REgional Drought Identification Module

REDIM
Antonino Cancelliere, Brunella Bonaccorso, Luca Cavallaro and Giuseppe Rossi

Extended Version 2005.1

October 2005
(Including computation of return period of drought events and SPI)

Reference Guide

Department of Civil and Environmental Engineering

University of Catania
Catania, Italy
Table of contents

1. INTRODUCTION ..................................................................................................................................... 3

2. PRELIMINARY ANALYSIS OF HYDROLOGICAL TIME SERIES ............................................... 3

3. DROUGHT IDENTIFICATION AND CHARACTERIZATION BY RUN METHOD ..................... 6

3.1 OVERVIEW ...................................................................................................................................... 6

3.2 DROUGHT IDENTIFICATION AT SITE ................................................................................................. 7
3.3 REGIONAL DROUGHT ANALYSIS ...................................................................................................... 8
3.4 PROBABILITY DISTRIBUTION OF DROUGHT CHARACTERISTICS AND RETURN PERIOD ..................... 12

4. DROUGHT ANALYSIS THROUGH THE STANDARDIZED PRECIPITATION INDEX (SPI) . 16

4.1 OVERVIEW .................................................................................................................................... 16

4.2 DROUGHT IDENTIFICATION AT SITE ............................................................................................... 17
4.3 REGIONAL DROUGHT ANALYSIS .................................................................................................... 18

5. DESCRIPTION OF THE SOFTWARE ................................................................................................ 20

5.1 OVERVIEW .................................................................................................................................... 20

5.2 SOFTWARE CONFIGURATION ......................................................................................................... 21
5.3 DATA IMPORT................................................................................................................................ 22
5.4 ANALYSIS OF STATIONARITY......................................................................................................... 22
5.5 DROUGHT ANALYSIS AT A SITE...................................................................................................... 27
5.6 REGIONAL DROUGHT ANALYSIS .................................................................................................... 30
5.7 DROUGHT ANALYSIS AT SITE THROUGH SPI INDEX ....................................................................... 34
5.8 REGIONAL DROUGHT ANALYSIS THROUGH SPI INDEX ................................................................... 36
5.9 GRAPHICAL OUTPUT ...................................................................................................................... 38

6. REFERENCES ........................................................................................................................................ 39

7. APPENDIX A: MS ACCESS DATA FILE FORMAT......................................................................... 41

8. APPENDIX B: TEXT DATA FILE FORMAT..................................................................................... 42

9. APPENDIX C: COMPUTATION OF SAMPLE QUANTILES ......................................................... 43

2
1. Introduction
Drought identification and characterization at a site and over a region is an important tool
both for planning and management of water systems as it provides useful information for
the assessment of water shortage risk and for implementing a appropriate mitigation
measures.. For example drought analysis performed on a streamflow series provides
important insights on the water availability, as well as on the expected length and severity
of water deficit periods. Similarly, regional drought analysis allows to estimate the areal
extent and severity of precipitation deficits.
In the present document, a new version of the software REDIM, which extends the original
release developed within the INCO-DC Project DSS-DROUGHT (Rossi et al., 2003), is
described. In particular the new version, besides changes in the graphics, includes new
tests for randomness and stationarity in hydrological series, the evaluation of return period
of historical drought events, and the drought analysis through the Standardize Precipitation
Index.

2. Preliminary analysis of hydrological time series

Since the method of run is based on the differences between the observed hydrological
variable and a constant (or periodic) threshold, it follows that results can be affected
adversely by the presence of non-stationarities in the series, such as trends. Therefore, a
correct analysis must be preceded by testing the series for the presence of trends, for
randomness, and for changes on the mean and the variance. Here, six tests have been
considered, namely the t-Student test (Helsel and Hirsch, 1992; Maidment, 1993) for linear
trends, the Kendall -test (Helsel and Hirsch, 1992; Maidment, 1993) and the turning point
test (Kottegoda, 1980) for randomness, the Rank sum test (Wilcoxon, 1945; Helsel and
Hirsch, 1992; Maidment, 1993), the F-test (Pollard, 1977) and t-test (Pollard, 1977;
Maidment, 1993) for the detection of change in the variance and mean respectively.
Testing for the presence of a linear trend can be carried out by fitting a linear
regression to the available data, using the time as independent variable. A slope coefficient
significantly different than zero will indicate the presence of a linear trend. The t-Student
test allows to verify statistically the difference from zero of the slope coefficient, by testing
whether the correlation coefficient is significantly different than zero. In particular, under

3
the null hypothesis that the population correlation coefficient = 0, the test statistic tc is
computed as:

r N 2
tc =
1 r2
where r is the sample correlation coefficient and N the sample size. The null hypothesis is
rejected on the basis of a two-tailed test at significance level if t c t1 / 2 ( N 2) where

t1 / 2 ( N 2) is the value drawn from a Students t distribution with N-2 degrees of

freedom and non exceedence probability (1-/2). Typical values of are 10, 5 , 2.5 and 1
percent.
The Kendall -test (also termed rank correlation test) is applied by computing for a
series xt (t=1, 2, .., n) the number p of pairs of observations (xt,xj) (with j>t) for which xj>xt
(Kottegoda, 1980). Under the null hypothesis of random series, the statistic:

zc =
var
where:
4p
= 1
N ( N 1)
2(2N + 5 )
var =
9N ( N 1)

converges rapidly to a standard normal distribution as N increases. Therefore, the null

hypothesis is rejected with a significance level if |zc|>zt1-/2 where zt1-/2 is a standard

normal variable corresponding to a non exceedence probability (1-/2).

The turning point test is based on counting in a series xt (t=1,2 .N) the number of
turning points m , i.e. for each t=2, 3 , N-1, the values for which xt is larger than xt-1 and
xt+1 or xt is smaller than xt-1 and xt+1 (Kottegoda, 1980).
Under the null hypothesis the series is assumed random and it can be shown that the
statistic:

zc =
[m E (m)]
var (m)

with:

4
 2 (N 2 )
E (m ) =
3

var (m ) =
(16 N 29 )
90
converges to a standard normal variable. Then, the null hypothesis is rejected with a
significance level if |zc|> zt1-/2 where zt1-/2 is a standard normal variable corresponding

to a non exceedence probability (1-/2).

The rank-sum test is applied by considering two sub-series with dimension N1 and N2
(with N1N2), and the sum of the ordinal of shorter series (with dimensions N1) Wrs = k =
1, ... N1 Rk.
Under the null hypothesis the two series are extracted by the same population and
Wrs tends to the normal distribution with standardized variable as:
zrs = [Wrs - d/2 - w ]/ w for Wrs > w
=0 for Wrs = w
= [Wrs + d/2 - w ]/w for Wrs < w
with d=1 minimum difference between two possible
value of the statistical variable Wrs
w = (N1/N) N(N+1)/2 = N1(N+1)/2 mean
w = [N1N2(N+1)/12]1/2 standard deviation
If the statistical variable Wrs is much greater (or smaller) than the expected value w,
the shorter series contains value which are appreciably greater (or smaller) than the other series
and hence the two series are not omogeneous. Then, the null hypothesis is rejected with a
significance level if |zrs|> zt1-/2 where zt1-/2 is a standard normal variable corresponding

to a probability (1-/2).
The F test allows to evaluate if two series (with dimensions N1 and N2, variances s12
and s22, and variances of the population 12 and 22) have not significant differences in
variance. Under the null hypothesis (12=22), the two series belong to the same population
and the statistical variable is equal to f = (s12/ 12) / (s22/22) = (s12/ s22). This statistical
variable follows a specific probability distribution (Pollard, 1977). Then, the null
hypothesis is rejected with a significance level if f>f1-/2(p,q), with:

p = N1-1 and q = N2-1 if f > 1,0

p = N2-1 and q = N1-1 if f < 1,0.

5
 The test on the mean homogeneity (t-test) allows to evaluate if two series (with
dimensions N1 and N2, means y1 and y2, variances s12 and s22, means of the population 1
and 2, and variances of the population 12 and 22) has significant difference in mean.
Under the null hypothesis the two series belong to the same population and have the same
variance, 12 = 22, (according the previous F test on variance) and the statistical variable t
is equal to:
t = | y1 - y2 | / [S(1/N1 + 1/N2)1/2]
with:
S = {[(N1-1)s12 + (N2-1)s22]/(N-2)}1/2
The variable t follows the t-student distribution, with = N1+N2-2 degrees of freedom.
Then, the null hypothesis is rejected with a significance level if t>t1-/2 ().

3. Drought identification and characterization by run method

3.1 Overview
Use of run analysis has been proposed as an objective method for identifying drought
periods and for evaluating the statistical properties of drought. According to this method a
drought period coincides with a "negative run", defined as a consecutive number of
intervals where a selected hydrological variable remains below a chosen truncation level or
threshold (Yevjevich, 1967).
Such a threshold can be a fixed value in the case of a non periodic (e.g. annual)
stationary time series, or a seasonally varying truncation level in the case of a stationary
periodic series. The truncation level in each time interval is somewhat arbitrary and it
must be selected on the basis of the objective of the study. Usually it is assumed equal to
the long-term mean (or median) of the variable of interest, while other possible choices
include a fraction of the mean (Clausen and Pearson, 1995), a value corresponding to a
given non-exceedence probability (Zelenhasic and Salvai, 1987, and Correia et al., 1987),
or a level defined as one standard deviation below the mean (Ben-Zvi, 1987). In any case,
the threshold should be chosen in such a way to be considered representative of the water
demand level (Yevjevich et al., 1983, Rossi et al., 1992).
The advantage of using run method for drought definition consists in the possibility of
deriving the probabilistic features of drought characteristics (such as duration, accumulated
deficit) analytically or by data generation once the stochastic properties of the basic

6
variable are known. This possibility is not limited to relatively simple cases where time
dependence of consecutive values can be neglected, but also when a Markov chain
structure is assumed for the underlying variable (Cancelliere et al., 1998; Fernandez and
Salas, 1999). Furthermore, procedures to assess the return period of droughts defined
according to the run method have been derived recently (Fernandez and Salas, 1999; Shiau
and Shen, 2001; Bonaccorso et al., 2003; Cancelliere and Salas, 2004), thus making the
method an ideal candidate to perform drought risk analysis.

3.2 Drought identification at site

Given a discrete time series h(i), i=1, 2, n and a truncation level ho, it is possible to
identify positive and negative deviations according to the sign of the difference between
the observed values of the variable and the fixed threshold.
Then, a run is defined as a sequence of intervals characterized by deviations with the
same sign preceded and followed by at least one interval with deviation of the opposite
sign. Negative deviations are usually referred as deficit, whereas the corresponding run is
termed as negative run or drought.
By analyzing a given drought s, it is possible to identify the following characteristics:
drought duration L(s), defined as the number of consecutive intervals where the
variable remains below the threshold;
accumulated deficit D(s), defined as the sum of the negative deviations, extended to
the whole drought duration;
drought intensity of ID(s), defined as the ratio between cumulated deficit and duration.

More precisely, by letting d(i) be the deficit in the interval i:

d(i) = [h0 h(i)] I[h(i)] where I[h(i)] = 1 if h(i) < ho

I[h(i)] = 0 if h(i) ho

the duration of the drought s is given by:

L(s) = if ii + 1

where if and ii are such that: d(i) > 0 for ii i if and d(ii - 1)=0, d(if +1) = 0 ;

the accumulated deficit can be expressed by:

7
 if
D ( s ) = d (i )
i =ii

and finally the drought intensity is given by:

ID(s) = D(s)/L(s)

Once the above characteristics have been computed for each drought, it is possible to
characterize the whole series by averaging them and/or by considering the maximum or the
minimum of each characteristic and the total number of droughts.
If the time series is aggregated at a yearly time scale, a constant threshold is usually
assumed. It can be given by the means or medians or computed as a function of the mean
hm and of the standard deviation s of the series (e.g. ho = hm s where is an empirical
coefficient or the value of a random variable (0,1) corresponding to a fixed probability
level). Alternatively the threshold can be computed as a sample quantile of the series
corresponding to a frequency level F or to a quantile of the distribution fitted to the series.
If the series is aggregated at a time scale shorter than the year, say season or month, it is
necessary to take into account the periodicity of the investigated series and to adopt a
variable threshold for the selected time scale.
The main advantages of defining drought in terms of run is the relative objectivity of
the identification and characterization process, once the threshold level is defined.
Furthermore, as shown below, it is possible to derive theoretical probability distribution
functions of the drought characteristics, on the basis of the stochastic properties of the
underlying series.

3.3 Regional drought analysis

The above analysis can be extended to the case of regional droughts, i.e. droughts
which affect large regions, by considering several series of the variable of interest and
selecting, besides of the truncation level at each site, an additional threshold, which
represents the value of the area affected by deficit above which a regional drought is
considered to occur. Thus, for a given interval i, two indices are computed, representative
of the areal extension and amount of the deficits over the investigated region.
More precisely, once the threshold levels h0(k) for each site k=1,..,K are fixed, it is
possible to identify for each interval i sites which present deficit, i.e. sites for which the
difference between the threshold and the observed value of the variable is positive:

8
 h0(k) h(i,k)>0

Then, it is assumed that the deficit at each site is extended to an influence area around
the observation station, which, for example, can be estimated by Thiessen polygons
method. Such area S(k) is usually expressed in terms of the total area under investigation
as:

A(k)=S(k)/Stot.

K
where the total area Stot is obviously: Stot = S (k )
k =1

By fixing the areal threshold Acrit, again expressed as a fraction of the total area, two
indices can be computed, namely:
the areal coverage of deficit Ad(i) :
K
Ad (i ) = I [h(i, k )] A( k )
k =1

where:
I [h(i, k )] = 1 if h(i, k ) < h0 (k)
I [h(i, k )] = 0 if h(i, k ) h0 (k)

the areal deficit d(i) in the interval i:

K
d ( i ) = [h0 ( k ) h( i , k )] I [h( i , k )]A( k ) if Ad ( i ) Acrit
k =1
d( i ) = 0 if Ad ( i ) < Acrit

The Ad(i) index is a measure of the area affected by deficit, expressed as a fraction of
the total area, and it ranges between 0 and 1. The second index provides some insight on
the total amount of the deficit in the area, being basically a sum of the deficits at each sites,
weighted by the corresponding influence areas.
Further, for each drought r:
the regional drought duration L(r) is defined as:

L(r) = if(r) - ii(r)+1

where if and ii are such that d(i)>0 for ii(r) i if(r) and d[ii(r) 1]=0, d[if(r) + 1]=0

the cumulated areal deficit is computed as:

9
 if
D ( r ) = d (i )
i =ii

the regional drought intensity is given by:

ID(r) = D(r)/L(r)

the mean areal coverage of drought can be computed as:

if
AD(r ) = Ad (i) / L(r )
i =ii

As in the case of at site analysis, the above characteristics can be further analyzed by
taking the average and by considering their minimum and maximum values, in order to
characterize the whole period of observations. A flow-chart depicting the regional drought
identification and characterization process is reported in Figure 1.

10
 Input:
- observed values at different stations over a region h(i,k), k=1, 2,..,K;
- threshold level h0(k) for each station (mean, median, etc...)
- areas of Thiessen polygons A(k) (as a fraction of the total area) under
investigation).
- areal threshold Acrit

For interval i

Compute areal coverage of deficit

K
Ad(i) = I [h( i , k )]A( k )
k =1

I [h( i , k )] = 1 if h( i , k ) < h0 (k)

where:
I [h( i , k )] = 0 if h( i , k ) h0 (k)

Compute areal deficit in interval i

K
d( i ) = [h0 ( k ) h( i , k )] I [h( i , k )]A( k )
k =1

No regional
drought in
Ad(i)>Acrit ? interval i
No (d(i))= 0

Yes

i=i+1

Identify beginning ii(r) and end if(r) of each drought r such that d(i) > 0 for ii(r) i if(r)
and d(ii(r) - 1) = 0, d(if(r) + 1) = 0

For each identified regional drought r compute:

if
regional drought duration L(r) = if(r) ii(r) +1; cumulated areal deficit D(r ) = d (i ) ;
i = ii

if
regional drought intensity ID(r)= D(r)/L(r) mean areal coverage AD(r ) = Ad (i ) / L(r )
i = ii

Compute overall characteristics of regional droughts for the whole observation period:
- No of drought periods;
- Mean, max and min values of duration, cumulated areal deficit, drought intensity,
mean areal coverage.

Figure 1. Regional drought identification and characterization procedure

11
3.4 Probability distribution of drought characteristics and return period
A reliable analysis of the probabilistic features of drought characteristics cannot be
properly carried out by fitting a parametric distribution to observed sequences of drought
characteristics, due to the limited number of drought events which can be identified even
on quite long historical series. In order to overcome such difficulty, analytical derivation of
the probability distributions of drought characteristics, based on the distribution of the
underlying hydrological series has been proposed (e.g., Downer et al., 1967; Llamas and
Siddiqui, 1969; Sen, 1976, 1977, 1980; Guven, 1983; Zelenhasic and Salvai, 1987; Mathier
et al., 1992; Sharma, 1995; Shiau and Shen, 2001). Analytical expressions have been
derived for the marginal distribution of accumulated deficit (Bonaccorso et al., 2003) and
for the bivariate distributions of duration and accumulated deficit or intensity (Cancelliere
et al., 2003; Salas et al., 2004).
The return period of droughts can be defined as the expected value of elapsed time or
interarrival time between occurrences of critical events (e.g. Lloyd, 1970; Loaciga and
Marino, 1991; Fernandez and Salas, 1999; Shiau and Shen, 2001). A description of the
concept of interrarival time is provided by Figure 2 for specific critical droughts.

Figure 2. Interarrival time Td between critical droughts characterized by accumulated

deficit D > dc and duration 4 (adapted from Cancelliere and Salas, 2004)

Although such definition is in agreement with the one traditionally adopted for flood
studies, however when dealing with droughts a different approach must be followed in
order to derive analytical expressions for its estimation. Indeed, in evaluating the return

12
period of multiyear droughts, it is necessary to consider both drought duration and
accumulated deficit (or duration and intensity), since in this case it is not possible to
identify an unique time unit (trial) with respect to which the probability of a critical
drought can be expressed, as one can usually do in flood frequency analysis. Thus, the
usual formula T=1/P[A], where P[A] is the exceedence probability with respect to a critical
event A, e.g. a variable value greater than a fixed one, cannot be adopted.
Shiau and Shen, (2001) have developed an analytical expression for return period of
droughts with accumulated deficit greater than a given value. Other authors have extended
such formulation to the more general case of different combinations of duration and
accumulated deficit or intensity (Cancelliere et al., 2003; Gonzales and Valdes, 2003), or
to the case of droughts identified on periodic series, such as monthly precipitation or
streamflows (Cancelliere and Salas, 2004). With reference to the generic critical drought
event A identified on stationary (annual) and serially independent series, the return period
can be written as:

1 1
T=
p1 p 0 P[A]

where p1 is the probability of observing a surplus (i.e. P[h(i)>h0]), while p0=1- p1. In the
case of Markov serial dependence, the above equation modifies as:

p01 + p10 1
T=
p01 p10 P[A]

where p01 represents the transition probability of observing a surplus at time i given that a
deficit occurred at time i-1, while p10 represents the transition probability of observing a
deficit at time i given that a surplus occurred at time i-1.
For analyzing the return period of droughts and the associated risks we need to specify
the drought event A under consideration. For instance, one may consider only the duration
of a drought regardless of the accumulated deficit (or drought intensity) or vice-versa, or
both duration and accumulated deficit (or drought intensity).
In Table I, a list of possible drought events and related occurrence probabilities are
reported (Bonaccorso et al., 2003, Cancelliere et al., 2003). Such probabilities have been
derived by considering drought accumulated deficit, as well as drought accumulated deficit
conditioned on drought duration, gamma distributed with parameters, respectively, r1,1
and r2,2. In the table, the symbol G refers to the incomplete gamma function (Abramowitz
and Stegun, 1970).

13
Table I. Occurrence probabilities of critical drought events
Drough event Formula

1) A = {Ld= lc (lc=1,2,)} P[Ld = lc ] = f ( L = lc ) = p1 (1 p1 )lc 1

2) A = {Ld lc (lc=1,2,)} P[Ld lc ] = (1 p1 )lc 1

d
3) A = {D>dc} P[Dc > dc ] = f Dc ( z) d z = 1 G r1, c

dc 1 x

4) A = {D>dc and d
P[Dc > dc , Ld = lc ] = f Dc , Ld ( z, lc ) d z = 1 G r2 , c p1(1 p1 )lc 1
Ld= lc (lc=1,2,)} dc 2

5) A = {D>dc and d c
Ld lc (lc=1,2,)}
P[Dc > dc , Ld lc ] =
d c l =l c
f Dc , Ld ( z, l ) d z = 1 G r ,
l = lc
2 p1(1 p1 )l 1

2

The parameters of the probability distributions of drought characteristics can be

determined according to a non parametric approach, as functions of the sample mean and
variance of single deficit Dt and drought duration Ld according to the following theoretical
equations:
E[Ld ]2 E[Dt ]2 E[L d ] Var[Dt ] + E[Dt ]2 Var[L d ]
r1 = and 1 =
E[Ld ] Var[Dt ] + E[Dt ]2 Var[Ld ] E[L d ] E[Dt ]

E [Dt ]2 Var [Dt ]

r2 = l c and 2 =
Var [Dt ] E [Dt ]

Alternatively, the same parameters can be derived as a function of the probability

distribution of the underlying hydrological variable and of the threshold level adopted for
drought identification. In particular, in Table II and Table III the parameters of the gamma
distribution of D and D|Ld are reported, by assuming the underlying variable h(i) normal,
lognormal and gamma distributed respectively and the threshold level parametrized as follows
(Yevjevich, 1967):

x0 = x x = x (1 C v )

with x, x e Cv respectively the mean, the standard deviation and the coefficient of
variation of h(i).

14
Table II. Parameters of the gamma distribution of D for different probability distribution of h(i)
Distribut. h(i) r1 1 Other parameters

( ) p
Normal + ( )
2
( ) 2 ( )
p0 x 0
C v + C v
+ 1
p1 p 0
p0 p 02

2
(x,x)

p0 +
( )
p1
( ) 2 ( )
+ 1 ( ) p 0 = ( )

+ p
p0 p0 p02 0

2 p 2 1 ln (1 C v )
Lognormal
1 C v

2 2 p0 = y +
p0
x 0
1 C v
+e y 2 y
p1 p0 p0 2 p0 1 ln (1 C v )
2 2 = y +
(y,y) 2 2 y
p0 1 C v p1 e y
p0 1 Cv ln (1 C v )
3
p0 p2 p0
= y +
0 2 y

2

Gamma 1 Cv p 2 p 0 = P[rx , rx (1 C v )]
p0 x 0

p1
1 C v



p0
2 2
2
+
p0
Cv + 1

( ) = P[rx + 1, rx (1 C v )]
2 2 p0

(rx,x) p0 1 Cv p1


2
Cv + 1 = P[rx + 2, rx (1 C v )]
p0 p 2 p0
0 1 C v
p0

Table III. Parameters of the gamma distribution of D|Ld for different probability distribution of h(i)
Distribut. h(i) r2 2 Other parameters
2 ( ) 2 ( )
( )
Normal l c + x C v + 1
p 0 p0 p 2
(x,x) ( ) 2 ( )
0 p 0 = ( )
+ 1 ( )
2 +
p0 p p0
0

2 1 ln (1 C v )
Lognormal p0 = +
l c 1 C v 2 2 2
y
y
p 0 x +e y
2 1 ln (1 C v )
2 p0 p0 = y +
(y,y) 2 2 y
+e y
p2 p0 3 ln (1 C v )
0 1 Cv = y +
p 2 y
0

( )
2
2 2 p 0 = P[rx , rx (1 C v )]
Gamma l c 1 C v x + Cv + 1
p 0 p02 p0 = P[rx + 1, rx (1 C v )]
2 = P[rx + 2, rx (1 C v )]
(rx,x)
p 2
+
2
p0
(
Cv + 1

)
1 Cv


0 p 0

15
4. Drought analysis through the Standardized Precipitation Index (SPI)

4.1 Overview
Among the several proposed indices for drought monitoring, the Standardized
Precipitation Index (SPI) has found widespread application (McKee et al., 1993; Heim,
2000; Wilhite et al., 2000; Rossi and Cancelliere, 2002).
Unlike other drought indices, the computation of the SPI only requires the availability
of monthly precipitation series. The main advantages of the SPI are (i) its standardized
nature, which makes it particularly suited to compare drought conditions among different
time periods and regions with different climatic conditions, and (ii) the possibility to
consider different aggregation time scale for drought analysis, which allows to take into
account all the possible drought impacts and the related affected components of the
hydrological cycle. The aggregation time scale should be properly selected according to the
aim of the study: i.e., from few months for studies oriented to analyse agricultural droughts
(as the soil water content is affected by reduction in precipitation on a short time period),
till one year or more for hydrological droughts (since streamflow, ground water, and water
volumes stored in the reservoirs are mainly affected by precipitation anomalies over a long
time scale).
Table IV reports the drought classification based on SPI values adopted by the US
National Drought Mitigation Center (NDMC, http://www.ndmc.unl.edu).

Table IV. Drought classification according to the National Drought Mitigation Center

SPI Class
2.00 Extremely wet
Da 1.50 a 1.99 Very wet
Da 1.00 a 1.49 Moderately wet
Da -0.99 a 0.99 Near normal
Da -1.00 a -1.49 Moderately dry
Da -1.50 a -1.99 Very dry
-2.00 Extremely dry

In the following sections, a description of the procedure based on the application of

SPI for drought identification either at site and at a regional scale is presented.

16
4.2 Drought identification at site
The Standardized Precipitation Index is based on an equi-probability transformation
of accumulated monthly precipitation observed at a given site into a standard normal
variable. In practice, computation of the index requires: (i) fitting a probability distribution
to monthly precipitation series aggregated at a specific time scale k (e.g. k= 3, 6, 12, 24
months, etc.), (ii) computing the non-exceedence probability related to such accumulated
values, and (iii) defining the corresponding normal standardized quantiles as the SPI series.
In Figure 3 a scheme of the procedure for drought identification at site based on SPI
computation is described.

Figure 3. SPI computation scheme

McKee et al. (1993) has assumed accumulated precipitation gamma distributed and
used maximum likelihood method to estimate the parameters of the distribution. The
gamma cumulative distribution function is defined as:
x
1
F (x ) = t 1e t / dt
( ) 0
where:
>0 is the shape parameter
>0 is the scale parameter

( ) = y
1 y
e dy is the gamma function.
0

The maximum likelihood solutions for parameters estimation provide (Thom, 1966):

) 1 4A
= 1 + 1 +
4 A 3

17
 x
=

where:
xi is the monthly rainfall;
N

x i
A = ln( x ) i =1
with N equal to the number of rainfall observations.
N
Since the gamma probability distribution is undefined for x=0 and a rainfall distribution
may contain zeros, the cumulative distribution function becomes:

H ( x) = q + (1 q) F ( x)

where q is the probability of 0 rainfall values. Thus, if M is the number of zeros in a

rainfall series aggregated at a time scale k, the probability q can be estimated as M/N. The
cumulative distribution function is finally transformed to the standard normal variable Z,
which is the value of the SPI.

4.3 Regional drought analysis

A regional drought analysis can be carried out based on the SPI values computed for
a given month i and a given aggregation time scale k at different sites.
In particular, a similar approach adopted for the regional run method (see section 3.3)
can be considered. More specifically, once that the SPI series for a fixed aggregation time
scale k are computed at several sites, local drought conditions at each site p and at each
month i can be identified if SPI (i, p, k) < SPIth, where SPIth is a fixed value of SPI
considered as a threshold level for drought identification.
Drought conditions detected at each site p can be extended to influence areas S(p) (or
polygons) around the rainfall stations, so that a drought areal coverage Ad(i) for each
month i can be determined by summing polygons corresponding to the stations affected by
drought according to the observed SPI values:
P
Ad (i ) = I [SPI (i , p , k )] A( p )
p =1

where:

18
 S ( p)
A( p ) =
S TOT
I [SPI (i , p , k )] = 1 if SPI (i , p , k ) < 0
I [SPI (i , p , k )] = 0 if SPI (i , p , k ) 0

Finally, the drought areal coverage Ad(i) is compared to a fixed areal threshold Acri,
representing the value of the area above which a regional drought is considered to occur. If
Ad(i) is greater than or equal to Acri, then a regional SPI series for the considered
aggregation time scale, is computed based on the areal rainfall hareal obtained as the
weighted rainfall mean with respect to the polygons of the stations under drought
conditions, namely:
SPI areal (i , k ) = 1 [H ( X (i .k ))]
where:
k 1
X (i , k ) = hareal (i j ) is the k-months aggregated areal rainfall series at month I;
j =0

T
hareal (i ) = h(i, p )A( p ) is the areal rainfall value at month i computed on the basis of the
p =1

contemporary rainfall values at the T sites for which

SPI (i, p, k) < SPIth;
H ( X (i, k )) is the exceedence probability of X(i,k) assumed gamma distributed;

1 () is the standard normal quantile corresponding to H ( X (i, k )) .

Further, for each drought r:

the regional drought duration L(r) is defined as:

L(r) = if(r) - ii(r)+1

where if and ii are such that:

SPI(i)< SPIth for ii(r) i if(r) and SPI [ii(r) 1] SPIth 0, SPI[if(r) + 1] SPIth.

the mean areal coverage of drought can be computed as:

if
AD(r ) = Ad (i) / L(r )
i =ii

19
 Again as in the case of at site analysis, the above characteristics can be further
analyzed by taking the average and by considering their minimum and maximum values, in
order to characterize the whole period of observations.

5. Description of the software

5.1 Overview
The package REDIM is an user friendly software which allows to perform drought analysis
on hydrological series both at a site and over a region using the method of run and the SPI
index. In addition, it allows to test statistically for the existence of non-stationarities in a
time series, whose presence would lead to misleading drought analyses. The software,
written in Visual Basic, runs under Windows platforms, and is structured as a succession of
dialog boxes which guide the user throughout the analysis of stationarity and drought
identification and characterization steps. The main features of REDIM can be listed as
follows:
Data files are in MS Access format, which allows flexibility and compatibility with
existing databases, as well as Windows based programs such as spreadsheets, word
processors, etc..;
The Data Import function allows to create automatically MS Access database files from
existing ASCII text files, thus avoiding tedious database creation processes;
Different aggregation time scales can be used (monthly, three months and yearly), with
the possibility to select the initial month of aggregation in order to take into account
water years;
Testing for stationarity in hydrological series is carried out by means of six different
statistical tests namely: Students t-test for linear trend, Kendalls or rank correlation
test and turning point test, Mann-Withney rank-sum test for detecting the homogeneity
of the series, F test for detecting change in variance, and t test for detecting change in
mean;
Identification and characterization of drought is performed by means of run method or
the Standardize Precipitation Index.
Return period of at site drought characteristics is computed;
Graphical output of results allows to easily identify droughts on a given time series,
and/or region.

20
 Text reports is produced in MsWord format, to facilitate the merging of the outputs
with other existing documents.

In what follows a more detailed description of REDIM features is reported.

5.2 Software configuration

Redim allows to configure its main parameters which are then automatically stored to be
retrieved every time the software starts. By selecting Configure from the Setting menu,
the dialog box reported in Figure 4 appears.

Figure 4. Configure dialog box

The dialog box contains two tabs, one for general settings, the other one for setting the
default path of database files. The General tab, allows to set the pathnames for MS
Access and MS Excel programs. The Database Library tab allows to define the path
where the hydrological databases are stored.
The Configure dialog box appears automatically when the software is run for the first
time after installation. Afterwards, the parameters can be changed at any time, according to
the users need.

21
5.3 Data import
The data import function allows to convert data files in ASCIII text format into MS
Access database format compatible with REDIM. After clicking on Data Import under
the File menu, the dialog box in Figure 5 appears:

Figure 5. Data import dialog box

The user is prompted to enter the pathname of the source ASCII file and of the output MS
Access (.mdb) file. In case the exact pathname is unknown, by clicking the Browse button
a file selection dialog box appears. The user must also select which type of data are being
converted. If the database already exists and contains already the selected type of data, a
warning message appears after clicking Ok asking whether the user wants to replace the
existing data table. Appendix B reports the requires format of the ASCII data file.

5.4 Analysis of stationarity

After clicking on New analysis in the File menu, the main dialog box, depicted in Figure
6, appears:

22
Figure 6. Main dialog box

The user is prompted to select the type of analysis to perform and the typology of
hydrological data. Note that regional drought analysis can be performed only with
precipitation data. After selecting Analysis of stationarity and the hydrological variable,
by clicking next the dialog box in Figure 7 appears:

23
Figure 7. Aggregation time scale selection dialog box

This dialog box allows to select the aggregation time scale for the analysis.
Note that for the analysis of stationarity only annual (calendar or water year) and three
months scales are present. Also, it should be pointed out that at the three months scale, the
analysis is performed only on the trimester starting at the month selected by the user. In
other words, suppose the user selects Trimester and initial month March. The analysis will
be performed on the series constructed by summing the variable of March+April+May for
the first year, March+April+May for the second year and so on. This allows to compare
trend features of different seasons of the year. By choosing instead the Year time scale, the
analyzed series is obtained by summing over 12 months the values, starting at the selected
month. For example, if the interest lies in analyzing data for the water year starting in
October, the Year scale and October initial month should be selected. After clicking Next,
the user is prompted to enter the pathname of the datafile, while as usual the Browse button
can be used to facilitate this task. The following two dialog boxes, which are accessed by
clicking Next, prompt the user to enter the initial and final year of the analysis, as well as
the type of tests to perform and the significance level. Then, the dialog box showing the
results of the selected tests appears (Figure 8 and 9):

24
Figure 8. Tests for trend detecting results dialog box

25
Figure 9. Tests for trend detecting change in variance and mean results dialog box

In particular the tab "Tests-Detection of trend" (Figure 8) reports the results of the
application of the Students t-test for linear trend, Kendalls or rank correlation test and
of the turning point test while the tab "Tests-Detection of changes in variance and mean"
shows the results of the Mann-Withney rank-sum test for detecting the homogeneity of the
series, F test for detecting change in variance, and t test for detecting change in mean. For
each test, the intermediate results, as well as the test outcome are reported. For the
explanation of the symbols please refer to chapter 2 of the present document. Note that for
each test a green label will appear if the hypothesis is accepted, otherwise the label color
will be red.

By clicking on the Plot command button the form containing the plot of hydrological
variable time series and the trend line appears (Figure 10).

26
Figure 10. Hydrological variable time series and the trend line

5.5 Drought analysis at a site based on run-method

After selecting drought analysis at site, the run method for drought identification and the
type of hydrological variable in the main dialog box, the user is prompted to select the
aggregation time scale (monthly, three months and yearly) and the initial month of
aggregation. For yearly analysis the user can also select the checkbox for the evaluation of
drought return period.
Note that for drought analysis (either regional or at site), aggregation at the three months
time scales is carried out differently than in the case of analysis of stationarity. Indeed in
the present case, the series obtained by aggregating at three months is simply the
succession of trimesters starting at the selected initial month. For example, if the user
select three months and initial month April, the aggregated series will be the obtained by
summing variables values of April+May+June, July+August+September, and so on. Thus
the resulting series will be in general periodic. After selecting the aggregation time scale
the user must specify the data file and the initial/final year of the analysis, as for the
analysis of stationarity case. Then dialog box in Figure 11 appears:

27
Figure 11. Drought analysis options dialog box

Such dialog box contains three sub-dialog boxes or tabs: Selected data, Statistics and
Options (which is the one shown). Each tab can be accessed by simply clicking on the
name. The Selected data displays the aggregated data used for the analysis and is mainly
used for checking that the analyzed data is the correct one. The Statistics tab shows the
main statistics of the analyzed series such as mean, standard deviations, etc.. The Option
tab (shown above) allows to select the threshold level as well as the options to compute
return period (only yearly time scale). Note that a sample quantile corresponding to a
frequency level 50% is (by definition) the median. Regarding the return period evaluation
the user can select either the non-parametric approach or the parametric approach to
compute the parameters of the gamma distribution for accumulated deficit (please refer to
par. 3.4 for the details). In the former case (non parametric), such parameters will be
computed from the sample moments of the single deficit identified on the series. In the
latter case (parametric), the parameters of the gamma will be estimated by assuming a
normal, log normal or gamma distribution for the underlying hydrological series. In this
case, the user is prompted to select the appropriate distribution. After clicking Next, the
drought analysis results dialog box appears (Figure 12).

28
Figure 12. Results of at site drought analysis

Again three tabs are present. The Drought characteristics tab (shown above) contains the
number of identified droughts, a detailed list of their characteristics, as well as their mean,
max and min values. Furthermore (only yearly time scale), for each drought, the return
period corresponding to different combinations of drought characteristics is also shown.
Note that by clicking on the appropriate Sort by buttons it is possible to show the list sorted
in descending order according to the selected characteristic. By clicking on Save Report
button, it is possible to save a report file. The Generic information tab contains plots of the
thresholds, while the Plots tab shows a graph of the analyzed series, of the threshold and
red/blue values indicating whether the interval is deficit or surplus respectively (Figure
13). Droughts can be easily identified as a consecutive series of red values preceded and
followed by at least one blue value.

29
Figure 13. Hydrological variable time series

5.6 Regional drought analysis based on extended run-method

The first steps for performing regional drought analysis are basically the same as in the
case of at site analysis. Regional drought analysis can be performed only on precipitation
data. After selecting the type of analysis in the main dialog box and the aggregation time
scale, the dialog box in Figure 14 appears.

30
Figure 14. Selection of precipitation stations for regional drought analysis

By clicking on the data file on the left box the user can select the hydrological stations to
be used for the analysis. The two buttons on the right allow to remove stations from the
selection. Once the stations have been selected, the user can save the station list to be
retrieved ad at later time by clicking on the button with the disk symbol. To reload the
station list, the user must click on the "Station list" check button in the upper part of the
dialog box and select the file.
Note that the file shown are located in the directory specified in the Setting dialog box at
installation. If necessary, the user can change such directory by accessing the Setting menu
any time.

By clicking next a dialog box appears prompting for the total area of the region of interest
and for the influence areas of each station, expressed as a percentage of the total area. By
clicking Next, the areas expressed in km2 appear, as well as the common period of
observation. The time span to be analyzed can be changed at this stage if needed. By
clicking Next again, the dialog box in Figure 15 appears.

31
Figure 15. Regional drought analysis options dialog box

Note that the user must provide the areal threshold expressed as a percentage of the total
area. As in the case of at site analysis, the Option tab allows to select the threshold, the
Statistics show the threshold values and the Selected data allows to check that the analyzed
data is correct. Again, the median can be selected as a threshold by assuming a sample
quantile corresponding to a frequency level 50%. By clicking Next, the drought analysis
results appear (Figure 16).

32
Figure 16. Regional drought analysis results

The windows has three tabs. The tables tab (shown above) contains the number of
identified droughts, a detailed list of their characteristics, as well as their mean, max and
min values. Note that the by clicking on the appropriate Sort by buttons it is possible to list
the results sorted in descending order according to the selected characteristic. By clicking
on Save Report button, it is possible to save a report text file which can be visualized or
printed by using any text editors such as Notepad. The Plots tab shows the areal coverage
and cumulated deficit plots. An example of such plots is reported in Figure 17. The
Generic Information tab contains plots of the thresholds and other information, such as the
total number of periods and the extensions of the critical area.

33
Figure 17. Example of regional drought identification

5.7 Drought analysis at site through SPI index

After selecting drought analysis at site, the SPI index method for drought identification
and the type of hydrological variable in the main dialog box, the user is prompted to
specify the data file and then a dialog box containing the original data and the statistic is
showed. By clicking on the Next button the dialog box showed in Figure 18 appears.

34
Figure 18. SPI analysis at site results.

Such dialog box contains five sub-dialog box or tabs which show the results of the analysis
for five different aggregation scales. Each tab contains a table with the identification of
drought period with SPI < -1.00 and a table with mean and minimum value of the SPI
index and the duration for different classes. By changing the aggregation scale and clicking
on the Evaluate button the analysis is repeated for the new aggregation scale. By clicking
on the Plot button the graphical representation of the SPI time series appears. By
clicking on the Table button the graphical representation of the SPI time series
disappears and the table results appears again. By clicking on Save Report button, it is
possible to save a report file showing the results of the aggregation time scales selected in
the dialog-box showed in Figure 19.

35
Figure 19. Selection of the aggregation scales

5.8 Regional drought analysis through SPI index

After selecting the type of analysis in the main dialog box, and selecting the precipitation
station and their influence area as showed in paragraph 4.6 the dialog box showed in
Figure 20 appears.

Figure 20. Selected data, statistics and selection of threshold dialog box.

36
This dialog box contains two tabs: the former contains the selected data, the latter the
statistics of the selected data on a monthly scale. The last one contains also two combo-box
for the selection of the areal and SPI thresholds for the identification of regional drought .
By clicking on the Next button the dialog box showed in Figure 21 appears.

Figure 21. Regional drought identification through SPI index.

Such dialog box contains five sub-dialog box or tabs which show the results of the analysis
for five different aggregation scale. Each tab shows two tables. The former contains a
detailed list of the drought period characteristics, the latter contains the mean, maximum
and minimum values of duration, SPI index and the drought areal coverage. By changing
the aggregation scale and clicking on the Evaluate button the analysis is repeated for the
new aggregation scale. By clicking on the Plot button the graphical representation of the
SPI time series appears showing the areal coverage and SPI index evaluates by taking into
account the areal hydrological variable computed by considering only the sites for wich the
SPI values are below the fixed threshold. An example of such plots is reported in Figure
22. By clicking on the Table button the graphical representation of the SPI time series
disappears and the table results appears again. By clicking on Save Report button, it is
possible to save a report file of the aggregation time scales selected in the dialog-box
showed in Figure 19.

37
Figure 22. Regional drought identification through SPI index graphical results.

5.9 Graphical output

Whenever the user clicks the Save report button, different graphics are automatically
created by Redim, according to the type of analysis. Such plots are automatically
embedded in a RTF document.

38
6. References
Abramowitz, M., Stegun, I.A. (eds.), (1970). Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Tables, U.S. Department of Commerce.
Ben-Zvi, A., (1987), Indices of hydrological drought in Israel, J. of Hydrology, 92.
Bonaccorso, B., Cancelliere, A., Rossi, G., (2003). An analytical formulation of return
period of drought severity, Stochastic Environmental Research and Risk Assessment,
publ. no. 17, pp. 157-174
Cancelliere, A. and Salas, J. D., (2004). Drought length properties for periodic-stochastic
hydrological data." Water Resources Research, 10(2).
Cancelliere, A., Bonaccorso, B., Rossi, G., Salas, J.D., (2003). On the probabilistic
characterization of drought events. In Proceedings of 23th Annual American
Geophysical Union Hydrology Days, Fort Collins, Colorado, USA, March 31 - April
2, pp. 33-44.
Cancelliere A., A. Ancarani, G. Rossi, (1995), Identification of drought periods on
streamflow series at different time scales, in Tsiourtis, N. X. (ed.) Proc. of the EWRA
Symposium on Water Resources Management under Drought or Water Shortage
Conditions, Nicosia, Cyprus, 14-18 March 1995, Balkema, Rotterdam.
Cancelliere A., A. Ancarani, G. Rossi, (1998), Distribuzioni di probabilit delle
caratteristiche di siccit. Atti del XXVI Convegno di Idraulica e Costruzioni
Idrauliche, Catania 9-12 settembre, CUECM, Catania.
Clausen, B., C. P. Pearson, (1995), Regional frequency analysis of annual maximum
streamflow drought, J. of Hydrology, 173.
Dracup, J. A., K. S. Lie, E. D. Paulson, (1980), On the definition of droughts, Water
Resources Research, 16(2).
Helsel, D. R., and R. M. Hirsch, (1992). Statistical methods in water resources, Elsevier,
Amsterdam.
Kottegoda, N.T., (1980), Stochastic Water Resources Technology, The Macmillan Press,
London.
Maidment, D. R., (ed.), (1993). Handbook of Hydrology, McGraw-Hill Inc., New York.
Rodrigues, R., M.A. Santos, F.N. Correia, (1993), Appropriate time resolution for
stochastic drought analysis, in: J. D. Salas, R. Harboe, and J. Marco-Segura (eds.),
Stochastic Hydrology and its Use in Water Resources Systems Simulation and
Optimization, Kluwer, Netherland.

39
Rossi G., et al., (1992), On regional drought estimation and analysis, Water Resources
Management, 6.
Rossi, G., Cancelliere, A., (2003). At-site and regional drought identification by REDIM
model, in G. Rossi, et al. (eds), Tools for drought mitigation in Mediterranean
regions, Kluwer Academic Publishing, Dordrecht, pp. 37-57.
Thom, H.C., (1966), Some Methods of climatological analysis, World Meteorogical
Organization Note no. 81, World Meteorogical Organization, pp. 1-53.
Wilcoxon, F. (1945). Individual Comparisons by Ranking Methods. Biometrics 1, 80-83.
Yevjevich, V., (1967). An objective approach to definitions and investigations of
continental hydrologic drought, Hydrology paper n. 23, Colorado State University,
Fort Collins, Colorado.
Yevjevich V., L. Da Cunha, E. Vlachos (eds.), (1983), Coping with Droughts, Water
Resources Publications, Littleton, Colorado.
Zelenhasic E., A. Salvai, (1987), A method of streamflow drought analysis, Water
Resources Research, 23(1).

40
7. Appendix A: MS Access data file format

The generic database file must contain at least one of the following tables:
rain
deflussi (streamflows)
rese (spring yield)

Each table contains the monthly observations in a row (year) by column( month) layout
and has the following record structure:

Field 1: Anno year of observation

Field 2: Mese1 January observation
Field 3: Mese2 February observation
:::
Field 13: Mese12 December observation
Field 14: Gener The first two record of this field must contain the units and the
station name respectively
Field 15: Area Empty. Kept for compatibility with other software

Note that it is important that the first month of observation is January, otherwise the
computation of the water year by Redim might be incorrect.

41
8. Appendix B: text data file format

The text file to be used in the Data Import function must contain the monthly observations
in a row (year) by column (month) layout. Data must be separated by at least one space and
the decimal separator must be the dot (.). The first two line must contain the units and the
name of the station.
An example of text file containing 20 years of monthly precipitation data is reported
hereafter.

Cerami, [mm]

1921 62.8 41.9 139.4 71.9 43.6 118.3 51 40 100.7 93.9 81.1 94.4
1922 242 105 47 20 36 0 0 0 22 38 65 99
1923 124 105 97 78 0 22 23 48 48 18 96 179
1924 97 112 79 32 0 26 16 0 0 152 59 78
1925 15 53 56 63 98 0 0 4 63 121 107 65
1926 48 5 52 84 36 71 7 0 80 20 75 92
1927 68 45 14 78 88 2 0 60 15 192 357 214
1928 122 81 247 123 0 0 1 3 43 41 18 112
1929 63 40 28 7 78 10 3 23 13 79 57 75
1930 142 301 31 23 34 6 30 0 67 62 38 51
1931 222 244 79 41 24 3 9 0 23 16 87 155
1932 5 77 177 35 1 2 2 0 57 36 290 54
1933 54 98 52 49 2 31 16 145 54 13 137 323
1934 147 67 47 76 14 23 0 0 36 120 138 110
1935 212 57 250 0 9 0 132 20 38 70 207 88
1936 17 89 11 61 20 106 0 81 90 86 136 193
1937 20 83 29 64 63 24 5 0 85 93 153 162
1938 100 56 32 67 66 12 7 19 33 210 108 160
1939 56 92 98 60 73 40 0 9 135 36 79 106
1940 187 28 34 86 62 22 0 5 13 108 14 97

Again, please note that it is important that the first month of observation is January,
otherwise the computation of the water year by Redim might be incorrect. Also no extra
lines should be inserted before the station name or after the last year as this might cause
unpredictable results.

42
9. Appendix C: computation of sample quantiles

With reference to a series xt (t=1,2 .n) and to a frequency level F, the sample quantile xF
is computed by letting i=F*(n+1), i1=int(i), i2=i1+1 where int(.) is the integer part of the
argument. Then:
xF=xi1+(xi2-xi1)*(i-i1)

43