Regional Hydrological Drought Monitoring Using Principal

Components Analysis
Rezgar Arabzadeh, S.M.ASCE 1; Mohammad Mehdi Kholoosi 2; and Javad Bazrafshan 3

Abstract: Drought quantification is a decision-making approach for water resources planners and managers. There are different statistical
methods for regional quantification of droughts, depending on the event being regionally discrete or continuous. Calculating a drought index
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

in the measured grid points of a given region is a prerequisite for the regionalization process of drought by these methods. The aim of
this study is the regional analysis of the streamflow drought using the multivariate technique of principal components analysis (PCA).
To this end, the streamflow drought index (SDI) was calculated in the seven hydrometric stations of the Sefid-Rud basin in Iran for the
period of 1984–2013 in seasonal (winter, spring, summer, and autumn), semiannual (October–March and May–September), and annual
(October–September) time scales. Regional monitoring of the SDI was carried out using the PCA technique summarizing the SDI series
of all stations into a new series, the so-called the multivariate streamflow drought index (MSDI). The MSDI series at each time scale were
analyzed from several statistical aspects. Results showed that there are relatively high correlations between the SDI series of the stations for
given time scales. The first principal component (PC1 ) explains 58–85% of the regional variations of the SDI series at the mentioned time
scales. The MSDI series at multiple time scales follow all stations’ SDI fluctuations and appropriately monitor droughts that occurred in the
region, especially in long dry periods. The dry and wet severity classes derived from the MSDI series greatly corresponded to those of the SDI
time series on different stations. The highest percentage of correspondence between MSDI and SDIs was calculated for the semiannual time
scale and the lowest for the spring season. DOI: 10.1061/(ASCE)IR.1943-4774.0000925. © 2015 American Society of Civil Engineers.
Author keywords: Hydrological drought; Principal components analysis (PCA); Streamflow; Regional analysis; Iran.

Introduction Drought quantifying is a managerial approach for water re-

Iran is one of the areas in the world that has suffered from severe
droughts over the last century. Drought environmental and socio-
economic bad effects are the only inheritance for Iran (Shahabfar
et al. 2011). Also, some studies that have been done recently show
both duration and frequency ascension in the drought index in this
region (Dai 2011).
Drought is a natural hazard that results from a precipitation de-
ficiency from expected rainfall, which is inadequate to supply the
demands of human activities and environment. This precipitation
deficit may arise quickly over a period of time, or it may take
months, even years before the deficiency begins to result in reduced
streamflows and reservoir levels. Drought causes chronic inequality
and transitory balances in basin hydrology (UN Secretariat General
1994). In another point of view, Gumbel (1963) has provided an-
other definition of drought, which states that drought is the smallest
streamflow in yearly annual time series. Some researchers say
drought is an abnormal phenomenon with a complex process that
is related to a specific climate region and local energy and water
balance status (Heim 2002).
1
M.Sc. Student, Dept. of Irrigation and Reclamation Engineering, Univ.
of Tehran, 31587-77871 Karaj, Iran (corresponding author). E-mail:
rezgararabzadeh@ut.ac.ir
2
M.Sc. Student, Dept. of Irrigation and Reclamation Engineering, Univ.
of Tehran, 31587-77871 Karaj, Iran. E-mail: mm.kholoosi@ut.ac.ir
3
Assistant Professor, Dept. of Irrigation and Reclamation Engineering,
Univ. of Tehran, 31587-77871 Karaj, Iran. E-mail: jbazr@ut.ac.ir
Note. This manuscript was submitted on September 3, 2014; approved
on April 27, 2015; published online on June 23, 2015. Discussion period
open until November 23, 2015; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Irrigation and Drai-
nage Engineering, © ASCE, ISSN 0733-9437/04015029(20)/$25.00.
source planners and managers. Acquisition of real-time drought in-
formation containing the initial time of drought and spatiotemporal
diffusion are the most decisive parameters for reducing drought
damage and assuring the success of preparedness and mitigation
(Özger et al. 2012). Streamflow decrease is the major effect of
drought on the hydroenvironmental system (Allen et al. 2011).
The effects drought have a large area of influence in the hydrology
of basin, including groundwater stability and water quality level
except surface water trend and precipitation conditions (Webster
et al. 1996). Monitoring and analysis of drought due to the lack
of a sufficient definition of this complex natural phenomenon
has been a difficult job up to now. The primary difficulty in drought
recognition is that dry periods must be compared among a variety
of scales of time and space. Using drought indices is one of the
solutions for overcoming this problem. This issue led the develop-
ment of drought indices, including simple methods for measuring
rainfall deficits to complicated models (Heim 2002). Usually a
drought index is essential for assessing the effect of drought and
defining different parameters, including intensity, duration, se-
verity, and spatial extent. A drought variable must be able to quan-
tify drought in a variety of scales of time. Most frequently, the time
scale is yearly and followed by a month. An annual time scale is
used to summarize regional drought information on the scale of
the basin, and large water management and a monthly time scale
are commonly used for monitoring the effects of drought on agri-
culture, water supply, and groundwater abstractions (Pnau and
Sharma 2002).
Although drought indices are mostly related to large-scale
climate indices and most of the research in this context has been
focused on the national or regional scales due to the nonhomoge-
neity in climate variation, study on macro-scale and local drought
monitoring is also applicable (Mishra and Singh 2010). A number

© ASCE 04015029-1 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 Table 1. Geographical and Statistical Characteristics of Selected Stations in the Sefid-Rud Basin
Number Stations UTM-X (m) UTM-Y (m) Elevation (m) Average (MCM) SD CV Skewness Kurtosis
1 Gezil-Ozan-Nisare 681,916 3971,981 1,725 40.71 65.88 1.62 3.10 11.71
2 Telwar-Mehr-Abad 744,151 3924,931 1,719 18.88 26.36 1.40 26.36 10.25
3 Heshta-Jift 759,597 4006,472 1,500 9.99 15.81 1.58 15.81 10.48
4 Telwar-Dehgolan 719,387 3906,714 1,825 1.96 3.74 1.91 3.74 17.17
5 Selan 676,209 3884,815 1,331 3.83 11.81 3.08 11.81 75.73
6 Remisht 679,585 3876,454 1,400 3.30 4.71 1.42 4.71 4.71
7 Chil-Gezi 677,241 3926,091 1,575 3.91 6.44 1.65 6.44 6.44

of different indices have been developed for quantifying drought; Supply Index (SWSI) monitors abnormalities in surface water
each one has its own strengths and weaknesses. In the following supply sources such as streamflow, reservoirs, snow pack stored
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

paragraph, some of them are briefly discussed.
The standardized precipitation index (SPI) was introduced by
McKee et al. (1993). The SPI is applicable for different regions
and is based on long-term rainfall data records. Moreover than pre-
cipitation Palmer (1965) introduced another drought index known
as the Palmer Drought Severity Index (PDSI), which uses temper-
ature for assessing drought on regional scales. The Surface Water
in mountainous regions, and precipitation (Shafer and Dezman
1982). Like the SWSI, the Reclamation Drought Index (RDI) as-
sesses drought using streamflow, temperature, snow pack, and pre-
cipitation to calculate drought on the river basin scale. Sharma and
Panu (2010) developed the Standardized Hydrological Index (SHI).
The concept of SHI is analogous to SPI (Mckee et al. 1993), except
that SPI represents a standardized entity and thus implicitly inherits

Fig. 1. Locations of the hydrometric stations selected in the Sefid-Rud basin in the northwest of Iran; for the names of stations corresponding to the
numbers written in the figure refer to Table 1 (data from Iranian Nation Organization of Surveying 2013)

© ASCE 04015029-2 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

© ASCE
04015029-3

J. Irrig. Drain Eng.

Fig. 2. Stations streamflow time series

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 2. (Continued.)

of the nonnormal characteristics of the flow series (Sharma and
Panu 2012).
Most of the indices described above are computationally diffi-
cult and data demanding in contrast the Streamflow Drought Index
(SDI) introduced by Nalbantis and Tsakiris (2009), which is very
simple in calculation and powerful in assessing at-site drought. The
SDI is only based on the streamflow values.
In drought studies, there is no deterministic time scale for
assessing drought severity, and the majority of researchers based
on their own assumptions, or other research work with time
frames of 3, 6, 9, 12, and 24 months proposed by different groups
of researchers, (Mishra and Desai 2005; Cancelliere et al. 2007)
or drought classification based on a 12-month time scale in order
to transition probability analysis also has been recommended
(Paulo and Pereira 2008). The purposes of this study are to ag-
gregate spatial and temporal patterns of drought for assessing
regional drought quantity using principal component analysis
(PCA), which reduces the dimensionality of the SDI data set.
In research related to drought, PCA has been used to assess the
spatiotemporal analysis of drought in different time and space
scales. (Bomaccorso et al. 2003; Bordi et al. 2004; Sonmez et al.
2005; Raziei et al. 2011; Bazrafshan et al. 2014). Based on Wilks
(2011), PCA is able to summarize several variables into fewer
important components.
The following sections discuss about how to implement and cal-
culate SDI and reduce it to its own major components. The paper

© ASCE 04015029-4 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 concludes with a discussion of the results of the study and the
authors’ conclusions.
Material and Method
Data and Study Area
The Sefid-Rud basin is one of the major basins in the Iran plateau
and is 50,280 km2 vast. In this basin, water is supplied to rivers,
such as Gezil-Ozan, Sefid-Rud, and Shahrud, largely by melting
snow and winter and spring precipitation. This area, with annual
precipitation of 200 to 400 mm for planes and 700–3,000 mm
in mountainous areas, has warm summers and freezing winters
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
and the temperature in mountain areas is as low as −30°C and
in the planes rises to 40°C. The majority of water flowing into
the Sefid-Rud basin is from basins situated in the Kurdistan
territory. Table 1 shows the geographic and hydrologic features
of stations used in the current study. The Gezil-Ozan-Nisare, Selan,
Remisht, and Chil-Gezi stations are situated in the upstream part of
Fig. 3. Locations of dams in study area and demand sites in the upstream of hydrometric stations (data from Iranian Nation Organization of
Surveying 2013)
© ASCE 04015029-5
J. Irrig. Drain Eng.
the study area, and Telwar-Mehr-Abad, Heshta-Jift, and Telwar-
Dehgolan are in the eastern part. The monthly streamflow data cov-
ers 1984–2013, equal to 30 hydrological years. All the stations’
natural streamflows were calculated based on drought indices der-
ivation. The naturalized streamflow was driven by adding the allo-
cations in stations upstream. In the case of Telwar-Dehgolan and
Telwar-Mehr-Abad, this issue was much more noticeable and had a
remarkable share of basin surface water (about 50 MCM per year).
In other stations, due to the geography of study area and poorly
developed of agricultural fields and other demanding sites, alloca-
tions in comparison to the current condition of the streamflow do
not represent a noticeable share, although the naturalized stream-
flow was calculated for the study and the analysis of the basin’s
drought. The skewness coefficients are positive for all stations that
show nonnormal distribution of the streamflow time series between
seasons and months. All kurtosis coefficients are also greater than
4.71. Fig. 1 shows the locations of the stations used in this study,
and Fig. 2 shows the naturalized streamflows used in this study.
During the last decade (2005–2015), two major dams have been
constructed in the basin, the Sang-Siyah Dam and Siyazagh Dam.
J. Irrig. Drain Eng.
 All of them are considered in the basin networks in 2011 onward
and have refilled during the last 2 years with no remarkable effects
on the entire basins’ discharge. Fig. 3 shows the geographical sit-
uation of the dams and hydrometric stations.
The Sang-Siyah dam has a volume of 23 MCM and regulates
35 MCM each year whereas the Telwar-Mehr-Abad hydrometric
station has recorded about 230 MCM per each year. In spite of an
overlap in the period between the study period and the refilling date
of the Sang-Siyah dam (2 years, 2011–2013), the data on the Sang-
Siyah dam showed no significant effect on the records of the Telwar
Mehr-Abad station. In addition to this issue, the regulated water
volumes are considered for the Telwar river streamflow and were
taken into account in the calculation of the SDIs. This issue is also
addressed for the Siyazagh dam, where the reservoir volume is
about 200 MCM and 500 MCM volume regulation annually.
It was found that most of the streamflow time series do not
follow a normal distribution, so the ability of log normal, gamma,
log Pearson type III, and Weibull distributions are investigated for
fitting on the time series. For all stations, the distributions men-
tioned above were fitted on their corresponding time series and
their goodness of fit was evaluated using two well-known tests:
Kolmogorov Smirnov and chi square. Then, based on a 5% level

Table 2. The p-Values Calculated Corresponding to the Statistics Values of Kolmogorov-Smirnov and Chi-Square Goodness-of-Fit Tests
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Kolmogorov Smirnov Chi-square

Station Time series N LN G W LP3 N LN G W LP3
Gezil-Ozan-Nisare Fall 0.69 0.36 0.72 0.68 0.89 0.43 0.98 0.02 0.77 0.30
Winter 0.41 0.91 0.86 0.81 0.92 0.24 0.50 0.63 0.38 0.51
Spring 0.69 0.36 0.72 0.68 0.89 0.43 0.98 0.02 0.77 0.30
Summer 0.32 0.36 0.20 0.73 0.62 0.32 0.27 0.36 0.75 0.67
First 6 months 0.50 0.88 0.92 0.67 0.91 0.70 0.59 0.46 0.41 0.49
Second 6 months 0.28 0.91 0.70 0.85 0.94 0.11 0.56 0.75 0.99 0.54
Annual 0.69 0.36 0.72 0.68 0.89 0.43 0.98 0.02 0.77 0.30
Chil-Gezi Fall 0.04 0.95 0.12 0.78 1.00 0.12 0.89 0.68 0.90 0.89
Winter 0.20 0.95 0.85 0.82 0.97 0.31 0.76 0.97 0.98 0.75
Spring 0.69 0.19 0.71 0.35 0.61 0.90 0.71 0.86 0.92 0.96
Summer 0.02 0.54 0.66 0.70 N/F 0.02 0.79 0.60 0.76 N/F
First 6 months 0.12 0.91 0.72 0.66 0.96 0.43 0.97 0.94 0.76 0.96
Second 6 months 0.66 0.22 0.76 0.41 0.62 0.89 0.71 0.88 0.91 0.99
Annual 0.67 0.58 0.90 0.83 0.87 0.52 0.87 0.89 0.98 0.90
Remisht Fall 0.08 0.67 0.60 0.71 0.60 0.02 0.93 0.29 0.30 0.32
Winter 0.30 0.98 0.76 1.00 0.98 0.32 0.95 0.87 0.93 0.95
Spring 0.44 0.23 0.84 0.45 0.70 0.03 0.97 0.97 0.96 0.86
Summer 0.42 0.02 0.51 0.08 0.51 0.04 0.86 0.84 0.05 0.75
First 6 months 0.11 0.88 0.80 0.98 0.87 0.40 0.98 0.73 0.97 0.98
Second 6 months 0.44 0.12 0.64 0.28 0.50 0.03 0.76 0.87 0.76 0.84
Annual 0.69 0.96 1.00 0.99 1.00 0.68 0.95 1.00 0.97 0.99
Telwar-Dehgolan Fall 0.01 0.00 0.00 0.00 N/F 0.07 0.00 0.06 0.05 N/F
Winter 0.31 0.66 0.94 0.92 1.00 0.31 0.83 0.68 0.84 0.98
Spring 0.53 0.62 0.99 0.78 0.93 0.78 0.66 0.98 0.95 N/A
Summer 0.00 0.00 0.00 0.00 N/F 0.00 0.00 0.04 0.01 N/F
First 6 months 0.14 0.56 0.97 0.87 0.97 0.21 0.93 0.76 0.62 0.89
Second 6 months 0.41 0.63 0.90 0.69 0.82 0.89 0.76 0.48 0.96 0.44
Annual 0.17 0.52 0.48 0.83 0.86 0.10 0.76 0.42 0.77 0.52
Selan Fall 0.00 0.41 0.00 0.46 0.50 0.00 0.29 0.00 0.03 0.14
Winter 0.00 0.58 0.00 0.56 0.87 0.00 0.48 0.00 0.44 0.97
Spring 0.00 0.56 0.00 0.69 0.62 0.05 0.98 0.00 0.97 0.98
Summer 0.01 0.90 0.20 0.95 N/F 0.01 0.94 0.35 0.98 N/F
First 6 months 0.00 0.57 0.00 0.56 0.89 0.00 0.19 0.11 0.35 N/A
Second 6 months 0.00 0.64 0.01 0.76 0.69 0.05 0.99 0.00 0.97 0.98
Annual 0.00 0.96 0.00 0.87 0.78 0.01 0.66 0.00 0.68 0.97
Telwar-Mehr-Abad Fall 0.15 0.13 0.34 0.28 0.42 0.72 0.03 0.10 0.05 0.20
Winter 0.80 0.74 0.90 0.97 0.98 0.57 0.92 0.96 0.99 0.97
Spring 0.17 0.47 0.88 0.75 0.63 0.61 0.30 0.94 0.98 0.10
Summer 0.02 0.44 0.49 0.83 N/F 0.02 0.11 0.20 0.79 N/F
First 6 months 0.87 0.26 0.56 0.48 0.53 0.77 0.09 0.66 0.67 0.63
Second 6 months 0.15 0.64 0.92 0.91 0.79 0.38 0.98 0.99 0.99 0.26
Annual 0.68 0.69 0.93 0.94 0.84 0.42 0.17 0.58 0.65 0.92
Heshta-Jift Fall 0.36 0.66 0.91 0.90 0.79 0.55 0.94 0.90 0.92 0.95
Winter 0.08 0.72 0.32 0.86 0.76 0.00 0.64 0.01 0.78 0.64
Spring 0.09 0.59 0.60 0.50 0.63 0.08 0.67 0.10 0.09 0.67
Summer 0.06 0.87 0.97 0.99 N/F 0.12 0.99 0.84 0.93 N/F
First 6 months 0.24 0.87 0.52 0.97 0.85 0.13 0.92 0.27 0.75 0.93
Second 6 months 0.12 0.58 0.62 0.50 0.62 0.18 0.67 0.25 0.89 0.67
Annual 0.09 0.60 0.64 0.47 0.64 0.11 0.90 0.70 0.82 0.92
Note: The bolded values are significant at the 5% level. Abbreviations N, LN, G, W, LP3 indicate normal, log normal, gamma, Weibull, and log Pearson type
III, respectively. N/A = not applicable, and N/F = not fitted.

© ASCE 04015029-6 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 4. Illustrative diagram of SDI calculation steps from a desired commutative probability distribution fitted on data set and a normal distribution
using principal of probability transformation

Fig. 6. MSDI percentile for fall season in the study area

Table 3. Classification of the SDI Values and Corresponding Probability

Fig. 5. Components of aggregated streamflow drought index; diagram Limits (Data from Nalbantis and Tsakiris 2009)
shows illustrative step-by-step procedure of MSDI derivation
Deception Criterion Probability limit (%)
No drought 0 ≤ SDI 50
Mild drought −1 ≤ SDI ≤ 0 34.1
the best distribution was selected to be fitted on the corresponding Moderate drought −1.5 ≤ SDI ≤ −1 9.2
time series. Results are presented in Table 2. In most of cases, log Severe drought −2 ≤ SDI ≤ −1.5 4.4
Extreme drought SDI ≤ −2 2.3
normal and gamma distributions due to time series skewness were
the best distributions, and Weibull, log Pearson type III, and normal
distributions were utilized occasionally.
October and so on), then the calculation of the ith hydrological
cumulative streamflow volume V i;j are used [Eqs. (1)–(3)]
Streamflow Drought Index
According to Nalbantis (2008), if there is a time series of stream- X
3k
flow Qi;j in which i denotes the hydrological year and j represents V i;j ¼ Qi;j ; k ¼ 1; 2; 3; 4 ð1Þ
a month within that hydrological year (i.e., j ¼ 1 corresponds to j¼3ðk−1Þþ1

© ASCE 04015029-7 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 X
6k
ðV i;k − V̄ k Þ
V i;j ¼ Qi;j ; k ¼ 1; 2 ð2Þ SDIi;k ¼ ; k ¼ 1; 2; 3; 4 ð4Þ
j¼6ðk−1Þþ1
Sk

in which V̄ k is the mean and Sk is the standard deviation of

X
12
the commutative streamflow volumes of the kth reference period,
V i;j ¼ Qi;j ð3Þ
j¼1
respectively. In this notation, the truncation level is set to V̄ k ,
although other values based on rational criteria could also be used.
where k in Eq. (1) shows kth reference periods in which k ¼ 1 According to principal of probability transformation, after calcula-
belongs to October–December, k ¼ 2 for January–March, k ¼ 3 tion of the commutative volumes for each period the value of SDI
for April–June, and k ¼ 4 for July–September. Also, in Eq. (2), can be quantified using the Eq. (4) as described in Fig. 4.
k ¼ 1 and k ¼ 2 are dedicated to the first and second 6-month peri-
ods in a hydrological year, respectively. For an annual time scale,
Multivariate Streamflow Drought Index
the discharge volume through the entire year is calculated using
Eq. (3) for the calculation of the yearly drought index. The multivariate streamflow drought index (MSDI) requires the
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

For each reference period k and ith hydrological year, the SDI is calculation of the time series at all stations for a given time scale.
calculated with Eq. (4) written as follows: In this study, a station’s SDI time series is aggregated using the

Fig. 7. Cross-correlation diagrams, representing correlation coefficient value among stations; each polygon shows correlation coefficients of a given
station with other stations

© ASCE 04015029-8 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 7. (Continued.)

Table 4. Results of Bartlett’s Sphericity Test for the SDI Series at Each technique of PCA to evaluate the regional drought in the study area
Time Scale in the Study Area for a particular time scale. For example, SDI was calculated for
Chi square Chi square each station at a scale of 6 months and by using the ability of multi-
MSDI series (calculated) (critical) p-value variate techniques, these drought indices were aggregated to one
Fall 217.237 32.671 <0.0001 component that represents the majority of all stations’ droughts.
Winter 84.334 32.671 <0.0001 Having a comprehensive view of drought and the vast regional in-
Spring 395.286 32.671 <0.0001 dex, which can be generalized to the whole study area for a specific
Summer 107.799 32.671 <0.0001 time scale, is much more reasonable and also applied. Also, various
First 6 months 204.129 32.671 <0.0001 drought indices for water resource managers and planners in mak-
Second 6 months 266.093 32.671 <0.0001 ing decisions about water resource systems in this situation is dif-
Annual 265.347 32.671 <0.0002
ficult, so having a unit and representative drought index for water

© ASCE 04015029-9 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

© ASCE
04015029-10

J. Irrig. Drain Eng.

Fig. 8. Scree plots of all time scales

J. Irrig. Drain Eng.

 resource planning is a crucial necessity. This aim is reachable using The elements of the eigenvector corresponding to the largest
a tool like PCA. The PCA, sometimes referred to as the orthogonal eigenvalue are in fact considered as the weighed coefficients of
empirical function (EOF), shows a linear combination of k varia- the first principal component (PC1 ). The index presented in this
bles that have strong relationships as shown as follows (Bazrafshan study is based on PC1 , which is able to explain the majority of
et al. 2014): variations in the hyper cloud.
Due to the characteristics of PC1 , unlike SDI, which has a mean
X
k
equal to 0 and a variance equal to 1, PC1 is not so (Bazrafshan et al.
PCi ¼ ETi X ¼ eij X k ; k ¼ 1; 2; 3; 4 ð5Þ
2014). Therefore, the PC1 time series will be standardized using
j¼1
the Eq. (8)
where PCi = ith principal component; ETi = ith eigenvector; X k =
kth original variable; and eij = kth element of the ith eigen com- PC1ym − PC1m PC1ym
ponent. These linear combinations have different features some of Z1ym ¼ ¼ ð8Þ
SD1m SD1m
which are described as follows (Sharma 1996):
• Mutually uncorrelated;
in which Z1ym = PC1 standardized value in the yth year and mth
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

• The components number are equal to the variables number; and

month or MSDI; PC1m = PC1 means in the mth month and SD1m =
• They are extracted in the way that the first principal component
PC1 standard deviation in the mth month.
equivalent to PC1 justifies a great amount of variation of hyper-
It can be proved that the value of PC1m is negligible and is close
cloud buildup by the variables.
to zero; hence, this term can be eliminated from Eq. (8) (Keyantash
Before PCA is implemented, Bartlett’s sphericity test (BST)
and Dracup 2004). Fig. 5 shows a step-by-step diagram that enables
(Snedecor and Cochran 1989) is checked to control for the suffi-
users to understand the overall stages of the aggregated drought
ciency of the hyper cloud’s correlation. The test statistic is calcu-
index using MSDI. This procedure clarifies the definitions and
lated using the correlation matrix derived from the SDI data sets as
separates MSDI’s derivations into two major steps: first, the SDI is
shown as follows:
calculated and then aggregation is performed.
P
2
ðN − kÞ lnðS2p Þ − ki¼1 ðni − 1Þ lnðS2i Þ The MSDI cumulative probability values were calculated and
X ¼ 1
P ð6Þ plotted on the diagram shown in Fig. 6. Then, using the classes
1 þ 3ðk−1Þ ½ ki¼1 ðni1−1Þ − N−k
1

presented in Table 3, thresholds were delineated and colored on the
figure to be much more representative and the MSDI values were
where k = number of stations with sample sizes ni (SDIs); S2i =
classified according to the same table.
SDI’s variance and N = summation of ni . This statistic follows ap-
A more explorative analysis was carried out for delineating the
proximately a X 2ðk−1Þ distribution.
stochastic relationship between drought periods in seasonal time
If the test shows a high value of correlation in the hyper cloud,
scale using MSDIs. In this step, the states of basins’ hydrology,
consequently the first component is able to justify a great part of
including classes from extremely dry to extremely wet, were ap-
the total variance in the original variables (Wilks 2011). For the
pointed using a transformation probability, constituting a desired
calculation of the PCA, first a correlation matrix of the SDI time
season state probability to the subsequent one, and was calculated
series is calculated. Then, the eigenvalues corresponding to each
for each pair of seasons (fall-winter, winter-spring, spring-summer,
component is calculated using the following equation:
and summer-fall) while considering the drought concept. The trans-
jC − λIj ¼ 0 ð7Þ formation probability matrix states that the probability of all prob-
able states in which the hydrologic system through a desired state
where C = correlation matrix of the SDI time series; λ = eigenvalue, will face a particular hydrologic condition in the subsequent states.
and I = identity matrix. As the eigenvalues were calculated, the In this study, the probability matrices of all pairs of time scales were
eigenvector will be formed to correspond to each eigenvalue. computed using Eq. (9)

Fig. 9. PC1 for each time scale

© ASCE 04015029-11 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. SDI’s biplots representing the contribution of the original SDI variants in PC1 and PC2

© ASCE 04015029-12 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 MSDIðu → vÞ Results and Discussion
Pu;v ¼ PD ð9Þ
v¼1 MSDIðu → vÞ
Correlation Analysis and Testing
where u = class of drought index in the first season, v = class of The SDI was calculated for all stations at different time scales.
drought index in the subsequent season, and D = number of classes Results were as four seasonal, two 6 month, and one annual time
that MSDI is categorized into them. series for all seven stations. For investigation of correlations
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 11. MSDI time series (solid line) and SDI time series (dark band) for each time scale

© ASCE 04015029-13 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 11. (Continued.)
between the stations’ SDI time series, the correlation matrices for
the 3-, 6-, and 12-month time scales were calculated and drawn
using the radar graphs. Fig. 7 shows the cross correlations of SDIs
among stations calculated for all time scales. As shown in Fig. 7,
correlations among variants are high and statistically significant
at the 5% level. The more the polygon became larger, a polygon
presents its correspondent station correlation coefficients with
other stations as being more statistically significant. The maxi-
mum correlation is between Telwar-Dehgolan and Selan equal
to 0.99 in spring and the minimum correlation of 0.29 was seen
between Chil-Gezi and Telwar-Mehr-Abad in the summer. In
the case of winter, the maximum correlation was 0.89 between
Telwar-Mehr-Abad and Gezil-Ozan-Nisare and the minimum
was between Telwar-Mehr-Abad and Chil-Gezi. For the fall sea-
son, the maximum and minimum correlations were 0.99 and 0.55
between Selan and Telwar-Dehgolan and Remisht and Heshta-Jift,
respectively. The results from the correlation matrices show a high
dependency of variation between the stations’ SDIs and revealed
© ASCE 04015029-14
J. Irrig. Drain Eng.
that all stations’ SDI time series have sufficient correlation. In
addition, the results of Bartlett’s sphericity test (Table 4) confirm
the sufficiency of the hyper cloud’s correlation at each time scale.
According to Table 4, chi square statistics are all significant at the
1% level.
Principal Component Analysis of the SDI Time Series
According to the stations’ correlation matrices, their eigenvalues
and corresponding eigenvectors were derived. Scree plots in Fig. 8
show the eigenvalues and their cumulative variability versus PCs’
number. As shown in Fig. 8, most of the hyper cloud correlation can
be explained just by the first PC and other components only cover
a little information of the data sets in which for the fall, winter,
spring, summer, first 6 months, and the second 6 months and for
annual periods can explain 72, 78, 80, 58, 75, 85, and 83% of the
variations, respectively, which shows the ability of the PC1 to
model regional streamflow drought.
J. Irrig. Drain Eng.
 In Fig. 9, the first components of each time scale have been
drawn schematically. As is shown, total variations that can be ex-
plained by PC1 increase from fall to spring and again decreases in
summer. This is due to a fairly even rainfall pattern in fall to
spring and one that is not evenly distributed in summer and for
6 months.
For analyzing the SDI’s variants’ behaviors in principal compo-
nents, biplots have been drawn in Fig. 10. A biplot shows the con-
tribution of each variant in forming the principal components. It is
common to draw biplots of PC1 versus PC2 . This type of picturing
checks the robustness of PC1 in comparison to variations explained
by other components in which the power of PC1 for modeling
the SDI time series for regional scales can be seen when there
is more of a tendency of correlation in biplot to PC1 axis. From
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
these biplots, both positive and large contributions (correlations)
of variants to PC1 are conspicuous. According to the diagrams
Fig. 12. Percentage of correspondence of the MSDI series with all stations’ SDI series for each time scale in the study area during the
period 1984–2013
© ASCE 04015029-15
J. Irrig. Drain Eng.
represented in Fig. 10, less contribution has been seen in PC1 for
3-month time scales as has been reported by other researchers
(e.g., Bazrafshan et al. 2014). In most of the 3-month time scales,
the Gezil-Ozan-Nisare has the minimum contribution to PC1 and
also a significant correlation to PC2 . In the time scales of 6 months
and annual, variants contributions for PC1 has been raised and cor-
relations are much more significant.
Regional Streamflow Drought Monitoring
For comparison, the MSDI and SDI were drawn on the same dia-
gram for all selected time scales. Results are shown in Fig. 11.
According to the figure, it is clearly seen that the MSDI follows
all stations’ drought index fluctuations and appropriately monitors
and gathers their indices into one index, especially during long dry
and wet periods. Also, the MSDI in the original SDI time series
J. Irrig. Drain Eng.
 Table 5. Classification of MSDI Series for the Period of 1984–2013 in the Study Area
First Second
Year Fall Winter Spring Summer 6 months 6 months Annual
1984 MiW MiW MiW MiW MiW MiW MiW
1985 MW MiD MW MW MiW MW MW
1986 MiW MiD MiW MiD MiD MiW MiD
1987 MW SW MiW MiW MW MiW MW
1988 MiW MiW MiW MiW MiW MiW MiW
1989 MiW MiW MiW MiD MiW MiW MiW
1990 MW SW MW MW SW MW SW
1991 MiW MiW MiW MiW MiW MiW MiW
1992 MiW MiW MiW MiD MiW MiD MiD
1993 MiD MiD MiD MiD MiD MiD MiD
1994 MiW MiD SW MW MiW SW MW
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.

1995 MiW MiW MiW MiW MiW MiW MiW

1996 SW EW MW SW EW MW SW
1997 EW MW MW SW SW MW SW
1998 MiW MiW MW MW MiW MW MiW
1999 MiW MiD MiD MiW MiD MiW MiD
2000 MiW MiW MiW MiW MiW MW MiW
2001 MiD MD SD MD MiD SD MD
2002 MD MD MiD MiD MD MiD MD
2003 MD MD SD SD MD SD SD
2004 SD SD MiD MiD MD MiD MiD
2005 MiD MiD MiD MiD MiD MiD MiD
2006 MiD MiW MiD MiD MiD MiD MiD
2007 MiD MiW MiD MiD MiW MiD MiD
2008 MiD MiD MiD MiD MiD MiD MiD
2009 MiD MiD MiW MiW MiD MiW MiD
2010 MiD MD SD ED MiD SD SD
2011 MiD SD MD MD MD MD MD
2012 MiD MiD MiD MiD MiD MiD MiD
2013 SD MD MD MD SD MD MD
Note: ED = extreme drought; EW = extremely wet; MD = moderate drought; MiD = mild drought; MiW = mildly wet; MW = moderately wet; SD = severe
drought; SW = severely wet.

Table 6. Frequencies of Drought/Non-Drought Classes Based on Different

MSDI Series in the Study Area
First Second
Status Class Fall Winter Spring Summer 6 months 6 months Annual
Drought ED 0 0 0 1 0 0 0
SD 2 2 3 1 1 3 2
MD 2 5 2 3 4 2 4
MiD 10 9 9 11 10 9 11
No EW 1 1 0 0 1 0 0
drought SW 1 2 1 2 2 1 3
MW 3 1 5 4 1 6 3
MiW 11 10 10 8 11 9 7

reduces rapid fluctuations, resulting in the reduction of the dry and

wet periods frequencies compared with the original SDI time series.
According to corresponding analysis of the SDI time series to
MSDI, an appropriate agreement is notable between the SDI time
series severity classes during the study period. Fig. 12 shows cor-
respondence percentages for all time scales. According to Fig. 12
and the results, both minimum and maximum correspondence per-
centages are in spring in Gezil-Ozan-Nisare and Telwar-Dehgolan
equal to 61.9 and 96.3%, respectively. The reason for this is that the
variance explained by the PC1 of the SDI time scales of 3 months
has the lowest value compared with other selected sets of time
scales. Altogether, the highest correspondence between the MSDI’s
Fig. 13. Pie chart of the percentages of the MSDI classes
dryness and wetness severity classes and each of the selected sets of

© ASCE 04015029-16 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

 the SDI time scales were obtained for the secondary 6-month
period and the lowest was summer.
The categorization of the MSDIs has been completed, and the
results are tabulated as presented in Tables 5 and 6. Based on the
results shown in Table 6, the frequencies in the drought classes are
17 for all time scales and a bit more than the frequencies for the
wet classes. Most frequencies have been calculated for the mild
classes, where mildly wet (MiW) and mild drought (MiD) are equal
to 7 and 11, respectively. In addition, the extreme classes, including
extremely (EW) and extremely dry (ED), were seen as very uncom-
mon and have not been seen for the annual time scale and just have
a frequency equal to 1 for ED in summer and EW in the first
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
6 months. The results show high percentages of occupation for the
mild classes in the hydrologic year and for percentages for the ex-
treme classes. The categorization of percentages by each class is
shown in Fig. 13.
In another statistical view, the probability transitions for both
MSDI and SDI were calculated. After computation of the transi-
tion probability matrices, the matrices are shown as diagrams in
Figs. 14 and 15. The results show a high value of transition prob-
ability for diagonal elements in the matrices, which belongs to
similar drought classes in the first and second transition seasons.
This issue is much more evident in case of spring-summer and
somewhat unclear in the summer-fall period in both simple and

Fig. 14. Probability transformation matrices of drought classes in two subsequent season in all stations

© ASCE 04015029-17 J. Irrig. Drain Eng.

J. Irrig. Drain Eng.

Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 14. (Continued.)
multivariate SDI time series. A notable case is that there is a
high density of probability in the lower-left corner in most of the
matrices except Telwar-Dehgolan. This shows that whenever the
hydrologic system traps in a severely dry season or extremely dry
period difficultly, it would be able to escape it or the system faces
another condition where the consequences of drought or its effects
are lower.
Conclusion
In this study, a new drought index is proposed based on EOFs to
assess the hydrological drought index (SDI), representing drought
dependent on streamflow. The features making MSDI a useful tool
for drought monitoring are
• MSDI is able to reduce SDI time series and gathers them into
one principal component, which is very useful in regional water
resources planning and management;
• Since MSDI is basically calculated on first principal component,
it contains a major part of all other components and has appro-
priate accuracy in regional drought monitoring;
© ASCE 04015029-18
J. Irrig. Drain Eng.
• Results obtained from MSDI are simple to illustrate and com-
parable to different time scales and spatial zones due to its abil-
ity to be standardized; and
• MSDI calculation is computationally simple and can be applied
to any arbitrary time scale and spatial zone.
The current study offers an MSDI calculation as a new drought
index for monitoring regional drought. First, a comprehensive
goodness of fit test was carried out for choosing the best distribu-
tion to model the probability function of the time series for all sta-
tions and each time scale. Then, the correspondent standard values
of each streamflow probability were extracted using normal distri-
bution, which is equal to the SDI time series. The SDIs of the sta-
tions were summarized using PCA into one PC1 and eventually
standardized to obtain the MSDI. According to results and dis-
cussed methods, the following conclusions have been made:
• The principal components analysis of the SDI time series
showed that the first PC1 could explain a large percentage of
variations in the original SDI time series in all selected stations;
• The dry and wet severity classes derived from the MSDI series
greatly corresponded to those of the SDI time series on different
stations;
J. Irrig. Drain Eng.
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 15. Probability transformation matrices of drought classes in two subsequent seasons
• The highest percentage of correspondence was calculated for
the 6-month time scales and the lowest was for the spring
season; and
• Altogether the season’s percentage of correspondence was lower
than the larger time scale’s percentage of correspondence.
References
Allen, P. M., Harmel, R. D., Dunbar, J. A., and Arnold, J. G. (2011).
“Upland contribution of sediment and runoff during extreme drought:
A study of the 1947–1956 drought in the Blackland Prairie, Texas.”
J. Hydrol., 407(1–4), 1–11.
Bazrafshan, J., Hejabi, S., and Rahimi, J. (2014). “Drought monitoring
using the multivariate standardized precipitation index (MSPI).” Water
Resour. Manage., 28(4), 1045–1060.
Bonaccorso, B., Bordi, I., Cancielliere, A., Rossi, G., and Sutera, A. (2003).
“Spatial variability of drought: An analysis of the SPI in Sicily.” Water
Resour. Manage., 17(4), 273–296.
Bordi, I., Fraedrich, K., Jiang, J. M., and Sutera, A. (2004). “Spatio-
temporal variability of dry and wet periods in eastern China.” Theor.
Appl. Climatol., 79(1–2), 81–91.
Cancelliere, A., Di Mauro, G., Bonaccorso, B., and Rossi, G. (2007).
“Drought forecasting using the standardized precipitation index.” Water
Resour. Manage., 21(5), 801–819.
© ASCE 04015029-19
J. Irrig. Drain Eng.
Dai, A. (2011). “Drought under global warming: A review.” Wiley Inter-
discip. Rev. Clim. Change, 2(1), 45–65.
Gumbel, E. J. (1963). “Statistical forecast of droughts.” Bull. Int. Assoc. Sci.
Hydrol., 8(1), 5–23.
Heim, R. R. (2002). “A review of twentieth-century drought indices used in
the United States.” Bull. Am. Meteorol. Soc., 83(8), 1149–1165.
Iranian Nation Organization of Surveying. (2013). “Technical service man-
agement datasets archives: Digital elevation model.” 〈http://tsm.ncc.org
.ir/〉 (May 1, 2014).
Keyantash, J., and Dracup, J. A. (2004). “An aggregate drought index:
Assessing drought severity based on fluctuations in the hydrologic cycle
and surface water storage.” Water Resour. Res., 40(9), W09304.
McKee, T. B., Doesen, N. J., and Kleist, J. (1993). “The relationship of
drought frequency and duration to time scales.” Proc., 8th Conf. on
Applied Climatology, Anaheim, CA, 179–184.
Mishra, A. K., and Desai, V. R. (2005). “Drought forecasting using stochas-
tic models.” Stochastic Environ. Res. Risk Assess., 19(5), 326–339.
Mishra, A. K., and Singh, V. P. (2010). “A review of drought concepts.”
J. Hydrol., 391(1–2), 202–216.
Nalbantis, I. (2008). “Evaluation of a hydrological drought index.” Eur.
Water, 23(24), 67–77.
Nalbantis, I., and Tsakiris, G. (2009). “Assessment of hydrological drought
revisited.” Water Resour. Manage., 23(5), 81–97.
Özger, M., Mishra, A. K., Singh, V. P. (2012). “Long lead time drought
forecasting using a wavelet and fuzzy logic combination model.”
J. Hydrometeorol., 13(1), 284–297.
J. Irrig. Drain Eng.
 Palmer, W. C. (1965). “Meteorological drought.” Technical Rep. No. 45,
U.S. Dept. of Commerce Weather Bureau Research, Washington, DC.
Panu, U. S., and Sharma, T. C. (2002). “Challenges in drought research:
Some perspectives and future directions.” J. Hydrol. Sci., 47, S19–S30.
Paulo, A. A., and Pereira, L. S. (2008). “Stochastic prediction of drought
class transitions.” Water Resour. Manage., 22(9), 1277–1296.
Raziei, T., Bordi, I., and Pereira, L. S. (2011). “An application of GPCC and
NCEP/NCAR datasets for drought variability analysis in Iran.” Water
Resour. Manage., 25(4), 1075–1086.
Shafer, B. A., and Dezman, L. E. (1982). “Development of a Surface Water
Supply Index (SWSI) to assess the severity of drought conditions
in snowpack runoff areas.” Proc., 50th Annual Western Snow Conf.,
Western Snow Conference, Reno, NV, 164–75.
Shahabfar, A., and Eitzinger, J. (2011). “Agricultural drought monitoring in
semi-arid and arid areas using MODIS data.” J. Agric. Sci., 149(04),
Downloaded from ascelibrary.org by University of Exeter on 07/18/15. Copyright ASCE. For personal use only; all rights reserved.
403–414.
Sharma, S. (1996). Applied multivariate techniques, Wiley.
© ASCE 04015029-20
J. Irrig. Drain Eng.
Sharma, T. C., and Panu, U. S. (2012). “Prediction of hydrological drought
durations based on Markov chains: Case of the Canadian prairies.”
Hydrol. Sci. J., 57(4), 705–722.
Snedecor, G. W., and Cochran, W. G. (1989). Statistical methods, 8th Ed.,
Iowa State Univ. Press, Ames, IA.
Sonmez, F. K., Komuscu, A. U., Erkan, A., and Turgu, E. (2005). “An
analysis of spatial and temporal dimension of drought vulnerability
in Turkey using the standardized precipitation index.” Nat. Hazards,
35(2), 243–264.
UN Secretariat General. (1994). “United Nations Convention to combat
drought and desertification in countries experiencing serious droughts
and/or desertification, particularly in Africa.” Paris.
Webster, K. E., Kratz, T. M., Bowser, C. J., and Adagnuson, J. J. (1996).
“The influence of landscape position on lake chemical responses to
drought in northern Wisconsin.” Limnol. Oceanogr., 41(5), 977–984.
Wilks, D. S. (2011). Statistical methods in the atmospheric sciences, 3rd
Ed., Academic, London.
J. Irrig. Drain Eng.