You are on page 1of 28

Water Resources Management

https://doi.org/10.1007/s11269-018-2089-z

Comparison of Drought Indices in a Semi-Arid River


Basin of India

Pawan S. Wable 1 & Madan K. Jha 1 & Ankit Shekhar 2

Received: 15 January 2018 / Accepted: 17 August 2018/


# Springer Nature B.V. 2018

Abstract
Due to the inherent complexity of drought phenomena, difference in hydro-climatic conditions
and watershed characteristics, there is a lack of a universal drought index for assessing drought
conditions in a particular region. Hence, the performance evaluation of different drought
indices is necessary for identifying a suitable drought index. In this study, the performance
of five drought indices was assessed for a semi-arid basin located in western India. The data
from nine raingauge stations and four climate stations for the period of 25 years (1985-2009)
were used. Based on the data availability, five meteorological drought indices were selected,
viz., Percent Departure from Normal (PDN), Effective Drought Index (EDI), Standardized
Precipitation Index (SPI), Reconnaissance Drought Index (RDI), and Standardized Precipita-
tion Evapotranspiration Index (SPEI). Among these drought indices, EDI is a time step
independent drought index and other four multi-time scale indices were defined at 1-, 3-, 6-,
9- and 12-month scales. The scale for the comparison of these drought indices was chosen
based on the correlation with EDI and within among the scales of multi-time scale drought
indices. The performance of drought indices during the historical drought was evaluated based
on the relative frequency of drought index in a particular drought severity class and the
response of these indices with multi-monthly rainfalls. In addition, the performance of these
drought indices was assessed using the decision criteria such as Robustness, Tractability,
Transparency, Sophistication, and Extendability. Analysis of the results indicated that the 9-
month scale is appropriate for comparing drought indices in the study area. SPEI-9 showed
maximum relative frequency in the ‘severe dry’ class and was found sensitive to 9-monthly
rainfall at most of the stations. Further, the results of the performance evaluation criteria
revealed that SPEI-9 has the highest total weighted score (136) followed by RDI-9, SPI-9,
EDI, and PDN-9. It is concluded that SPEI-9 is the most suitable drought index for monitoring
drought conditions in the study area.

Keywords Drought Indices . Percent Departure from Normal . Effective Drought Index .
Standardized Precipitation Index . Reconnaissance Drought Index . Standardized Precipitation
Evapotranspiration Index . Semi-arid region

* Pawan S. Wable
pawan.wable@gmail.com

Extended author information available on the last page of the article


Wable P.S. et al.

1 Introduction

Drought assessment plays an important role in the planning and management of water
resources. For assessment, some numerical standard is needed so that drought measures
can be compared between regions and can be compared with past drought events (Heim Jr
2002). For this, numerous drought indices have been developed in the last century and new
indices are being established based on new technological advances for quantifying drought
impacts, with each index having its own advantages and limitations. A drought index
value is typically a single number, which helps in taking decisions regarding mitigating
impacts of drought based on drought severity values (http://drought.unl.edu/
Planning/Monitoring/ComparisonofDroughtIndices.aspx). Drought indices use several
hydro-meteorological data such as rainfall, streamflow, reservoir storage, soil moisture,
groundwater, and water supply indicators either independently or collectively (WMO
1975). Generally, based on these physical datasets, drought indicators are classified into
three main indicators: meteorological drought indicators, agricultural drought indicators,
and hydrological drought indicators. The popular meteorological drought indicators are
Deciles Index (DI), Palmer Drought Severity Index (PDSI), Percent Departure from
Normal (PDN), Effective Drought Index (EDI), Standardized Precipitation Index (SPI),
Reconnaissance Drought Index (RDI) and Standardized Precipitation Evapotranspiration
Index (SPEI); agricultural drought indicators are Aridity Index (AI), Moisture Adequacy
Index (MAI), Crop Moisture Index (CMI), Crop Water Stress Index (CWSI); and hydro-
logical drought indicators are Standardized Water Level Index (SWLI), Surface Water
Supply Index (SWSI), Streamflow Drought Index (SDI) and Standardized Hydrological
Index (SHI). In the past few decades, several indices have also been developed using
remote sensing data such as Normalized Difference Vegetation Index (NDVI), Enhanced
Vegetation Index (EVI), Vegetation Condition Index (VCI), etc. These indices can be
integrated with the meteorological indices and can provide useful information if the
ground data are not available. However, the frequent cloud-cover during monsoon seasons
creates difficulty in interpreting vegetation cover from remote sensing images. Reviews of
drought indices in general and according to region/country are presented in Heim Jr (2002),
Smakhtin and Hughes (2004), Niemeyer (2008), and Mishra and Singh (2010).
Nevertheless, a drought index suitable for one region may not be applied specif-
ically for other regions due to the inherent complexity of drought phenomena,
difference in hydro-climatic conditions and watershed characteristics (Redmond
2002; Smakhtin and Hughes 2007). Therefore, several researchers across the globe
have compared different drought indices with an aim of finding suitable drought index
for a particular region or basin (e.g., Morid et al. 2006; Pandey et al. 2008; Roudier
and Mahe 2010; Barua et al. 2011; Vicente-Serrano et al. 2012; Jain et al. 2015).
Morid et al. (2006) compared seven indices, viz., EDI, SPI, PDN, DI, Z-index, China-
Z index (CZI) and modified CZI (MCZI) for six stations representing different
climates in Teheran province, Iran. They found that the performance of EDI was
better and it was more sensitive to the emerging drought condition compared to other
drought indices. The EDI was also found to be the best drought index in semi-arid,
sub-humid temperate and dry regions of India and Western Africa (Pandey et al.
2008; Jain et al. 2015; Roudier and Mahe 2010). Barua et al. (2011) compared five
drought indices, viz., PDN, DI, SPI, SWSI, and Aggregated Drought Index (ADI) for
the Yarra River Catchment, Australia. Based on decision criteria, they concluded that
Comparison of Drought Indices in a Semi-Arid River Basin of India

the ADI performed better in detecting historical drought events. In two different
studies on the comparative evaluation of SPI with RDI, representing different climate
conditions, the RDI was recommended in drought monitoring systems of Iran
(Jamshidi et al. 2011; Zarch et al. 2011). Vicente-Serrano et al. (2012) compared
SPEI, PDSI and SPI at the global scale for the 1901-2009 period and reported that
the SPEI had better capability in identifying drought conditions and suggested its use
if the drought variables for drought assessment are not a priori known.
Along with the comparison of drought indices, a couple of studies also focused on
the selection of scale for multi-time scale drought indices as it plays a vital role in the
evaluation of regional drought characteristics (Dogan et al. 2012; Jain et al. 2015).
These studies compared multi-month based drought indices, namely PDN, DI, Z-
Score, CZI, SPI, and a time-step independent index (EDI) to identify an appropriate
drought index for the semi-arid region of Konya closed basin (Dogan et al. 2012),
and the semi-arid and dry sub-humid climatic regions of Ken River basin, Central
India (Jain et al. 2015). Based on the correlation of EDI and multi-month drought
indices, they recommended to use at least 6- or 9-month time scale drought index for
evaluating droughts in arid/semi-arid regions and to avoid 1-month scale for compar-
ison unless there is a specific reason. Both of these studies indicated EDI as a more
suitable drought index for the respective study areas.
The aforementioned review of past literature reveals that although many studies on
comparison of drought indices have been carried out worldwide, but very few studies
have been reported from India up to now (e.g., Pandey et al., 2008; Jain et al. 2015).
Therefore, it calls for more and more studies in this direction in different agro-
climatic regions of India. Also, earlier studies did not consider multi-time scale
evapotranspiration-based indices such as RDI and SPEI, even if evapotranspiration
causes significant water resource losses in semi-arid/arid regions. Considering these
research gaps and increasing incidences of droughts in several parts of India (Chary
et al. 2010), including Maharashtra, the present study was carried out in an agrarian,
drought-prone and semi-arid river basin of Maharashtra located in Western India. The
specific objectives of the study are: (i) to identify a suitable scale for the comparison
of drought indices, (ii) to evaluate the performances of salient drought indices during
historical droughts, and (iii) to identify a suitable drought index for the study area.

2 Overview of Drought Indices

Based on the data availability in the study area, five meteorological drought indices/indicators,
viz., PDN (Percent Departure from Normal), EDI (Effective Drought Index), SPI (Standard-
ized Precipitation Index), SPEI (Standardized Precipitation Evapotranspiration Index), and
RDI (Reconnaissance Drought Index) and one hydrological index SDI (Streamflow Drought
Index) were selected to carry out this study. A brief description about these drought indices is
provided in the subsequent sub-sections.

2.1 Percent Departure from Normal

Percent departure from normal (PDN) is a simple and easy-to-use easy indicator of dry/wet
condition over a specified area for a given time. India Meteorological Department (IMD) uses
Wable P.S. et al.

this indicator to declare drought on a weekly/monthly/annual basis. PDN is calculated by


estimating rainfall departure from its long-term averages. According to IMD, if the percent
departure of rainfall in a year from the normal rainfall is <0%, 0-25%, 25-50% and > 50%; the
drought is classified as ‘No Drought’, ‘Mild Drought’, ‘Moderate Drought’ and ‘Severe
Drought’, respectively.

2.2 Standardized Precipitation Index

The understanding of a fact that a deficit of precipitation has different impacts on


groundwater, reservoir storage, soil moisture, snowpack, and streamflow led to the
development of Standardized Precipitation Index (SPI) (McKee et al. (1993). It has
been accepted worldwide and is recommended by the World Meteorological Organi-
zation (WMO) as the primary meteorological index for tracking meteorological
droughts (Hayes et al. 2012).
Conceptually, SPI is equivalent to the Z-score used in statistics and is formulated as,
X ij −μij
SPI ij ≈ ð1Þ
σij

where, SPIij is the SPI of ith month at jth time-scale, Xij is precipitation total for ith
month at jth time-scale, μijand σij are long-term mean and standard deviation associ-
ated with ith month at jth time-scale. As precipitation is typically not normally
distributed, the SPI overcomes this disadvantage by fitting to suitable distribution
(gamma) then transforming into normal or Gaussian distribution, and further calculat-
ed in a way as presented in Eq. (1).
Based on the SPI values, severity of drought are classified as ‘Extremely Wet’ for SPI > 2,
‘Very Wet’ for SPI = 1.5 to 1.99 and ‘Moderately Wet’ for SPI = 1.0 to 1.49, ‘Near Normal’ for
SPI = −0.99 to 0.99, ‘Moderately Dry’ for SPI = −1 to −1.49, ‘Severely Dry’ for SPI = −1.5 to
−1.99 and ‘Extremely Dry’ for SPI < −2.

2.3 Reconnaissance Drought Index

Reconnaissance Drought Index (RDI) was developed by Tsakiris and Vangelis (2005)
to overcome the limitations of SPI and PDSI, as the former considers only precipi-
tation for water deficit, while the latter is not “sensitive enough” for monitoring of
drought. It is based on the ratio between cumulative values of precipitation and
evapotranspiration (PET). The initial value of the RDI is obtained by calculating a
ratio ‘ak’ between the precipitation in a given area and the total potential evapotrans-
piration for each consecutive period of the k months in a year. It is mathematically
expressed as:
j¼k
∑ Pij
j¼1
ak ¼ j¼k
ð2Þ
∑ PET ij
j¼1

Where, Pij and PETij are the precipitation and the potential evapotranspiration, re-
spectively for the jth month of the ith year.
Comparison of Drought Indices in a Semi-Arid River Basin of India

In several locations, the values of ak has been found to follow either lognormal or
Gamma distribution. This ratio is standardized similar to the equations used for
standardizing SPI to obtain the values of RDI. It has same drought severity classifi-
cation as that of SPI and it can also be defined at multiple scales.

2.4 Standardized Precipitation Evapotranspiration Index

Standardized Precipitation Evapotranspiration Index (SPEI) was proposed by Vicente-Serrano


et al. (2010a) to combine the sensitivity of PDSI to changes in evaporation demand and multi-
time scale property of SPI. They also stated that SPEI is particularly suited for detecting and
monitoring the effects of global warming on drought condition. It represents a simple climatic
water balance computed as the weekly or monthly difference between cumulative precipitation
and potential evapotranspiration. It is mathematically expressed as follows:

Di ¼ Pi −PET i ð3Þ

The calculated ‘D’ values are aggregated at different time scales as:
k−1
Dkn ¼ ∑ ðPn−i −PET n−i Þ ð4Þ
i¼0

Where, k (months) is the timescale of the aggregation and n is the calculation month.
SPEI is calculated similar to SPI. However, a three-parameter distribution is needed to
standardize D-series as D-values can have negative values. Globally, the three-parameter log-
logistic distribution has been found to be a better fit for SPEI at all time scales using
Kolmogorov–Smirnov test (Vicente-Serrano et al. 2010b). The drought severity classification
based on SPEI values is similar to the SPI classification and it can be defined at multiple
scales.

2.5 Effective Drought Index

Effective Drought Index (EDI) was formulated by Byun and Wilhite (1999), which is an
effectively non-parametric index. It was originally developed for daily time step, but can
also work on monthly scale (Smakhtin and Hughes 2004; Pandey et al. 2008). It is based
on the concept of effective precipitation (EP), which is a function of current day and
previous day’s rainfall, but with lower weights [Eq. (5)]. It is very clear from Eq. (5) that
the computation of the EP, which is the most important aspect of EDI, utilizes the
concept of available water as a function of precipitation and time. According to the
concept of EP, the current day’s precipitation contributes fully (weight of 1) to the
available water, whereas the second day’s contribution is less (weight of 0.85) and the
third day’s contribution is even less (weight of 0.77) and so on with the contribution of
the precipitation that occurred a year ago being the least (weight of 0.000423). This
duration is either 365 days, a representative value of the total water resources available
or stored for a longer period, or it can be 15 days, a representative of a short period. It is
noteworthy that rainfall-runoff models exhibit a similar effect on EP in representing the
decay of available water resources over the passage of time (Deo et al. 2015). This
conceptualization allows for the robust analysis of drought risk due to water scarcity
conditions (Pandey et al. 2008; Dogan et al. 2012).
Wable P.S. et al.

  
i n
EPi ¼ ∑ ∑ Pm =n
n¼1
 m¼1 
1 1 1
¼ P1 1 þ þ þ … þ
2 3 365 ð5Þ
   
1 1 1 1
þ P2 þ þ…þ þ … þ P365
2 3 365 365
≈P1 þ 0:85P2 þ 0:77P3 þ … þ 0:000423P365

Further, following equations were used to formulate EDI:


DEPi ¼ EPi −MEP ð6Þ

DEPi
EDI i ¼ ð7Þ
ST ðDEPi Þ

Where MEP is the mean EP, ST is the standard deviation derived for each day’s deviation of
EP (DEP) and ‘i’ is the particular day.
EDI requires at least 30 years data. Like other indices, EDI is a standardized index, which
facilitates comparison of severity between two regions despite of different climates. It has also
similar severity classes as those of SPI (Byun and Kim 2010).

2.6 Streamflow Drought Index (SDI)

Streamflow Drought Index (SDI) was proposed by Nalbantis (2008) for characterizing
hydrological drought. It is based on the monthly streamflow volumes, (Qi,j) available, where
i denotes the hydrological year and j denotes the month within hydrological year. The
cumulative streamflow (Vi,k) volume for the ith hydrological year and the kth reference period
with four overlapping time periods is given as:
3k
V i ;k ¼ ∑ Qi; j i ¼ 1; 2; …; j ¼ 1; 2; …; 12; k ¼ 1; 2; 3; 4 ð8Þ
j¼1

The following equation is used to calculate SDI:

V i;k −V k
SDI i ;k ¼ ð9Þ
sk
Where, V k and sk are respectively the mean and standard deviation of cumulative streamflow
volumes of reference period k. In this definition, V k is the truncation level set, whereas other
values could be used.
Usually, Gamma and lognormal distributions are used for representing streamflow. How-
ever, in case of lognormal distribution, the normalization is easy. In this case, SDI is defined as:

yi;k −yk
SDI i ;k ¼ ; i ¼ 1; 2; …; k ¼ 1; 2; 3; 4 ð10Þ
sy;k


yk ¼ ln V i;k ; i ¼ 1; 2; …; k ¼ 1; 2; 3; 4 ð11Þ
Comparison of Drought Indices in a Semi-Arid River Basin of India

Where, yk is the natural logarithm of cumulative streamflow with mean yk and the standard
deviation Syk. SDI also has same drought-severity classes as those of SPI.

3 Methodology

3.1 Study Area and Data Collection

For this study, the Sina River basin was selected as the study area. Sina River is the largest
tributary of Bhima River, which originates near Ahmednagar city and flows southeast for
300 km through Ahmednagar and Solapur districts to join Bhima River in the south of Solapur
(near the Maharashtra and Karnataka boundary). This basin is located between 17° 28′ N and
19° 16′ N latitude, 74° 28′ E and 76° 7′ E longitude and falls in the drought prone region of
Maharashtra, Western India (Fig. 1). The basin has an area of 12,244 km2 with the topographic
elevation ranging from 420 to 964 m (above MSL). The climate of the study area is
characterized as semi-arid. The study area has three seasons: summer, winter, and rainy
(monsoon). The summer season commences from 1st week of March and lasts until 1st week
of June. The winter season begins from November and ends in February. The average
maximum and minimum ambient temperatures are 40.5 °C in the month of May and
10.5 °C in the month of December, respectively. The rainy season extends from mid-June to
October-end. The average annual rainfall of the study area is 644 mm.
Hydro-meteorological data used in this study were collected from various govern-
ment organizations/agencies. Figure 1 shows the locations of the climate and
raingauge stations, whereas Tables 1 and 2 present the information related to their
latitude, longitude, elevation, and annual time series characteristics. Daily rainfall data
of nine rain gauge stations for the period of 1985-2009 were collected from India
Meteorology Department (IMD), Pune and State Data Storage Center, Hydrology
Project (HP), Nashik, India. The daily temperature data of Solapur station were
collected from the IMD, Pune. However, the temperature data at Kashti and Rosa
meteorological stations were collected from the State Data Storage Centre, Nashik and
those for the Rahuri meteorological station were acquired from the Mahatma Phule
Agricultural University, Rahuri. The streamflow data of Wadakbal gauging station,
located at the extreme downstream of Sina River (Fig. 1), for the period of 1985-2009
were procured from Centre Water Commission, Hyderabad, India.

3.2 Performance Evaluation of Drought Indices

3.2.1 Calculation of Drought Indices

In this study, five meteorological indices/indicators, viz., PDN (Percent Departure from
Normal), EDI (Effective Drought Index), SPI (Standardized Precipitation Index), SPEI (Stan-
dardized Precipitation Evapotranspiration Index), and RDI (Reconnaissance Drought Index)
and one hydrological index SDI (Streamflow Drought Index) were used to assess their
performance during historical drought periods.
Thiessen polygons were created for the nine rainguage stations in the study area as shown
in Fig. 2a. The stations representing smaller area (Supa, Kolgaon, Bandalgi stations) were
merged into neighboring larger station’s area by employing paired t-test for testing their mean
Wable P.S. et al.

INDIA Gujarat
Madhya Pradesh
China
Pakistan
Nepal

Maharashtra Chhattis-
-garh
Andhra Pradesh
Arabian Sea
Arabian Sea Bay of Bengal

Karnataka

Fig. 1 Map of the study area showing block boundaries and the locations of monitoring sites

Table 1 Location of climate station and annual temperature series (1985–2009)

Station Latitude (oN) Longitude (°E) Elevation (m MSL) Annual Tmax (°C) Annual Tmin (°C)

1. Kashti 18.57 74.57 523.15 32.76 (± 3.51) 18.52 (± 4.6)


2. Rahuri 19.34 74.65 527.73 32.27 (± 3.30) 16.99 (± 4.79)
3. Rosa 18.28 75.43 509.71 32.69 (± 3.74) 19.88 (± 4.26)
4. Solapur 17.67 75.90 462 34.07 (± 3.42) 20.75 (± 3.24)
Comparison of Drought Indices in a Semi-Arid River Basin of India
Table 2 Location of meteorological station and statistics of annual rainfall series (1985–2009)

Station Latitude (oN) Longitude (°E) Elevation (m MSL) Mean (mm) Standard deviation (mm) Skewness coefficient Kurtosis coefficient

1. Alni 18.33 76.00 677 765.9 285.11 1.35 2.16


2. Bandalgi 17.48 75.92 425 621.5 217.32 1.10 0.36
3. Chinchondipatil 18.99 74.96 721 441.9 209.78 0.47 0.19
4. Jamkhed 18.74 75.33 628 615.6 187.82 0.99 1.15
5. Kasegaon 17.82 75.97 496 603.1 274.49 1.03 1.69
6. Kolgaon 18.81 74.67 634 387.3 156.76 0.33 −0.90
7. Solapur 17.67 75.90 462 628.4 228.63 1.23 1.81
8. Supa 18.99 74.53 775 388 181.66 0.26 −0.65
9. Temburni 18.03 75.22 514 535.2 184.55 0.36 0.04
Wable P.S. et al.

Fig. 2 Thiessen polygons for a Raingauge stations and b Climate stations

annual rainfall significance. Such stations pair are Chinchondipatil-Supa, Chinchondipatil-


Kolgaon, Solapur–Bandalgi and their calculated t-test values for the mean significance are
2.052, 1.490, and 0.290, respectively. Notably, no significant difference is found, as the
calculated values are less than critical value (2.064) at 5% level of significance. Hence, the
results obtained from performance analysis of drought indices at Chinchondipatil station were
assigned for the Kolgaon and Supa stations and those of Solapur station were assigned for the
Bandalgi station.
In order to define Reconnaissance Drought Index (RDI) [Eq. (2)] and Standardized
Precipitation Evapotranspiration Index (SPEI) [Eq. (3)], the estimation of Potential Evapo-
transpiration (PET) is needed. It is worth to mention that RDI is not dependent on the PET
estimation method (Vangelis et al. 2013). However, in case of SPEI, the use of Thornthwaite
method proposed by Vicente-Serrano et al. (2010a) was later revisited by Beguería et al.
(2014). They showed that on using different PET methods, there is a significant difference in
the estimated SPEI series. Hence, they primarily proposed Penman-Monteith method for the
estimation of SPEI, while suggested alternative (temperature based) methods like Hargreaves
method as first choice and the Thornthwaite method as second choice. In this study, since only
maximum and minimum temperature data are available for the study area, PET was estimated
using Hargreaves method. The Hargreaves equation is given as (Hargreaves and Samani
1985):

PET ¼ 0:0023Ra  ðTmean þ 17:8Þ  ðTmax ‐Tmin Þ0:5 ð12Þ

Where, PET = Potential Evapotranspiration, Ra = extraterrestrial radiation (mm/day), Tmax =


maximum ambient temperature (°C), Tmin = minimum ambient temperature (°C). It is worth to
mention that the concept of PET has been modified (Allen et al. 1998) and PET is replaced
with Reference Evapotranspiration (ETo).
Comparison of Drought Indices in a Semi-Arid River Basin of India

However, the temperature data were not available at all the raingauge stations.
Therefore, the Thiessen polygons of climate stations were created and their
weights were assigned to the corresponding raingauge stations as depicted in
Fig. 2b. Thus, the Rahuri climate station represents Chinchondipatil rainguage
station, the Rosa climate station represents Alni, Jamkhed, and Temborni
raingauge stations, and the Solapur climate station represents Solapur and
Kasegaon raingauge stations.
Percent Departure from Normal (PDN) values for multiple time scales (1, 3, 6, 9,
and 12 months) were calculated by comparing the moving monthly sum (of a
particular scale) with the long-term moving month sum. For example, PDN for the
3-month scale (PDN-3) of March month of a particular year will be the departure
value of sum of rainfalls of January, February, and March (3 months) of that year
with the long-term rainfall sum of the specified 3 months.
For defining EDI, a dummy water deficit period of 365 days was selected as it
represents dominant precipitation cycle (Morid et al. 2006). For standardization, the
remaining four indices, i.e., SPI, SPEI, RDI, and SDI need fitting of a probability
distribution. The variables used for distribution fitting and the selected distributions
for these indices are shown in Table 3. These selected distributions were tested by the
nonparametric goodness-of-fit test, Kolmogorov-Smirnov (K-S) test, which has been
widely used (e.g., Mishra et al. 2009; Vicente-Serrano et al. 2010b). In the K-S test,
if the maximum deviation between empirical distribution and specified cumulative
distribution (i.e., KS statistic value) is greater than or equal to the critical tabulated
value of the KS statistic, the hypothesis of following the specified probabilistic
distribution is rejected.
Software from Spanish National Research Council website (available at
http://digital.csic.es/handle/10261/10006;http://digital.csic.es/handle/10261/10002), and
Korean Daily Drought Monitoring website (available at http://atmos.pknu.ac.
kr/~intra) were used to calculate drought indices. For calculating EDI, 30 years data
are needed. However, at some stations initial five years rainfall data (1980-84) was
missing. Hence, as suggested by the experts, missing values of rainfall were filled by
the normal values of the corresponding month.
Similar to PDN, the drought indices SPI, RDI, and SPEI were also defined for 1,
3, 6, 9, and 12 months, whereas EDI, which is a time-step independent index, was
calculated using monthly time step. Thus, the use of four indices at five different time
scales and EDI with self-defined one-time step resulted in a total of 21 drought index
series. As EDI is scale-free index, it can be compared with any scale of other drought

Table 3 Variables used for distribution fitting and probability distribution selected for SPI, RDI, SPEI and SDI

Drought Index Variable Probability Distribution


Selected

1. Standardized Precipitation Index (SPI) Rainfall Gamma


2. Reconnaissance Drought Index (RDI) Ratio of Rainfall to ET Gamma
3. Standardized Precipitation Evapotranspiration Difference of Rainfall and Log-Logistic (3-parameter)
Index (SPEI) ET
4. Streamflow Drought Index (SDI) Streamflow Log-normal
Wable P.S. et al.

indices. Further, to select a suitable scale out of 1, 3, 6, 9, and 12 months, the


Pearson correlation coefficients with EDI and among the different scales of all
drought indices were evaluated (Jain et al. 2015). Using four indices and five time
scales, the number of drought time series was 20 and by considering one monthly
step of EDI, a correlation matrix of 21 rows and 21 columns was created. The first
drought severity value for multi-scalar drought indices at 1, 3, 6, 9, 12 months were
defined from Jan-1985, Mar-1985, June-1985, Sept-1985 and Dec-1985, respectively.
Consequently, to have equal number values for calculating correlation among drought
indices at different scales was computed from Jan-1986 to Dec 2009. It was computed
jointly between all 21-drought index series by creating correlation matrix. Pearson
correlation coefficient was calculated as:
  
1 n xi −xm yi −ym
r¼ ∑ ð13Þ
n i¼1 σx σy

Where, xi and yi represent the values of arrays with ‘n’ number of elements being
compared, and xmand ym are respectively mean values of two arrays, and σx and σy
are the standard deviations of xi and yi, respectively.

3.2.2 Comparative Evaluation of Drought Indices

After selecting a suitable scale, the meteorological indices were examined for their
ability to model historical drought events. The ability of the drought indices was
evaluated based on their ‘relative frequency’ in drought-severity class and their
response to the variation in rainfall. ‘Relative frequency’ indicates percentage of
drought index that remains in a particular drought-severity class for the considered
period. As, drought index have different ranges for defining severity of drought
events, they are classified into different wet and dry classes for the comparison
purpose (Table 4). It is evident from Table 4 that the four indices, SPI, RDI, SPEI,
and EDI have same severity classification. However, as mention in Section 2.1, PDN
is categorized for only drought or dry classes. Therefore, in order to compare with
other indices, PDN classification was modified to include three wet classes. For this,
25% was increased to each wet class to yield ‘Moderately Wet’, ‘Extremely Wet’, and
‘Severely Wet’ classes with 125 to 150%, 150 to 175%, and > 175% departures from
the normal rainfall, respectively. The modified classes of PDN with its value of ranges
in decimals is shown in the last column of Table 4.

Table 4 Drought severity classification

Class Values of SPI, RDI, SPEI and EDI Value of PDN

1. Extremely Wet (EW) ≥2 >1.75


2. Severe Wet (SW) 1.5 to 1.99 1.50 to 1.75
3. Moderately Wet (MW) 1.0 to 1.49 1.25 to 1.50
4. Normal (N) 0.99 to 0.99 0.75 to 1.25
5. Moderately Dry (MD) −1 to −1.49 0.50 to 0.75
6. Severely Dry (SD) −1.5 to −1.99 0.25 to 0.50
7. Extremely Dry (ED) ≤ −2 < 0.25
Comparison of Drought Indices in a Semi-Arid River Basin of India

3.2.3 Selection of Suitable Drought Index

Although numerous drought indices have been developed, relatively less focus has
been applied to its practical decision making application as each index has its own
limitations (Steinemann et al. 2015). Thus, drought indices used for assessing drought
conditions in the study area were evaluated based on the desirable properties of a
good drought index. For this evaluation, six decision criterion namely, robustness,
tractability, transparency, sophistication, extendabilty and dimensionality wereused
(Keyantash and Dracup 2002). These criteria owe their origin to Redmond (1991),
which proposed list of desirable qualities a drought indicator should possess to be
useful for practical decision making. However, as suggested by Keyantash and Dracup
(2002) and Barua et al. (2011), transparency covers dimensionality and hence, the
dimensionality was not included in the evaluation criteria. In order to consider the
relative importance of decision criterion, relative weights were defined based on
professional experience and personal judgment. It is important to note that researchers
and practitioners are free to modify these weights to suit their own perspectives
(Keyantash and Dracup 2002; Barua et al. 2011). Raw scores ranging from 1
(weakest) to 5 (strongest) were assigned to all the criteria. These scores were
multiplied by their respective weights and were added to get total weighted score
for the evaluation of drought indices. The drought index with highest total weighted
score was considered as a best drought index. A brief description about these each
evaluation criteria is given below (Keyantash and Dracup 2002; Barua et al. 2011).

Robustness Robust decision, a decision that is as resistant to uncertainty as is possible


and looks good to all constituents long after it is made (https://en.wikipedia.org/wiki/
Robustness). Ideally, it means the index should be responsive but not temperamental
to drought condition. It is an important criterion for suitable drought selection, but it
is not sufficient.

Tractability It is ease of obtaining a mathematical solution (https://en.wikipedia.org/wiki/


Tractable). Tractability deals with the feasible expression of the drought index. A tractable
index should not be high-level computational demanding as well as should not require more
extensive database.

Transparency It represents general utility. This evaluation criterion implies that drought index
should be not just understandable to research field only but also to common people, who get
affected by the drought. It gives more weightage to the index, which is more used for
communication purpose in the publications or broadcasts.

Sophistication This evaluation criterion relates to the theory behind developing a particular
drought index. It is partially contrary to transparency, but has to be considered due to
conceptual advantage. A drought index may not be transparent, but it may be sophisticated.
The level of sophistication included in defining a drought index should also be supported by
the quality of available data and fundamental accuracy of assessment methods.

Extendability Extendability specifically covers one of the aspects of tractability. It is the


degree to which the index may be extended across time to predict or forecast future drought
Wable P.S. et al.

scenarios. Thus, the extendability of a drought index is primarily dependent upon the avail-
ability of the historical data used for the construction of the index and secondarily on its
calculation procedure.

4 Results and Discussion

4.1 Suitability of Distribution for the Drought Variables

Suitability of distribution fitting to the rainfall, ratio of rainfall to ETo, and difference of rainfall
and ETo (needed for defining SPI, RDI and SPEI), was tested using K-S goodness-of-fit test.
The suitability of gamma distribution was checked for the rainfall and the ratio of rainfall to
ETo, whereas that of log-logistic distribution was examined for the difference in rainfall and
ETo. The results summarized in Table 5 reveal that at all the stations, variables follow the stated
distributions at 1% level of significance as the values of K-S are less than the corresponding
critical values, except at Kasegaon station, where the variable (Rainfall − ETo) does not follow
log-logistic distribution. As the K-S value is slightly greater than the critical value at the
Kasegaon station for Rainfall − ETo, the log-logistic distribution is considered suitable. It
should be noted that the sample size varies for the ‘rainfall’ and ratio of ‘rainfall to ETo’, since
the gamma distribution can be defined for only non-zero values. Hence, all the meteorological
indices were defined at six raingauge stations, viz., Alni, Chinchondipatil, Jamkhed,
Kasegaon, Solapur and Tembhurni. Further, streamflow volume needed for defining SDI
during monsoon season was also found to follow lognormal distribution at 1% level of
significance.

4.2 Selecting Scale for Drought Indices

SPI, RDI and SPEI are multi-time scale indices, whereas EDI is a time-step independent index.
PDN can also be calculated for 1, 3, 6, 9 and 12 months, in which the departure is calculated
from the corresponding multi-monthly long-term mean. For example, for 3-month rainfall, a
sum of 3 months for a particular year is compared with the long-term mean of sum of 3 months.
In India, there is a high seasonal variability from one month to another. Therefore, these multi-
time scale indices were defined at 1, 3, 6, 9 and 12 months’ time scales (Jain et al. 2015).
To select a suitable scale among five time scales (1, 3, 6, 9 and 12) of SPI, RDI, SPEI, and
PDN for comparison, their correlation with EDI was evaluated (Jain et al. 2015). Thus, with
total 21 drought series, the correlation matrix of 21 rows and 21 columns was created for all
the six stations. The correlation matrix for the Jamkhed station as an example is shown in
Table 6. It is important to note that the values of correlation coefficient of EDI with other
indices are mostly higher than all-scale indices, except for similar time-scale indices (orange
color shading). Also, in multi-time scale indices for a given month scale, the correlation
between similar time scales is higher than that of dissimilar scales. For example, SPI-1
(Column 1 of Table 6) with RDI-1, SPEI-1, and PDN-1 has correlation coefficient (r) values
greater than 0.55 as compared to other scale indices. Similarly, for 3, 6, 9, and 12 months of
SPI, the correlation coefficient values are greater than 0.80, 0.88, 0.94 and 0.94, respectively as
compared to dissimilar time scales. This suggests that the estimates obtained from similar time
scales are comparable with similar scales estimates only. The correlation between SPI and RDI
Comparison of Drought Indices in a Semi-Arid River Basin of India
Table 5 Results of the goodness-of-fit test for distribution fitting to the drought variables

Station ‘Rainfall’ and ‘Ratio of Rainfall to ETo’ ‘Rainfall − ETo’

Sample Size Critical Value K-S Value (Rainfall) K-S Value (Rainfall/ETo) Sample Size Critical Value K-S Value

1. Alni 201 0.115 0.087 0.072 300 0.094 0.069


2. Chinchondipatil 147 0.134 0.046 0.041 300 0.094 0.082
3. Jamkhed 176 0.123 0.098 0.099 300 0.094 0.073
4. Kasegaon 139 0.138 0.06 0.073 300 0.094 0.097
5. Solapur 223 0.109 0.052 0.077 300 0.094 0.068
6. Tembhurni 156 0.131 0.100 0.087 300 0.094 0.068

Bold values indicate the designated distribution does not fit to ‘Rainfall − ETo’
Table 6 Correlation matrix for all scales drought indices along with EDI for Jamkhed station

Jamkhed Station 1-Month 3-Month 6-Month 9-Month 12-Month

SPI RDI SPEI PDN SPI RDI SPEI PDN SPI RDI SPEI PDN SPI RDI SPEI PDN SPI RDI SPEI PDN EDI

1-Month SPI 1.00 1.00 0.74 0.55 0.50 0.51 0.45 0.38 0.36 0.36 0.33 0.30 0.31 0.32 0.31 0.32 0.24 0.24 0.24 0.24 0.49
RDI 1.00 1.00 0.75 0.56 0.51 0.52 0.46 0.39 0.37 0.37 0.35 0.30 0.32 0.33 0.33 0.32 0.25 0.26 0.26 0.26 0.49
SPEI 0.74 0.75 1.00 0.56 0.52 0.54 0.66 0.42 0.39 0.39 0.46 0.28 0.38 0.42 0.45 0.37 0.31 0.35 0.39 0.30 0.54
PDN 0.55 0.56 0.56 1.00 0.41 0.40 0.31 0.40 0.25 0.23 0.25 0.23 0.27 0.26 0.27 0.27 0.25 0.24 0.23 0.25 0.40
3-Month SPI 0.50 0.51 0.52 0.41 1.00 0.99 0.81 0.80 0.64 0.62 0.59 0.50 0.54 0.54 0.53 0.52 0.48 0.47 0.47 0.48 0.71
RDI 0.51 0.52 0.54 0.40 0.99 1.00 0.82 0.80 0.65 0.64 0.62 0.50 0.55 0.56 0.56 0.52 0.49 0.50 0.50 0.50 0.72
SPEI 0.45 0.46 0.66 0.31 0.81 0.82 1.00 0.68 0.60 0.61 0.73 0.44 0.57 0.61 0.65 0.53 0.52 0.56 0.60 0.51 0.70
PDN 0.38 0.39 0.42 0.40 0.80 0.80 0.68 1.00 0.48 0.45 0.46 0.44 0.40 0.40 0.42 0.40 0.39 0.39 0.39 0.39 0.58
6-Month SPI 0.36 0.37 0.39 0.25 0.64 0.65 0.60 0.48 1.00 0.97 0.89 0.88 0.76 0.75 0.73 0.72 0.67 0.66 0.65 0.67 0.83
RDI 0.36 0.37 0.39 0.23 0.62 0.64 0.61 0.45 0.97 1.00 0.90 0.86 0.75 0.76 0.75 0.71 0.67 0.68 0.67 0.65 0.82
SPEI 0.33 0.35 0.46 0.25 0.59 0.62 0.73 0.46 0.89 0.90 1.00 0.79 0.78 0.81 0.85 0.72 0.72 0.76 0.78 0.70 0.83
PDN 0.30 0.30 0.28 0.23 0.50 0.50 0.44 0.44 0.88 0.86 0.79 1.00 0.64 0.63 0.62 0.66 0.58 0.57 0.55 0.59 0.76
9-Month SPI 0.31 0.32 0.38 0.27 0.54 0.55 0.57 0.40 0.76 0.75 0.78 0.64 1.00 0.98 0.94 0.96 0.86 0.84 0.80 0.85 0.89
RDI 0.32 0.33 0.42 0.26 0.54 0.56 0.61 0.40 0.75 0.76 0.81 0.63 0.98 1.00 0.97 0.94 0.85 0.87 0.84 0.83 0.88
SPEI 0.31 0.33 0.45 0.27 0.53 0.56 0.65 0.42 0.73 0.75 0.85 0.62 0.94 0.97 1.00 0.90 0.85 0.89 0.90 0.83 0.86
PDN 0.32 0.32 0.37 0.27 0.52 0.52 0.53 0.40 0.72 0.71 0.72 0.66 0.96 0.94 0.90 1.00 0.83 0.81 0.77 0.84 0.88
12-Month SPI 0.24 0.25 0.31 0.25 0.48 0.49 0.52 0.39 0.67 0.67 0.72 0.58 0.86 0.85 0.85 0.83 1.00 0.98 0.94 0.99 0.83
RDI 0.24 0.26 0.35 0.24 0.47 0.50 0.56 0.39 0.66 0.68 0.76 0.57 0.84 0.87 0.89 0.81 0.98 1.00 0.97 0.96 0.81
SPEI 0.24 0.26 0.39 0.23 0.47 0.50 0.60 0.39 0.65 0.67 0.78 0.55 0.80 0.84 0.90 0.77 0.94 0.97 1.00 0.92 0.78
PDN 0.24 0.26 0.30 0.25 0.48 0.50 0.51 0.39 0.67 0.65 0.70 0.59 0.85 0.83 0.83 0.84 0.99 0.96 0.92 1.00 0.83
EDI 0.49 0.49 0.54 0.40 0.71 0.72 0.70 0.58 0.83 0.82 0.83 0.76 0.89 0.88 0.86 0.88 0.83 0.81 0.78 0.83 1.00
Average correlation without EDI 0.43 0.45 0.48 0.36 0.60 0.61 0.61 0.49 0.65 0.65 0.67 0.57 0.68 0.68 0.69 0.66 0.64 0.65 0.64 0.64

Wable P.S. et al.


Bold values show correlation values between similar time steps for various drought indices
Comparison of Drought Indices in a Semi-Arid River Basin of India

is found very strong (r > 0.97) at all month timescales, thereby indicating that ETo is not
sensitive for drought analysis if the ratio of ETo and rainfall is used. Further, the fact that both
SPI and RDI follow the same Gamma distribution implies that they have at least same values
during non-monsoon seasons.
The correlation of EDI with other drought indices for 1, 3, 6, 9 and 12-month at all stations
is shown in Figs. 3a-f. Surprisingly, the correlation with EDI increases from 1-month to 9-
month time scales but it decreases for the 12-month time scale. The correlation of EDI for 1-

Fig. 3 Correlation of EDI with various drought indices for 1-month (red line), 3-month (brown line), 6-month
(pink line), 9-month (orange line), 12-month (blue line) at six stations: a Alni, b Chinchondipatil, c Jamkhed, d
Kasegaon, e Solapur, and f Tembhurni stations
Wable P.S. et al.

month and 3-month time scales varies between 0.4 to 0.5 and 0.6 to 0.7, respectively, whereas
it is greater than 0.7 for 6-, 9- and 12-month time scales. Among all the drought indices, PDN
has the lowest correlation for 1-, 3- and 6-month time scale, thereby suggesting that PDN is
comparable with EDI for higher scales only. This finding is in agreement with that reported by
Dogan et al. (2012).
The average value of correlation of drought indices for a given time scale by different
indices (excluding EDI) is shown in the last line of Table 6 (bold values). The results reveal
that the average value of 0.43 is the paired correlation of SPI-1 with RDI-1, SPEI-1, PDN-1,
SPI-3, RDI-3, SPEI-3, PDN-3, SPI-6, RDI-6, SPEI-6, PDN-6, SPI-9, RDI-9, SPEI-9, PDN-9,
SPI-12, RDI-12, SPEI-12, and PDN-12. Similarly, average paired correlation values were
obtained for all indices time scales and were plotted as shown in Fig. 4a-f. It is evident from
Fig. 4a-f that the value of average correlation is higher for the 9-month scale. In addition, mean
of average correlation coefficient of all indices for a given scale is shown with a continuous
line in Fig. 4a-f. For example, at Jamkhed station (last line of Table 6) for 1-month time scale
(SPI = 0.43, RDI = 0.45, SPEI = 0.48, PDN = 0.36), the average correlation is 0.43. It is clear
from the figure that even excluding EDI, the maximum correlation value is for 9-month scales
followed by 6 and 12 months. Hence, for the comparison of drought indices, 9-month is
selected, which is reasonable since the major part of India receives rainfall only after 9 months,
once the monsoon is terminated in September (Jain et al. 2015) or early October and the same
is true for the study area as well.

4.3 Performance of Drought Indices with Respect to Historical Droughts

According to a report from the Government of Maharashtra (EIS 2006), the study area was in
the grip of drought with all the districts of the study area facing acute shortage of water in one
or other year and in some places, the situation was so worst that the drinking water had to be
supplied by water tankers. As the Sina River is an ephemeral river, the streamflow during
monsoon season (June to October) months are used for defining Streamflow Drought Index
(SDI). The time-series of SDI for 1985-2009 years is shown in Fig. 5. It is evident from the
figure that during 2002 and 2003 years there were ‘severe’ to ‘extreme’ hydrological droughts
in the study area. The occurrence of hydrological drought suggests that during this period
agricultural and meteorological drought also occurred in the study area during 2002 and 2003,
which had emergence before 2002 year.
Moreover, to evaluate the performance of selected meteorological drought indices (SPI,
RDI, SPEI and PDN at 9-month scale, and EDI) with respect to historical droughts, time series
of 2001-2004 period was considered. The relative frequencies of these indices during this
period for the monsoon months are shown in Fig. 6a-f. Relative frequency indicates percentage
of the drought index that remains in the particular drought severity class during the considered
drought spell. The total wetness is the sum of relative frequencies in ‘Extreme Wet (EW)’,
‘Severe Wet (SW)’, and ‘Moderate Wet (MW)’ classes, whereas, total dryness is the sum of
relative frequency in ‘Extreme Dry (ED)’, ‘Severe Dry (SD)’, and ‘Moderate Dry (MD)’
classes.
It can be seen from Fig. 6a-f that the SPI-9 has detected 45% (Jamkhed station) to 85% (for
Kasegaon station) of the months as ‘Normal’ situation and a total of 10-55% in dry classes for
the latter and former stations. EDI shows the least relative frequency for total dry classes
ranging from 5% (Chinchondipatil station) to 50% (Jamkhed station). On the other hand, the
maximum total relative frequencies for SPEI-9, RDI-9, and PDN-9 in dry classes are found to
Comparison of Drought Indices in a Semi-Arid River Basin of India

Fig. 4 Average correlation of a particular time-step drought index with all other indices time scales at: a Alni, b
Chinchondipatil, c Jamkhed, d Kasegaon, e Solapur, and f Tembhurni stations

be 65%, 70%, and 75%, respectively for the Jamkhed station. The main difference between
these three indices is the relative frequency in the wet classes. The wet classes frequency
identified by the PDN-9 varies from 0 to 25%, even though there was a dry spell during this
period (2001-2004). A closer look on the dry classes reveals that SPEI-9 shows maximum
relative frequency in ‘severe dry’ classes for all the stations, except the Kasegaon station [Fig.
6d]. The performance of PDN-9 is the worst, since most of its relative frequencies during the
2001-2004 drought spell are in ‘Normal’ class. Thus, the dry spell in the study area is well
identified by SPEI-9.
Wable P.S. et al.

3
SDI Mod. Dry Sev. Dry Ext. Dry
2

0
SDI

-1

-2

-3

-4
1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

2009
Year
Fig. 5 Annual variation of SDI during 1985-2009 period

For better understanding the performance of drought indices, comprehensive figures were
also created for all the six stations as depicted in Fig. 7a-f. For the multi-monthly rainfall, 9-
month scale was chosen to examine the response of drought indices with respect to 9-monthly
average rainfall. The 9-monthly rainfall for a particular month is the sum of rainfall in the
current month and that in preceding eight months. For instance, the 9-month rainfall for the
June month is total of June and preceding 8 months rainfalls and the 9-monthly average is the
average of all such June-months in the 9-monthly rainfall. For a particular month, this average
value is constant and hence, this value repeats month, which is shown by cross mark for each
month in Fig. 7a-f. It is evident from the figures that the deficit of monsoon as 9-monthly
rainfall being less than the corresponding normal values is well indicated by all the drought
indices. SPEI-9 is mostly under ‘Severe Dry’ and ‘Extreme Dry’ classes with its variation
clearly depicts the drought emergence highlighted by lowest index value even few months
before as ‘severe’ or ‘extreme’ drought conditions (Fig. 7). Also, as end of drought event is a
slow process the SPEI curve shows a gentle rise in index value (still lowest among all the
indices) after the extreme or drought conditions unlike EDI, which depicts a sharp rise. Other
stations show more or less similar condition except at Kasegaon station [Fig. 7d], where
there are least drought incidences, i.e., drought indices values are mostly in ‘Normal’ class
with only few months in ‘Moderate Dry’ class. Overall, SPEI-9 identifies drought char-
acteristics such as emergence, development, and end of drought condition better than other
drought indices.
For instance, at the Jamkhed station [Fig. 7c], it is obvious that there is a deficient of rainfall
as the 9-monthly rainfall is less than the corresponding normal values, which is well indicated
by all the drought indices. There is a deficit of monsoon rainfall as the 9-monthly rainfall is
less than the corresponding normal values such as from April-2001 to May-2002, November-
2002 to March-2003 and June-2003 to June-2004. During dry spells of April-2001 to May-
2002 and November-2002 to March-2003, SPEI-9 identifies drought characteristics better than
other indices. In case of June-2003 to June-2004 spell, the emergence of drought indicated by
SPI-9 and RDI-9 is better compared to other drought indices, but the development and end of
drought event are well represented by SPI-9, RDI-9 and SPEI-9. The drought indices SPI-9
and RDI-9 are found to follow the same trend due to the fact that the correlation between them
Comparison of Drought Indices in a Semi-Arid River Basin of India

Fig. 6 Histogram of the drought frequency classes of the selected indices during monsoon months of 2001-2004
dry spell at a Alni, b Chinchondipatil, c Jamkhed, d Kasegaon, e Solapur, and f Tembhurni stations

is very strong (r ≈ 1). During this spell, EDI is mostly found in the ‘Normal’ and ‘Moderate
Dry’ classes. Among all the drought indices, PDN-9 is the least responsive index as during
these dry spells; it is found mostly in the ‘Normal’ class category.
On the other hand, at Kasegaon station [Fig. 7d], EDI and SPEI-9 are able to identify the
emergence and development of drought events (October-2003 to April-2004) better than other
indices. Nevertheless, EDI shows a sudden peak after the end of drought in May-2004, which
is unusual. This sudden peak is also seen at Alni, Solapur and Tembhurni stations. Here also
like other stations, SPI-9 and RDI-9 are found to have similar trend. For the Kasegaon station,
PDN-9 is the least sensitive index.
Wable P.S. et al.

Hence, within the basin there is a difference in the opinion for choosing suitable drought
index. This can be attributed to variability in climatic condition in the study area. However, as
the water management is done on a watershed basis, the best performing index for a majority
of stations should be selected. Accordingly, SPEI-9 drought index is suitable for drought
assessment in the study area as it is the best for modeling historical drought events at five
stations out of six stations.

4.4 Comparative Assessment of Drought Indices

Comparative assessment of drought indices was performed to evaluate the indices for their
practical applicability. In order to perform comparative evaluation of the drought indices, the
weights were assigned to the five decision criteria based on literature review (Keyantash and
Dracup 2002; Barua et al. 2011) and professional judgment as shown in Table 7. The fact that
‘robustness’ of an index is the most important in identifying the condition of a drought, it has
been assigned maximum weight, followed by ‘transparency’, ‘sophistication’, ‘tractability’,
and ‘extendability’. Accordingly, the relative importance of these assigned weights varies from
14 to 26% (Table 7).
Furthermore, raw scores ranging from 1 to 5 were assigned to all the five decision criteria
(Table 8). The raw scores were assigned based on the qualitative and quantitative assessment
of drought indices. The qualitative assessment deals with the findings from other studies for
evaluating drought indices and their theoretical and computational aspects, whereas the
quantitative assessment deals with the performance of these drought indices in identifying
the historical droughts in the study area. The total scores were used for a comparative
evaluation of drought indices in the study area. The maximum possible total weighted score
of any drought index can be 175. The calculated total weighted scores for all the drought
indices are also shown in Table 8. The justification for the assigned raw scores to the decision
criteria for the study area is provided below.

Robustness Based on the relative frequency analysis (Fig. 6a-f) and comprehensive
time series plots (Fig. 7a-f), the SPEI-9 index is found to detect historical droughts
well as compared to other drought indices in terms of identifying drought character-
istics and dryness during the 2001-2004 dry spell. Hence, it is given the highest score
of 5. Thereafter, SPI-9 and RDI-9, which during the historical dry spell follow the
same trend and have their severity values in ‘Moderate Dry’, to ‘Severe Dry’ classes,
are assigned a score of 4. Although EDI indicates ‘Moderate Dry’ condition, it shows
assigned the capability of drought emergence condition better than other indices, and
hence it is assigned a score of 3. On the other hand, PDN-9 is given a score of 1,
because its performance is very poor.

Tractability SPI-9, EDI and PDN-9 require only precipitation data for computation,
but the PDN-9 needs lower level of numerical computation; and therefore it is given a
maximum score of 5. SPI-9 needs distribution fitting, so it is assigned a score of 4.
On the other hand, EDI needs calculation of effective precipitation, which is compu-
tationally complex, hence it is assigned a score of 3. RDI-9 and SPEI-9 need
temperature data along with precipitation data, but during distribution fitting, RDI
needs 2-parameter distribution and SPEI needs 3-parameter distribution. Hence, they
are assigned scores of 3 and 2, respectively.
Comparison of Drought Indices in a Semi-Arid River Basin of India

Fig. 7 Time series of drought indices and 9-monthly rainfall at a Alni, b Chinchondipatil, and c Jamkhed,
stations, d Kasegaon, e Solapur, and f Tembhurni stations

Transparency For the maximum score of 5 is assigned to PDN-9, since it is easy to


understand as compared to other indexes for both the scientific community and general public
(mostly farmers in this case). Other indices are easy understandable by researchers and
professionals, but not by public. Hence, SPI-9, RDI-9, and SPEI-9, which need distribution
fitting, they each are assigned a score of 3. Since the concept of effective precipitation used by
EDI is new and somewhat difficult to comprehend as compared to distribution fitting, it is
assigned a score of 2.
Wable P.S. et al.

(d)

(e)

(f) Tembhurni

Fig. 7 (continued)

Sophistication The EDI is less tractable and transparent, but it is conceptually sound as it uses
the concept of EP (Byun and Wilhite 1999) and hence, it is allocated a score of 5. The SPEI-9
needs 3-parameterdistribution fitting, whereas SPI-9 and RDI-9 need 2-parameter distribution.
It is a well-known fact that more parameters a probabilistic model has, the more flexible it
becomes in adjusting the data, thus making the model more sophisticated. Therefore, SPEI-9 is
assigned a score of 5 and, SPI-9 and RDI-9 are assigned a score of 3 and 4, respectively. It is to
be noted that although both RDI-9 and SPI-9 need 2-parameter distribution, RDI-9 is given a
Comparison of Drought Indices in a Semi-Arid River Basin of India

Table 7 Assigned weights for the drought index decision criteria

Decision Criteria Weight Relative Importance (%)

1. Robustness 9 26
2. Tractability 6 17
3. Transparency 8 23
4. Sophistication 7 20
5. Extendability 5 14

higher score than SPI-9 owing to the incorporation of temperature data in the former. The score
of 1 is assigned to PDN-9 since it is based on the concept of average or normal, which is not
conceptually sound.

Extendability On all the index data availability, the extendability of PDN-9 is easy as
compared to other indices as it only employs precipitation data and it is easiest to calculate;
hence, it is assigned maximum score of 5.However, the probabilistic indices, viz., SPI, RDI
and SPEI, need distribution fitting of precipitation and/or temperature data and thus can be
easily extendable with the availability of dataset (Barua et al. 2011); and therefore, each of
them are given an equal score of 4. In addition, EDI can also be easily extendable as that of
probabilistic indices as it requires only precipitation data for its calculation. Consequently, EDI
is also allocated a score of 4.
It is apparent from Table 8 that the highest total weighted score (136) is obtained for SPEI-9
and PDN-9 has the lowest score (111). Hence, based on total weighted score the overall
rankings of the drought indices are SPEI-9 followed by RDI-9, SPI-9, EDI, and PDN-9. Thus,
comparative evaluation also indicates that SPEI-9 is the best drought index. As discussed in
the previous section, SPEI-9 was also found to have relatively better historical drought
modeling capability in terms of drought emergence, development, and termination than other
drought indices. The superiority of SPEI-9 suggests that the variable difference of rainfall and
ETo, i.e., climatic water balance, helps in the better identification of drought characteristics in
the study area. Therefore, SPEI-9 is recommended for drought characterization as well as for
modeling and management of droughts in the study area. The finding of this study is contrary
to the findings obtained in similar studies carried out in eastern India (Pandey et al. 2008) and
in central India (Jain et al. 2015). This disagreement is attributed to the fact that these studies
did not consider evapotranspiration-based indices, thereby neglecting a vital component of the
hydrologic cycle. This study emphasizes the consideration of evapotranspiration-based
drought indices for drought studies performed in arid and semi-arid regions.

Table 8 Relative weights for decision criteria and raw scores for drought indices

Performance Criteria Relative Weights Raw Score for Drought Index

PDN-9 SPI-9 RDI-9 SPEI-9 EDI

1. Robustness 9 1 4 4 5 3
2. Tractability 6 5 4 3 2 3
3. Transparency 8 5 3 3 3 2
4. Sophistication 7 1 3 4 5 5
5. Extendability 5 5 4 4 4 4
Total Weighted Score 111 125 126 136 116
Wable P.S. et al.

5 Conclusions

In any area/region, the selection of a suitable drought index is vital for developing
effective strategies for the mitigation of drought conditions. In this study, performance
of five meteorological drought indices, namely, SPI, RDI, SPEI, PDN, and EDI was
examined with an aim of identifying a suitable drought index for the semi-arid river
basin of western India. Based on the results of this study, following conclusions can
be drawn:

& The correlation matrix of time step independent EDI and 1-, 3-, 6-, 9- and 12-month scale
indices of SPI, RDI, SPEI, and PDN indicated that all the indices are strongly correlated
for similar time scales and poorly correlated for dissimilar time scales.
& Among all the month scales considered in this study, 9-month scale is found to have the
highest correlation with EDI and among the similar scale indices. Thus, 9-month scale is
appropriate for the comparison of drought indices in the study area.
& The performance of drought indices during the historical drought periods (2001-2004)
revealed that SPEI-9 performed better in identifying drought characteristics than other
drought indices as well as it has the maximum relative frequency in ‘severe dry’ classes at
all the stations except Kasegaon station.
& The results of evaluation of drought indices based on the five decision criteria raw score
indicated that the rank of drought indices based on the total weighted score is in the order
of SPEI-9 (136), RDI-9 (126), SPI-9 (125), EDI (116), and PDN-9 (111). This finding
emphasizes that the water balance difference in terms of rainfall and PET is important for
effective drought monitoring in the study area.
& RDI-9 and SPI-9 have similar total weighted score, thereby indicating that PET is not
sensitive if the ratio of rainfall to PET is used for drought monitoring in the study area. On
the other hand, EDI has a total weighted score of 116, suppressing the concept of effective
precipitation for the study area.

On the whole, SPEI-9 performs better in detecting historical droughts and identify-
ing drought characteristics as compared to other drought indices. Hence, SPEI at 9-
month scale is recommended in drought monitoring system for the effective plan-
ning and management of water resources in the study area. Due to changing climate
and socio-economic conditions, it is recommended for testing other probabilistic
distributions for the probability-based drought indices in the future as a follow-up
to this study. Also, with the availability of more drought-index defining variables, a
study focusing on the development of a comprehensive drought index is suggested
for the study area.

Acknowledgements The authors are very grateful to India Meteorology Department, Pune (Maharashtra); State
Data Storage Centre, Nashik (Maharashtra) and Central Water Commission, Hyderabad (Andhra Pradesh) for
providing required hydro-meteorological data for this study. The authors are also grateful to the anonymous
reviewers for their useful comments/suggestions that improved the manuscript significantly.

Compliance with Ethical Standards

Conflict of Interest None.


Comparison of Drought Indices in a Semi-Arid River Basin of India

References

Abramowitz M, Stegun A (1965) Handbook of Mathematical Formulas, Graphs, and Mathematical Tables.
Dover Publications, Inc., New York
Allen RG, Perevia LS, Raes D, Smith M (1998) Crop Evapotranspiration: Guidelines for Computing Crop Water
Requirements. FAO Irrigation and Drainage Paper No. 56, Food and Agriculture Organization, Rome, Italy
Barua S, Ng AWM, Perera BJC (2011) Comparative evaluation of drought indexes: case study on the Yarra River
catchment in Australia. J Water Resour Plan Manag 137(2):215–226
Beguería S, Vicente-Serrano SM, Reig F, Latorre B (2014) Standardized precipitation evapotranspiration index
(SPEI) revisited: parameter fitting, evapotranspiration models, tools, datasets and drought monitoring. Int J
Climatol 34(10):3001–3023
Byun HR, Kim DW (2010) Comparing the effective drought index and the standardized precipitation index.
Options méditerranéennes, Serie A, no. 95- Economics of Drought and Drought Preparedness in a Climate
Change context, 27-33 pp
Byun HR, Wilhite DA (1999) Objective quantification of drought severity and duration. J Climatol 12:2747–
2756
Chary GR, Vittal KPR, Venkateswarlu B, Mishra PK, Rao GGSN, Pratibha G, Rao KV, Sharma KL, Rao GR
(2010) Drought Hazards and Mitigation Measures. In: Jha MK (ed) Natural and Anthropogenic Disasters:
Vulnerability, Preparedness and Mitigation. Springer, Netherlands, pp 197–236
Deo RC, Byun HR, Adamowski JF, Kim DW (2015) A real-time flood monitoring index based on daily effective
precipitation and its application to Brisbane and Lockyer Valley flood events. Water Resour Manag 29(11):
4075–4093
Dogan S, Berktay A, Singh VP (2012) Comparison of multi-monthly rainfall-based drought severity indices,
with application to semi-arid Konya closed basin, Turkey. J Hydrol 470:255–268
EIS (2006) Disaster Management in Maharashtra. Environmental Information System (EIS), Environmental
Department, Environmental Information System (ENVIS) Newsletter, Government of Maharashtra, India, 1:
1-10
Hargreaves GH, Samani ZA (1985) Reference crop evapotranspiration from temperature. Appl Eng Agric 1(2):
96–99
Hayes MJ, Svoboda MD, Wardlow B, Anderson MC, Kogan F (2012) Drought monitoring: Historical
and current perspectives. In: Wardlow BD, Anderson MC, Verdin JP (eds) Remote Sensing for
Drought: Innovative Monitoring Approaches. Taylor and Francis Group, CRC Press, Boca Raton,
USA, pp 1–19
Heim RR Jr (2002) A review of twentieth-century drought indices used in the United States. Bull Am Meteorol
Soc 83(8):1149–1165
Hosking JR (1990) L-moments: analysis and estimation of distributions using linear combinations of order
statistics. J R Stat Soc Ser B Methodol 52(1):105–124
Jain VK, Pandey RP, Jain MK and Byun HR. (2015). Comparison of drought indices for appraisal of drought
characteristics in the Ken River Basin. Weather Clim Extrem 8:1–11
Jamshidi H, Khalili D, Zadeh MR, Hosseinipour EZ (2011) Assessment and comparison of SPI and RDI
meteorological drought indices in selected synoptic stations of Iran. Proceedings of the World
Environmental and Water Resources Congress 2011: Bearing Knowledge for Sustainability. 22-26
May 2011, Palm Springs, California, pp. 1161-1173
Keyantash J, Dracup JA (2002) The quantification of drought: an evaluation of drought indices. Bull Am
Meteorol Soc 83(8):1167–1180
McKee TB, Doesken NJ, Kleist J (1993) The relationship of drought frequency and duration to time scales.
Proceedings of the 8th International Conference on Applied Climatology, 17-22 January 1993, Anaheim,
California, pp. 179-184
Mishra AK, Singh VP (2010) A review of drought concepts. J Hydrol 391(1):202–216
Mishra AK, Singh VP, Desai VR (2009) Drought characterization: a probabilistic approach. Stoch Env Res Risk
A 23(1):41–55
Morid S, Smakhtin V, Moghaddasi M (2006) Comparison of seven meteorological indices for drought monitor-
ing in Iran. Int J Climatol 26(7):971–985
Nalbantis I (2008). Evaluation of a hydrological drought index. Eur Water 23(24): 67–77
Niemeyer, S., 2008: New drought indices. Proceedings of the 1 International Conference on Drought
Management: Scientific and Technological Innovations, Options Méditerranéennes, Series A, No. 80,
Zaragoza, Spain, pp. 267–274
Pandey RP, Dash BB, Mishra SK, Singh R (2008) Study of indices for drought characterization in KBK districts
in Orissa (India). Hydrol Process 22:1895–1907
Rao AR, Hamed KH (2000) Flood frequency analysis, CRC Press. In: Boca Raton. Florida, USA
Wable P.S. et al.

Redmond KT (1991) Climate monitoring and indices. In: Wilhite DA, Wood DA, Kay PA (eds) Drought
Management and Planning: Proceedings of the Seminar and Workshop. University of Nebraska–Lincoln,
Lincoln, pp 29–33
Redmond KT (2002) The depiction of drought: a commentary. Bull Am Meteorol Soc 83(8):1143–1147
Roudier P, Mahe G (2010) Study of water stress and droughts with indicators using daily data on the Bani river
(Niger basin, Mali). Int J Climatol 30(11):1689–1705
Smakhtin VU, Hughes DA (2004) Review of Automated Estimation and Analyses of Drought Indices in South
Asia. Working Paper 83, International Water Management Institute (IWMI), Colombo, Sri Lanka
Smakhtin VU, Hughes DA (2007) Automated estimation and analyses of meteorological drought characteristics
from monthly rainfall data. Environ Model Softw 22(6):880–890
Steinemann A, Iacobellis SF, Cayan DR (2015) Developing and evaluating drought indicators for decision-
making. J Hydrometeorol 16(4):1793–1803
Tsakiris G, Vangelis H (2005) Establishing a drought index incorporating evapotranspiration. Eur Water 9–10:1–9
Vangelis H, Tigkas D, Tsakiris G (2013) The effect of PET method on Reconnaissance Drought Index (RDI)
calculation. J Arid Environ 88:130–140
Vicente-Serrano SM, Beguería S, López-Moreno JI (2010a) A multiscalar drought index sensitive to global
warming: the standardized precipitation evapotranspiration index. J Clim 23(7):1696–1718
Vicente-Serrano SM, Beguería S, López-Moreno JI, Angulo M, El Kenawy A (2010b) A new global 0.5 gridded
dataset (1901–2006) of a multiscalar drought index: comparison with current drought index datasets based
on the Palmer Drought Severity Index. J Hydrometeorol 11(4):1033–1043
Vicente-Serrano SM, Beguería S, Lorenzo-Lacruz J, Camarero JJ, López-Moreno JI, Azorin-Molina C, Revuelto
J, Morán-Tejeda E, Sanchez-Lorenzo A (2012) Performance of drought indices for ecological, agricultural,
and hydrological applications. Earth Interact 16(10):1–27
WMO (1975) Drought and Agriculture. World Meteorological Organization, Technical Note 138, WMO
Publication No. 392, Geneva, Switzerland, 127 pp
Zarch MAA, Malekinezhad H, Mobin MH, Dastorani MT, Kousari MR (2011) Drought monitoring by
reconnaissance drought index (RDI) in Iran. Water Resour Manag 25(13):3485–3504

Affiliations

Pawan S. Wable 1 & Madan K. Jha 1 & Ankit Shekhar 2

Madan K. Jha
madan@agfe.iitkgp.ernet.in

1
Agricultural & Food Engineering Department, IIT Kharagpur, Kharagpur 721 302, India
2
Centre for Oceans, Rivers, Atmosphere and Land Sciences, IIT Kharagpur, Kharagpur 721 302, India

You might also like