Atmospheric Research 181 (2016) 220–235

Contents lists available at ScienceDirect

Atmospheric Research

journal homepage: www.elsevier.com/locate/atmosres

The 20th century transitions in basic and extreme monsoon rainfall

indices in India: Comparison of the ETCCDI indices
Dileep K. Panda ⁎, P. Panigrahi, S. Mohanty, R.K. Mohanty, R.R. Sethi
Indian Institute of Water Management (ICAR-IIWM), Chandrasekharpur, Bhubaneswar 751023, Odisha, India

a r t i c l e i n f o a b s t r a c t

Article history: The mean and extreme matrices of the monsoon rainfall in India not only play an important role in depicting the
Received 10 March 2016 global monsoon climate, but also their spatiotemporal patterns influence the socio-economic profile of a major
Received in revised form 1 July 2016 proportion of the country's huge population. Given the reported conflicting trends at the global and national
Accepted 5 July 2016
scales, the present study investigates the 20th century (1901–2004) changes in monsoon rainfall of India, partic-
Available online 6 July 2016
ularly focusing the indices developed by the Expert Team on Climate Change Detection and Indices (ETCCDI) to
Keywords:
facilitate a global comparison. Result of this comprehensive analysis, which includes the response of fifteen
Extreme rainfall indices indices over two study periods (i.e., 1901–1940 and 1961–2004), indicates clear signals of change with respect
Spatiotemporal pattern to the period and region of study and the choice of the ETCCDI indices. While wet day frequency, low-to-
Indian monsoon season moderate events and consecutive wet days (CWD) exhibit a prominent transition from a pre-1940 wetting to
PDSI a post-1960 drying tendency over a large part of the central-north India (CNI), both the wet and dry extremes
TIO SST have occurred in a spatially less consistent manner during the recent decades. For consecutive dry days (CDD),
the reported less clear global signals could be related to the timescale of analysis, as our sub-seasonal scale results
display consistent changes compared to that of the seasonal and annual scales. The Palmer Drought Severity
Index (PDSI) provides clear indications of a post-1960 non-stationarity, showing changes in the mean as well
as variance. Based on the partial Mann–Kendall test (PMK), some of the identified rainfall trends during 1961–
2004 are found to be influenced more by the tropical Indian Ocean sea surface temperatures than the El Niño–
Southern Oscillation index. These results have important implications for formulating the water resource man-
agement strategy, particularly over the drying central and northern parts of the country.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction 2013; Krishnan et al., 2013). The potential response of the Indian
monsoon circulation, identified as a complex feature of the global
The south Asian countries, home to about one-fifth of the world's climate system, poses challenges of both scientific and societal interest
population, rely heavily upon the summer monsoon rainfall during because of its influence on the economy by providing subsistence
June to September for their socio-economic welfare and environmental to nearly two-third of the country's over 1.2 billion population
sustenance. However, erratic spatiotemporal distributions in the form (Revadekar and Preethi, 2012).
of extreme rainfall events, as illustrated in the recent Intergovernmental The global mean monsoon rainfall has experienced multidecadal
Panel on Climate Change (IPCC) Fifth Assessment Report (Stocker et al., variations during the 20th century (1901–2001), with an overall
2013), have been causing immense economic losses, including the loss increasing pattern up to 1955 driven particularly by the increases over
of human beings. Given the profound implications of the summer the North African, Indian and East Asian monsoon belt, and followed
monsoon rainfall, several research efforts have been undertaken in the by a decreasing pattern up to 2001 (Zhang and Zhou, 2011). In general,
recent past, reflecting substantial spatial variations of the statistically a weakening of the monsoon during the second half of the 20th century
significant rainfall trends (Turner and Annamalai, 2012). The 21st has been reported by several other studies (Zhou et al., 2008; De Luis
century projected increases in both wet and dry extremes, particularly et al., 2009; Wang et al., 2012). Consistently, the Indian monsoon
at a higher rate over the Asian monsoon region due to anthropogenic rainfall has increased (decreased) during the first (second) half,
warming induced intensification of the monsoon rainfall (Kitoh et al., resulting in no apparent trend in the century-scale analysis (Joshi and
Pandey, 2011; Turner and Annamalai, 2012). Several global studies
provide clear indications of warming induced increases in frequency
⁎ Corresponding author. and intensity of extreme rainfall since the 1950s (Stocker et al., 2013;
E-mail address: [email protected] (D.K. Panda). Westra et al., 2013; Zhang et al., 2013).

http://dx.doi.org/10.1016/j.atmosres.2016.07.002
0169-8095/© 2016 Elsevier B.V. All rights reserved.
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 221

During the post-1950 period, the Indian summer monsoon rainfall underscore the importance of our current effort to undertake a compre-
shows a complex spatiotemporal and intra-seasonal pattern in its hensive assessment.
mean and extreme measures. While the aggregated daily rainfall Therefore, the purpose of this study is to examine the spatiotempo-
(R) over central India has displayed a rise in the frequency and intensity ral and regional transitions of the 20th century monsoon rainfall in India
of heavy (R70/R100, R N 70/100mm) and very heavy rain events (R150, using a multitude of statistical measures that encompass the basic
R N 150 mm), along with decreases in the low-to-moderate rainfall rainfall characteristics, threshold- and percentile-based extreme
events (LMR, 5 mm b R ≤ 100 mm) in the background of an warming measures, and duration-based dry and wet spells, mostly defined by
environment (Goswami et al., 2006; Rajeevan et al., 2008; Dash et al., the ETCCDI. Since statistical detection of extreme rainfall events is
2009), a robust drying signal from the weakened monsoon circulation more challenging because of its inherent quality of local occurrences,
is also reported for the same region due to the likely influence of the non-Gaussianity and large interannual variability, it is advisable, in
south Asian aerosol emissions (Bollasina et al., 2011). general, to investigate a wide range of rainfall diagnostics from a
Preceding the 1950s, assessment of changes in rainfall extremes has dense observational network (Groisman et al., 2005; Schmidli and
not been possible in many parts of the world because of scarcity in high Frei, 2005). Therefore, a suit of indices are examined in this study
resolution daily records (Alexander et al., 2006), for which understand- using a high-resolution uniformly gridded daily rainfall dataset during
ing on multidecadal changes in extreme climate over the entire past 1901–2004, thus avoiding the pre-1950 period influence of the irregular
century remains less explained (Moberg et al., 2006; Griffiths and station density as highlighted by Sen Roy and Balling (2004) and Klein
Bradley, 2007). Using a newly updated observational dataset with im- Tank et al. (2006).
proved spatial and temporal coverage since the beginning of the 20th Several previous studies have clearly indicated the varied response
century, Donat et al. (2013) assessed the global patterns of climatic in- of the extreme rainfall indices, particularly defined by the ETCCDI, to
dices developed by the Expert Team on Climate Change Detection and natural modes of climate variability in different parts of the world
Indices (ETCCDI). Although they observed more areas with significant (Kenyon and Hegerl, 2010). For India, however, response of the mean
increasing trends in extreme rainfall, a less clear signal has emerged monsoon rainfall is studied extensively (e.g., Joshi and Pandey, 2011),
for some of the indices, such as the heavy rainfall intensity although more understanding is required on how the ocean forcing
(i.e., maximum 5-day rainfall total, Rx5day) and also in dry spell lengths mechanisms influence their extreme states. During 1901–2004,
(i.e., maximum number of consecutive dry days, CDD). It can also be Rajeevan et al. (2008) observed that the interdecadal variability and
noticed that R10 (R N 10mm), defined by the ETCCDI as a heavy rainfall trend of the central India R150 is modulated by the tropical Indian
event, has decreased uniformly over central-north India (CNI), which Ocean sea surface temperature (TIO SST). Extending to other ETCCDI in-
appears to be the reflection of the LMR (low-to-moderate rainfall dices, the current study investigates the potential influence of local as
events) tendency, based on the LMR definition of the Indian monsoon well as remote SST forcing on trends of the 20th century monsoon ex-
climate. Some continental and country-specific assessments of the tremes. We examine the trend and variability for the sub-periods
twentieth century changes in extreme rainfall include the following: 1901–1940 and 1961–2004, which coincides with the two major
Moberg et al. (2006) for Europe, Kunkel et al. (2003) and Griffths and warming phases of the 20th century, separated by a cooling period
Bradley (2007) for the USA, Gallego et al. (2011) for the Iberian (i.e., 1941–1960). This will provide important insights about the transi-
Peninsula, Brunetti et al. (2004) for Italy, Gallant and Karoly (2010) tion in monsoon rainfall characteristics under two contrasting back-
for Australia, and Villafuerte et al. (2015) for the Philippines. Results ground, as far as anthropogenic warming is concerned. Given the 21st
of these studies suggest that the whole 20th century contains some century uncertainty in the model projections of the mean monsoon
important information, which has not been reflected in the post-1950 rainfall and the ETCCDI indices (Kitoh et al., 2013; Saha et al., 2014;
changes. Hasson et al., 2016), this paper presents for the first time a full picture
For the Indian subcontinent, however, the existing studies focusing of the century scale (i.e., 1901–2004) changes in the ETCCDI indices in
on extreme indices for the whole 20th century are very few, particularly India, which should serve as an useful observational baseline for the fu-
in the context of a large spatial and temporal variability of the monsoon ture model projection and validation.
rainfall. But, valuable contributions have been made with respect to the In the following section, the datasets and indices are described.
analysis of mean rainfall on seasonal or annual scales for different states Section 3 presents the methodology. The results are discussed in
(e.g., Subash et al., 2011; Patra et al., 2012; Duhan and Pandey, 2013; Section 4, and concluding remarks are made in Section 5.
Panda et al., 2013; Goyal, 2014; Pingale et al., 2014). For the whole
country, Sen Roy and Balling (2004) assessed the linear regression 2. Dataset and extreme indices
trends for six extreme rainfall (Rx1day, Rx5day, Rx30day, 90th, 95th,
and 97.5th percentiles) frequencies using daily rainfall over 129 stations The daily gridded rainfall dataset employed in this study for the
for the period 1910–2000, revealing decreases in the eastern part of the period 1901–2004 at a spatial resolution of 1° latitude × 1° longitude
Gangetic Plain and increases in the southern peninsular region of India. has been developed by the India Meteorological Department (IMD)
In homogeneous regions of India, Guhathakurta et al. (2015) reported a employing the Shepard's interpolation method. The observed rainfall
multidecadal variability of different phases in the monthly rainfall data from N 1800 gauges was used to develop this gridded dataset,
datasets during 1901–2011, while the core monsoon region experi- and multi-stage quality control steps were followed to minimize
enced significant decreases in the moderate rainfall events (i.e., LMR) temporal inhomogeneity due to varying station densities, which is
but no change in R100 and R150 during 1951–2010. In central India, detailed in Rajeevan et al. (2008). We opt this dataset over that of the
however, Rajeevan et al. (2008) observed increases in R150 during recently updated rainfall dataset at a spatial resolution of 0.25°
2001–2004. Comparing the pre- and post-1950 extreme rainfall events, latitude × 0.25° longitude, because the employed dataset is validated
Vittal et al. (2013) attributed the changes in the post-1950 period to the with a range of applications in climate assessment studies until now
impacts of rapid urbanization. In contrast, Ali et al. (2014) noted no such (e.g., Rajeevan et al., 2008; Joshi and Pandey, 2011; Krishnamurthy
trends during 1901–2010, considering four different measures extreme and Krishnamurthy, 2013; Vittal et al., 2013). Moreover, our results
rainfall. In the Himalayan domain of the northeast India, Singh and can be compared with the previous ones without anticipated bias
Goyal (2016) highlighted the influence of elevation on the mean and from different datasets. For a more recent qualitative comparison of
extreme measures of rainfall. Spatially and sub-seasonally less IMD rainfall datasets at different spatial resolutions, Guhathakurta
consistent patterns reported in the literature particularly due to use of et al. (2015) may be referred. To calculate the monsoon rainfall indices
different study period, datasets, spatial domains and varying definitions defined in Table 1, the time series from 306 evenly distributed grids
of extreme rainfall (Ghosh et al., 2012; Panda and Kumar, 2014) over the Indian subcontinent are used.
222 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

Table 1
List of rainfall indices used in this study for the monsoon season (June to September).

Category Index Definition Unit

Absolute indices Rx1day Maximum 1-day rainfall total mm

Rx5day Maximum 5-day rainfall total mm
Threshold indices R70/100/150 (nn) Number of days with rainfall 70/100/150 mm or more Days
Duration indices CWD Consecutive wet days. Maximum number of consecutive wet days (i.e. when R ≥ 1 mm) Days
CDD Consecutive dry days. Maximum number of consecutive dry days (i.e. when R b 1 mm) Days
Percentile-based indices R95p Very wet days. Total rainfall due to events exceeding the 95th percentile of the 1961–1990 base mm
period
R99p Extremely wet days. Total rainfall due to events exceeding the 99th percentile of the 1961–1990 mm
base period
R95pTOT Contribution from very wet days. Fraction of total rainfall due to events exceeding the 95th %
percentile of the 1961–1990 base period
R99pTOT Contribution from extremely wet days. Fraction of total rainfall due to events exceeding the 99th %
percentile of the 1961–1990 base period
Basic indices WD Wet days. Number of days with rainfall (R) of ≥1 mm Days
PRCPTOT Total rainfall (precipitation) on days with ≥1 mm rain mm
LMR Low-to-moderate rainfall events. Number of days with rainfall between 5 mm and 100 mm Days
(i.e. 5 mm b R ≤ 100 mm)
SDII Simple daily intensity index. Total rainfall divided by the number of wet days (i.e. average rainfall of mm day−1
the on days with rainfall ≥1 mm)

Full definitions are available from the ETCCDI website http://cccma.seos.uvic.ca/ETCCDI/.

The SST-based indices are derived from the National Oceanic and below this value is a dry day, while comparisons with the IMD defined
Atmospheric Administration National Climate Data Center (NOAA wet day (R ≥ 2.5 mm) yield no apparent difference. The extreme wet
NCDC) monthly Extended Reconstructed Sea Surface Temperature and dry rainfall indices investigated in this study (Table 1) fall roughly
version 3 dataset (ERSST.v3b) (Smith et al., 2008). Specifically, we con- into five different categories as presented in Alexander et al. (2006).
struct TIO SST index by averaging the SST anomalies over the tropical Nevertheless, the threshold-based frequency of heavy (R10) and very
Indian Ocean (20° S–30° N, 50–120° E), and the El Niño–Southern heavy rainfall events (R20) are replaced by the corresponding appropri-
Oscillation (ENSO) index by averaging the monsoon season SST anoma- ate definition of R70, R100 and R150 for India (i.e., Rnn, monsoon count
lies over the NINO3.4 region (5° S–5° N, 170°–120° W). These two of rainfall above a user-defined threshold). In order to spatially compare
indices are widely used in the literature to link variations of the mon- the extreme rainfall within the diverse climatology of India (Fig. 1) and
soon rainfall. also for the international comparisons, the percentile-based rainfall
Consistent with the ETCCDI definitions of extreme indices in Table 1 intensity, such as the total rainfall from very and extreme wet days
(Zhang et al., 2011; Donat et al., 2013), daily rainfall (R) ≥ 1 mm in the (R95p and R99p) (i.e., above the 95th and 99th percentile of the base
monsoon season (June to September) is considered as a wet day and period 1961–1990), have relative advantage over the fixed thresholds

Fig. 1. Spatial pattern of the mean monsoon (June–September) rainfall rates (mm day−1) during the (a) 1901–1940 and (b) 1961–2004 periods in India. The central-northern India (CNI),
indicated by the box (20°–28° N, 76°–87° E), encompassing parts of the core monsoon belt and the Ganga River basin, shows a noticeable rainfall decline during the recent period. Other
regions indicated by boxes are the northwest India (NWI), southwest India (SWI) and southeast India (SEI).
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 223

indices (Klein Tank and Kӧnnen, 2003). Even if the indices R95pTOT and Under the null hypothesis, for n ≥ 10, the test statistics S is
R99pTOT, representing the fraction of monsoon rainfall due to very and approximately normally distributed. Using a two-sided test, the trend
extreme wet days are not included in the ETCCDI definitions, they are is statistically significant at the level α if the absolute value of the
identified to have significant social impacts by the Hadley Center Global standard normal variate, |Z| N Z1-α/2, where Z is given as
Climate Extremes Index 2 (HadEX2) (Donat et al., 2013). Furthermore, 8
the magnitude of intense rainfall events are assessed through the > S−1
>
> pffiffiffiffiffiffiffiffi2 for SN0
indices Rx1day, Rx5day and the average amount of rainfall received >
< σs
on wet days (i.e., SDII). Z¼ 0 for S¼0 ð3Þ
>
> Sþ1
The basic rainfall characteristics, such as total monsoon rainfall from >
> pffiffiffiffiffiffiffiffi Sb0
: for
wet days (PRCPTOT), wet days frequency (WD) and low-to-moderate σ s2
rainfall events (LMR), are important, since they are closely tied with
the upper and lower tails (extremes) of the distribution; for example, A positive (negative) value of Z indicates an upward (downward)
under global warming increases in rainfall intensity concurs with trend. In this study, the significant trends are detected at the 5% and
increases in dry spells through decreases in WD and LMR, leading to a 10% significance level (i.e., p-value of ≤0.05 and 0.1) unless mentioned
little or no change in the mean rainfall (PRCPTOT) (Goswami et al., otherwise. Presence of serial correlation in time series has been
2006; Giorgi et al., 2011). Although decreases in consecutive wet days addressed through the trend-free pre-whitening (TFPW) procedure
(CWD) and/or increases in consecutive dry days (CDD) reflect the (Yue et al., 2002). The magnitude (β) of trends is determined using
length of the driest part of the monsoon season from climate change the Theil–Sen approach (Zhang et al., 2001), which is also a robust non-
prospective, but the agricultural and hydrological droughts are directly parametric method given as
linked to LMR. Since droughts have posed serious challenges to water  
resources and ecosystems of India, to supplement the above indices, X j −X i
β ¼ Median ∀i b j: ð4Þ
we use the monthly self-calibrated Palmer Drought Severity Index j−i
(PDSI) with a spatial resolution 2.5° latitude × 2.5° longitude spanning
over January 1961 to December 2010, which is a prominent index for We assess the common trend over the regions depicted in Fig. 1
characterization of a meteorological drought taking into account the employing the analytical method developed by Douglas et al. (2000).
cumulative effect of rainfall and temperature changes (Dai et al., Using the test statistics S of individual grid points, the regional average
2004). In order to understand the contrasting regional diversity of the of the MK test statistics is defined as
monsoon climatology, spatially aggregated dry and wet extremes are
compared for four regions of India (Fig. 1): the central-northern India X
m
Sm ¼ Sg ð5Þ
(CNI; 20°–28° N, 76°–87° E), northwest India (NWI; 23.5°–31.5° N, g¼1
71.5°–76.5° E), southwest India (SWI; 11.5°–21.5° N, 73.5°–76.5°
E) and southeast India (SEI; 8.5°–16.5° N, 77.5°–80.5° E). The grid where Sg represents the MK test statistics for the grid g of a region with
scale standardized anomalies, based on the 1961–1990 climatology,
m grids. For an independent and identically distributed dataset, Sm
are filtered through a 10-year running mean to remove the sub-
follows a normal distribution with mean 0 and variance σ2s m− 1, and
decadal variability (Joshi and Pandey, 2011), and the regionally
their ratio gives the normalized test statistics Zr. However, for a spatially
averaged anomalies are plotted to facilitate a visual comparison of the
correlated domain, as pointed by Kulkarni et al. (2012) for the monsoon
existing multidecadal variability.
rainfall of India, the variance of Sm becomes
3. Methodology   σ 2h i
s
Var Sm ¼ 2 1 þ ðm−1Þρg;gþl ; and ð6Þ
m
3.1. Trend detection
X m−g
m−1 X
Presence of monotonic trends in rainfall indices have been evaluated ρg;gþl ¼ 2m−1 ðm−1Þ−1 ρg;gþl ð7Þ
using the Mann-Kendall (MK) nonparametric test (Sneyers, 1999; g¼1 l¼1
Zhang et al., 2001), which does not make assumptions about the
underlying distribution, and also is insensitive to effects of outliers in represents the average cross-correlation for the region (Douglas et al.,
the extreme indices. Since world-wide analysis suggests that several 2000).
extreme indices do not necessarily follow a Gaussian distribution The partial Mann–Kendall test (PMK) (Libiseller and Grimvall, 2002)
(Alexander et al., 2006), this test is one of the most widely used statisti- has been used to detect significant trends in the mean and extreme
cal tests to detect the hydro-climatological signals of climate change and indices of monsoon rainfall after accounting the correlation with the
variability given its advantages over the standard regression trend test SST-based large-scale climatic modes (i.e., TIO SST and NINO3.4 indices)
(e.g., Westra et al., 2013). However, the null hypothesis of the MK test as covariates. Using the conditional mean and variance of the dependent
requires that data (X1, X2, …, Xn) are a sample of n independent and variable (rainfall indices), the PMK is computed as
identically distributed random variables.
The MK test statistics, S, is defined as Sy −ρ ^ Sx
PMK ¼ r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi ð8Þ
2
1−ρ ^ nðn−1Þð2n þ 5Þ=18
X
n−1 X
n  
S¼ sgn X j −X i ð1Þ
i¼1 j¼iþ1 where Sx is the MK statistics of the covariate and Sy is the MK statistics of
the dependent variable and ρ ^ represents the correlation Sx and Sy. Under
where Xi and Xj are the data values at times i and j of the time series of the null hypothesis, the PMK follows a normal distribution with mean 0
length n. The sgn is evaluated as and standard deviation 1. A significant (p ≤ 0.1) PMK statistics suggests
that the dependent variable co-varies with the trend of the independent
8
< þ1 for θ N0 SST indices. Additionally, we compare with the results from the method
sgnðθÞ ¼ 0 for θ ¼ 0 ð2Þ followed by Vincent et al. (2015) that first assess the influence of the
:
−1 for θ b 0 independent variables by regressing with the grid scale response
224 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

Fig. 2. (a) Time series of the 10-year running mean of normalized anomalies of the monsoon rainfall indices during 1901–2004 averaged over geographical regions: all India (AI), central-
north India (CNI; 20°–28° N, 76°–87° E), northwestern India (NWI; 23.5°–31.5° N, 71.5°–76.5° E), southwest India (SWI; 11.5°–21.5° N, 73.5°–76.5° E), and southeast India (SEI; 8.5°–
16.5° N, 77.5°–80.5° E), arranged in respective rows. The grid point time series are normalized by their means and standard deviations using the base period 1961–2000, and then are
smoothed with a 10-year running mean to remove the subdecadal variability. (b) The monsoon season normalized SST anomalies over the tropical Indian Ocean (TIO SST; 30°N–20° S,
50° E–120° E) and over the NINO3.4 region (5° S–5° N, 170°–120° W) during 1901–2004, along with the embedded lines of the 10-year running means.
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 225

variable and then applying the MK test to the residual time series to 1961–2004 (Fig. 4). This spatial predominance rainfall increasing trends
identify trends. during 1901–1940 is reflected through the identification of a statistical-
Detection of trend is indicative of a non-stationary climate, which ly significant (p b 0.1) regional trend (Zr) after accounting their spatial
could have arisen due to a change in mean, in variability, or both. In correlations in PRCPTOT, WD, LMR over the CNI sector only. In
order to better diagnose these changes, we compare the probability particular, significant rise in LMR appears to have contributed to
density function (PDF) of the pooled grid scale meteorological indices the rise of 39.75 mm decade−1 in PRCPTOT (Table 2). From the post-
in different sub-periods of the 20th century. Furthermore, even if trends 1960 frequency distribution in Fig. 3, it is clear that most parts of the
are not detected due to the high interannual and interdecadal variability country (i.e., about 70% of 306 grid points) have experienced a drying
of the monsoon climate, then its extreme nature can be reflected tendency in wet days (WD) and low-to-moderate rainfall events
through the heavy tails in the PDFs. Significance of the difference in (LMR), though the distribution of the significant trends (Fig. 4a and
mean and variability has been evaluated using the Student's t-test and d) are location-specific, without identification of any regionally signifi-
Levene's F-test, respectively. cant pattern.
In general, LMR, an important basic rainfall index that primarily
4. Result and discussion drive changes in PRCPTOT not only for India but also for the whole
Asian continent, has declined over southwest Asia (Yao et al., 2008).
4.1. Trends in basic indices From the Indian socioeconomic prospective, LMR events (contribute
about 85% of PRCPTOT) sustains the monsoon season agriculture,
Fig. 2a illustrates the nationally and regionally averaged normalized improves the recharge rates and base flows by increasing the opportu-
anomalies, with the sub-decadal variability smoothed by the 10-year nity time, and also ensures security of engineering structures in compar-
moving mean, in the basic and extreme measures of the monsoon ison to that of the extreme events. Between the pre-1940 and the post-
(JJAS) rainfall in India during the 20th century (1901–2004). The most 1960 periods, a clear signal of transition in WD and LMR does not
interesting feature of the multidecadal oscillation is the conspicuous commensurate with the changes in PRECPTOT, particularly in the
difference in the trend and variability between the pre-1940 (1901– frequency and spatial consistency of trends during 1961–2004
1940) and the post-1960 (1961–2004) periods, which corresponds to (Figs. 3–4). This raises the question if the extreme matrices have also
the two observed warming phases of the TIO SST index (Fig. 2b). The undergone any perceptible change, as explored in the following
basic indices (PRCPTOT, WD, LMR) appear to have two inverted-U sections.
shaped above-normal phases during the pre-1970 period, and followed
by the presence of simultaneous below- and above-normal rainfall 4.2. Trends in wet and dry spells
epochs (Fig. 2a), with significant (p b 0.05) correlation coefficients
ranging from 0.70 to 0.94 for most of the regions. For the unsmoothed Consistent with the wet day (WD) pattern, it is worth noticing that
total monsoon rainfall (i.e., PRCPTOT), the pre-1940 (post-1960) period the average longest wet spells, measured through the consecutive wet
has experienced a significant increasing (nonsignificant decreasing) day (CWD) index, have increased (decreased) during the pre-1940
trend (Table 1), with the mean rainfall of 900 mm (896 mm) and (post-1960) period, particularly over CNI and NWI, comprising the
standard deviation of 87 mm (96 mm). This nonlinear century-scale northern parts of the country (Fig. 2a). Regionally averaged time series
behaviour of the monsoon rainfall, with a near-neutral trend of show significant downward trend in CWD for all the four domains and
0.1 mm decade− 1, has also been earlier reported (e.g., Joshi and the whole country during the post-1960 period (Table 2). This reversal
Pandey, 2011). The patterns observed in this study are identical to the of trend evolution, from a wetting spell to that of a drying spell between
global and Northern hemispheric land monsoon precipitation during the first and later phase of the last century, is more evident from the
1901–2001 identified by Zhang and Zhou (2011) using the MK test. frequency distribution in Fig. 3 as well as the widespread spatial cover-
Therefore, this study confirms the reported major contributions of the age of the significant trends in Fig. 5; particularly, those post-1960
Indian monsoon rainfall to the global monsoon trends. decreases have a regionally significant trend (Zr, p b 0.1) for all the
Noteworthy is the substantial regional differences of the basic and domains. However, trend analysis over the whole 20th century
extreme indices, which have not been clearly reflected in the nationally (1901–2004) appear to have reflected to a large degree the post-1960
averaged time series (Fig. 2a). Whereas the northern parts of India, in- drying pattern in WD, LMR and CWD, specifically over central-north
cluding central-north India (CNI) and northwest India (NWI), have ex- India (CNI) and northwest India (NWI), thus offsetting the wetting sig-
perienced a nonsignificant decline in total rainfall (PRCPTOT) during nals of the pre-1940 period.
1901–2004, an opposite tendency can be noticed in southwest India More important, however, is to find that the temporal evolution of
(SWI) and southeast India (SEI) (Table 2). But all the four regions the 20th century longest dry spells (CDD) does not reflect clearly the
have a common negative trend during 1961–2004, with the largest signals that we observe using the CWD index, although both the indices
decline of 45.98 mm decade−1 in the southwest region, part of the do- are expected to complement each other through opposite signs, as seen
main through which the moisture-laden monsoon wind enters the during the pre-1940 period (Fig. 3). Particularly, the post-1960 CWD
Indian subcontinent. The core monsoon region (18°–28° N, 73°–82° has dropped significantly in grid scale and regional scale, but no clear
E), identified for having experienced a drying tendency along with signal in dry spells (Table 2 and Fig. 3). This type of inconsistency
high spatiotemporal variability (Singh et al., 2014), also shows a post- among extreme indices has already been highlighted in the literature
1960 decline of 12.81 mm decade− 1. In central-north India (CNI), (Kiktev et al., 2003; Alexander et al., 2006), and detections of the region-
which is a densely populated and intensely cultivated landscape al trends here can partly be attributed to the filtering of noises in clima-
encompassing of the core monsoon belt and the Ganga River basin, tologically heterogeneous settings of India. Interestingly, global scale
the monsoon rainfall has reduced from 1000 mm in the pre-1940 to analysis has displayed a marked decline in the post-1950 globally
959 mm in the post-1960 period, with respective standard deviation averaged annual CDD (e.g., Alexander et al., 2006; Donat et al., 2013;
of 104 mm and 129 mm. Fischer and Knutti, 2014), with a noticeable congregation of the down-
Results of the grid scale trend analysis reveal no clear change in ward trends over central and north India. This contradicts the observed
PRCPTOT during the pre-1940 period, but there is a higher proportion drying trends evident from WD, LMR and CWD over the same region
of increasing trends in WD and LMR (Fig. 3). Map of the significant (Fig. 3) and also the previously reported rainfall declines over the core
(p b 0.1) trends manifests a noticeable regional distinction, in which monsoon region (e.g., Bollasina et al., 2011; Singh et al., 2014).
most of the pre-1940 rising trends concentrated over central and Moreover, comparing different HadEX datasets, Donat et al. (2013)
north India have turned out to be neutral or decreasing trends during show a neutral pattern in CWD during 1951–2003.
226 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

Table 2
All India and regional trends (per decade) in the unsmoothed time series of the monsoon rainfall indices.

Indices Period All India (AI) Central-north-India (CNI) Northwest India (NWI) Southwest India (SWI) Southeast India (SEI) Core monsoon regiona

PRCPTOT 1901–1940 23.98 39.75 5.14 21.36 −5.92 39.05

1961–2004 −11.89 −4.80 −9.25 −45.98 −5.98 −12.81
1901–2004 0.10 −3.30 −0.20 0.12 3.05 −1.04
WD 1901–1940 1.01 1.60# −0.21 1.13 −0.74 1.07
1961–2004 −0.53 −1.03 −1.01 −1.04 −1.24 −0.92
1901–2004 −0.10 −0.25 −0.18 −0.04 −0.03 −0.14
SDII 1901–1940 0.13 0.24 −0.03 0.13 0.07 0.31
1961–2004 0.03 0.09 0.11 −0.32# 0.17 −0.03
1901–2004 0.05 0.02 0.06 0.03 0.09 0.06
LMR 1901–1940 0.77 1.04# 0.12 1.00 −0.36 0.89
1961–2004 −0.55 −0.77 −0.64 −1.38 −0.68# −0.89
1901–2004 −0.08 −0.15 −0.04 −0.05 0.09 −0.02
CDD 1901–1940 −0.34 −0.55 0.26 −0.40 0.08 −0.44
1961–2004 0.14 0.04 0.42 0.23 0.85# 0.00
1901–2004 −0.04 0.03 −0.08 0.06 0.11 −0.02
CWD 1901–1940 0.22 0.49 0.00 0.82 −0.09 0.32
1961–2004 −0.38 −0.93 −0.37 −1.28 −0.37 −0.78
1901–2004 −0.13 −0.26 −0.13 −0.19 0.04 −0.20
R70 1901–1940 0.06 0.14 −0.01 0.05 0.00 0.10
1961–2004 0.01 0.05 0.00 −0.15 0.03 0.01
1901–2004 0.02 0.01 0.00 0.01 0.01 −0.01
R100 1901–1940 0.03 0.06 −0.01 0.03 0.00 0.04#
1961–2004 0.01 0.04# 0.00 −0.03 0.00 0.02
1901–2004 0.01 0.00 0.00 0.00 0.01 0.005
R150 1901–1940 0.01 0.02 0.00 0.00 0.00 0.01
1961–2004 0.008 0.007 0.00 0.008 0.00 0.01
1901–2004 0.007 0.003 0.00 0.00 0.00 0.004
Rx1day 1901–1940 1.08# 0.91 0.34 −1.75 1.40 1.12
1961–2004 1.13 2.27# 0.96 −0.72 −2.56 2.55#
1901–2004 0.40 0.78 0.94 −0.28 −0.44 0.84
Rx5day 1901–1940 2.39 3.44 −0.61 −0.38 2.48 2.23
1961–2004 0.66 1.69 1.27 −3.53 −4.97 2.10
1901–2004 0.11 0.71 1.15 −0.03 −1.36# 1.14
R95p 1901–1940 4.34 20.70 −0.43 5.45 −0.16 16.30
1961–2004 6.19 6.19 1.34 −6.51 5.02 2.82
1901–2004 1.84# −0.14 0.81 1.60 2.03# 1.44
R99p 1901–1940 −0.21 9.49 −0.37 0.97 1.23 4.97
1961–2004 8.65 6.06 2.85 1.85 3.94 4.98
1901–2004 1.35 0.47 0.65 0.22 1.11 0.88
R95pTOT 1901–1940 −0.08 0.79# 0.10 0.43 0.53 0.84#
1961–2004 0.59 0.39 0.91 −0.10 1.15 0.33
1901–2004 0.14# 0.00 0.24 −0.02 0.26 0.11
R99pTOT 1901–1940 −0.21 0.47# −0.25 0.23 0.27 0.31
1961–2004 0.55 0.23 0.44 0.27 0.59# 0.42
1901–2004 0.08# 0.04 0.04 −0.05 0.11 0.06

Bold and bold with # values correspond to the trends significant at the 5% and 10% levels, respectively (i.e., p b 0.05 and 0.1).
a
Core monsoon region (18°–28° N, 73°–82° E).

It is, therefore, important to discuss whether the CDD index is about 66% of the grid points during July to September, similar to the
relevant enough to assess the dry spells of the monsoon season, or the whole monsoon scale results (Fig. 3), while an opposite tendency with
CWD index is more informative, given their sensitivity to the dichotomy similar proportions is seen in June.
of rainfall distribution within the season. In fact, there is a likelihood Consistently, several previous researches have pointed out this
that the longest wet spell (CWD) could occur in the rainiest month of dichotomy within the monsoon season, a wetting June and a drying
July and/or August (both the months contribute about 62% of PREPTOT), July to September months (e.g. Gautam et al., 2009; Lau and Kim,
while the longest dry spell (CDD) could be in the relatively dryer 2010; Panda and Kumar, 2014). Thus, it appears logical to infer that
months of June or September. While there is strong agreement in the the CWD index is pertinent to capture the observed drying patterns of
frequency and spatial locations of significant trends between the pre- the recent decades. Moreover, agricultural droughts during the critical
1940 CDD and CWD shown in Figs. 3 and 5, the contradiction of the growth stages of the monsoon rain-dependent crops, for example in
post-1960 period can be rooted to the recent non-uniform rainfall 2000 and 2002 causing significant socioeconomic crisis in India
distribution. A part of the post-1960 decreases in dry spells could be a (Gadgil and Gadgil, 2006), is mainly due to a drop in WD and CWD in
manifestation of the increased wetness of June rather than the drying July and August.
tendency from July to September. Monthly analysis of CDD and CWD
provides reasonable explanation to such presumption, as the nationally 4.3. Trends in extreme indices
averaged June CDD has exhibited a significant downward trend during
1961–2004 (also complemented by a significant rise in CWD), but Noteworthy feature of the 20th century monsoon climate is the
contradicts the nonsignificant increases in other months of the mon- conspicuous rise in the frequency and intensity of the extreme rainfall
soon season. This finding is also corroborated by the grid scale results, indices during the post-1960 period, as depicted in the nationally
with the dry spell (CDD) declines in 74% of the geographical area in averaged smoothed time series of the wet extremes (e.g., SDII, R100,
June, whereas the rest of the monsoon months indicate increases over Rx1day, and R95pTOT) (Fig. 2a). Nationally, an initial low-frequency
55% of grid points. For CWD, a decreasing tendency is observed in and below-average oscillation can be seen before a high frequency rise
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 227

Fig. 3. Comparison of the percentage of 306 grid points with positive (P) and negative (N) trends along with the percentage of their significant (p b 0.1) trends (shaded) in the basic and
extreme monsoon rainfall indices (Table 1) during the (a) 1901–1940, (b) 1961–2004 and (c) 1901–2004 period.

since the 1970s. However, the patterns over central-north India (CNI), some indices over CNI and all India (Table 2). However, it is important
where most of the extreme events generally occur, stand out clear in to note how the level of spatial aggregation and definition of extremes
terms of two sharp rises, one each during the first and last decades of have yielded different results, which has led to disagreement over the
the study period. Although identification of grid scale trends in extreme spatiotemporal evolution in the recent studies (e.g., Ghosh et al.,
matrices (Fig. 3) is naturally difficult, spatial averaging, as suggested by 2012). For example, analysing the south Asian summer rainfall, Yao
Goswami et al. (2006), appears to have yielded significant trends for et al. (2008) reported a downward trend in extreme and heavy rainfall
228 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

Fig. 4. The spatial map of the significant (p b 0.1) trends in (a) PRCPTOT, (b) wet days, (c) SDII and (d) LMR during 1901–1940, 1961–2004, and 1901–2004. Blue triangles show the
increasing trends and red inverted triangles show the decreasing trends, with the sizes of the triangles proportionate to the rate of change (i.e. Kendall slope) per decade.

during 1979–2002 over India using the index R50, contradicting our From the frequency distribution of the grid scale results in Fig. 3, the
findings in R70 (Table 2). This means that R50 has reflected a large significant increasing trends are more often detected during the 1961–
part of the LMR events, which is not captured in R70. 2004 and 1901–2004 periods in the percentile-based rainfall intensity
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 229

Fig. 5. Same as Fig. 4, but for (a) CDD, (b) CWD, (c) R95p and (d) R95pTOT.

indices (R95p, R99p, R95pTOT and R99pTOT), which are importance for distribution of the significant trends of average wet day rainfall intensity
the global climate change assessment and comparisons. Their spatial index (i.e., SDII) and wet day frequency (WD) shown in Fig. 4b and c
occurrence (Figs. 4–6) indicates that the stronger upward trends, partic- suggests that the increases in SDII could be due to the decreases in
ularly in contribution from very and extreme wet days, R95pTOT and WD. However, a significant rise in R95pTOT and R99pTOT, which are
R99pTOT (Fig. 5d), have a larger geographic coverage. A similar spatial insensitive to WD, explains that both these episodes of decreases in
230 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

Fig. 6. Same as Fig. 4, but for (a) Rx1day, (b) Rx5day, (c) R70, and (d) R100.

WD and increases in extreme events have co-occurred. A spatially less India only. Similarly, although Rx1day and Rx5day are spatially repre-
regular pattern is observed in the threshold-based frequency of heavy sentative indices of absolute rainfall intensity, they only capture a part
rainfall events, R70 and R100 (Fig. 6c and d), because these are local of the entire monsoon period with a high year-to-year variability.
events confined to certain parts of central, northeast and southwest Although the spatial outlines of the core monsoon and CNI regions are
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 231

slightly different, R99p and R99pTOT have a regionally significant trend are likely to be driven by the trends during 1961–2004 for most of the
(Zr) over the core monsoon region, while Rx1day is significant over CNI indices. From the spatial distribution of the 1901–2004 period trends,
during the post-1960 period. however, it appears that the large-scale pre-1940 rising patterns in
Nevertheless, use of a high resolution gridded dataset has allowed us WD and LMR have been masked by that of the post-1960 period trends,
to identify the patches where the indices with potential threat to human while a part of the pre-1940 increases in PRCPTOT is retained,
and infrastructure have increased significantly in the post-1960 period. possibly due to the cancellation of the observed dry and wet extremes
In particular, irrespective of the indices, parts of the central and north- during 1961–2004. Moreover, with the increasing wet extremes during
west India appear to be more vulnerable, where most of the upward 1961–2004, the observed decreases in wet days can be attributed to the
trends of high magnitudes have congregated (Figs. 4–6). These results declining LMR frequency. In general, statistical detection of trends in ex-
are consistent with the findings of Vittal et al. (2013) who used the treme rainfall indices remains challenging due to its nonlinearity and its
same dataset but a different methodology on percentile-based very local nature of occurrence (Groisman et al., 2005), and this could be
measures of rainfall extremes. The 20th century (1901–2004) changes the primary reason for the spatially less consistent patterns observed in

Fig. 7. Probability density functions (PDFs) of some selected monsoon rainfall indices (a to f) during the 1901–1940 (pre-1940, blue continuous line), 1961–1985 (pre-1985, discrete line)
and 1986–2010 (post-1985, discrete with dot line) periods.
232 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

this study and also in other parts of the world during the second half of 4.5. Trends in palmer drought severity index (PDSI)
the 20th century (e.g., Zolina et al., 2008; Donat et al., 2013; Fischer and
Knutti, 2014). To draw further insights into the drying monsoon climate of recent
decades in India, we use the Palmer Drought Severity Index (PDSI),
which has been used widely as a meteorological drought index (Dai,
4.4. Changes in probability density function (PDF) 2011). Fig. 8 illustrates the monsoon season spatial average and
variance of the monthly PDSI index over all India (AI) and central-
Fig. 7 illustrates the spatiotemporal changes within the second half north India (CNI) during 1961–2010. The observed high interannual
of the 20th century and departure from that of the first half through variability of the average PDSI is consistent with the rainfall (PRCPTOT)
the probability density function (PDF) during the 1901–1940 (pre- variability, with a significant (p b 0.05) correlation coefficient of 0.86
1940), 1961–1985 (pre-1985) and 1986–2010 (post-1985) periods and 0.77 for AI and CNI, respectively during the common period
from the grid scale monsoon rainfall indices over central-north India 1961–2004. Occurrence of both the negative (dry) and positive (wet)
(CNI), which has been highlighted in the literature because of its climat- anomalies, with some century scale extreme droughts since 2000, has
ic evolutions (Bollasina et al., 2011; Singh et al., 2014). Note that, in the resulted in a nonsignificant drying trend in the mean PDSI of the AI
first 40 years (pre-1940), the PDFs of the basic rainfall indices and CNI time series. A significant rise in the spatial variance of PDSI,
(PRCPTOT, WD and LMR) resemblances a normal distribution, but in with marked increases since the mid-1980s (Fig. 8), is indicative of
the later periods (i.e., pre-and post-1985) a clear asymmetry is evident the lack of uniform pattern of both dry and wet conditions, as highlight-
with a change towards the dryer part. For PECPTOT, changes in the ed by Ghosh et al. (2012) and Singh et al. (2014). This increased spatial
mean between the former (pre-1940) and later periods are statistically variability in PDSI is consistent with the post-1985 period increased
significant (p b 0.05), but not the changes in variance. For WD and LMR, mean and variances of the extreme rainfall indices over CNI (Fig. 7).
however, both the mean and variance have changed significantly We, therefore, compare the pre-1985 (1961–1985) and post-1985
among the sub-periods; similar is also the changes in the PDFs of (1986–2010) period PDFs of all India (AI) and central-north India
CWD (Fig. 7d). In extreme indices, the change in Rx1day is more notice- (CNI) monthly PDSI index to detect if such non-stationarity exists
able compared to R99pTOT (Fig. 7e and f), with the recent period (Fig. 9). From the national distribution, signal of an increased
displaying a significant rise in the mean Rx1day over that of the pre- occurrence of both dry and wet extremes is evident from the raised
1940 and pre-1960 periods. Moreover, Rx1day is also characterized by opposite PDF tails, with a higher frequency of the post-1985 extreme
a significant change in variance. These changes are in agreement with droughts. In general, a change in the tails of the PDF is more important
the expected shifts from the normal to weak and strong monsoons from climate change impacts and risk assessment prospective rather
(i.e., change in tails of the PDFs) in a warmer climate (Turner and than the mean. However, the major mode of the PDF (i.e., peak),
Annamalai, 2012). which corresponds to the negative anomalies, suggests the prevalence
of a drying condition during the whole 1961–2010 period (Fig. 9),

Fig. 8. Time series of the monsoon season spatial average and variance of monthly PDSI for Fig. 9. Probability density function (PDF) of the pre-1985 (1961–1985, continuous line)
(a) all India (AI) and (b) central-north India (CNI) during 1961–2010, along with their 10- and the post-1985 (1986–2010, discrete line) period: (a) all India (AI) and (b) central-
year running mean (bold line for the average and discrete line for the variance). north India (CNI) monthly PDSI.
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 233

thereby lending support to the widely reported weakening of the mon- b), except partially for R100 (r = 0.30, p b 0.05) and R150 (r = 0.40,
soon circulation. p b 0.05) with TIO SST. But removal of the decadal variability through
In particular, the spatially prominent drying patterns in the central smoothing the TIO SST time series with a 10-year running mean, as
and north-central parts discussed in the previous section are clearly followed by Rajeevan et al. (2008), led to an improvement of the linear
reflected in its PDSI, with a negatively skewed PDF (elongated left tail) relationship with R100 (r = 0.71) and R150 (r = 0.83); moreover, a
and negative modes (Fig. 9b). Results of the global scale PDSI analysis significant inverse relationship is observed with WD, LMR and CWD in
have also indicated a drying trend over the northern parts of India, as contrast to a significant positive relationship with most of the wet
an integral part of a general rise in aridity in south Asia since 1950 extremes.
(Dai, 2011). Moreover, for the adjacent country China, Sun et al. While the smoothed pre-1940 time series of all the considered
(2016) observed more drying trends in the Loess Plateau during rainfall indices reflect a relatively high degree of positive (negative for
1986–2013 compared to that of 1960–1985 due to a weaker eastern CWD) correspondence with the TIO SST anomalies (average r of 0.62),
Asian summer monsoon. Nevertheless, it can be seen that the frequency a transition in the sign of correlation is observed during the post-1960
and intensity of droughts in India have increased in recent years, partic- period for PRCPTOT, WD, LMR and CWD, with an average r of − 0.32.
ularly the occurrence of century scale droughts in 2002 and 2009, Important to note that the TIO SST (Fig. 2b) has experienced a stronger
leading to changes in the mean PDSI state over CNI, as the post-1985 warming rate of 0.12 °C decade− 1 during 1961–2004 in comparison
PDF indicates a slight negative shift along with a noticeable rise in the to 0.07 °C decade− 1 during 1901–1940, while a cooling trend
PDF peak (Fig. 9b); in particular, the 2009 drought effect, as reflected (i.e., −0.11 °C decade−1) is observed during 1941–1960. In the back-
in terms of an anomalous PDSI value (Fig. 8b), also led to pronounced ground of a little warming of the land surface temperature during
terrestrial water storage (TWS) losses in the groundwater-irrigated 1901–1940, with Tmax (daytime) experiencing a nonsignificant trend
northern and central parts of India (Panda and Wahr, 2016). Consistent- of 0.03 °C decade− 1 and Tmin (night) a neutral trend, the pre-1940
ly, Singh et al. (2014) observed more decreases in the mean rainfall and rainfall changes is likely to be associated with the increasing tropical
wet spell frequency in the core monsoon belt covering a major portion depression frequency (i.e., 0.45 decade−1) that generally form over
of CNI during 1981–2011 compared to that of the period 1951–1980. the Bay of Bengal and propagate northwestward into the monsoon
For the all India (AI) PDSI values shown in Fig. 9a, a change in the trough, contributing to the mean and extreme rainfalls over CNI. In
upper and lower tail of the PDF with no clear difference in the mean addition to the ocean warming, both Tmax and Tmin have shown a
suggests the increased occurrence of opposite extremes. pronounced warming rate of 0.11 and 0.13 °C decade−1, respectively
during the post-1960 period, a period during which most of the conflict-
4.6. Linkages with SST ing trend results have been reported.
It is therefore important to find the extent to which the observed
Identification of the causes of the trends and variability of the grid scale trends co-vary with the TIO SST and NINO3.4 indices during
monsoon rainfall of India remains an active area of research because 1961–2004. The PMK test results show that the strength of the trends
of the complex interaction between the Ocean and the landmass. In are better modulated by the TIO SST index with a notable decrease in
particular, a strong warming of the adjacent tropical Indian Ocean sea the proportion of significant (p = 0.1) trends (Fig. 10). Its comparison
surface temperature (TIO SST) and also that of the remote NINO3.4 with the results obtained from the original time series (Fig. 3b), about
region (Fig. 2b) have been reported to have influenced the Asian mon- 2% and 6% grids are found to be significantly increasing and decreasing,
soon in general. Even in absence of a strong El Niño (NINO3.4) condi- respectively for PRCPTOT after accounting the TIO SST index, corre-
tion, the major drought years of 2000 and 2002 have been attributed sponding to an equal proportion of 10% grids in the original time series.
to the rise in SST over the equatorial Indian Ocean, which induced For wet days and LMR, the proportion of significant negative trends has
intense monsoon breaks through modulating the strength of the dropped from N 20% of grids in the original time series (Fig. 3b) to an
monsoon Hadley circulation (Krishnan et al., 2006). However, correla- average of 10% and 14% with the association of the TIO SST and
tion analysis indicates that the basic and extreme matrices of the NINO3.4 indices, respectively. However, the TIO SST index accounts for
monsoon rainfall, which have undergone non-linear to multidecadal a large part of the wet spell (CWD) trends with the identification of
changes, exhibiting no direct correspondence on a year-to-year basis significant decreasing trends in 11% of grids only, which is about two-
with the TIO SST and NINO3.4 indices during 1901–2004 (Fig. 2a and and-half time less than the proportion observed in the original time

Fig. 10. Percentage of 306 grid points with significant (p b 0.1) positive (P) and negative (N) trends in the basic and extreme monsoon rainfall indices after accounting the TIO SST and
NINO3.4 indices (shaded differently) through the PMK test during 1961–2004.
234 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235

series (Fig. 3b), but for the NINO3.4 index it is 23% of grids. Other nota- Acknowledgements
ble changes include the drop in the significant upward trend propor-
tions from 10% to 3.6% in Rx1day, 6% to 2.6% in R100, 15.6% to 5.5% in Grant of the Indian Council of Agricultural Research (ICAR), New
R99p, and 16.3% to 7.9% in R99pTOT after removing the influence of Delhi to undertake this research is duly acknowledged. We thank the
the TIO SST index (Fig. 10). Interestingly, the observed influence of the anonymous reviewers for their comments and suggestion to improve
TIO SST index assessed through the PMK test has been found to be ro- the contents of this paper.
bust, as we also obtained similar proportion of significant trends using
the method followed by Vincent et al. (2015).
In general, the post-1950 weakening of the southwest summer References
monsoon flow through the increased frequency and duration of Alexander, L.V., Zhang, X., Peterson, T.C., Caesar, J., Gleason, B., Klein Tank, A.M.G., Haylock,
monsoon break phases has been attributed to the SST warming over M., Collins, D., Trewin, B., Rahimzdeh, F., Tagipour, A., Kumar, K.R., Revadekar, J.V.,
the tropical eastern IO (Ramesh Kumar et al., 2009). Note that the Griffiths, G., Vincent, L., Stephenson, D.B., Burn, J., Aguilar, E., Brunet, M., Taylor, M.,
New, M., Zhai, P., Rusticucci, M., Vazquez, A.J.L., 2006. Global observed changes in
CWD index reflects rainfall in a single spell of days compared to other
daily climate extremes of temperature and precipitation. J. Geophys. Res. 111,
indices, typically representing the active phase of the monsoon rainfall. D05109. http://dx.doi.org/10.1029/2005JD006290.
The increased sensitivity of CWD to the TIO SST index therefore Ali, H., Mishra, V., Pai, D.S., 2014. Observed and projected urban extreme rainfall events in
supports the reported frequent monsoon breaks due to warming of India. J. Geophys. Res. 119. http://dx.doi.org/10.1002/2014JD022264.
Bollasina, M.A., Ming, Y., Ramaswamy, V., 2011. Anthropogenic aerosols and the weaken-
the adjacent Ocean, which has led to more occurrences of drying ing of the South Asian summer monsoon. Science 334, 502–505.
tendencies. Moreover, the observed significant drop in the depression Brunetti, M., Maugeri, M., Monti, F., Nanni, T., 2004. Changes in daily precipitation
frequency (i.e., −0.60 decade−1) supports the widely reported weak- frequency and distribution in Italy over the last 120 years. J. Geophys. Res. 109,
D05102. http://dx.doi.org/10.1029/2003JD004296.
ening of the monsoon circulation. Although some of the extreme fea- Dai, A., 2011. Characteristics and trends in various forms of the palmer drought severity
tures of the monsoon rainfall are showing similar responses to what index (PDSI) during 1900–2008. J. Geophys. Res. 116, D12115. http://dx.doi.org/10.
has been reported as an expected outcome of the anthropogenic 1029/2010JD015541.
Dai, A., Trenberth, E., Qian, T., 2004. A global data set of palmer drought severity index for
warming (Giorgi et al., 2011; Westra et al., 2013; Zhang et al., 2013), 1870–2002: relationship with soil moisture and effects of surface warming.
the observed spatial inconsistency could be due to local factors, such J. Hydrometeorol. 5, 1117–1130.
as urbanization, land-use change and orography. A large number of Dash, S.K., Kulkarni, M.A., Mohanty, U.C., Prasad, K., 2009. Changes in the characteristics of
rain events in India. J. Geophys. Res. 114, D10109. http://dx.doi.org/10.1029/
driving factors appear to have contributed to the extremes of varying
2008JD010572.
nature and extent observed in this study, thus posing a challenge to De Luis, M., Gonzalez-Hidalgo, J.C., Longares, L.A., Stepanek, P., 2009. Seasonal precipita-
the validation and attribution studies. tion trends in the Mediterranean Iberian peninsula in second half of 20th century.
Int. J. Climatol. 29, 1312–1323.
Donat, M.G., Alexander, L.V., Yang, H., Durre, I., Vose, R., Dunn, R.J.H., Willett, K.M., Aguilar,
E., Brunet, M., Caesar, J., Hewitson, B., Jack, C., Klein Tank, A.M.G., Kruger, A.C.,
5. Conclusions Marengo, J., Peterson, T.C., Renom, M., Oria, R.C., Rusticucci, M., Salinger, J., Elrayah,
A.S., Sekele, S.S., Srivastava, A.K., Trewin, B., Villarroel, C., Vincent, L.A., Zhai, P.,
Zhang, X., Kitching, S., 2013. Updated analyses of temperature and precipitation ex-
This study presents the 20th century (1901–2004) changes in the treme indices since the beginning of the twentieth century: the HadEX2 dataset.
mean and extreme monsoon rainfall indices, developed by the ETCCDI, J. Geophys. Res. 118, 2098–2118.
of India using robust statistical tools. Nationally and regionally averaged Douglas, E.M., Vogel, R.M., Kroll, C.N., 2000. Trends in floods and low flows in the United
States: impact of spatial correlation. J. Hydrol. 240 (1–2), 90–105.
anomalies show multidecadal variability with marked differences, Duhan, D., Pandey, A., 2013. Statistical analysis of long term spatial and temporal trends of
suggesting the fact that the country-wide time series does not depict precipitation during 1901–2002 at Madhya Pradesh, India. Atmos. Res. 122, 136–149.
the spatial diversity of the monsoon rainfall. Comparison of the rainfall Fischer, E.M., Knutti, R., 2014. Detection of spatially aggregated changes in temperature
and precipitation extremes. Geophys. Res. Lett. 41. http://dx.doi.org/10.1002/
matrices between the pre-1940 and post-1960 periods exhibits a note- 2013GL058499.
worthy transition from a wetting condition to a drying tendency, Gadgil, S., Gadgil, S., 2006. The Indian monsoon, GDP and agriculture. Econ. Polit. Wkly.
particularly in PRCPTOT, WD, LMR and CWD of central-north India 41, 4887–4895.
Gallant, A.J.E., Karoly, D.J., 2010. A combined climate extremes index for the Australian re-
(CNI). Moreover, the probability density function (PDF) of rainfall and
gion. J. Clim. 23, 6153–6165.
PDSI indices indicate a significant change in the spatiotemporal variabil- Gallego, M.C., Trigo, R.M., Vaquero, J.M., Brunet, M., García, J.A., Sigró, J., Valente, M.A.,
ity during the post-1985 period. The first decade of the 21st century 2011. Trends infrequency indices of daily precipitation over the Iberian Peninsula
during the last century. J. Geophys. Res. 116, D02109. http://dx.doi.org/10.1029/
(i.e., 2001–2010) has witnessed the worst droughts since the beginning
2010JD014255.
of the 20th century in the years 2002 and 2009, along with the region- Gautam, R., Hsu, N.C., Lau, K.M., Kafatos, M., 2009. Aerosol and rainfall variability over the
ally devastating floods of 2005 and 2008, leading to drop in food grain Indian monsoon region: distributions, trends and coupling. Ann. Geophys. 27,
production of about 10%. These events appear to be highly unlikely 3691–3703.
Ghosh, S., Das, D., Kao, S.-C., Ganguly, A.R., 2012. Lack of uniform trends but increasing
without the role of the post-1960 trends, though the fundamental nat- spatial variability in observed Indian rainfall extremes. Nat. Clim. Chang. 2, 86–91.
ural variability of the monsoon rainfall cannot be understated. If such Giorgi, F., Im, E.-S., Coppola, E., Diffenbaugh, N.S., Gao, X.J., Mariotti, L., Shi, Y., 2011. Higher
opposite extremes continue in the future, generally anticipated in a hydroclimatic intensity with global warming. J. Clim. 24, 5309–5324.
Goswami, B.N., Venugopal, V., Sengupta, D., Madhusoodanan, M.S., Xavier, P.K., 2006. In-
warmer climate, the spatial variability would rise but not detectability creasing trend of extreme rain events over India in a warming environment. Science
of trends, as reflected in the PDSI index. 314, 1442–1445.
Compared to the recently reported global scale identification of Goyal, M.K., 2014. Statistical analysis of long term trends of rainfall during 1901–2002 at
Assam, India. Water Resour. Manag. 28 (6), 1501–1515.
significant trends in extreme indices, we find a less proportion of Griffiths, M.L., Bradley, R.S., 2007. Variations of twentieth-century temperature and pre-
significant trends, with some indices even displaying a contradictory cipitation extreme indicators in the northeast United States. J. Clim. 20, 5401–5417.
tendency. This, in general, underscores the importance of the season Groisman, P.Y., Knight, R.W., Easterling, D.R., Karl, T.R., Hegerl, G.C., Razuvaev, V.N., 2005.
Trends in intense precipitation in the climate record. J. Clim. 18 (9), 1326–1350.
of analysis and the choice of the extreme indices, particularly for the
Guhathakurta, P., Rajeevan, M., Sikka, D.R., Tyagi, A., 2015. Observed changes in southwest
highly heterogeneous monsoon climate of India. Although warming monsoon rainfall over India during 1901–2011. Int. J. Climatol. 35, 1881–1898.
of the TIO SST index explains a degree of association, a detailed attribu- Hasson, S., Pascale, S., Lucarini, V., Böhner, J., 2016. Seasonal cycle of precipitation over
major river basins in South and Southeast Asia: a review of the CMIP5 climate models
tion, which is beyond the scope of this study, will enhance the under-
data for present climate and future climate projections. Atmos. Res. 180, 42–63.
standing in the context of the reported multiple drivers of the Joshi, M.K., Pandey, A.C., 2011. Trend and spectral analysis of rainfall over India during
monsoon rainfall changes. Given the vulnerability of India to the 1901–2000. J. Geophys. Res. 116, D06104. http://dx.doi.org/10.1029/2010JD014966.
impending water and food crisis, the results of the current study can Kenyon, J., Hegerl, G., 2010. Influence of modes of climate variability on global precipita-
tion extremes. J. Clim. 23, 6248–6262.
be used as a reference to implement the location-specific adaptation Kiktev, D., Sexton, D., Alexander, L., Folland, C., 2003. Comparison of modeled and ob-
and mitigation measures. served trends in indices of daily climate extremes. J. Clim. 16, 3560–3571.
 D.K. Panda et al. / Atmospheric Research 181 (2016) 220–235 235

Kitoh, A., Endo, H., Krishna Kumar, K., Cavalcanti, I.F.A., Goswami, P., Zhou, T., 2013. Mon- Schmidli, J., Frei, C., 2005. Trends of heavy precipitation and wet and dry spells in
soons in a changing world: a regional perspective in a global context. J. Geophys. Res. Switzerland during the 20th century. Int. J. Climatol. 25, 753–771.
118. http://dx.doi.org/10.1002/jgrd.50258. Sen Roy, S., Balling, R.C., 2004. Trends in extreme daily precipitation indices in India. Int.
Klein Tank, A.M.G., Kӧnnen, G.P., 2003. Trends in indices of daily temperature and precip- J. Climatol. 24, 457–466.
itation extremes in Europe, 1946–99. J. Clim. 16, 3665–3680. Singh, V., Goyal, M.K., 2016. Analysis and trends of precipitation lapse rate and extreme
Klein Tank, A.M.G., Peterson, T.C., Quadir, D.A., Dorji, S., Zou, X., Tang, H., Santhosh, K., indices over North Sikkim eastern Himalayas under CMIP5ESM-2 M RCPs experi-
Joshi, U.R., Jaswal, A.K., Kolli, R.K., Sikder, A.B., Deshpande, N.R., Revadekar, J.V., ments. Atmos. Res. 167, 34–60.
Yeleuova, K., Vandasheva, S., Faleyeva, M., Gomboluudev, P., Budhathoki, K.P., Singh, D., Tsiang, M., Rajaratnam, B., Diffenbaugh, N.S., 2014. Observed changes in
Hussain, A., Afzaal, M., Chandrapala, L., Anvar, H., Amanmurad, D., Asanova, V.S., extreme wet and dry spells during the south Asian summer monsoon season. Nat.
Jones, P.D., New, M.G., Spektorman, T., 2006. Changes in daily temperature and pre- Clim. Chang. 4, 456–461. http://dx.doi.org/10.1038/NCLIMATE2208.
cipitation extremes in Central and South Asia. J. Geophys. Res. 111, D16105. http:// Smith, T.M., Reynolds, R.W., Peterson, T.C., Lawrimore, J., 2008. Improvements to NOAA's
dx.doi.org/10.1029/2005JD006316. historical merged land-ocean surface temperature analysis (1880–2006). J. Clim. 21,
Krishnamurthy, L., Krishnamurthy, V., 2013. Decadal Scale Oscillations and Trend in the 2283–2296.
Indian Monsoon Rainfall. COLA Technical Report 322. Center for Ocean-Land- Sneyers, R., 1999. On the Statistical Analysis of Series of Observations. Technical Note 143,
Atmosphere Studies, Fairfax, VA. WMO-No 415 Vol. 192.
Krishnan, R., Ramesh, K.V., Samala, B.K., Meyers, G., Slingo, J.M., Fennessy, M.J., 2006. Stocker, T.F., Qin, D., Plattner, G.-K., Tignor, M., Allen, S.K., Boschung, J., Nauels, A., Xia, Y.,
Indian ocean-monsoon coupled interactions and impending monsoon droughts. Bex, V., Midgley, P.M., 2013. Climate Change 2013: The Physical Science Basis, Inter-
Geophys. Res. Lett. 33, L08711. http://dx.doi.org/10.1029/2006GL025811. governmental Panel on Climate Change. Working Group I Contribution to the IPCC
Krishnan, R., Sabin, T., Ayantika, D., Kitoh, A., Sugi, M., Murakami, H., Turner, A., Slingo, J., Fifth Assessment Report (AR5). Cambridge Univ. Press, New York.
Rajendran, K., 2013. Will the south Asian monsoon overturning circulation stabilize Subash, N., Singh, S.S., Priya, N., 2011. Variability of rainfall and effective onset and length
any further? Clim. Dyn. 40, 187–211. of the monsoon season over a sub-humid climatic environment. Atmos. Res. 99,
Kulkarni, M.A., Singh, A., Mohanty, U.C., 2012. Effect of spatial correlation on regional 479–487.
trends in rain events over India. Theor. Appl. Climatol. 109, 497–505. Sun, W., Mu, X., Song, X., Wu, D., Cheng, A., Qiu, B., 2016. Changes in extreme temperature
Kunkel, K.E., Easterling, D.R., Redmond, K., Hubbard, K., 2003. Temporal variations of ex- and precipitation events in the loess plateau (China) during 1960–2013 under global
treme precipitation events in the United States: 1895–2000. Geophys. Res. Lett. 30, warming. Atmos. Res. 167, 34–60.
1900. http://dx.doi.org/10.1029/2003GL018052. Turner, A.G., Annamalai, H., 2012. Climate change and the south Asian summer monsoon.
Lau, W.K.M., Kim, K.-M., 2010. Finger printing the impacts of aerosols on long-term trends Nat. Clim. Chang. 2, 587–595.
of the Indian summer monsoon regional rainfall. Geophys. Res. Lett. 37, L16705. Villafuerte II, M.Q., Matsumoto, J., Kubota, H., 2015. Changes in extreme rainfall in the
http://dx.doi.org/10.1029/2010GL043255. Philippines (1911–2010) linked to global mean temperature and ENSO. Int.
Libiseller, C., Grimvall, A., 2002. Performance of partial Mann–Kendall tests for trend de- J. Climatol. 35, 2033–2044. http://dx.doi.org/10.1002/joc.4105.
tection in the presence of covariates. Environmetrics 13, 71–84. Vincent, L.A., Zhang, X., Brown, R.D., Feng, Y., Mekis, E., Milewska, E.J., Wan, H., Wang, X.L.,
Moberg, A., Jones, P.D., Lister, D., Walther, A., Brunet, M., Jacobeit, J., Alexander, L.V., Della- 2015. Observed trends in Canada's climate and influence of low-frequency variability
Marta, P.M., Luterbacher, J., Yiou, P., Chen, D., Klein Tank, A.M.G., Saladié, O., Sigró, J., modes. J. Clim. 28, 4545–4560.
Aguilar, E., Alexandersson, H., Almarza, C., Auer, I., Barriendos, M., Begert, M., Vittal, H., Karmakar, S., Ghosh, S., 2013. Diametric changes in trends and patterns of ex-
Bergstrom, H., Öbohm, R., Butler, C.J., Caesar, J., Drebs, A., Founda, D., Gerstengarbe, treme rainfall over India from pre-1950 to post-1950. Geophys. Res. Lett. 40,
F.W., Micela, G., Maugeri, M., Österle, H., Pandzic, K., Petrakis, M., Srnec, L., Tolasz, 3253–3258. http://dx.doi.org/10.1002/grl.50631.
R., Tuomenvirta, H., Werner, P.C., Linderholm, H., Philipp, A., Wanner, H., Xoplaki, E., Wang, B., Liu, Y., Kim, H.-J., Webster, P.J., Yim, S.-Y., 2012. Recent change of the global
2006. Indices for daily temperature and precipitation extremes in Europe analyzed monsoon precipitation (1979–2008). Clim. Dyn. 39, 1123–1135.
for the period 1901–2000. J. Geophys. Res. 111, D22106. http://dx.doi.org/10.1029/ Westra, S., Alexander, L.V., Zwiers, F.W., 2013. Global increasing trends in annual maxi-
2006JD007103. mum daily precipitation. J. Clim. 26, 3904–3918.
Panda, D.K., Kumar, A., 2014. The changing characteristics of monsoon rainfall in India Yao, C., Yang, S., Qian, W.H., Lin, Z.M., 2008. Regional summer precipitation events in Asia
during 1971–2005 and links with large scale circulation. Int. J. Climatol. 34, and their changes in the past decades. J. Geophys. Res. 113, D17107. http://dx.doi.org/
3881–3899. 10.1029/2007JD009603.
Panda, D.K., Wahr, J., 2016. Spatiotemporal evolution of water storage changes in India Yue, S., Pilon, P., Phinney, B., Cavadias, G., 2002. The influence of autocorrelation on the
from the updated GRACE-derived gravity records. Water Resour. Res. 52, 135–149. ability to detect trend in hydrological series. Hydrol. Process. 16 (9), 1807–1829.
http://dx.doi.org/10.1002/2015WR017797. Zhang, L., Zhou, T., 2011. An assessment of monsoon precipitation changes during 1901–
Panda, D.K., Kumar, A., Singandhupe, R.B., Sahoo, N., 2013. Hydroclimatic changes in a 2001. Clim. Dyn. 37, 279–296.
climate-sensitive tropical region. Int. J. Climatol. 33, 1633–1645. Zhang, X., Hervey, K.D., Hogg, W.D., Yuzyk, T.R., 2001. Trends in Canadian stream flow.
Patra, J.P., Mishra, A., Singh, R., Raghuwanshi, N.S., 2012. Detecting rainfall trends in twen- Water Resour. Res. 37, 987–998.
tieth century (1871–2006) over Orissa state, India. Clim. Chang. 111 (3), 801–817. Zhang, X., Alexander, L., Hegerl, G.C., Jones, P., Klein Tank, A., Peterson, T.C., Trewin, B.,
Pingale, S.M., Khare, D., Jat, M.K., Adamowski, J., 2014. Spatial and temporal trends of Zwiers, F.W., 2011. Indices for monitoring changes in extremes based on daily tem-
mean and extreme rainfall and temperature for the 33 urban centers of the arid perature and precipitation data. WIREs Clim. Change 2, 851–870. http://dx.doi.org/
and semi-arid state of Rajasthan, India. Atmos. Res. 138, 73–90. 10.1002/wcc.147.
Rajeevan, M., Bhate, J., Jaswal, A., 2008. Analysis of variability and trends of extreme rain- Zhang, X., Wan, H., Zwiers, F.W., Hegerl, G.C., Min, S.K., 2013. Attributing intensification of
fall events over India using 104 years of gridded daily rainfall data. Geophys. Res. Lett. precipitation extremes to human influence. Geophys. Res. Lett. 40, 5252–5257.
35, L18707. http://dx.doi.org/10.1029/2008GL035143. http://dx.doi.org/10.1002/grl.51010.
Ramesh Kumar, M.R., Krishnan, R., Sankar, S., Unnikrishnan, A.S., Pai, D.S., 2009. Increasing Zhou, T., Yu, R., Li, H., Wang, B., 2008. Ocean forcing to changes in global monsoon precip-
trend of ‘Break-Monsoon’ conditions over India — role of ocean atmosphere processes itation over recent half-century. J. Clim. 21, 3833–3852.
in the Indian Ocean. IEEE Geosci. Remote Sens. Lett. 6 (2), 332–336. Zolina, O., Simmer, C., Kapala, A., Bachner, S., Gulev, S., Maechel, H., 2008. Seasonally
Revadekar, J.V., Preethi, B., 2012. Statistical analysis of the relationship between summer dependent changes of precipitation extremes over Germany since 1950 from a very
monsoon precipitation extremes and food grain yield over India. Int. J. Climatol. 32, dense observational network. J. Geophys. Res. 113, D06110. http://dx.doi.org/10.
419–429. 1029/2007JD008393.
Saha, A., Ghosh, S., Sahana, A., Rao, E., 2014. Failure of CMIP5 climate models in simulating
post-1950 decreasing trend of Indian monsoon. Geophys. Res. Lett. 41, 7323–7330.