You are on page 1of 14

Hydrological Sciences Journal

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/thsj20

Can multiscalar meteorological drought indices


detect soil moisture droughts? A study of Indian
regions

Prabir Kumar Das, S. K. Mohinuddin, Subrata Kumar Midya, Dilip Kumar Das,
Richa Sharma & Soumya Bandyopadhyay

To cite this article: Prabir Kumar Das, S. K. Mohinuddin, Subrata Kumar Midya, Dilip Kumar
Das, Richa Sharma & Soumya Bandyopadhyay (2021) Can multiscalar meteorological drought
indices detect soil moisture droughts? A study of Indian regions, Hydrological Sciences Journal,
66:9, 1475-1487, DOI: 10.1080/02626667.2021.1942475

To link to this article: https://doi.org/10.1080/02626667.2021.1942475

Published online: 09 Jul 2021.

Submit your article to this journal

Article views: 400

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=thsj20
HYDROLOGICAL SCIENCES JOURNAL
2021, VOL. 66, NO. 9, 1475–1487
https://doi.org/10.1080/02626667.2021.1942475

Can multiscalar meteorological drought indices detect soil moisture droughts? A


study of Indian regions
Prabir Kumar Dasa, S. K. Mohinuddinb, Subrata Kumar Midyac, Dilip Kumar Das†d, Richa Sharmab
and Soumya Bandyopadhyaya
a
Regional Remote Sensing Centre-East, NRSC, Kolkata, India; bDepartment of Remote Sensing, Birla Institute of Technology, Mesra, India;
c
Department of Atmospheric Sciences, University of Calcutta, Kolkata, India; dAgricultural Chemistry and Soil Science, University of Calcutta, Kolkata,
India

ABSTRACT ARTICLE HISTORY


The present study aims to explore the potential of multiscalar meteorological drought indices in Received 16 June 2020
detecting soil moisture drought events. The standardized soil moisture index (SSMI), standardized Accepted 28 April 2021
precipitation index (SPI), standardized evapotranspiration index (SEI), standardized precipitation evapo­ EDITOR
transpiration index (SPEI) and multivariate moisture anomaly index (MMAI) were computed using long- A. Fiori
term (1980–2015) MERRA-2 soil moisture, precipitation and/or evapotranspiration data products. The
performances of the meteorological indices were evaluated based on a zone-wise and spatial correlation ASSOCIATE EDITOR
approach along with failure rate (FR) and false alarm rate (FAR) values. The spatial correlation was highest D. Bressiani
in SEI, followed by MMAI, in comparison to SPI and SPEI. FR and FAR values indicated that SEI is the best KEYWORDS
index for detecting soil moisture drought events, whereas MMAI outperformed the other indices in standardized soil moisture
representing combined drought events, i.e. meteorological or/and soil moisture droughts. The outcome index; multivariate moisture
of the study may be useful in retrieving information about soil moisture drought over a region using only anomaly index;
meteorological parameters. meteorological drought;
combined drought; soil
moisture drought; MERRA-2

1 Introduction (iii) hydrological and (iv) socio-economic droughts. However,


Drought, a widespread hydroclimatic extreme leading to agri­ recent studies have highlighted the importance of soil moisture
cultural as well as societal catastrophe, is considered the cost­ in regulating drought conditions (Seneviratne 2012, Teuling
liest of all natural disasters (Wilhite 2000, Dai 2013, NCEI et al. 2013, AghaKouchak 2014). Being a key component of
2016). Although there is some controversy over changes in both the water and energy cycles, alterations in soil moisture
drought during the last few decades (Sheffield et al. 2012, Dai directly impact the agricultural system as well as water avail­
2013, Damberg and AghaKouchak 2014), increasing drought ability (Wang et al. 2011). Hence, “soil moisture drought” is
intensity and magnitude are evident in most parts of the world a new addition to the list of drought categories. It generally
(Masih et al. 2014). The rising temperature from global warm­ arises as a consequence of meteorological drought and may
ing may not lead directly to drought, but drought incidences lead to agricultural drought (Wang et al. 2011, Mishra et al.
may produce concomitant alterations in the rate of global 2015, Halwatura et al. 2017, Zhang et al. 2017, Manning et al.
warming (Trenberth et al. 2014). 2018, Blyverket et al. 2019). Soil moisture drought is generally
At the global scale, the economic losses due to drought may defined as a deficit in moisture within the upper soil layer
amount to several billions of US dollars (Wilhite 2000). relative to its normal value (Manning et al. 2018). It generally
According to Murthy and SeshaSai (2010), agriculture over arises due to deviations in the supply and demand components
58.6 million hectares in India was damaged due to drought of the water cycle, i.e. the antecedent precipitation and evapo­
incidence in 1987, and more than 250 million people were transpiration (Teuling et al. 2013). However, the individual
directly or indirectly affected. Similarly, one of the worst influence of these two components is regulated by the complex
drought incidences in the Indian region during 2002 led to interactions of several land-atmospheric processes. Hence, the
a reduction in the crop sown area by ~12 million hectares and quantification and monitoring of soil moisture drought may
an associated food grain production loss of ~48 million tons enhance our understanding of the transformation of meteor­
(Murthy and SeshaSai 2010). Hence, clear understanding, ological to agricultural drought and may improve simulations
proper monitoring and objective damage assessments of of agricultural drought scenarios based on meteorological
drought incidences are essential for developing sustainable records and information.
management plans towards minimizing the impacts of Soil moisture drought information is generally derived
drought. from physical land-surface models and empirical indices
For ease of understanding and management, droughts are (Mitchell et al. 2004, Sheffield et al. 2014, Manning et al.
broadly categorized into (i) meteorological, (ii) agricultural, 2018). Several soil moisture drought indices incorporating

CONTACT Prabir Kumar Das prabir111@gmail.com Regional Remote Sensing Centre-East, NRSC, Kolkata, India
© 2021 IAHS
1476 P. K. DAS ET AL.

soil moisture information have been proposed by researchers,


such as the soil moisture drought index (Hollinger et al. 1993),
soil moisture deficit index (Narasimhan and Srinivasan 2005),
standardized soil moisture index (Hao and AghaKouchak
2013), soil water deficit index (Martínez-Fernández et al.
2015), drought severity index (Cammalleri et al. 2016), soil
moisture anomaly percentage index (Mao et al. 2017), etc.
However, the applications of these drought indices are limited
due to the sparse density or unavailability of long-term soil
moisture data (Zawadzki and Kędzior 2016). On the other
hand, the physical models used to simulate soil moisture are
very data intensive, requiring e.g. location-specific soil proper­
ties, information regarding irrigation availability and infra­
structure, long-term meteorological parameters, etc.
Moreover, the approximations of several processes and the
unavailability of reliable datasets add more uncertainty to the
estimation of accurate soil moisture data (Chowdhary and
Singh 2010, Wang et al. 2016). Hence, it is necessary to develop
simpler drought indices that are less data intensive, that
involve simple computation and, more importantly, can be
computed using meteorological parameters only. Figure 1. The study area with its meteorological regions, viz. northwest, central,
northeast and peninsular India. NW: Northwest; NE: Northeast.
The multiscalar drought indices, such as the standardized
precipitation index (SPI), standardized precipitation evapo­
transpiration index (SPEI), etc., were found to perform better
Indian mainland is generally classified into four homogeneous
than other drought indices considering a two-layer bucket
regions, i.e. northeast, northwest, central and peninsular India
model (Wang et al. 2015). Recent studies have also highlighted
(Guhathakurta et al. 2014). The annual average rainfall values
the importance of integrating temperature in drought indices
(% standard deviation) are 1324.6 (11%), 618.7 (19%), 976.4
to address the multivariate nature of soil moisture drought
(14%) and 730.5 (15%) mm for northeast, northwest, central
(Teuling et al. 2013, AghaKouchak 2014, Manning et al. 2018).
and peninsular India, respectively.
Drought indices such as the Palmer drought severity index
(Palmer 1965), SPEI (Vicente-Serrano et al. 2010), reconnais­
sance drought index (RDI; Tsakiris and Vangelis 2005) and 3 Data and methodology
multivariate moisture anomaly index (MMAI; Das et al. 2017)
3.1 Data processing
consider the temperature component in terms of evapotran­
spiration while capturing the impact of both precipitation and The long-term (1980–2015) total surface precipitation
temperature on drought conditions in changing climate (PRECTOT), evapotranspiration (EVLAND) and total profile
scenarios. soil moisture content (PRMC) from the Modern-Era
Therefore, the present study aims to explore the potential of Retrospective analysis for Research and Applications-2
several multiscalar meteorological drought indices in detecting (MERRA-2) were downloaded from the Goddard Earth
soil moisture droughts over the Indian region using long-term Sciences Data and Information Services Centre (http://disc.
soil moisture and meteorological parameters. sci.gsfc.nasa.gov/mdisc/). The soil moisture data from
MERRA-Land reanalysis were used, as it provides improved
modelled soil moisture data compared with the regular
2 Study area
MERRA model for hydrological studies (Reichle et al. 2011).
The study was been conducted over the Indian mainland The use of more accurate precipitation forcing along with
(Fig. 1), i.e. between 08°04ʹ and 37°06ʹN latitude and between improved catchment surface models is the key to the improved
68°07ʹ and 97°25ʹE longitude. Agriculture, being the direct output over the regular MERRA model. The datasets are avail­
source of livelihood for more than 50% of the population able in a ½° (latitude) by ⅔° (longitude) grid at a monthly
and a major contributor to the national economy (i.e. 17– interval for the period 1979–2015. The datasets are down­
18% of gross value added), plays a key role in India’s sustain­ loaded in netcdf (.nc) format, and further spatial subsetting
ability and development. Around 68% of the country’s agri­ was carried out to restrict our study to the Indian mainland.
cultural area – covering ~278 districts of 11 states – is prone to Many previous studies have been carried out on the valida­
drought-like situations (Nandakumar 2009, Nagaraja 2012). tion of the MERRA-2 soil moisture products using in situ soil
Winter (January–February), pre-monsoon (March–May), moisture data (Reichle et al. 2011, 2017, Yi et al. 2011, Liu et al.
monsoon (June–September) and post-monsoon (October– 2014, Bosilovich et al. 2015). The studies revealed that the
December) are the major seasons of the Indian mainland MERRA-derived soil moisture corresponds favourably with
(www.imdpune.gov.in). Around 80% of the total rainfall is in situ observations; moreover, its advantages over Advanced
generally received during monsoon season (Sahai et al. 2003). Microwave Scanning Radiometer on the Earth Observing
Based on the mean long-term rainfall and its variability, the System (AMSR-E) soil moisture products were also established
HYDROLOGICAL SCIENCES JOURNAL 1477

because it is less limited by dense vegetation (Yi et al. 2011). SEI, using a joint probability distribution function. For details
Another study, carried out by Liu et al. (2014), showed that the of MMAI one may refer to Das et al. (2017).
MERRA data were closest to the climatological observations of Unlike SPI and SEI, the standardized soil moisture index
all five models derived from re-analysed soil moisture pro­ (SSMI) was calculated by fitting the 6-month aggregated soil
ducts, viz. MERRA, ERA-Interim (ERA), Japanese 25-year moisture data into a non-parametric empirical probability
Reanalysis (JRA), Climate Forecast System Reanalysis (CFSR) distribution function, rather than any specific distribution
and National Centers for Environmental Prediction (NCEP). function, i.e. parametric (Farahmand and AghaKouchak
Similarly, validation of MERRA-2 soil moisture products 2015). In the present study, the marginal distribution of soil
using 220–320 in situ measurements across North America, moisture was derived using the empirical Gringorten plotting
Europe and Australia demonstrated its efficiency in represent­ position (Gringorten 1963):
ing surface as well as root zone soil moisture (Reichle et al.
i 0:44
2017). Based on the average anomaly correlation and unbiased p ðx i Þ ¼ (1)
Root Mean Square Error (RMSE) calculated using the in situ n þ 0:12
soil moisture data from 140 network sites, the MERRA-2 soil where n and i represent the sample size and rank of non-zero
moisture was found to be best, with improved soil moisture soil moisture data in ascending order. p(xi) is the correspond­
dynamics, among various reanalysed data products (Reichle ing empirical probability. The SSMI was computed by fitting
et al. 2011, Bosilovich et al. 2015). A recent study by Modanessi the output of Equation (1) into the following equation:
et al. (2020) over parts of India concluded that MERRA-
2-derived soil moisture is capable of explaining drought inci­ SSMI ¼ ϕ 1 ð pÞ (2)
dence and its impact on crops. Similarly, high agreement −1
Where ϕ represents the normal distribution function, and
between the agricultural drought information derived from p denotes the probability values derived from Equation (1).
satellite images and soil moisture drought derived from The scale of all the standardized indices, i.e. SPI, SEI, SPEI,
MERRA-2 data for several drought and normal years during MMAI and SSMI, ranges between −4 and +4. Negative values
the last 34 years was reported for all of India (Das et al. 2020). denote drought or anomalous moisture conditions, whereas
positive values indicate normal or wetter conditions.
3.2 Preparation of standardized indices
The long-term precipitation and evapotranspiration data were 3.3 Evaluation of drought indices with respect to soil
transformed into a 6-month standardized precipitation index moisture droughts
(SPI) and standardized evapotranspiration index (SEI), respec­
tively. A very small time period may not able to capture the The agreement between the soil moisture drought informa­
drought occurrence, whereas a much longer period may pro­ tion, i.e. SSMI, and other meteorological drought indices, i.e.
vide redundant information (Tornros and Menzel 2014). SPI, SEI, SPEI, MMAI, were analysed using both zone-wise
Hence, many researchers have recommended a time period and spatial correlation techniques. Subsequently, the efficien­
of 3 to 6 months for soil moisture drought studies (Tornros cies of all meteorological indices to detect soil moisture
and Menzel 2014, Manning et al. 2018). According to the drought were calculated using the failure rate (FR) and the
World Meteorological Organization (WMO 2012), soil moist­ false alarm rate (FAR), as suggested by Halwatura et al. (2017).
ure or agricultural conditions respond to precipitation anoma­ FR denotes the proportion of drought events that were not
lies on a relatively short time scale, which generally varies from detected by meteorological indices, i.e. SPI or SPEI or SEI or
3 to 6 months. Hence, in the present study the 6-month time MMAI, but that were soil moisture drought conditions based
period was selected, considering the probable resilience of the on SSMI values of less than minus one, i.e. SSMI < −1.
soil systems. The gamma probability distribution was adopted Minimum FR values indicate a better capability of detecting
for calculating SPI, whereas log-logistic distribution was soil moisture drought conditions.
selected for SEI (McKee et al. 1993, Das et al. 2017). As SSMIn SIn
suggested by Das et al. (2017), the SEI values were multiplied FR ¼ (3)
SSMIn
by −1 to bring them into the same scale (sign) as SPI.
Subsequently, two recently developed drought indices, i.e. where SSMIn is the number of droughts based on SSMI and SIn is
SPEI and MMAI, integrating both precipitation and evapo­ the number of drought occurrences based on meteorological
transpiration to represent anomalous drought or moisture indices.
conditions, were computed using the methodology described On the other hand, FAR represents the proportion of drought
by Vicente-Serrano et al. (2010) and Das et al. (2017), respec­ events that were detected by meteorological indices, i.e. SPI or
tively. The SPEI, representing a simple climatic water balance, SPEI or SEI or MMAI, but that were not actually soil moisture
was calculated by fitting 6-monthly aggregated differences drought conditions. Again, lower FAR values denote the ability
between precipitation and evapotranspiration into a log- to reduce error in detecting soil moisture drought incidents.
logistic probability distribution function. According to Das SIn SSMIn
et al. (2017), the MMAI is more efficient than SPEI in captur­ FAR ¼ (4)
SIn
ing the impact of both precipitation and evapotranspiration in
representing anomalous moisture conditions. The MMAI was Hence, FR and FAR are tools for estimating under- and over­
computed by coupling two univariate variables, i.e. SPI and performances of the target index.
1478 P. K. DAS ET AL.

3.4 Lag-correction in time-series drought indices with SEI, were estimated to find the agreement among the different
respect to SSMI indices over an individual grid. Higher (lower) correlation
values generally denote a better (worse) representation of soil
Based on the resilience of systems, meteorological drought
moisture drought conditions. The correlation coefficient
may or may not transform into soil moisture drought.
values generally ranged between 0.5 and 0.9 over the Indian
Moreover, the transformation may not be immediate; it may
mainland for SPI and SPEI (Fig. 3). It was interesting to find
be delayed. Hence, in the present study, the existing response
that the correlation values as well as their spatial distribution
lags for all meteorological indices to SSMI time-series data
were similar for SPI and SPEI. The lower values of 0.5–0.6 were
were estimated over each grid. The lag existing in the time-
distributed over the western coast and parts of eastern as well
series meteorological indices for an individual grid was tested
as central India, whereas the higher values of >0.7 were mainly
based on maximum correlation values within the range of −5
confined to northwestern Indian and parts of peninsular India.
to +5 months with respect to the time-series SSMI. Based on
However, in major parts of the Indian mainland, the correla­
the lag information, the time-series meteorological indices
tion values between SSMI and SPI or SPEI were within 0.6
were adjusted for all grids, to eliminate the response lag
to 0.7.
towards overcoming the probable reason for underperfor­
In the case of MMAI, the values were >0.8 over major parts
mance of some of the indices. Further, the lag-adjusted
of India, except for northeastern India where correlation
meteorological indices were tested for their agreement and
values varied between 0.5 and 0.8 (Fig. 3). The northeastern
their efficiency in representing soil moisture drought, using
region is dominated by high orographic rainfall associated
correlation statistics and estimated FR and FAR values. Details
with steep slopes due to the mountainous landscape, which
of the entire methodology are provided in Fig. 2.
may be the reason for the relatively poor association between
precipitation and soil moisture over that region. The correla­
4 Results and discussion tion between SSMI and SEI was found to be the highest, i.e.
>0.9, among all meteorological drought indices over the Indian
4.1 Spatial correlation of meteorological indices with
region except for parts of northeastern India.
SSMI
According to Halwatura et al. (2017), the estimated correla­
Coefficients of correlation between SSMI and the other tion values between soil moisture drought and SPI were within
meteorological drought indices, i.e. SPI, SPEI, MMAI and a range of 0.36 to 0.69 for different sites over Eastern Australia,

Figure 2. Schematic diagram of the overall methodology.


HYDROLOGICAL SCIENCES JOURNAL 1479

Figure 3. Maps of the spatial correlation between SSMI and (a) SPI, (b) SPEI, (c) MMAI and (d) SEI.

whereas for RDI (an index incorporating precipitation and and parts of peninsular India had values of 0.6–0.8 (Fig. 4). In
evapotranspiration), the correlation coefficient was found to the case of SPEI, the FR values were found to be relatively
be lower. Similar results, with correlation values of less than lower, i.e. 0.6–0.8, in the majority of the area, with some
0.70 between SPI and soil moisture, were obtained by Gwak isolated patches of higher FR values, i.e. >0.8, over the western
et al. (2017) and Budi et al. (2008). The SPEI works similarly to coast and lower FR values, i.e. <0.5, over the northeastern
or better than SPI to represent soil moisture drought, with region. When using MMAI to detect the soil moisture drought,
correlation values ranging from 0.6 to 0.9 for different sites at the FR values were found to be less than 0.2 over the entire
optimum time scales (Wang et al. 2015). Das et al. (2017) Indian mainland except for some parts of northeastern India.
found that SPEI largely represented the SPI only and was In the case of SEI, the FR values were much lower than those
unable to capture the impact of evapotranspiration in charac­ for SPI and SPEI, i.e. <0.2 to 0.3, representing soil moisture
terizing moisture anomaly conditions over the Indian region. droughts with better accuracy than SPI or SPEI (Fig. 4).
However, FR values of >0.6 were also observed in some parts
of northeastern India. Hence, based on FR values, MMAI was
4.2 Evaluation of meteorological drought indices
found to be best suited for detecting soil moisture drought
The FR values of different indices represent their ability to incidents, followed by SEI.
detect soil moisture drought using meteorological parameters Similar to FR, lower FAR values indicate a better ability of
alone. Lower (higher) FR values indicate greater (worse) cap­ the indices to represent soil moisture drought conditions.
ability to represent soil moisture drought. In the case of SPI, Figure 5 shows that according to FAR also, the performances
the FR values were mainly above 0.8 in parts of eastern as well of SPI and SPEI were poorer among the four selected meteor­
as central India, whereas some portions of northwestern India ological drought indices. The values mainly ranged from 0.6 to
1480 P. K. DAS ET AL.

Figure 4. Maps of the failure rate of (a) SPI, (b) SPEI, (c) MMAI and (d) SEI in detecting soil moisture drought.

>0.8, with some lower FAR values mainly confined to northern SPI corroborate the findings of Halwatura et al. (2017). It
and peninsular India. In the case of MMAI, the FAR values is interesting that MMAI was found to be the most efficient
were found to be lower than those for SPI and SPEI; however, in detecting soil moisture drought incidences with more than
the distribution of patches of higher and lower FAR values was 80% accuracy; however, it overestimates drought conditions
almost identical to that of SPI and SPEI (Fig. 5). In contrast to on the order of 50–80%. On the other hand, the SEI could
the FR results, the FAR values for SEI indicated it outper­ identify soil moisture drought occurrences very efficiently,
formed MMAI in representing soil moisture drought. The with both FR and FAR values of less than 30% over major
values mainly varied between <0.2 to 0.3, indicating fewer parts of the Indian mainland. Hence, for detecting soil moist­
overestimations of soil moisture drought incidences in com­ ure drought incidences using meteorological parameters alone,
parison to the other indices. SEI was found to be the most efficient, followed by MMAI.
According to Halwatura et al. (2017), the aim is to achieve The false prediction of soil moisture drought events, i.e.
lower FR and FAR rates for meteorological indices to detect overestimating the drought frequency, remains the major con­
the soil moisture drought incidences. In the present study, cern for using meteorological indices (Touchan et al. 2005,
based on the analysis of both FR and FAR values, it was Törnros and Menzel 2014). An extreme rainfall deficit and
inferred that SPI and SPEI were unable to detect all of the high potential evapotranspiration (PET) are generally preva­
soil moisture drought occurrences – which were as high as 80% lent over transition climate regions between wet and dry cli­
of the total incidences. Moreover, the SPI and SPEI detected mates, such as the Indian region, where the land-atmospheric
some drought events that were not actually drought incidences interactions have a major impact on soil moisture droughts
based on SSMI values. The overestimations were on the order (Seneviratne et al. 2006, Seneviratne 2012). However, over­
of 60–80% of the total estimations. The FR and FAR values for simplification of the land-atmosphere interactions may be
HYDROLOGICAL SCIENCES JOURNAL 1481

Figure 5. Maps of the false alarm rate of (a) SPI, (b) SPEI, (c) MMAI and (d) SEI in detecting soil moisture drought.

one reason for the underperformance of SPI and SPEI (Shiau meteorological and soil moisture droughts may occur simulta­
2006, Seneviratne 2012). The soil type specific to a site has neously, or only one may occur in a given season or year. The
immense importance in governing the soil moisture drought concept of combined or cumulative drought was proposed to
phenomenon (Wang et al. 2015). Unlike SPEI, MMAI is devel­ address the overall drought occurrences, i.e. either meteorolo­
oped by optimizing the interactions between precipitation and gical (SPI < −1) or soil moisture drought (SSMI < −1). Hao and
PET in an individual grid over the entire study region (Das AghaKouchak (2013) proposed a multivariate standardized
et al. 2017). The better performance of SEI, on the other hand, drought index (MSDI) for characterizing combined drought
can perhaps be attributed to the stronger and linear influence conditions, by coupling SPI and SSMI.
of PET on the soil moisture minima over the region (Khalili In the present study, the combined drought occurrences
et al. 2011, Asadi Zarch et al. 2015, Halwatura et al. 2017). were estimated by considering a union of meteorological and
soil moisture droughts – that is, a drought was considered
a combined drought if it was either a meteorological or a soil
4.3 Performances of meteorological drought indices for
moisture drought. The combined drought conditions were
representing combined drought conditions
analysed to address the overestimations of soil moisture
Generally, different kinds of droughts – meteorological, soil drought incidences by different meteorological indices.
moisture, agricultural and hydrological droughts – occur in Subsequently, the FR and FAR values were estimated for all
a sequence. However, one kind of drought does not necessarily the meteorological drought indices to evaluate their capability
transform into another, based on the resilience or buffering in retrieving combined drought information.
capacity of the systems. Hence, although these kinds of Similar to soil moisture drought, both the SPI and SPEI
droughts are interdependent, the occurrences of one kind of were unable to detect the combined drought conditions; they
drought are independent. Likewise, in the present study produced FR values of more than 60% in most parts of India
1482 P. K. DAS ET AL.

(Fig. 6). In contrast to our earlier finding that SEI was capable
of representing soil moisture drought, SEI failed to detect
combined drought in more than 50% cases over almost the
entire Indian region. In the case of combined drought condi­
tions, meteorological drought is contributing along with soil
moisture drought. In meteorological drought, although evapo­
transpiration has a role to play, precipitation makes a greater
contribution. In SEI computation, the direct precipitation
component is missing. Hence SEI was unable to represent
meteorological drought or the combined drought very effi­
ciently, as it could perform only in the case of soil moisture
drought events.
However, it was observed that MMAI could successfully
capture the combined drought conditions with an accuracy of
generally more than 80% (Fig. 6). Unlike soil moisture
drought, the FAR values for MMAI were found to be very
low, signifying that there was hardly any overestimation of
combined drought conditions (Fig. 7). In the case of MMAI
computation, both the evapotranspiration and the precipita­
tion were used. Hence, MMAI was able to represent the
combined drought more efficiently compared to other Figure 7. Maps of the false alarm rate of MMAI in detecting combined drought
indices. events, i.e. meteorological or/and soil moisture drought.

Figure 6. Maps of the failure rate of (a) SPI, (b) SPEI, (c) MMAI and (d) SEI in detecting combined drought events, i.e. meteorological or/and soil moisture drought.
HYDROLOGICAL SCIENCES JOURNAL 1483

Drought is considered a multivariate compound event, It was observed that the maximum agreement with SSMI, in
which generally develops due to the joint dependent concur­ terms of correlation coefficients, was achieved by adjusting the
rence of different extreme events, such as precipitation, PET, lag by 1–2 months for different indices (Fig. 8). In the case of
etc. (Leonard et al. 2014, Bevacqua 2017, Manning et al. 2018). SPI, a 2-month lag was observed over major parts of the Indian
According to Hao and AghaKouchak (2013), information on region, whereas for the other three indices only a 1-month lag
different variables, such as precipitation and PET, is important was seen, except in the northeastern region where 2–3 months’
for proper characterization of drought. The present study lag was observed.
demonstrates that compound drought events, combining The spatial correlation, FR and FAR were estimated using
extremes of precipitation, evapotranspiration and soil moist­ lag-adjusted meteorological drought indices to evaluate their
ure, can be successfully detected by MMAI. The findings are in potential in representing both soil moisture and combined
line with a previous study carried out by Das et al. (2017), in drought conditions. It was observed that the spatial correlation
which the improved efficiency of MMAI over other drought increased in both SPI and SPEI by around 10%, but this was
indices in capturing the impact of both rainfall and PET for mainly confined to the grids with correlation coefficients of 0.5
characterizing drought events was established. to 0.7. In grids with higher correlation coefficients, along with
SEI and MMAI, hardly any differences were observed (figure
not shown). Interestingly, there was also no significant
4.4 Lag-adjustments of different meteorological indices improvement in the FR and FAR values after adopting the lag-
with respect to time-series SSMI adjusted indices.
The temporal adjustments of four meteorological indices were Hence, it can be inferred that the temporal adjustments
carried out based on their maximum correlation coefficient could increase the spatial correlation of some indices with
values for a window of −5 to +5 months with respect to SSMI. SSMI over some selected parts of the Indian region; however,

Figure 8. Lag (month) of time-series (a) SPI, (b) SPEI, (c) MMAI and (d) SEI with respect to SSMI.
1484 P. K. DAS ET AL.

the performances of the meteorological drought indices in meteorological zones for both non-adjusted and lag-adjusted
representing the soil and combined drought conditions remain datasets. Among the different zones, northeastern India
virtually unaltered. The lag analysis confirmed that the showed relatively lower correlation values with SSMI for all
response lag between SPI/SPEI and SSMI may not be the the meteorological indices, whereas after adjusting the lag the
reason for its underperformance when representing the soil differences were negligible. The performances of all indices
moisture drought events. were comparable in the other three zones, i.e. northwestern,
central and peninsular India, in both non-adjusted and lag-
4.5 Zone-wise performances of meteorological drought
adjusted indices. Interestingly, the correlation values increased
indices
for SPI and SPEI after lag adjustments in all zones, but this
The web plot in Fig. 9 shows the performances of the meteor­ change was not significant for MMAI or SEI. These observa­
ological drought indices over different homogeneous tions corroborate our findings discussed in Section 4.4.

Figure 9. Zone-wise correlation curves between SSMI and (a) SPI, (b) SPEI, (c) MMAI and (d) SEI for both non-adjusted and lag-adjusted datasets.
HYDROLOGICAL SCIENCES JOURNAL 1485

Table 1. Zone-wise failure rate and false alarm rate of different meteorological drought indices for representing soil moisture drought and combined drought events.
Zones SPI SPEI SEI MMAI
FR_sm FAR_sm FR_cd FR_sm FAR_sm FR_cd FR_sm FAR_sm FR_cd FR_sm FAR_sm FR_cd FAR_cd
NE 0.78 0.72 0.73 0.68 0.68 0.67 0.41 0.40 0.61 0.16 0.63 0.24 −0.07
NW 0.70 0.66 0.69 0.62 0.64 0.69 0.26 0.28 0.59 0.07 0.58 0.19 −0.03
Central 0.88 0.85 0.72 0.80 0.79 0.70 0.16 0.17 0.58 0.04 0.62 0.16 −0.15
Peninsular 0.75 0.71 0.70 0.68 0.69 0.68 0.22 0.24 0.58 0.07 0.58 0.20 −0.03
Lag-adjusted
NE 0.68 0.71 0.74 0.95 0.65 0.67 0.37 0.32 0.60 0.01 0.60 0.22 −0.04
NW 0.69 0.64 0.69 0.97 0.61 0.68 0.38 0.30 0.59 0.02 0.59 0.21 −0.04
Central 0.69 0.80 0.72 0.95 0.75 0.70 0.24 0.17 0.58 0.02 0.61 0.20 −0.02
Peninsular 0.54 0.66 0.70 0.94 0.64 0.68 0.30 0.26 0.59 0.01 0.58 0.23 −0.04
FR: Failure rate; FAR: False alarm rate; sm: soil moisture drought; cd: combined drought.

The zone-wise mean FR and FAR statistics for both soil improved, no significant improvements in the performances
moisture drought and combined drought are presented in of all indices were observed based on FR and FAR values.
Table 1. The results agree with our earlier findings that the The zone-wise correlation results corroborated the earlier
FR values were lowest in MMAI when representing both kinds findings.
of droughts, i.e. soil moisture drought and combined drought;
however, it overestimated the soil moisture drought. Hence,
SEI was selected as the best meteorological index for represent­ Acknowledgements
ing soil moisture drought, based on its lower FR and FAR
The authors are thankful to Dr Santanu Chowdhury, Dr Uday Raj and
values. But in the case of combined drought conditions, the Dr D. Dutta for their continuous support and encouragement during the
MMAI outperformed all other indices, with the lowest FR and investigation. We also acknowledge the support received from BITS,
negligible FAR values in all four meteorological zones. The Mesra and RRSC-East staffs during manuscript preparation. We duly
trends were similar in the non-adjusted and lag-adjusted acknowledge NASA GMAO for providing long-term MERRA-2 data.
datasets. The authors are grateful to the anonymous reviewers for their construc­
tive suggestions.

5 Conclusions Disclosure statement


In the present study, the performances of multiscalar meteor­ No potential conflict of interest was reported by the authors.
ological drought indices in detecting soil moisture drought
conditions were evaluated. The spatial correlations of SPI References
and SPEI with SSMI were found to be lower, i.e. 0.5 to 0.7,
over major parts of the Indian mainland, whereas correlation AghaKouchak, A., 2014. A baseline probabilistic drought forecasting
coefficients were highest for SEI, i.e. >0.9, in most areas (except framework using standardized soil moisture index: application to the
2012 United States drought. Hydrology and Earth System Sciences,
some isolated parts of northeastern India). For MMAI, the 18 (7), 2485–2492. doi:10.5194/hess-18-2485-2014
correlation coefficient values were slightly lower (>0.8) than Asadi Zarch, M.A., Sivakumar, B., and Sharma, A., 2015. Droughts in
for SEI; however, they were significantly higher than those for a warming climate: a global assessment of Standardized Precipitation
both SPI and SPEI. It was interesting to find that SPI and SPEI Index (SPI) and Reconnaissance Drought Index (RDI). Journal of
displayed almost identical behaviour, along with similar lower Hydrology, 526, 183–195. doi:10.1016/j.jhydrol.2014.09.071
Bevacqua, E., 2017. CD VineCopulaConditional: sampling from condi­
and higher correlation patches. tional C- and D-vine copulas. R package, version 0.1.0. Availabe from:
In terms of the multivariate meteorological drought https://CRAN.R-project.org/package5CDVineCopulaConditional
indices’ performance in representing soil moisture drought Blyverket, J., et al., 2019. Monitoring soil moisture drought over northern
based on FR and FAR, both SPI and SPEI failed, with FR and high latitudes from space. Accessed 26 June 2019. Availabe from:
FAR values >80% in major parts of the Indian region. In the https://www.preprints.org/manuscript/201904.0009/v1
Bosilovich, M.G., et al., 2015. MERRA-2: initial evaluation of the climate.
case of MMAI, although the FR values were the lowest In: R.D. Koster eds. Technical report series on global modeling and data
among the indices, the FAR values were much higher, indi­ assimilation. Vol. 43. NASA/TM–2015-104606/Vol. 43. Maryland:
cating overestimation of soil moisture drought events. Goddard Space Flight Centre Greenbelt, 139.
Hence, SEI was selected as the best index to detect soil Budi, H.N., Sudibyakto, H.A., and Jetten, V.G. 2008. Drought monitoring
moisture drought, based on its lower FR and FAR values. using rainfall data and spatial moisture modeling. Thesis submitted
Gadjah Mada University, International Institute for Geo-information,
However, the MMAI was found to outperform the other Science and Earth Observation.
indices in simulating the combined drought events, i.e. either Cammalleri, C., Micale, F., and Vogt, J., 2016. A novel soil moisture-based
meteorological or soil moisture drought events. Thus, we Drought Severity Index (DSI) combining water deficit magnitude and
conclude that SEI may be used to detect soil moisture frequency. Hydrological Processes, 30 (2), 289–301. doi:10.1002/
drought, whereas MMAI would be a better choice for repre­ hyp.10578
Chowdhary, H. and Singh, V.P., 2010. Reducing uncertainty in estimates
senting combined drought events. The time-series lags in the of frequency distribution parameters using composite likelihood
indices with respect to SSMI were adjusted, and we found approach and copula-based bivariate distributions. Water Resources
that although the spatial correlations of SPI and SPEI were Research, 46 (11), 1–23, W11516. doi:10.1029/2009WR008490
1486 P. K. DAS ET AL.

Dai, A.G., 2013. Increasing drought under global warming in observations and partners in a continental distributed hydrological modeling
and models. Nature Climate Change, 3 (1), 52–58. doi:10.1038/ system. Journal of Geophysical Research, 109, D07S90. doi:10.1029/
nclimate1633 2003JD003823
Damberg, L. and AghaKouchak, A., 2014. Global trends and patterns of Modanessi, S., et al., 2020. Do satellite surface moisture observations
drought from space. Theoretical and Applied Climatology, 117 (3–4), better retain information about crop-yield variability in drought
441–448. doi:10.1007/s00704-013-1019-5 conditions? Water Resources Research, 56, e2019WR025855.
Das, P.K., et al., 2017. Characterizing Indian meteorological moisture doi:10.1029/2019WR025855
anomaly condition using long-term (1901–2013) gridded data: Murthy, C.S. and SeshaSai, M.V.R., 2010. Agricultural drought monitor­
a multivariate moisture anomaly index approach. International ing and assesment. In: P.S. Roy, R.S. Dwivedi, and D. Vijayan, eds.
Journal of Climatology, 38 (S1), 144–159. doi:10.1002/joc.5359 Remote sensing applications. National Remote Sensing Centre, Indian
Das, P.K., et al., 2020. Quantification of agricultural drought over Indian Space Research Organization, Department of Space, Government of
region: a multivariate phenology-based approach. Natural Hazards, India, 303–330.
101, 255–274. doi:10.1007/s11069-020-03872-6 Nagaraja, B.C., 2012. Impact of drought on agriculture: challenges facing
Farahmand, A. and AghaKouchak, A., 2015. A generalized framework for poor farmers of Karnataka, south India. Karnataka: Department of
deriving nonparametric standardized drought indicators. Advances in Environmental Science, Bangalore University.
Water Resources, 76, 140–145. doi:10.1016/j.advwatres.2014.11.012 Nandakumar, T., 2009. Manual for drought management. New Delhi:
Gringorten, I.I., 1963. A plotting rule for extreme probability paper. Department of Agriculture and Cooperation, Ministry of Agriculture,
Journal of Geophysical Research, 68 (3), 813–814. doi:10.1029/ Government of India.
JZ068i003p00813 Narasimhan, B. and Srinivasan, R., 2005. Development and evaluation of
Guhathakurta, P., et al., 2014. Observed changes in southwest monsoon soil moisture deficit index (SMDI) and evapotranspiration deficit index
rainfall over India during 1901–2011. International Journal of (ETDI) for agricultural drought monitoring. Agriculture and Forest
Climatology. doi:10.1002/joc.4095 Meteorology, 133, 69–88. doi:10.1016/j.agrformet.2005.07.012
Gwak, Y.S., et al., 2017. The relationships between drought indices (SPI, NCEI, 2016. Billion-dollar weather and climate disasters. Accessed 26
AQPI) and in-situ soil moisture in forested hill slopes. WIT June 2019. Available from: www.ncdc.noaa.gov/billions
Transactions on Ecology and the Environment. 220. doi:10.2495/ Palmer, W.C., 1965. Meteorological drought. Research Paper No. 45. US
WRMI170211 Department of Commerce. Washington, DC: Weather Bureau.
Halwatura, D., et al., 2017. Capability of meteorological drought indices Reichle, R.H., et al., 2011. Assessment and enhancement of MERRA land
for detecting soil moisture droughts. Journal of Hydrology: Regional surface hydrology estimates. Journal of Climate, 24, 6322–6338.
Studies, 12, 396–412. doi:10.1175/JCLI-D-10-05033.1
Hao, Z. and AghaKouchak, A., 2013. Multivariate standardized drought Reichle, R.H., et al., 2017. Assessment of MERRA-2 land surface hydrol­
index: a parametric multi-index model. Advances in Water Resources, ogy estimates. Journal of Climate, 30 (8), 2937–2960. doi:10.1175/JCLI-
57, 12–18. doi:10.1016/j.advwatres.2013.03.009 D-16-0720.1
Hollinger, S.E., Isard, S.A., and Welford, M.R., 1993. A new soil moisture Sahai, A.K., et al., 2003. Long-lead prediction of Indian summer monsoon
drought index for predicting crop yields. In: Eighth conference on rainfall from global SST evolution. Climate Dynamics, 20, 855–863.
applied climatology. Anaheim, CA: American Meteorological Society, doi:10.1007/s00382-003-0306-8
187–190. Preprints. Seneviratne, S.I., et al., 2006. Land–atmosphere coupling and climate
Khalili, D., et al., 2011. Comparability analyses of the SPI and RDI change in Europe. Nature, 443, 205–209. doi:10.1038/nature05095
meteorological drought indices in different climatic zones. Water Seneviratne, S.I., 2012. Climate science: historical drought trends
Resource Management, 25 (6), 1737–1757. doi:10.1007/s11269-010- revisited. Nature, 491, 338–339. doi:10.1038/491338a
9772-z Sheffield, J., et al., 2014. A drought monitoring and forecasting system for
Leonard, M., et al., 2014. A compound event framework for understand­ sub-Sahara African water resources and food security. Bulletin of
ing extreme impacts. Wiley Interdiscip. Rev.: Climate Change, 5, American Meteorological Society, 95, 861–882. doi:10.1175/BAMS-
113–128. doi:10.1002/wcc.252 D-12-00124.1
Liu, L., Zhang, R., and Zuo, Z., 2014. Intercomparison of spring soil Sheffield, J., Wood, E.F., and Roderick, M.L., 2012. Little change in global
moisture among multiple reanalysis data sets over eastern China. drought over the past 60 years. Nature, 491 (7424), 435–438.
Journal of Geophysical Research: Atmospheres, (119), 54–64. doi:10.1038/nature11575
doi:10.1002/2013JD020940 Shiau, J.T., 2006. Fitting drought duration and severity with
Manning, C., et al., 2018. Soil moisture drought in Europe: a compound two-dimensional copulas. Water Resource Management, 20, 795–815.
event of precipitation and potential evapotranspiration on multiple doi:10.1007/s11269-005-9008-9
time scales. Journal of Hydrometorology, 19, 1255–1271. doi:10.1175/ Teuling, A.J., et al., 2013. Evapotranspiration amplifies European summer
JHM-D-18-0017.1 drought. Geophysical Research Letters, 40 (10), 2071–2075.
Mao, Y., et al., 2017. Spatio-temporal analysis of drought in a typical plain doi:10.1002/grl.50495
region based on the soil moisture anomaly percentage index. Science of Törnros, T. and Menzel, L., 2014. Addressing drought conditions
Total Environment, 576, 752–765. doi:10.1016/j.scitotenv.2016.10.116 under current and future climates in the Jordan River region.
Martínez-Fernández, J., et al., 2015. A soil water based index as a suit-able Hydrology and Earth System Science, 18 (1), 305–318. doi:10.5194/
agricultural drought indicator. Journal of Hydrology, 522, 265–273. hess-18-305-2014
doi:10.1016/j.jhydrol.2014.12.051 Touchan, R., et al., 2005. Standardized precipitation index reconstructed
Masih, I., et al., 2014. A review of droughts on the African continent: from Turkish tree-ring widths. Climate Change, 72 (3), 339–353.
a geospatial and long term perspective. Hydrology and Earth System doi:10.1007/s10584-005-5358-9
Sciences, 18 (9), 3635–3649. doi:10.5194/hess-18-3635-2014 Trenberth, K.E., et al.. 2014. Global warming and changes in drought.
McKee, T.B., Doesken, N.J., and Kliest, J., 1993. The relationship of Nature Climate Change, 4 (1), 17–22. doi:10.1038/nclimate2067
drought frequency and duration to timescales. In: Proceedings of the Tsakiris, G. and Vangelis, H., 2005. Establishing a drought index incor­
8th conference on applied climatology, January 17–22. Anaheim, CA: porating evapotranspiration. European Water, 9 (10), 3–11.
American Meteorological Society, 179–184. Boston, MA. Vicente-Serrano, S.M., Beguera, S., and López-Moreno, J.I., 2010.
Mishra, A.K., et al., 2015. Anatomy of a local-scale drought: application of A multiscalar drought index sensitive to global warming: the standar­
assimilated remote sensing products, crop model, and statistical meth­ dized precipitation evapotranspiration index. Journal of Climate, 23,
ods to an agricultural drought study. Journal of Hydrology, 526, 15–29. 1696–1718. doi:10.1175/2009JCLI2909.1
doi:10.1016/j.jhydrol.2014.10.038 Wang, A., Lettenmaier, D.P., and Sheffield, J., 2011. Soil moisture drought
Mitchell, K.E., et al., 2004. The multi-institution North American Land in China, 1950–2006. Journal of Climate, 24 (13), 3257–3271.
Data Assimilation System (NLDAS): utilizing multiple GCIP products doi:10.1175/2011JCLI3733.1
HYDROLOGICAL SCIENCES JOURNAL 1487

Wang, H., et al., 2016. Monitoring winter wheat drought threat in Yi, Y., Kimball, J.S., and Jones, L.A., 2011. Evaluation of MERRA land
Northern China using multiple climate- based drought indices and surface estimates in preparation for the soil moisture active passive
soil moisture during 2000–2013. Agricultural and Forest Meteorology, mission. Journal of Climate, 24, 3797–3816. doi:10.1175/2011JCLI4034.1
228–229 (15), 1–12. doi:10.1016/j.agrformet.2016.06.004 Zawadzki, J. and Kędzior, M., 2016. Soil moisture variability over Odra
Wang, H., Rogers, J.C., and Munroe, D.K., 2015. Commonly used drought watershed: comparison between SMOS and GLDAS data. International
indices as indicators of soil moisture in China. Journal of Journal of Applied Earth Observations and Geoinformation, 45 (Part B),
Hydrometeorology, 16 (3), 1397–1408. doi:10.1175/JHM-D-14-0076.1 110–124. doi:10.1016/j.jag.2015.03.005
Wilhite, D.A., ed., 2000. Drought: a global assessment. London, UK: Zhang, X., et al., 2017. Soil moisture drought monitoring and forecasting
Routledge. using satellite and climate model data over Southwestern China.
WMO, 2012. Standardized precipitation index user guide. Available from: Journal of Hydrometeorology, 18 (1), 5–23. doi:10.1175/JHM-D-16-
http://www.wamis.org/agm/pubs/SPI/WMO_1090_EN.pdf 0045.1

You might also like