi, S., Rao, A.R., Govindaraju, R.S., 2008. Regional flood Zhang, J., Hall, M.J., 2004. Regional flood frequency analysis for the Gan-Ming River
frequency analysis by combining self-organizing feature maps and fuzzy basin in China. J. Hydrol. 296, 98–117.
clustering. J. Hydrol. 348, 148–166. Zrinji, Z., Burn, D.H., 1994. Flood frequency analysis for ungauged sites using a
Svensson, C., Olsson, J., Berndtsson, R., 1996. Multifractal properties of daily rainfall region of influence approach. J. Hydrol. 153, 1–21.
in two different climates. Water Resour. Res. 32, 2463–2472.

You might also like