Theoretical and Applied Climatology (2021) 145:673–686

[Link]

ORIGINAL PAPER

Spatio-temporal rainfall variability over different meteorological

subdivisions in India: analysis using different machine learning
techniques
Gyanendranath Mohapatra 1,2 & V. Rakesh 1,2 & Smrati Purwar 1,2 & A. P. Dimri 3

Received: 26 May 2020 / Accepted: 3 May 2021 / Published online: 20 May 2021
# The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2021

Abstract
Understanding and quantifying long-term rainfall variability at regional scale is important for a country like India where
economic growth is very much dependent on agricultural production which in turn is closely linked to rainfall distribution.
Using machine learning techniques viz., cluster analysis (CA) and principal component analysis (PCA), the spatial and temporal
rainfall patterns over the meteorological subdivisions in India are examined. Monthly rainfall data of 117 years (1901–2017)
from India Meteorological Department over 36 meteorological subdivisions in India is used in this study. Using hierarchical
clustering method, six homogeneous rainfall clusters were identified in India. Among the rainfall clusters, Group 1 had 30%
dissimilarity with Groups 2, 3, and 4 while Group 5 and Group 6 are highly dissimilar (more than 90% dissimilarity) with the rest
of the groups. Rainfall seasons in each group were further classified into dry, wet, and transition periods. The duration of dry
period is smaller in group which consists of subdivisions from southern part of the country. The transition period between dry and
wet period was found to be smaller for subdivisions in the coastal region. Both CA and PCA showed high rainfall variability in
Groups 5 and 6, which comprise subdivisions from north east, Kerala, Konkan, and costal Karnataka and low rainfall variability
in Groups 1 and 2 which comprise subdivisions from east, north, and central part of the country. Strong negative trend in annual
and Indian summer monsoon rainfall is seen in northeast India and Kerala while positive trend is observed over costal Karnataka
and Konkan region. The negative trend in post monsoon rainfall particularly over the peninsular and northeast India indicates
weakening of northeast monsoon rainfall in the country.

1 Introduction et al. 2013; Kantharao and Rakesh 2018; Prabhakar et al.

2019; Deepesh et al. 2019). Time series correlations of rain
Several studies have been carried out in the past to understand gauge observations indicate that there is no coherency in rain-
large spatial heterogeneity in the rainfall pattern at different fall among different stations within India (Gadgil et al. 1988).
temporal scales over the Indian region (Ramamurthy 1969; Hence, average rainfall index over India may not be an ade-
Krishnamurthy and Shukla 2008; Gadgil 2003; Goswami quate indicator for various applications like agriculture, water,
et al. 2006; Rao et al. 2009; Malik et al. 2010; Kumar et al. and river basin management. Earlier studies showed that co-
2010; Rajeevan et al. 2010; Kulkarni et al. 2011; Saikranthi herency and homogeneity in rainfall is more evident at smaller
spatial scales of the order of a few hundred kilometers (Gadgil
et al. 1993; Iyengar and Basak 1994). While studying spatio-
temporal variability in rainfall, it is advisable to analyze spa-
* V. Rakesh tially averaged data over homogenous regions for a longer
rakesh@[Link]; rakeshv82@[Link] time period, rather than inspecting time series of individual
station data for longer term. Such a strategy not only reduces
1
CSIR Fourth Paradigm Institute, NAL Wind tunnel Road, the data volume to be analyzed but also enhances the signal
Bangalore 560037, India variation at larger spatial scales (Nicholson 1986).
2
Academy of Scientific and Innovative Research (AcSIR), For operational convenience, Indian land area is divided
Ghaziabad 201002, India into 36 meteorological subdivisions by the India
3
School of Environmental Sciences, Jawaharlal Nehru University, Meteorological Department (IMD). Primarily, such a classifi-
New Delhi 110067, India cation is meant for generating weather forecasts over political
674 G. Mohapatra et al.

boundaries like states (Saikranthi et al. 2013). However, rain- yielded universally acceptable results, many methods are still
fall variability over these subdivisions does not show homo- in use for regionalization of rainfall data.
geneity, to be more precise, inter-series correlations among Also, it is always better to analyze rainfall trend at re-
subdivisions are not significant (Gadgil et al. 1993; gional rainfall clusters rather than looking all India rainfall
Parthasarathy et al. 1993). Identifying homogenous rainfall as a single unit in the perspective of tackling local rainfall
clusters among these meteorological subdivisions based on extremes. In addition, it is worth noting that coherent rain-
rain-gauge station data is a challenging task due to tremendous fall zone for one application (e.g., watershed studies) may
spatio-temporal variability in rainfall owing to complex to- not necessarily be coherent for other applications (e.g.,
pography, climate variability, and coarser spatio-temporal agricultural planning or precipitation enhancement)
coverage of rain gauges (Bharath and Srinivas 2015). One of (Saikranthi et al. 2013). Previous studies in this direction
the ways to identify regions of homogeneous rainfall variabil- in India had identified rainfall coherent zones for applica-
ity is deploying advance mathematical techniques over the tions like watershed (national and river basin level) man-
entire study area like fuzzy technique used by Bharath and agement, flood and drought analysis, agricultural manage-
Srinivas 2015. Such categorization of homogeneous rainfall ment, and improving regional forecast (Parthasarathy et al.
zones is useful for numerous applications primarily in sectors 1993) based on analysis of rainfall spectral signature (Azad
like agriculture and water management (Gregory 1989; Gadgil et al. 2010), seasonality in rainfall (Malik et al. 2010),
et al. 1993; Parthasarathy et al. 1993; Singh et al. 2018). rainfall variability (Gadgil et al. 1993), and similarity in
In addition, understanding the linkages between global/ synoptic characteristics (Satyanarayana and Srinivas
regional circulation and homogeneous rainfall zones at re- 2008, 2011; Rao and Srinivas 2008) even though the ob-
gional scale is an important aspect in the context of improving jective of identifying rainfall coherent zones were to un-
forecasts at finer spatial resolution which is more useful than derstand the processes/mechanisms that cause rainfall var-
all-India forecast as one unit in the application point of view iability (Saikranthi et al. 2013).
(Parthasarathy et al. 1993; Rakesh et al. 2011; Kashid and In earlier studies, those addressed the spatio-temporal rain-
Maity 2012). The first attempt of identifying homogenous fall variability over different subdivisions in India, the homo-
rainfall zones was by Blanford (1886), in which based on geneous rainfall groups were selected based on climatology or
station data Indian land area is divided into 21 subdivisions. regional characteristics. In this study, identification of homo-
Subsequently, there were many investigators who put their geneous rainfall groups was carried out in an objective way
effort on classifying rainfall zones in India for various appli- using machine learning techniques and also the analysis was
cations (Gregory 1989; Iyengar and Basak 1994; Singh and carried at different temporal scales (pre-monsoon, monsoon,
Singh 1996; Venkatesh and Jose 2007; Satyanarayana and and post-monsoon). Previous studies (Gadgil et al. 1993;
Srinivas 2008, 2011; Malik et al. 2010; Azad et al. 2010; Parthasarathy et al. 1993; Iyengar and Basak 1994; Panda
Ratna 2012; Deepesh et al. 2019). and Sahu 2019; Deepesh et al. 2019) utilized time series of
Clustering of spatial meteorological subdivision is useful in individual rain gauge data. In most cases, the number of sta-
understanding the duration, intensity, and frequency of the tions that they used for the analysis is not large (few hundred)
annual rainfall at regional scale. Individual cluster has its or limited over specific area of interest and in some other
own characteristics in terms rainfall pattern, trend, and decadal cases, these stations are heterogeneously distributed across
spatial and temporal variation compared to stations falls in that the country. A major problem with such approaches is identi-
particular cluster. Previously, a number of techniques were fying the boundaries between homogeneous regions from
deployed for identifying homogeneous rainfall zones. such sparse rain gauge network (Saikranthi et al. 2013). To
Though they differed among themselves in terms of applica- reduce this problem, the present study utilizes a large dataset
tion and study region, prominent techniques used were corre- for a long time period with its own characteristics from 36
lation analysis (Nicholson 1986; Gadgil et al. 1993;Saikranthi meteorological subdivision spread across the India. Ward’s
et al. 2013), principal component analysis (PCA; Iyengar and clustering analysis (CA), which uses partitional clustering pro-
Basak 1994; Singh and Singh 1996), spectral analysis (Azad cedure to identify homogenous rainfall region by refining the
et al. 2010; Joshi and Pandey 2011), clustering method clusters derived from hierarchical clustering algorithms, and
(Matulla et al. 2003; Venkatesh and Jose 2007; Rao and PCA techniques are used to examine their effectiveness in
Srinivas 2008; Lyra et al. 2014; Deepesh et al. 2019), inverse regionalization of rainfall variability.
techniques (Gupta et al. 2017), and clustering followed by The paper is organized as follows. Section 2 describes the
PCA analysis (Satyanarayana and Srinivas 2008, 2011). data used and methods for delineating homogeneous regions.
Each method has its own advantage in understanding rainfall The homogeneous regions are evaluated at different temporal
variation over the homogeneous rainfall zone. However, as no scales and the results are discussed Section 3. The major con-
procedure has emerged as a single unique method which clusions are given in Section 4.
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using... 675

2 Data and methodology temporal variability of precipitation (Guhathakurta et al.

2015; Sahany et al. 2018; Nandargi and Mahto 2019; Gadgil
2.1 Data used et al. 2019; Guntu et al. 2020, Khouider et al. 2020), analyzing
trend and intrinsic predictability of Indian monsoon rainfall
Primarily, based on variability in meteorological parameters, (Guntu et al. 2020; Praveen et al. 2020; Kurths et al. 2019).
IMD divided Indian subcontinent into 36 meteorological sub- We also used 0.25°×0.25° all India-gridded daily rainfall data
divisions (Fig. 1) to generate rainfall prediction at regional for the period 1901–2013 from IMD ([Link])
scale. We analyzed 117 years (1901–2017) long time series for analyzing the climatology of spatial distribution of rainfall
data of monthly average rainfall over 36 meteorological sub- for different seasons. This gridded data is prepared from daily
divisions in India ([Link] rainfall records of 6995 rain gauge stations (Pai et al. 2014).
divisional-monthly-rainfall-1901-2017) taken from India The station data were interpolated to fixed spatial grid points
government open data portal, generated by IMD (http:// of 0.25° × 0.25° resolution by the simplest form of inverse
[Link]). Rain gauge stations have irregular spatial distance-weighted interpolation (IDW) scheme proposed by
density throughout the country with relatively high density in Shepard (1968). This method assumes that the interpolating
the south Peninsula as compared to eastern parts of central area should be affected most by the nearest neighbor stations
India and northern parts particularly data sparse regions like and less by the more distant stations. Analysis using the
Jammu, and Kashmir. The subdivision wise monthly rainfall Modified Similarity Index (MSI) and novel Conformity
data for 1901–2017 were constructed applying the area Index (CI) showed that this data set depicts better representa-
weighted method on district wise rainfall data for each month tion of rainfall in India compared to other existing data sets
taken from the archival of National Data Center (NDC) of (Meher and Das 2019; Das et al. 2020).
India Meteorological Department (Guhathakurta et al. 2011;
Kothawale and Rajeevan 2017). From more than 6000 rain
gauge stations distributed non-uniformly over India, 2599 sta- 2.2 Methodology
tions having 30 years or more consistent rainfall records were
used (Fig. S1) for generating subdivisional rainfall data 2.2.1 Cluster analysis
(Guhathakurta et al. 2011). There were only three rain gauge
stations in Jammu representing the Jammu and Kashmir sub- Cluster analysis is a technique which is used to club similar
division having more than 30 years of rainfall records of con- objects in one group and separates out dissimilar objects in
sistent rainfall observation (Fig. S1). The monthly area different groups. CA is used for identifying the homogeneous
weighted subdivision wise rainfall data was generated by rainfall regions based on their monthly rainfall pattern.
assigning district area as the weight for different rain gauge The average values of monthly rainfall for each subdivision
stations (Adarsh et al. 2019). The station rainfall data were are arranged in a matrix Xm×n, where the element xi,k
pre-processed and quality controlled before generating representing the values of average rainfall of ith subdivision
subdivisional rainfall data (Rajeevan et al. 2006; Kothawale in the kth month. Thus, each row vector of this matrix X
and Rajeevan 2017). In the past, several studies have demon- represents a subdivision and each column vector represents a
strated the usefulness of this dataset in studying spatio- month. Ward’s hierarchical clustering (Ward Jr 1963), which

Fig. 1 Map showing all the 36

meteorological subdivision of
India (source: [Link].
in)
676 G. Mohapatra et al.

considers the Euclidean distance (straight line distance) as a During winter (December, January, February), the western
measure of dissimilarity. disturbances (WDs) moving across the Himalaya and north
So, the dissimilarity between two clusters C1 and C2 is India bring significant amount of rainfall in the hilly region

(Rakesh et al. 2009). At the same time, other parts of north and
DC1 ∪C2 ¼ ∑ D x; μC1 ∪C2 central India receive small amount of rainfall due to induced
x∈C1 ∪C2
low pressure systems associated with the WDs, which is clear-
Where μ = centroid of all the objects in C1 and C2 cluster ly visible in Fig. 2. During the Indian summer monsoon (ISM-
D = the Euclidean distance between two objects defined as June, July, August, September) the whole country receives
 1= 2 about 75% (1070.4 mm) of the total annual rainfall (1426.2
n
2 mm) with large spatial and temporal variability. The west
Dðx1 ; x2 Þ¼ ∑ x1;k −x2;k
k¼1 coast receives significant amount of rainfall primarily due to
orographic lifting since the prevailing westerly winds are per-
pendicular to the Western Ghats (Shastri et al. 2015).
2.2.2 Principal component analysis Similarly, during northeast monsoon (NEM-October,
November, and December) the same is experienced in the
The principal component analysis (PCA) is applied Eastern Ghats with increased precipitation.
(Preisendorfer 1988; Hannachi and O’Neill 2001) on monthly
average rainfall to understand the spatial variability over the 3.2 Cluster analysis of rainfall
36 meteorological subdivisions in India.
In PCA analysis, X is defined as the data matrix whose The CA is applied on monthly averaged (1901–2017) rainfall
elements xi,k represents the average rainfall at the ith subdivi- for 36 meteorological subdivisions to identify homogenous
sion (i=1, 2..., m) and kth month (k=1, 2..., n). Let the element rainfall regions in India. The 36 subdivisions were clustered
ri, k represents the departure of the rainfall from the annual into 06 (six) (hereafter Group 1 to Group 6) homogeneous
average rainfall of the ith subdivision and kth month. clusters based on monthly rainfall (Fig. 3a). Among the rain-
The principal component (PC) is defined as fall clusters, the dissimilarity was least between Group 3 and
Group 4. Group 1 showed ~30% dissimilarity with Group 2,
p ¼ Xϕ
Group 3, and Group 4 while Group 5 and Group 6 were totally
Where ϕ is eigenvector of the covariance matrix C, whose dissimilar (> 90% dissimilarity) with the rest of the Groups.
elements are defined as The spatial distribution of different rainfall clusters identified
over India is shown in Fig. 3b. Group 1, composed of eastern
1 n India (Gangetic West Bengal, Orissa, Jharkhand, Bihar) with
ci; j ¼ ∑ ri;k r j;k ;
n k¼1 Chhattisgarh, eastern Madhya Pradesh, and Uttarakhand, re-
where ði ¼ 1; 2; …; mÞ and ð j ¼ 1; 2; …; nÞ ceives ~100–150 mm annual average rainfall where 70–80%
contribution is from Indian Summer monsoon rainfall
This transforms the original input variables xi,k into the (ISMR). Earlier study showed that there is an increase in
new variables pi,k, which reflects the spatial variation of the heavy rainfall events over the central northeast region, where-
subdivisions (Hannachi et al. 2007). as moderate rainfall events are found to be decreasing (Dash
et al. 2009). Group 2, consists of subdivisions from northern
(Jammu and Kashmir, Himachal Pradesh, Uttar Pradesh) and
3 Results and discussion central (Eastern Rajasthan, western Madhya Pradesh, Madhya
Maharashtra, Marathwada, Vidarbha) part with Gujarat and
In the following section, for general understanding a brief Telangana, receives annual average rainfall of ~ 50–100 mm
discussion on rainfall climatology over India is provided and having 80–90% contribution from the ISMR. Group 3, com-
then discussion of the objective of the rainfall homogenization posed of subdivisions in the north and western part (Haryana,
based on CA and PCA is deliberated. Delhi, Chandigarh, Punjab, West Rajasthan, and Saurashtra
and Kutch), receives lowest annual average rainfall (less than
3.1 Rainfall climatology over India 50 mm) where 80–90% contribution is from the ISMR similar
to Group 2. In this region, heavy rainfall events are reported to
Spatial distribution of monthly averaged (1901–2013) rainfall be increasing while moderate rainfall events are found to be
(Fig. 2) showed that during pre-monsoon (March, April, decreasing (Dash et al. 2009). Group 4, composed of subdivi-
May), maximum rainfall of the order of 150 mm/month was sions from southern part (coastal Andhra Pradesh,
observed over north east India. The west and south India re- Rayalseema, Tamil Nadu, north and south interior
ceived comparatively less rainfall during pre-monsoon. Karnataka) with Lakshadweep, receives ~50–100 mm annual
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using... 677

Fig. 2 Spatial distribution of monthly climatology of rainfall (mm) over different meteorological subdivisions in India averaged over the period 1901–
2013

average rainfall where 50–70% contribution is from the blocks rain bearing winds of the ISM and hence the ISMR is
ISMR. This peninsular region is sheltered by Western Ghats weaker over the leeward region. Group 5 consists of only two
(orographic barrier along the west coast of India) which subdivisions (Coastal Karnataka and Konkan & Goa) with
678 G. Mohapatra et al.

Fig. 3 (a) Tree diagram from

cluster analyses based on
normalized distance between
groups and spatial distribution.
The vertical axis represents the
distance or dissimilarity between
different cluster groups. The
spatial groups are based on cluster
analysis of average monthly
rainfall for 36 meteorological
subdivisions over India for the
period 1901–2017. (b) Spatial
distribution of the homogeneous
rainfall region; (c) Average
annual rainfall and group wise
contribution of seasonal rainfall
(%) to the annual rainfall (mm)

high annual average rainfall (> 200 mm) where 85–95% con- months of July and August constituted the wet period and
tribution is from the ISMR. In most of these areas, the agri- having mean rainfall of 352.42± 51.47 mm and it contributed
culture and water management are regulated by ISMR. Group 52% of the annual rainfall in that Group (Table 1). The tran-
6 was composed of northeastern region (Arunachal Pradesh, sition period consisted of 2 months (June and September), 1
Assam & Meghalaya, Nagaland, Manipur, Mizoram, Tripura, month each before and after the wet period with a mean
West Bengal and Sikkim) with Kerala and Andaman Nicobar monthly rainfall amount of 204.46± 60.57 mm and it contrib-
Islands and has the highest annual average rainfall (~200–300 uted 30.17% of the annual rainfall amount (Table 1).
mm) where contribution of ISMR is 60–70%. For Group 2 and Group 3 also, the dry, transition, and wet
Cluster analysis is extended to understand the monthly periods were similar to Group 1. The dry period accounted for
rainfall pattern over different rainfall Groups (Fig. 4). mean monthly rainfall of 23.51±14.87 mm and contributed
Identifying rainfall seasons based on monthly rainfall distri- 19.84% to the annual rainfall in Group 2. Similarly, the tran-
bution at subdivisional scale using CA method over homoge- sition period accounted for mean monthly rainfall of 131.01±
nous groups is useful for agriculture and irrigation manage- 46.64 mm and contributed 27.64% to the annual rainfall and
ment at regional scale. Dendrogram (Fig. 4) shows monthly the wet period accounted for mean monthly rainfall of 249.01
rainfall variability and different rainfall seasons within each ± 52.57 mm and contributed 52.53% to the annual rainfall. For
Cluster. Using CA, three rainfall periods are identified viz., Group 3, the dry period accounted for 15.2% of annual rainfall
dry, wet, and transition for each of the 06 Groups. For Group with monthly average rainfall of 9.07± 11.20 mm. The transi-
1, the periods were classified into dry (October to May), tran- tion period accounted for 25.54% of annual rainfall with
sition (June and September), and wet (July and August). The monthly mean rainfall of 61± 42.78 mm. The wet period
dry period contributed 17.82% to the annual rainfall with accounted for 59.26% of annual rainfall with mean monthly
mean monthly rainfall of 30.19±20.51 mm (Table 1) and ma- rainfall of 141± 57.12 mm (Table 1). Similarly, subdivisions
jority (8 months) of the months fall under this category. The under Group 4, Group 5, and Group 6 were also classified into
Table 1 Cluster wise decadal mean, standard deviation (SD) and coefficient of variation (CV) in monthly, seasonal, and annual rainfall during the period 1901–2017

JAN FEB MARCH APRIL MAY JUN JUL AUG SEP OCT NOV DEC Annual rainfall JF MAM JJAS OND

Group 1
Mean (1901–2017) 20.27 25.72 23.51 24.68 51.63 189.16 357.82 347.02 219.75 73.53 14.33 7.85 1355.28 45.99 99.81 1113.76 95.72
S.D. 15.39 19.11 18.05 15.51 23.68 67.20 56.18 46.75 53.94 46.92 16.24 9.18 143.76 24.34 35.23 111.63 49.54
C.V. 75.94 74.28 76.78 62.83 45.87 35.53 15.70 13.47 24.55 63.80 113.35 116.93 10.61 52.93 35.29 10.02 51.76
Group 2
Mean (1901–2017) 23.97 25.29 27.15 19.95 23.09 112.86 262.17 235.86 149.17 41.39 14.14 13.11 948.09 49.28 70.20 760.15 68.68
S.D. 12.92 13.78 15.13 9.60 11.51 39.78 54.37 50.76 53.50 29.73 15.07 11.23 133.81 18.31 22.81 117.74 33.65
C.V. 53.90 54.48 55.72 48.12 49.86 35.25 20.74 21.52 35.86 71.84 106.64 85.63 14.11 37.16 32.49 15.49 49.00
Group 3
Mean (1901–2017) 11.64 12.50 10.51 6.20 10.74 49.85 152.36 130.66 72.16 11.58 3.74 5.68 477.60 24.14 27.42 405.03 21.01
S.D. 10.01 12.29 11.46 8.44 11.16 30.82 57.80 56.43 54.75 21.38 7.39 7.49 121.19 14.62 19.05 113.75 23.04
C.V. 86.02 98.31 108.99 136.02 103.92 61.82 37.94 43.19 75.87 184.61 197.46 131.72 25.37 60.56 69.49 28.09 109.69
Group 4
Mean (1901–2017) 12.35 9.03 12.05 33.37 80.58 134.72 164.03 146.25 144.86 149.89 92.94 34.16 1004.42 21.27 125.66 589.01 277.31
S.D. 11.86 9.91 11.55 16.75 31.51 28.07 38.18 37.63 35.85 45.13 46.20 22.34 114.71 14.87 34.63 78.91 64.17
C.V. 96.06 109.80 95.85 50.19 39.10 20.84 23.27 25.73 24.75 30.11 49.71 65.40 11.42 69.88 27.56 13.40 23.14
Group 5
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using...

Mean (1901–2017) 1.57 1.02 3.82 17.35 77.98 766.58 1098.34 697.55 326.39 148.67 43.54 8.58 3190.96 2.54 99.15 2888.86 200.80
S.D. 3.35 3.02 9.91 14.72 87.75 175.49 257.97 220.59 145.90 74.20 41.03 17.74 428.90 4.41 87.39 401.23 86.57
C.V. 212.88 297.57 259.39 84.83 112.53 22.89 23.49 31.62 44.70 49.91 94.23 206.73 13.44 173.55 88.14 13.89 43.11
Group 6
Mean (1901–2017) 25.32 35.53 68.12 153.48 304.29 538.48 556.19 439.06 357.25 206.97 86.25 40.30 2803.68 60.70 525.36 1890.50 332.82
S.D. 17.79 14.90 25.05 37.32 61.06 85.51 86.08 74.52 61.11 44.50 30.55 26.00 268.22 24.86 76.29 199.72 60.83
C.V. 70.29 41.94 36.77 24.31 20.07 15.88 15.48 16.97 17.11 21.50 35.42 64.52 9.57 40.95 14.52 10.56 18.28
679
680 G. Mohapatra et al.

Fig. 4 Temporal variation in

rainfall within spatial groups
based on cluster analysis of
average monthly rainfall for the
period 1901–2017

three categories (dry, transition, and wet), although the vari- 218 mm mean monthly rainfall. The dry period for Group 6
ance and the months falling in each rainfall seasons were also was from October to April, while the transition months
different for each Group. The dry period duration was the were May, August, and September and wet months were June
smallest (December to April) for Group 4 which contributed and July. The dry period has mean monthly rainfall of 87.99
9.95% to annual rainfall with 20.19±14.48 mm mean monthly ±28.02 mm and contributed 21.91% to the annual rainfall. The
rainfall, whereas the wet period consists of 5 months (June to transition period contributed for 27.44% to the annual rainfall
October) and contributed 72.94% to the annual rainfall with with mean monthly rainfall of 366.46 ± 65.56 mm while the
147.95 ±36.97 mm mean monthly rainfall. The transition pe- wet period accounted for 40.93% of annual rainfall with mean
riod includes May and November having a mean monthly monthly rainfall of 547.34 ± 85.79 mm.
rainfall of 86.76 ± 38.86 mm and it contributed 17.11% of
the annual rainfall (Table 1). In Group 5, the duration of the 3.3 Box–whisker plot analysis
dry period was similar to Group 1, Group 2, and Group 3
(October to May), the transition period accounted only for The box–whisker plot analysis of different spatial clusters for
one month (September) and the wet period ranged from June monthly rainfall is depicted in Fig. 5. It also indicated rainfall
to August. The annual cycle of monthly rainfall for Group 5, distribution having significant variability among different
where all stations were close to the coast (< 45 km), had only 1 clusters and contributions from ISM in annual rainfall for all
month in the dry-to-wet transition (March) and 1 month for the clusters. The mean rainfall among the months was lowest
the wet-to-dry transition (August). The dry periods in this for Group 3 and highest for Group 5. High amount of rainfall
Group contributed 9.48% to the annual rainfall with average during post-monsoon period is received in Groups 4, 5, and 6.
monthly rainfall of 37.82 ± 31.46 mm. The transition period There was significant variance of rainfall in Group 6 where the
contributed 10.23% to the annual total rainfall with average gap between the second and third quartile is wider with me-
monthly rainfall of 326.39 ±145.9 mm while the wet period dian value >500 mm in the month of July. Similarly, the
contributed 80.3% to the annual rainfall with 854.16 ± smallest spatial rainfall cluster, Group 5 exhibited tremendous
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using... 681

Fig. 5 Box–whisker plots of

average monthly rainfall for
different spatial groups for the
period 1901–2017

variability in monthly rainfall distribution with rainfall peaked the total variance and the distribution of the points
during the month of July (median value >1100). The driest representing the 36 subdivisions in the new coordinate space
rainfall cluster with relatively less variability was Group 3 of PC1 and PC2 is shown in Fig. 6.
(peak rainfall during July <200 mm) which covered mostly
the northwest India. Group 1 and Group 2 showed similar
monthly distributions with consistent ranges of quartiles dur-
ing the ISM though the magnitude of rainfall was higher in
Group 1.

3.4 Principal component analysis (PCA) for justifica-

tion of homogeneous clusters

PCA is applied to understand the spatial variability of rainfall

in different homogeneous clusters identified by clustering
through defining the variance by taking linear combinations
of average monthly rainfall. The relative importance of one Fig. 6 First factorial plane of the principal components PC1 and PC2 for
specific PCA is explained by using the variance fraction per- different spatial groups showing rainfall variability for the period 1901–
centage. It is found that the first two PCs could explain 93% of 2017
682 G. Mohapatra et al.

The PC1 explained 82% of the total variance in rainfall with the largest trend seen in Group 6 (−0.668 mm/year) while
pattern of all the 06 (six) Groups. It shows the high positive Group 2, Group 3, and Group 5 showed slight positive trend.
values for Group 5 and Group 6, having higher annual rainfall It is interesting to note that during pre-monsoon any signifi-
variability and the lowest negative values for Group 3 having cant trend in the last 117 years rainfall pattern is not observed
lowest annual rainfall variability in spatial coverage. The rain- except for Group 6 (Table 2). Here, the convective activity at
fall variability was relatively less in Group 1 and Group 2 (Fig. regional scale which drives the pre-monsoon rainfall is pri-
6). The PC2 explained 11% of total rainfall variability. PC2 marily induced by low pressure systems by and large through-
had the lowest value for Group 5 where the contribution of out the coastal area. Rainfall during ISM (Fig. 9) also showed
ISMR to annual rainfall was very high (85–95%) while posi- a negative trend in Group 6 (−4.1mm/year), which consists of
tive values with high variability were observed for Group 4 the northeast India, while positive trend was observed in
and Group 6 where contribution of ISMR to annual rainfall Group 5 (3.08 mm/year) which include Coastal Karnataka,
was relatively less (50–70%). The PC2 also showed lower and Konkan, and Goa. Slight positive trend in ISMR was observed
similar rainfall variability for Group 1, Group 2, and Group 3 in Group 3 and Group 4 while negative trend was seen in
where the ISMR contributes 70–80% toward the annual rain- Group 1. There was hardly any trend seen in ISMR over
fall which is important for agriculture and rainfed crops par- regions coming in Group 2. In summary, there was clear sig-
ticularly over the central, northeast, northwest, and hilly nature of negative trend in ISMR over regions falling under
Indian regions. In summary, the PC1 and PC2 analysis also Group 6 that constitutes mainly the northeast part and Kerala;
revealed high similarity in monthly rainfall variability among and positive trend over the regions falling under Group 5
Group 1, Group 2, and Group 3 whereas Group 4, Group 5, which consists of costal Karnataka and Konkan regions
and Group 6 had lesser similarity. (Table 2).
During the post-monsoon (Fig. S2), there is negative trend
3.5 Trend analysis in each spatial cluster in rainfall for most of the rainfall clusters except Group 5
(Table 2). The negative trend in rainfall for Group 2 and
India as a whole is too large to be treated as a single unit and Group 3 were marginal. Since NEM over the country falls in
rainfall in many areas are negatively correlated each other this season, the regions in Group 4 and Group 6 like peninsu-
(Normand 1953). If you consider the country as a whole, lar and northeast India are the regions where significant rain-
ISMR does not show any significant trend but there is signif- fall activity is observed during this season. It is important to
icant variability at regional scale (Guhathakurta and Rajeevan note that in both of these Groups, rainfall is found to be de-
2008). For the rest of the season, the magnitude of trend (mm/ creasing significantly. Rainfall has shown negative trend in
year) is quite small, as magnitude of rainfall is much smaller almost all the subdivisions except over Himachal Pradesh,
compared to the ISM. We examined annual as well as pre- Jharkhand and Nagaland, Manipur, Mizoram, and Tripura
monsoon, ISM, and post-monsoon rainfall trends for each of which is consistent with earlier studies (Guhathakurta and
the six Groups identified using CA. Rajeevan 2008). The number of moderate rainfall events is
The trend in annual rainfall for each spatial Group is shown found to be decreasing in the central northeast zone and in-
in Fig. 7. It is clear that rainfall trend showed large variations creasing in the peninsular region (Dash et al. 2009), which
at regional scale. It is noted that annual rainfall has negative indicates there is role of easterlies which brings moisture to
trend in Group 1, Group 2, and Group 6, whereas positive the peninsular region in modulating rainfall variability during
trends are observed for Group 3, Group 4, and Group 5. the post-monsoon rather than localized convection. It is noted
While the negative trend in Group 2 was marginal (−0.061 that there is consistent negative trend in rainfall over Group 1
mm/year), negative trend was moderate in Group 1 (−1.05 irrespective of season which includes most of the central India
mm/year) and large (−5.41mm/year) in Group 6. Group 3 region where agriculture is still largely rainfed. This decline in
and Group 4 showed slight positive trend (0.269 mm/year the rainfall can be attributed to the weakening trend in mon-
and 0.367 mm/year respectively) while positive trend was soon circulation (Sandeep et al. 2018) owning to a combina-
much larger in Group 5 (3.32mm/year). However, it should tion of factors including the warming over the Indian Ocean,
be noted here that the magnitude of annual rainfall in Group 5 increase in frequency and magnitude of El Nino events, and
and Group 6 was much higher compared to other Groups. land use changes over the subcontinent.
There are several reasons attributed to this trend like warming
at regional scale and regional changes in the land use pattern 3.6 Decadal variability in annual and seasonal rainfall
during past century which led to increased frequency of heavy
rainfall events and decrease of moderate rainfall events at The decadal trend (% departure from mean) in annual, pre-
regional scale (Goswami et al. 2006). monsoon, ISM, and post-monsoon rainfall variability for each
Figure 8 shows trends in pre-monsoon rainfall for each spatial cluster is shown in Fig. S3. The decadal variability in
cluster. Group 1, Group 4, and Group 6 showed negative trend annual rainfall did not show any specific trend except for
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using... 683

Fig. 7 Temporal variation in annual rainfall for different spatial groups with trend lines show positive/negative trend in the rainfall during the period
1901–2017

Group 6 where positive trend was observed till mid of 20th owing to its prime importance in water supply, agriculture
century and thereafter rainfall trend was found to be mostly management, renewable energy production, and economic
positive. In general, decadal variability in annual and seasonal growth. Considering the complexity and heterogeneity in
rainfall was relatively high over regions in Group 3 and Group spatio-temporal rainfall distribution in India, rainfall variabil-
5. The annual and ISMR showed relatively less decadal vari- ity needs to be investigated at finer spatial scale. Advanced
ability over all the spatial Groups. Group 2, Group 3, and data mining techniques are useful tools to understand the var-
Group 5 showed significant decadal variability in pre- and iability and to further explore possible causes for this variabil-
post-monsoon seasonal rainfall, whereas there was not much ity. In this study, using machine learning techniques like clus-
departure in other Groups. The pre-monsoon rainfall showed ter analysis (CA) and principal component analysis (PCA), the
negative departure in last two decades (2001–2017) over most spatio-temporal rainfall patterns over the meteorological sub-
of the Groups. There was negative decadal departure from divisions in India is examined using long term observed data.
normal in ISMR during recent decades (2001–2017) except Using clustering technique homogenous rainfall regions in
Group 5. Among different rainfall seasons, post-monsoon India are identified and clustered into six spatial groups.
rainfall showed the highest decadal departure over different Further, for each homogeneous spatial group, using the clus-
spatial clusters with vary high variability in Group 3 and tering method, rainfall seasons were identified like dry, wet,
Group 5. and transition; such classification is relevant and useful for
agriculture and irrigation management at regional scale.
Three out of six groups identified has similarity in rainfall
4 Conclusions seasons with longer dry period (October to May) and 2
months each of wet and transition periods. Overall, the prom-
Quantifying the spatio-temporal variability in rainfall is one of inent rainfall season for majority of homogenous regions was
the most important aspects for a developing country like India from April to July which contributed at least 60% of annual

Table 2 Rainfall trend over

different rainfall clusters Group-1 Group-2 Group-3 Group-4 Group-5 Group-6
identified in India for different
seasons. Arrow↑ (↓) represents Annual −1.054↓ −0.062↓ ↑0.270 ↑0.367 ↑3.319 −5.410↓
the positive (negative) trends in Pre-monsoon −0.029↓ ↑0.037 ↑0.042 −0.084↓ ↑0.099 −0.668↓
rainfall South-west monsoon −0.737↓ −0.071↓ ↑0.310 ↑0.590 ↑3.081 −4.10↓
Post-monsoon −0.078↓ 0.001↑ −0.014↓ −0.207↓ ↑0.174 −0.505↓
684 G. Mohapatra et al.

Fig. 8 Temporal variation in pre-monsoon rainfall for different spatial groups with trend lines show positive/negative trend in the rainfall during the
period 1901–2017

rainfall. The dry period duration was the smallest over the Among the rainfall clusters, high variability in rainfall distri-
subdivisions in southern part of the country and transition bution was in the coastal regions of southern India and north
period in rainfall was smallest over the coastal regions. east India whereas low variability seen in central and northern
The mean rainfall during south-west monsoon season is part. Both CA and PCA showed that rainfall trend has large
lowest for subdivisions in north-west India and highest over variations at regional scale. The annual rainfall had slight neg-
Konkan and coastal Karnataka regions while post-monsoon ative trend in central and northern India whereas strong posi-
season rainfall is prominent in the southern part of the country. tive trends were observed over coastal India. Strong negative

Fig. 9 Temporal variation in south west monsoon rainfall for different spatial groups
Spatio-temporal rainfall variability over different meteorological subdivisions in India: analysis using... 685

trend in annual and south-west monsoon season rainfall was References

observed in Group 6 which includes Kerala and north east part
of the country. The pre-monsoon season rainfall did not show Adarsh S, Kumar DN, Deepthi B, Gayathri G, Aswathy SS, Bhagyasree S
(2019) Multifractal characterization of meteorological drought in
any significant trend in the last 117 years except the negative
India using detrended fluctuation analysis. Int J Climatol 39(11):
trend seen in Group 6. The post-monsoon rainfall showed 4234–4425
negative trend in most of the regions except coastal Azad S, Vignesh TS, Narasimha R (2010) Periodicities in Indian mon-
Karnataka and Konkan regions. Significant negative trend in soon rainfall over spectrally homogeneous regions. Int J Climatol
rainfall over the regions like peninsular and northeast India 30:2289–2298
Bharath R, Srinivas VV (2015) Delineation of homogeneous hydromete-
indicates weakening of NEM rainfall activity in India. It is orological regions using wavelet-based global fuzzy cluster analysis.
noted that there is consistent decrease in rainfall in Group 1 Int J Climatol 35(15):4707–4727
irrespective of seasons which include most of the east central Blanford HF (1886) Rainfall of India. Mem India Meteor Dept 2:217–
India regions where agriculture is still largely rainfed. The 448
decadal variability in rainfall did not show any specific trend Das S, Narula P, Sarkar K (2020) Design of intermittent rainfall-pattern
for structures with gridded data: validation and implementation.
except for Group 6 where positive trend is observed till mid of Journal of Building Engineering 27:100939
20th century and thereafter rainfall trend was found to be Dash SK, Kulkarni MA, Mohanty UC, Prasad K (2009) Changes in the
mostly negative. The pre-monsoon rainfall showed negative characteristics of rain events in India. J Geophys Res 114:D10109
departure in last two decades (2001–2017) over most parts in Deepesh M, Kumar S, Meena HM, Santra P, Singh RK, Singh DV (2019)
the country. Clustering of rainfall stations and distinguishing influential factors
using PCA and HCA techniques over the western dry region of
India. Meteorol Appl 26(2):300–311
Supplementary Information The online version contains supplementary
Gadgil S (2003) The Indian monsoon and its variability. Annu Rev Earth
material available at [Link]
Planet Sci 31:429–467
Gadgil S, Gowri R, Yadumani (1988) Coherent rainfall zones: case study
Acknowledgements The authors gratefully acknowledge the India
for Karnataka. J Earth Syst Sci 97:63–79
Meteorological Department for providing the data in the open govern-
ment data portal and also providing 0.25°×0.25° gridded daily rainfall Gadgil S, Yadumani, Joshi NV (1993) Coherent rainfall zones of the
data from 1901 to 2013 for all India to do the cluster analysis. The authors Indian region. Int J Climatol 13:547–566
also acknowledge the open source software like NCL and R program- Gadgil S, Rajendran K, Pai DS (2019) A new rain-based index for the
ming used in this manuscript to generate figures from the data. The au- Indian summer monsoon rainfall. Mausam 70(3):485–500
thors acknowledge the Head, CSIR 4PI, for support and encouragement. Goswami BN, Venugopal V, Sengupta D, Madhusoodanan MS, Xavier
PK (2006) Increasing trend of extreme rain events over India in a
Author contribution G N Mohapatra contributed in conceptualization, warming environment. Science 314:1442–1445
design, and data analysis of manuscript. V Rakesh contributed in concep- Gregory S (1989) Macro-regional definition and characteristics of Indian
tualization, design, data analysis, and drafting the manuscript. Mrs. summer monsoon rainfall: 1871-1985. Int J Climatol 9:465–483
Smrati Purwar contributed in analyzing data and generating figures. A Guhathakurta P, Rajeevan M (2008) Trends in the rainfall pattern over
P Dimri contributed in manuscript editing and proof reading. India. Int J Climatol 28:1453–1469
Guhathakurta P, Koppar A, Krishan U, Menon P (2011) New rainfall
Funding The research was not supported by any funding other than in- series for the districts, meteorological sub-divisions and country as
stitutional support from CSIR, India. whole of India. National Climate Centre Research Report 2
Guhathakurta P, Rajeevan M, Sikka DR, Tyagi A (2015) Observed
changes in southwest monsoon rainfall over India during 1901–
Availability of data and material (data transparency) Openly available.
2011. Int J Climatol 35(8):1881–1898
Guntu RK, Rathinasamy M, Agarwal A, Sivakumar B (2020)
Code availability (software application or custom code) Openly Spatiotemporal variability of Indian rainfall using multiscale entro-
available py. J Hydrol 587:124916
Gupta A, Kamble T, Machiwal D (2017) Comparison of ordinary and
Declarations Bayesian kriging techniques in depicting rainfall variability in arid
and semi-arid regions of north-west India. Environ Earth Sci 76:512
Ethics approval Complied with the Ethical Standards of TAAC Journal. Hannachi A, O’Neill A (2001) Atmospheric multiple equilibria and non-
Gaussian behavior in model simulation. Q J R Meteorol Soc 127:
939–958
Consent to participate All listed authors have approved the manuscript
Hannachi A, Jolliffe IT, Stephenson DB (2007) Empirical orthogonal
before submission.
functions and related techniques in atmospheric science: a review.
Int J Climatol 27:1119–1152
Consent for publication All authors agreed with the content and gave Iyengar RN, Basak P (1994) Regionalization of Indian monsoon rainfall
explicit consent to submit and that they obtained consent from the respon- and long-term variability signals. Int J Climatol 14:1095–1114
sible authorities at the institute/organization where the work has been
Joshi MK, Pandey AC (2011) Trend and spectral analysis of rainfall over
carried out, before the work is submitted.
India during 1901–2000. J Geophys Res-Atmos 116(D6)
Kantharao B, Rakesh V (2018) Observational evidence for the relation-
Conflict of interest/Competing interests The authors declare no com- ship between spring soil moisture and June rainfall over the Indian
peting interests. region. Theor Appl Climatol 132(3-4):835–849
686 G. Mohapatra et al.

Kashid SS, Maity R (2012) Prediction of monthly rainfall on homoge- Rajeevan M, Bhate J, Kale JD, Lal B (2006) High resolution daily
neous monsoon regions of India based on large scale circulation gridded rainfall data for the Indian region: analysis of break and
patterns using Genetic Programming. J Hydrol 454:26–41 active monsoon spells. Curr Sci 91(10):296–306
Khouider B, Sabeerali CT, Ajayamohan RS, Praveen V, Majda AJ, Pai Rajeevan M, Gadgil S, Bhate J (2010) Active and break spells of the
DS, Rajeevan M (2020) A novel method for interpolating daily Indian summer monsoon. Indian Academy of Sciences J Earth
station rainfall data using a stochastic lattice model. J Syst Sci 119:229–247
Hydrometeorol 21(5):909–933 Rakesh V, Singh R, Yuliya D, Pal PK, Joshi PC (2009) Impact of vari-
Kothawale DR, Rajeevan M (2017) Monthly, seasonal, annual rainfall ational assimilation of MODIS thermodynamic profiles in the sim-
time series for all-india, homogeneous regions, meteorological sub- ulation of western disturbance. Int J Remote Sens 30(18):4867–
divisions: 1871-2016, ISSN 0252-1075, Research Report No. RR- 4887
138, ESSO/IITM/STCVP/SR/02(2017)/18. Rakesh V, Singh R, Joshi PC (2011) Evaluation of short-range forecasts
Krishnamurthy V, Shukla J (2008) Seasonal persistence and propagation from a mesoscale model over the Indian region during monsoon
of intraseasonal patterns over the Indian summer monsoon region. 2006. Pure Appl Geophys 168(5):919–943
Clim Dyn 30:353–369 Ramamurthy K (1969) Monsoon of India: Some aspects of the “break” in
Kulkarni A, Kriplani R, Sabade S, Rajeevan M (2011) Role of intra- the Indian southwest monsoon during July and August. India
seasonal oscillations in modulating Indian summer monsoon rain- Meteorological Department FMU Rep. IV-18-3, 13 pp
fall. Clim Dyn 36:1005–1021
Rao AR, Srinivas VV (2008) Regionalization of watersheds - an ap-
Kumar V, Jain SK, Singh Y (2010) Analysis of long-term rainfall trends
proach based on cluster analysis. Series: Water Science and
in India. Hydrol Sci J 55(4):484–496
Technology Library, 58, ISBN: 978-1-4020-6851-5, Springer
Kurths J, Agarwal A, Shukla R, Marwan N, Rathinasamy M, Caesar L,
Publishers, pp248
Krishnan R, Merz B (2019) Unravelling the spatial diversity of
Indian precipitation teleconnections via a non-linear multi-scale ap- Rao TN, Uma KN, Satyanarayana TM, Rao DN (2009) Differences in
proach. Nonlinear Process Geophys 26(3):251–266 draft core statistics from the wet spell to dry spell over Gandaki,
Lyra GB, Oliveira-Júnior JF, Zeri M (2014) Cluster analysis applied to India (13.58N, 79.28E). Mon Weather Rev 137:4293–4306
the spatial and temporal variability of monthly rainfall in Alagoas Ratna SB (2012) Summer monsoon rainfall variability over Maharashtra,
state, Northeast of Brazil. Int J Climatol 34:3546–3558 India. Pure ApplGeophys 169:259–273
Malik N, Marwan N, Kurths J (2010) Spatial structures and directional- Sahany S, Mishra SK, Pathak R, Rajagopalan B (2018) Spatiotemporal
ities in monsoonal precipitation over South Asia. Nonlinear Process variability of seasonality of rainfall over India. Geophys Res Lett
Geophys 17:371–381 45(14):7140–7147
Matulla C, Edouard KP, Haas P, Formayer H (2003) Comparative anal- Saikranthi K, Rao TN, Rajeevan M, BhaskaraRao SV (2013)
ysis of spatial and seasonal variability: Austrian precipitation during Identification and validation of homogeneous rainfall zones in
the 20th century. Int J Climatol 23:1577–1588 India using correlation analysis. J Hydrometeorol 14(1):304–317
Meher JK, Das L (2019) Gridded data as a source of missing data re- Sandeep S, Ajayamohan RS, Boos WR, Sabin TP, Praveen V (2018)
placement in station records. Journal of Earth system Science Decline and poleward shift in Indian summer monsoon synoptic
128(3):1–4 activity in a warming climate. Proc Natl Acad Sci 115(11):2681–
Nandargi SS, Mahto SS (2019) Frequency and intensity of tropical dis- 2686
turbances over the Indian region and its neighboring seas with asso- Satyanarayana P, Srinivas VV (2008) Regional frequency analysis of
ciated rainfall during the monsoon season: a perspective. precipitation using large-scale atmospheric variables. J Geophys
Engineering Reports 1(5):e12069 Res 113:D24110
Nicholson SE (1986) The nature of rainfall variability in Africa South of Satyanarayana P, Srinivas VV (2011) Regionalization of precipitation in
the Equator. J Clim 6(5):515–530 data sparse areas using large scale atmospheric variables–a fuzzy
Normand C (1953) Monsoon seasonal forecasting. Quart J Roy Meteoro clustering approach. J Hydrol 405(3-4):462–473
Soc 79:463–473 Shastri H, Paul S, Ghosh S, Karmakar S (2015) Impacts of urbanization
Pai DS, Sridhar L, Rajeevan M, Sreejith OP, Satbhai NS, Mukhopadhyay B on Indian summer monsoon rainfall extremes. J Geophys Res-
(2014) Development of a new high spatial resolution (0.25× 0.25) long Atmos 120(2):496–516
period (1901–2010) daily gridded rainfall data set over India and its Singh KK, Singh SV (1996) Space-time variation and regionalization of
comparison with existing data sets over the region. Mausam 65(1):1–8 seasonal and monthly summer monsoon rainfall of the sub-Himalayan
Panda A, Sahu N (2019) Trend analysis of seasonal rainfall and temper- region and Gangetic plains of India. Clim Res 6(3):251–262
ature pattern in Kalahandi, Bolangir and Koraput districts of Odisha,
Singh SK, Meshram SG, Meshram C, Deo RC, Ambade B (2018)
India. Atmos Sci Lett 20(10):e932
Statistical evaluation of rainfall time series in concurrence with ag-
Parthasarathy B, Kumar KR, Munot AA (1993) Homogeneous Indian
riculture and water resources of Ken River basin, Central India
Monsoon rainfall: variability and prediction. Proc Indian Acad Sci
(1901-2010). Theoretical & Applied Climatology 134
(Earth Planet Sci) 102(1):121–155
Prabhakar AK, Singh KK, Lohani AK, Chandniha SK (2019) Venkatesh B, Jose MK (2007) Identification of homogeneous rainfall
Assessment of regional-level long-term gridded rainfall variability regimes in parts of Western Ghats region of Karnataka. J Earth
over the Odisha State of India. Appl Water Sci 9:93 Syst Sci 116:321–329
Praveen B, Talukdar S, Mahato S, Mondal J, Sharma P, Islam ARMT, Ward JH Jr (1963) Hierarchical grouping to optimize an objective func-
Rahman A (2020) Analyzing trend and forecasting of rainfall chang- tion. J Am Stat Assoc 58(301):236–244
es in India using non-parametrical and machine learning approaches.
Sci Rep 10(1):1–21 Publisher’s note Springer Nature remains neutral with regard to jurisdic-
Preisendorfer RW (1988) Principal component analysis in meteorology tional claims in published maps and institutional affiliations.
and oceanography. In: Development in Atmospheric Science 17
Elsevier Amsterdam:425pp