European Water 23/24: 67-77, 2008.

© 2008 E.W. Publications

Evaluation of a Hydrological Drought Index

I. Nalbantis
Laboratory of Reclamation Works and Water Resources Management, School of Rural and Surveying
Engineering, National Technical University of Athens
9 Heroon Polytecniou 15773 Zographou – Greece

Abstract: Indices for characterising hydrological drought are, in general, data demanding and computationally intensive. A very
simple and effective index, the Streamflow Drought Index (SDI), has been recently proposed. It is based on
cumulative streamflow volumes for overlapping periods of three, six, nine and twelve months within each
hydrological year. It allows defining drought states which are modelled as a non-stationary Markov chain. The
methodology is validated using data from two river basins in Greece (Evinos and Boeoticos Kephisos). Water from
these basins is diverted for water supply of the Athens Metropolitan Area. Thus, the methodology is tested on a real-
world system which allows for assessing its applicability within a Drought Watch System in river basins with
significant storage works.

Key words: Hydrological drought; drought prediction; SDI; Markov chain; Evinos basin, Boeoticos Kephisos basin.

1. INTRODUCTION
Drought is a naturally occurring phenomenon related to a significant decrease of water
availability during a significant period of time and over a large area. It affects man’s economic
activities, human lives and various elements of the environment such as the Earth’s ecosystems. The
origin of drought is impossible to define much as the starting point of the global hydrological cycle.
Conventionally, decrease of precipitation is considered as the origin of drought. This leads to a
reduction of storage volumes and fluxes involved in the hydrological cycle. Depending on the
choice of the hydrological variable or variables of interest, drought is characterised as
meteorological, hydrological or agricultural (Beran and Rodier, 1985).
Hydrological drought is defined as a significant decrease in the availability of water in all its
forms appearing in the land phase of the hydrological cycle. Various hydrological variables are used
to describe these forms but streamflow is, by far, the most significant variable from the viewpoint of
quantity of water. Hence, a hydrological drought episode is related to streamflow deficit with
respect to normal conditions. Each drought event is characterised through four attributes: (a) its
severity expressed by a drought index, (b) its time of onset and its duration, (c) its areal extent, and
(d) its frequency of occurrence.
According to the methodology proposed by Nalbantis and Tsakiris (2008) the four-dimensional
relationship of drought severity-duration-frequency-area is reduced into a much simpler two-
dimensional relationship of severity versus frequency. First, an index called Streamflow Drought
Index (SDI) was proposed which characterises drought severity while fulfilling all requirements of
such indices. Second, the time of onset and duration of drought events was eliminated through
properly treating time. Third, the frequency of drought occurrence was kept as a significant
parameter. Last, the areal extent of a drought event, although very useful for meteorological
droughts, is not of interest for hydrological droughts since water managers are interested on
streamflow only at a small number of points in space (basin outlets, reservoir inlets and outlets etc.);
evidently, streamflow at these points provides an integrated measure of spatially distributed runoff;
furthermore, the river basin is proposed by the Water Framework Directive as the unit for applying
measures for water resources protection and management.
68 I. Nalbantis

The aim of this paper is to validate the methodology based on SDI through the application on
data from two river basins in Greece (Evinos and Boeoticos Kephisos). Water from these basins is
diverted for water supply of the Athens Metropolitan Area. So, application on a real-world system
serves assessing the applicability of the methodology within a Drought Watch System (Wilhite et
al., 2007). To achieve the above goal, a suitable testing framework is set up.

2. METHODOLOGY

2.1 General

Indices for characterising a hydrological drought such as Palmer Hydrological Drought Index
(PHDI), Surface Water Supply Index (SWSI) or the index proposed by Palfai (2002) are, in general,
data demanding and computationally intensive. On the contrary, the proposed index SDI keeps the
advantages of simplicity and effectiveness found in indices of meteorological droughts such as the
Standardised Precipitation Index (SPI) (McKee et al., 1993; Tsakiris and Vangelis, 2004;
Cancelliere et al., 2007) or the Reconnaissance Drought Index (RDI) (Tsakiris and Vangelis, 2005;
Tsakiris et al., 2007). Exclusive use of streamflow is made as the key variable for assessing
hydrological droughts.

2.2 The treatment of time

The fact that drought is a natural phenomenon which is slowly developing and is identified only
after it has been well established dictates the use of coarse time steps for drought assessments. The
typical time step used is monthly which is also employed in this study.
The onset of a drought episode is defined as the time when a drought index falls below a certain
truncation level. The truncation level has been defined in various ways. For stationary processes a
fixed value has been used while for periodic processes a set of seasonally varying values is more
appropriate. The mean over a long period of time was chosen in this work.
In the classical approach in treating time, successive non-overlapping time intervals are used. In
the proposed methodology, time is treated as follows: (1) October the first is considered the
beginning of the hydrological year which is typical in the Mediterranean region; (2) every three
months (31st December, 31st March, 30th June, 30th September) a drought assessment is made
regarding the time interval from the start of the hydrological year up to that time; thus time intervals
of duration of three, six, nine and twelve months are used; (3) at the above dates, predictions are
issued regarding drought conditions for future time intervals.
The overlapping time periods used within each hydrological year are reported as reference
periods. These are October-December, October-March, October-June, and October-September (one
complete hydrological year).

2.3 The Streamflow Drought Index (SDI)

It is assumed that a time series of monthly streamflow volumes Qi,j is available where i denotes
the hydrological year and j the month within that hydrological year (j = 1 for October and j = 12 for
September). Based on this series we obtain

3k
Vi ,k = ∑ Qi , j i = 1, 2, … j = 1, 2, …, 12 k = 1, 2, 3, 4 (1)
j =1
European Water 23/24 (2008) 69

where Vi,k is the cumulative streamflow volume for the i-th hydrological year and the k-th reference
period, k = 1 for October-December, k = 2 for October-March, k = 3 for October-June, and k = 4 for
October-September.
Based on cumulative streamflow volumes Vi,k the Streamflow Drought Index (SDI) is defined for
each reference period k of the i-th hydrological year as follows

Vi,k − Vk
SDI i,k = i = 1, 2, …, k = 1, 2, 3, 4 (2)
sk

where Vk and sk are respectively the mean and the standard deviation of cumulative streamflow
volumes of reference period k as these are estimated over a long period of time. In this definition
the truncation level is set to Vk although other values could be used.
The hydrological drought index of equation 2 is identical to the standardised streamflow volume.
This is not entirely new since Ben-Zvi (1987) made use of the standardised annual streamflow
volumes.
Generally, for small basins, streamflow may follow a skewed probability distribution which can
well be approximated by the family of the Gamma distribution functions. The distribution is then
transformed into normal. In this work we use the two-parameter log-normal distribution for which
the normalisation is simple: it suffices taking natural logarithms of streamflow. The SDI index is
defined as

y i ,k − y k
SDI i ,k = i = 1, 2, …, k = 1, 2, 3, 4 (3)
s y ,k

where

yi,k = ln(Vi,k), i = 1 , 2, …, k = 1, 2, 3, 4 (4)

are the natural logarithms of cumulative streamflow with mean yk and standard deviation s y ,k as
these statistics are estimated over a long period of time.
Based on SDI, states of hydrological drought are defined which are identical to those used in the
meteorological drought indices SPI and RDI. Five states are considered which are denoted by an
integer number ranging from 0 (non-drought) to 4 (extreme drought) and are defined through the
criteria of Table 1.

Table 1: Definition of states of hydrological drought with the aid of SDI

State Description Criterion Probability (%)
0 Non-drought SDI ≥ 0.0 50.0
1 Mild drought −1.0 ≤ SDI < 0.0 34.1
2 Moderate drought −1.5 ≤ SDI < −1.0 9.2
3 Severe drought −2.0 ≤ SDI < −1.5 4.4
4 Extreme drought SDI < −2.0 2.3

The problem of treating intermittent or ephemeral flows is of importance when dealing with
hydrological droughts. Three cases can be distinguished: (1) watercourse with perennial flow, (2)
watercourse with ephemeral flow and without complete dryness throughout a whole hydrological
year, (3) watercourse with no flow in some hydrological years. According to our definition of SDI,
case 2 becomes irrelevant since cumulative streamflow will always possess some positive value.
Only the case of completely dry hydrological years (case 3) remains which is arbitrarily classified
as extreme drought (state equal to 4).
70 I. Nalbantis

2.4 The methodological steps

Starting from historical streamflow series, an SDI series is computed which yields a series of
drought states. The underlying state process is assumed to possess the structure of a non-stationary
Markov chain. Markov chains have been widely applied to predicting droughts (mainly
meteorological ones) (Lohani and Loganathan, 1997; Lohani et al., 1998; Ochola and Kerkides,
2003; Paulo and Pereira, 2006).
Let Qi,j (i = 1, 2, …, N; j = 1, 2, …, 12) be the observed time series of monthly streamflow
volumes for the river basin under study, where N is the number of hydrological years.
First, the cumulative streamflow volumes Vi,k (i = 1, 2, …, N; k = 1, 2, 3, 4) are calculated via
equation 1. Second, the series SDIi,k of the SDI index is calculated based on equation 2 or 3. Third,
the series of states xi,k (i = 1, 2, …, N; k = 1, 2, 3, 4) is obtained according to the criteria of Table 1.
For each k, the related state process Xi,k takes discrete values m ∈ [0, 1, 2, 3, 4]. Fourth, the
frequency of appearance of each state m in each reference period k, Fm,k, is estimated as

n m,k
Fm,k = (5)
N

where nm,k is the number of occurrences of state m in reference period k within the available sample
of N years. This is an estimate of the marginal probability pm,k of appearance of state m in reference
period k, i.e.

p m,k = P(X i,k = m ) m ∈ [0, 1, 2, 3, 4] ∀i (6)

where P(.) denotes probability. For each k, probabilities pm,k (m = 0, 1, 2, 3, 4) form a 5×1 column
vector pk.
Fifth, the frequency of state transition Fm,m’,k from state m in reference period k to state m’ in
reference period k+1 is

nm,m′,k
Fm,m′,k = (7)
∑ nm,m′,k
m′

where nm,m’,k is the number of occurrences of state m in reference period k and state m’ in reference
period k+1. This is an estimate of the transition probability pm,m’,k which is defined as

p m,m′,k = P(X i,k +1 = m′ | X i ,k = m ) m ∈ [0, 1, 2, 3, 4] m’ ∈ [0, 1, 2, 3, 4] ∀i (8)

where P(.|.) denotes conditional probability. For each k, transition probabilities form a 5×5 matrix
denoted as Pk.
Assume now that the current time interval is (i,k). Before characterising current drought state,
one can predict the marginal probabilities for the next reference period k+1 as

p k +1 = Pk p k (9)

Within an operational context, at the end of time interval (i,k), all historical data up to that time
are assumed to become available. This allows classifying the current interval. Thus, vector pk of
equation 9 has now one element equal to one and all other elements zero. It follows from equation 9
that the only information needed for drought prediction is the matrix Pk as this is approximated by
its estimate, the matrix of state transition frequency. This is the main output of the methodology
when working off-line on historical series. In real-time situations, the output is (a) a single value of
European Water 23/24 (2008) 71

current state and (b) the probabilities of future states as obtained from a stored matrix of state
transition probability.
Nalbantis and Tsakiris (2008) have also treated the case with lack of streamflow information. For
this they devised a special methodology based on SPI. This was not used in this paper.
A rigorous testing framework was set up to further validate the above methodology.

3. CASE STUDIES

3.1 The study area

The methodology proposed was applied to two river basins from which water is diverted for
supplying the Athens Metropolitan Area. The first basin is the Evinos river basin which is located in
the West Sterea Hellas Water District in central Greece (Figure 1). At the site of the Agios
Demetrios dam it has an upstream basin area of 352 km2, mean elevation of 990 m above the mean
sea level, and steep slopes. Its average annual streamflow is 297×106 m3 (Efstratiadis et al. 2000).
Water from the Agios Demetrios reservoir is diverted eastwards to the adjacent Mornos reservoir
for the water supply of the Greater Athens area. The reservoir possesses a surface area of 3.6 km2
(at the elevation of the spillway crest) and an active storage capacity of 112.1×106 m3. The Agios
Demetrios reservoir and the adjacent Mornos reservoir form the main storage facilities of the water
supply system of Athens.
The second basin is the Boeoticos Kephisos River Basin which is located in the East Sterea
Hellas Water District (Figure 1). It has an upstream area of 1956 km2 and an average annual
streamflow 285×106 m3. Water is collected in the natural Lake Hyliki and thereafter is pumped to
Athens. The lake suffers from considerable leakage losses which may reach up to 50% of its total
annual inflow due to karstic background. Spills are directed to the adjacent Lake Paralimni. The
lake shows a surface area of 27.74 km2 (at the elevation of the spillway crest) and an active storage
capacity of 584.8×106 m3.
Monthly streamflow data at the Agios Demetrios dam have been obtained through processing
raw data of concurrent velocity and stage measurements combined with stage recordings at the daily
or the hourly time step, mainly at the Poros Reganiou hydrometric station, downstream of the dam
site. Data covering the period from 1970-71 to 1999-2000 have been used by Nalbantis and Tsakiris
(2008). To include the drought around 2000-01, the record is now extended up to 2001-2002.
Regarding the Boeoticos Kephisos River Basin, monthly streamflows from 1907-08 were available.
To minimise the effect of any kind of hydrological changes and make a fair comparison of the
results from the two basins, the period 1970-71 up to 2002-03 is finally selected for the second
basin.

Figure 1: The Evinos river at the Agios Demetrios Reservoir and the Boeoticos Kephisos river at the Karditsa Tunnel
(outlet)
72 I. Nalbantis

3.2 Results

In the earlier work by Nalbantis and Tsakiris (2008) the statistical significance of the skeweness
coefficient of the cumulative streamflow volumes was tested. The authors concluded that using
natural logarithms of volume was necessary only for the October-December reference period. This
led to using the definition of equation 3 for SDI of October-December while keeping equation 2 for
the other reference periods. The tests were repeated in this paper also. The skewness coefficients of
the initial data, of the natural logarithms of the data and of the final data are shown in Table 2 for
the two basins. According to the test of Snedecor and Cochran (1967) the critical upper limits of the
absolute value of the skewness coefficient are equal to 0.986 and 0.662 respectively at 0.02 and
0.10 significance level. Hence, taking logarithms was only necessary for the October-December
reference period and the Evinos basin. No logarithms were used for the Boeoticos Kephisos
streamflows.

Table 2: Skewness coefficient of cumulative streamflow and of its natural logarithm.

Basin Calculation basis Oct-Dec Oct-Mar Oct-Jun Year
Evinos Initial data 0.857 -0.230 -0.207 -0.240
Logarithms -0.550 -1.127 -0.952 -0.987
Final data -0.550 -0.230 -0.207 -0.240
B. Kephisos Initial data 0.384 0.143 0.034 0.118
Logarithms -0.982 -1.051 -1.145 -1.051
Final data 0.384 0.143 0.034 0.118
Statistically significant values (at the 0.10 probability) are in italics.

For the Evinos basin the evolution of SDI from one hydrological year to another and each
reference period is depicted in Figures 2, 3, and 4 where the SDI series were grouped by two.
Similar graphs are given in Figures 5, 6 and 7 for the Boeoticos Kephisos basin. As expected,
significant discrepancies are observed only when passing from the first three-month period
(October-December) to the first semester (October-March). This is due to the typical Mediterranean
hydrological regime which is manifested as a wet period of six months of the hydrological year and
a mostly dry period thereafter. As a result, high predictive capacity of drought state is expected
when, in the end of March the following question is raised: will the nine-month period of the
running year be considered as drought period? The same holds for the assessment in the end of
June. The above behaviour was similar in both test basins. The SDI series of the two basins were
graphically compared for each reference period separately. Significant differences were shown in
the hydrological regime between the western and the eastern part of the Sterea Hellas Region. For
the three historical droughts that affected the water supply of Athens one can observe the following:
a. In the hydrological year 1976-77 the B. Kephisos basin has undergone a moderate drought
while in the Evinos basin no-drought conditions prevailed.
b. During the drought from 1988-89 to 1994-95, we observe that: (1) in 1988-89 the drought
was mild in both basins; (2) in 1989-90 the drought became extreme in the Evinos basin and
severe in the B. Kephisos basin; (3) in 1990-91 a mild drought in B. Kephisos is combined
with no-drought conditions in the west; (4) in 1991-92 the drought was severe in Evinos and
moderate in B. Kephisos; (5) in 1992-93 the situation of 1991-92 was reversed; (6) in the
next two hydrological years 1993-94 and 1994-95 drought can be characterised as mild.
c. In 1999-2000 a mild drought was observed in B. Kephisos; in 2000-2001 the drought
became severe in both basins while in 2001-2002 a moderate drought in the west is
accompanied with a mild drought in the east.

The above drought assessments confirm the results obtained at the time of drought in a more or
less qualitative manner (Nalbantis et al, 1994). The conclusions are drawn from comparisons for all
European Water 23/24 (2008) 73

reference periods which show a similar behaviour to that of the whole hydrological year (Figure 9)
with the exception of the October-December period (Figure 8) for which the differences between
the two basins are clearly more significant.

2
1
SDI

0
-1
Oct-Dec
-2
Oct-Mar
-3
19 71

19 74

19 77

19 80

19 83

19 86

19 89

19 92

19 95

20 98
1
-0
-

-
70

73

76

79

82

85

88

91

94

97

00
19

Figure 2: SDI series for the Evinos basin and the reference periods October-December and October-March.

2 Oct-Mar
1 Oct-June
SDI

0
-1
-2
-3
1
20 98
19 71

19 74

19 80

19 83

19 89
19 92

19 95
19 77

19 86

-0
-
-

-
-
-

00
91
94

97
82

85

88
70

73

76

79
19

Figure 3: SDI series for the Evinos basin and the reference periods October-March and October-June.

2 Oct-June
1 Oct-Sep
SDI

0
-1
-2
-3
19 74
19 71

19 77

19 80

19 83
19 86

19 89

19 92

19 95

20 98
1
-0
-
-

-
-

-
70

73

76

79

82

85
88

91

94

97

00
19

Figure 4: SDI series for the Evinos basin and the reference periods October-June and the hydrological year (October-
September).
74 I. Nalbantis

2 Oct-Dec
1 Oct-Mar
SDI 0
-1
-2
-3
19 71

19 74

19 77

19 80

19 83

19 86

19 89

19 92

19 95

20 98
1
-0
-

-
70

73

76

79

82

85

88

91

94

97

00
19

Figure 5: SDI series for the Boeoticos Kephisos basin and the reference periods October-December and October-
March.

2 Oct-Mar
1 Oct-June
SDI

0
-1
-2
-3
19 71

19 74

19 77
19 80

19 83

19 86

19 89
19 92

19 95

20 98
1
-0
-

-
-

-
-

-
00
70

73

76

79
82

85

88

91
94

97
19

Figure 6: SDI series for the Boeoticos Kephisos basin and the reference periods October-March and October-June.

2 Oct-June
1 Oct-Sep
SDI

0
-1
-2
-3
20 98
1
19 89

19 92

19 95
19 77

19 80

19 83

19 86
19 71

19 74

-0
-

-
-

-
-

-
-

97

00
91

94
82

85

88
70

73

76

79
19

Figure 7: SDI series for the Boeoticos Kephisos and the reference periods October-June and the hydrological year
(October-September).
European Water 23/24 (2008) 75

2
1

SDI
0
-1
Evinos
-2
B.Keph.
-3

1 9 89

19 92

19 95

20 98
1
19 86
1 9 71

19 74

19 77

19 80

19 83

-0
-

-
-
-

-
-

00
88

91

94

97
85
70

73

76

79

82
19

Figure 8: Comparison of SDI series for the reference period October-December and the two test basins.

2
1
0
SDI

-1
Evinos
-2
B.Keph.
-3
19 83
19 71

19 74

19 77

19 80

19 86

19 89

19 92

19 95

20 98
1
-0
-

-
70

73

76

79

82

85

88

91

94

97

00
19

Figure 9: Comparison of SDI series for the reference period October-September and the two test basins.

Regarding the frequency of state transition, preliminary tests showed very small observed
number of occurrences of states 2 (moderate drought), 3 (severe drought) and 4 (extreme drought).
To remedy this problem, the above three states were grouped into one state to which we assigned
number 2. In Table 3 we present the matrices of state transition frequency for all pairs of reference
periods when passing from one period to the next lengthier period (three pairs in all). These
matrices are the main tools for predicting drought state in real-time in the case of availability of
streamflow data (see subsection 2.4. In both basins the state transition frequency shows a significant
“dispersion” between different states when the starting period is October-December. When passing
to other starting periods, the state transition frequency tends to be stabilised to a value close to 1
when staying in the same state and 0 for all state changes. In other words, the respective matrix
tends to the identity matrix. Differences were encountered for the Evinos basin between frequencies
found by Nalbantis and Tsakiris (2008) and by the present analyses which are due to the extension
of the observation period in this work.

4. CONCLUDING REMARKS
This paper aims at validating a methodology for forecasting hydrological droughts within an
operational context regarding river basins with works of large total storage capacity.
76 I. Nalbantis

Table 3: Frequency of state transition as estimated from data of the test basins.
(a) Evinos Basin (b) Boeoticos Kephisos basin
State for State for Oct-Mar State for State for Oct-Mar
Oct-Dec 0 1 2 Oct-Dec 0 1 2
0 0.706 0.294 0.000 0 0.579 0.368 0.053
1 0.455 0.273 0.273 1 0.500 0.400 0.100
2 0.000 0.500 0.500 2 0.000 0.250 0.750
State for State for Oct-Jun State for State for Oct-Jun
Oct-Mar 0 1 2 Oct-Mar 0 1 2
0 1.000 0.000 0.000 0 0.875 0.125 0.000
1 0.100 0.900 0.000 1 0.000 1.000 0.000
2 0.000 0.000 1.000 2 0.000 0.000 1.000
State for State for Oct-Sep State for State for Oct-Sep
Oct-Jun 0 1 2 Oct-Jun 0 1 2
0 0.944 0.056 0.000 0 1.000 0.000 0.000
1 0.000 1.000 0.000 1 0.000 0.929 0.071
2 0.000 0.000 1.000 2 0.000 0.000 1.000

The methodology is based on a proposed index called Streamflow Drought Index (SDI) which is
calculated for overlapping reference periods within each hydrological year thus allowing for
categorising droughts into a small number of categories of drought state. The latter is modelled
through a Markov chain which allows for estimating state transition frequency within an off-line
context and predicting future drought state within a real-time context.
A rigorous testing framework was set up which allowed for validating the methodology on a
real-world water resources system. Two river basins from those contributing to water supply of
Athens, Greece were selected as the test basins. These cover both the western part (Evinos basin)
and the eastern part (Boeoticos Kephisos basin) of the system and have different hydrological
regimes in three respects: precipitation regime, the geological background and the degree of
anthropogenic intervention. Analysis has shown that SDI can well discover the main droughts
known to have occurred in the region: 1976-77, 1988-95 and 1999-2001. The matrix of frequency
of state transition proved a useful tool which allows high predictive capacity of drought when the
end of wet season is close (end of March) as well as in later times within the hydrological year. In
the middle of the wet season predictions are far less certain.
The methodology used can be easily applied within a Drought Watch System. However, it
requires streamflow data of high quality and of sufficient length to accurately estimate the
frequency of rare drought phenomena.

REFERENCES
Ben-Zvi, A., 1987. Indices of hydrological drought in Israel. J. Hydrol.; 92(1-2): 179-191.
Beran, M.A., Rodier, J.A., 1985. Hydrological aspects of drought. In: UNESCO-WMO Studies and Reports in Hydrology 39:149 pp.
Cancelliere, A., Di Mauro, G., Bonaccorso, B., Rossi, G., 2007. Drought forecasting using the Standardized Precipitation Index.
Water Resour. Manage.; 21: 801–819.
Efstratiadis, A., Nalbantis, I., Mamassis, N. 2000. Hydrometeorological data processing. In: Modernisation of the supervision and
management of the water resource system of Athens, Report 8:129 pages. Department of Water Resources, Hydraulic and
Maritime Engineering - National Technical University of Athens, Athens.
Lohani, V.K., Loganathan, G.V., 1997. An early warning system for drought management using the Palmer drought index. Journal of
the American Water Resources Association; 33(6): 1375–1386.
Lohani, V.K., Loganathan, G.V., Mostaghimi, S., 1998. Long-term analysis and short-term forecasting of dry spells by the Palmer
drought severity index. Nordic Hydrology; 29(1): 21–40.
McKee, T.B., Doeskin, N.J., Kleist, J., 1993. The relationship of drought frequency and duration to time scales. Proceedings of the
Eighth Conference on Applied Climatology, Anaheim, CA, January 17-23, 1993. American Meteorological Society Boston MA,
pp 179-184.
Nalbantis, I., Mamassis, N., Koutsoyiannis, D., 1994. Le phémomène récent de sécheresse persistante et l'alimentation en eau de la
cité d'Athènes. Comptes-rendus du 6ème colloque de l'Association Internationale de Climatologie, Théssalonique, 20-25
Septembre 1993, pages 8..
European Water 23/24 (2008) 77

Nalbantis, I., Tsakiris, G., 2008. Assessment of Hydrological Drought Revisited. Water Resour. Manage.; DOI 10.1007/s11269-008-
9305-1
Ochola, W.O., Kerkides, P., 2003. A Markov chain simulation model for predicting critical wet and dry spells in Kenya: Analysing
rainfall events in the Kano plains. Irrigation and Drainage; 52(4): 327–342.
Palfai, I., 2002. Probability of drought occurrence in Hungary. IdŰjárás (edition of Hungarian Meteorological Service); 106(3-4):
265-275.
Paulo, A.A., Pereira, L.S., 2007. Prediction of SPI Drought Class Transitions Using Markov Chains. Water Resour. Manage.; 21:
1813-1827.
Snedecor, G.W., Cochran, W.G., 1967. Statistical Methods. The Iowa University Press. Iowa, USA.
Tsakiris, G., Vangelis, H., 2004. Towards a Drought Watch System based on Spatial SPI. Water Resour. Manage.; 18: 1–12.
Tsakiris, G., Vangelis, H., 2005. Establishing a Drought Index Incorporating Evapotran-spiration. European Water; 9/10: 3-11.
Tsakiris, G., Pangalou, D., Vangelis, H., 2007. Regional Drought Assessment based on the Reconnaissance Drought Index (RDI).
Water Resour. Manage.; 21: 821-833.
Wilhite, D.A., Svoboda, M.D., Hayes, M.J., 2007. Understanding the complex impacts of drought: A key to enhancing drought
mitigation and preparedness. Water Resour. Manage.; 21: 763–774.