WATER RESOURCES RESEARCH, VOL. 49, 3589–3600, doi:10.1002/wrcr.20301, 2013
Enhancement of inland penetration of monsoon depressions in the Bay
of Bengal due to prestorm ground wetness
C. M. Kishtawal,1,2 Dev Niyogi,1 Balaji Rajagopalan,3 M. Rajeevan,4 N. Jaiswal,2 and U. C. Mohanty5
Received 28 December 2011; revised 2 May 2013; accepted 2 May 2013; published 20 June 2013.
[1] Observations of 408 monsoon low-pressure systems (MLPSs) including 196 monsoon
depressions (MDs) that formed in the Bay of Bengal during the 1951–2007 period, and the
gridded analysis of daily rainfall fields for the same period, were used to identify the
association of antecedent rainfall (1 week average rainfall prior to the genesis of MLPS)
with the genesis of MLPS and length of inland penetration by MDs. Prestorm rainfall is
treated as a surrogate to prestorm ground wetness conditions due to unavailability of
historical soil-moisture data over the monsoon region. These observations were analyzed
using self-organizing maps (SOMs) to group nine different prestorm monsoon rainfall
patterns into different transition states like active, active-to-break, break-to-active, break,
etc. The analysis indicates that MLPS are four times more likely to form on a day during
active monsoon state compared to break state. Analysis of MLPSs linked to each monsoon
state represented by SOM nodes shows that MDs with higher inland penetration were
associated with higher antecedent rainfall. On the other hand, there was no significant
difference in low-level atmospheric circulation for MDs with shortest and longest inland
penetration.
Citation: Kishtawal, C. M., D. Niyogi, B. Rajagopalan, M. Rajeevan, N. Jaiswal, and U. C. Mohanty (2013), Enhancement of inland
penetration of monsoon depressions in the Bay of Bengal due to prestorm ground wetness, Water Resour. Res., 49, 3589–3600,
doi:10.1002/wrcr.20301.
1.
Introduction
[2] Monsoon lows and monsoon depressions (MDs hereafter), together termed as monsoon low-pressure systems
(MLPSs) are arguably the most important rain-bearing
weather systems for the Indian subcontinent during the
Indian summer monsoon (ISM) season. Yoon and Chen
[2005] reported that MDs contribute to about 45%–55% of
the total monsoon seasonal rainfall. About 9–10 MLPSs
form in each monsoon season providing copious amounts
of rainfall along their tracks [Krishnamurti, 1979; Sikka,
1977]. The rainfall structure in a MLPS is maintained by
moisture convergence coupled with lower tropospheric circulation. The unique topography of Indian peninsula and
Indo-China/Myanmar region favors the formation and development of MDs in warm and moist air over the Bay of
Bengal [Holt and Sethuraman, 1986]. Latent heat release
1
Department of Agronomy and Department of Earth and Atmospheric
Sciences, Purdue University, West Lafayette, Indiana, USA.
2
Space Applications Centre, Indian Space Research Organization,
Ahmedabad, India.
3
Department of Civil, Environmental and Architectural Engineering and
Co-operative Institute for Research in Environmental Sciences, University
of Colorado, Boulder, Colorado, USA.
4
National Atmospheric Research Laboratory, Gadanki, India.
5
Indian Institute of Technology Delhi, New Delhi, India.
Corresponding author: C. M. Kishtawal, Space Applications Centre, Indian Space Research Organization, Ahmedabad 380015, India.
([email protected])
©2013. American Geophysical Union. All Rights Reserved.
0043-1397/13/10.1002/wrcr.20301
due to organized convection [Shukla, 1978], barotropic
instability [Krishnamurti et al., 1980; Nitta and Masuda,
1981] as well as moist baroclinic instability [Arakawa and
Moorthi, 1988; Krishnakumar et al., 1992; Aravequia
et al., 1995] has been shown to be an important mechanisms for the genesis and development of MDs. After their
genesis in the Bay of Bengal, the MLPS move on the westnorthwest track along the monsoon trough to the warmer
and drier heat low regions of northwest India and Pakistan.
Goswami [1987] concluded that the west-northwest movement of the MLPS is due to the generation of mixed
Rossby-gravity waves to the west of the initial diabatic
heat source over the Bay of Bengal. This creates maximum
moisture convergence in the west-northwest direction,
leading to a continuous positive feedback loop.
[3] Due to the significance of the MLPS in the ISM rainfall, the distance of their inland penetration and the amount
of time spent by these systems over the land can often lead
to widespread flooding and loss of life and property and is
of significant interest for hydrological, meteorological, and
agricultural applications. Tropical systems weaken rapidly
after landfall due to the lack of surface moisture fluxes
[Kaplan and DeMaria, 1995]. Heterogeneities in the landscape structure (e.g., soil moisture, surface roughness,
albedo, vegetated land cover, and stomatal conductance)
tend to create mesoscale boundaries that can impact mesoscale circulation, convection, and precipitation [Anthes,
1984; Avissar and Liu, 1996; Segal and Arritt, 1992;
Pielke, 2001; Pielke and Niyogi, 2010]. Soil moisture plays
a predominant role because of its key influence on the partitioning of energy into sensible and latent heat fluxes at the
3589
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
ground surface [Alapaty et al., 1997]. Dastoor and
Krishnamurti [1991] reviewed the role of soil wetness on a
Bay of Bengal MD using a mesoscale numerical weather
prediction model and suggested that the extremely dry land
surface would stall the storm motion, while a extremely
wet land surface would weaken the storm, as well as produce weaker but more widespread precipitation. In their
study, the improved ground wetness parameterization
resulted in simulated rainfall closer to observation over
coastal and inland regions in the Indian monsoon region.
Recently, Chang et al. [2009] developed a process-based
assessment of three landfalling MDs and concluded that
wetter antecedent land surface can intensify the landfalling
MDs over the Indian monsoon region. In this study, we
seek to assess if this relation between the antecedent land
condition and landfalling MDs evolution can be detected in
the climatological observations. Historical records of soilmoisture observations over the monsoon region do not
exist. Therefore, we tested the viability of using the antecedent rainfall as a surrogate to the ground wetness conditions over the monsoon region. The underlying hypothesis
we will test using observations is that ground wetness represented by antecedent rainfall increases the lifespan and
intensity of MLPSs, particularly MDs. In section 2 we have
described the data used for the analysis of MLPSs and the
antecedent rainfall. This section also describes the methods
used for extracting dominant patterns of variability in antecedent rainfall and atmospheric circulation. Characterization of monsoon states based on self-organizing map
(SOM) analysis of antecedent rainfall, the transition of
monsoon among these states, and a brief analysis of MLPSs
formed during different monsoon states are described in
section 3. Section 4 provides an analysis on the comparative role of atmospheric circulation and the antecedent rainfall on the postlandfall behavior of MDs. The main
conclusion of the study is summarized in section 5.
2.
Data and Methods
2.1. Monsoon Low-Pressure Systems
[4] MLPSs have weak cyclonic circulation and a diameter of about 2000 km that form within the monsoon trough
over the Bay of Bengal and generally move in the westnorthwestward direction. India Meteorological Department
(IMD) defines a system as ‘‘low’’ if surface pressure deficit
(with respect to background) at the center of the system is
up to 1 hPa, and maximum wind speed at surface is below
8.5 m s1. MDs and deep depressions are characterized by
surface pressure deficit of 1–4.5 hPa, and maximum surface
wind speed is between 8.5 and 17.5 m s1 [World Meteorological Organization Technical Document, 2012]
[5] For the present study, MD tracks for monsoon
months (June-September, JJAS) during the period 1951–
2007 were obtained from the electronic archive of IMD
[IMD-e-atlas, 2008]. The region of the genesis of MDs in
the Bay of Bengal is in vicinity of the land mass, and the
quality of historical MD location data (e.g., latitude/longitude positions) is considered good owing to the existence
of a meteorological observational network in colonial India
since 1857. Figure 1 shows the tracks of 196 MDs that
formed over the Bay of Bengal during monsoon seasons
(JJAS) between the years 1951 and 2007. Of these 196
MDs, 190 made landfall and crossed the landmass over the
Indian subcontinent. The 190 MDs that crossed the land
were considered for further analysis and inland penetration
for these storms were determined using six-hourly track
positions available in IMD data.
[6] Data for ‘‘monsoon lows’’ are not available in the
IMD e-atlas ; therefore, we used the daily fields of 850 hPa
winds and sea level pressure from National Centres for
Environmental Prediction (NCEP) and National Centre for
Atmospheric Research (NCAR) reanalysis data [Kalnay et
al., 1996] during the period 1951–2007 for automated identification of MLPSs. Our algorithm first identifies a cyclonic circulation over the Bay of Bengal by matching 850
hPa wind fields with a 7 7 grid-point template (17.5 17.5 area) containing anticlockwise circulation. A high
correlation between the cyclonic template and the reanalysis winds indicates low-level circulation. Next, the difference of Sea Level Pressure (SLP) at the center of
circulation and a farthest point 10 away in the ocean is
used to define the pressure anomaly, and pressure anomaly
tendency at the center of the circulation. A negative pressure anomaly and its negative tendency during next 2 days
were then used as indicators of low-pressure systems. The
earlier procedure could identify 408 MLPSs that included
188 of the 196 MDs reported in the IMD e-atlas.
2.2. Rainfall Data
[7] Daily gridded rainfall data at 1 grid spacing from
IMD [Rajeevan et al., 2006] were used. This data set is
based on quality-controlled rainfall data from 2140 stations
over India with a minimum 90% data availability during
the period 1951–2007. Using these station data, the gridpoint analysis of daily accumulated rainfall was prepared
following Shepard’s directional interpolation method [Shepard, 1968] over the Indian subcontinent (6.5 N–37.5 N,
66.5 E–101.5 E). Standard quality controls were performed before carrying out the interpolation analysis. Due
to averaging, the gridded rainfall data are smoother compared to individual station data. These data are considered
of high quality and representative of the monsoon region
rainfall and have been used in various assessments and climatic studies over the Indian monsoon region [e.g.,
Goswami et al., 2006; Rajeeven et al., 2006; Dash et al.,
2009; Kishtawal et al., 2010; Niyogi et al., 2010].
2.3. Land Rainfall Antecedent to MLPSs (Antecedent
Rainfall)
[8] Monsoon lows and MDs (referred as MLPS henceforth) travel primarily along the seasonal monsoon trough
(Figure 1). Lateral and surface fluxes of heat and moisture
over this region can be expected to directly impact the
MLPSs traversing through it [Krishnamurti et al., 1998].
Since long-term soil-moisture observations do not exist for
the Indian monsoon region, we tested different averaging
times to adapt antecedent rainfall as a surrogate for soil
wetness. We considered the 1 week average rainfall over
the monsoon-trough region (area and mean prestorm rainfall shown in Figure 2a) preceding the time of formation of
a MLPS over the Bay of Bengal as the antecedent rainfall
for that MLPS. In other words, if To is the date of the formation of a MLPS, then antecedent rainfall is defined as
the average rainfall over monsoon-trough region for a
3590
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 1. Tracks of all the MDs originating from the Bay of Bengal during 4 months (JJAS) of the
monsoon season (1951–2007).
period To7 to To1 days. The choice of averaging period
for rainfall (1 week) was made considering the average
time between two consecutive MDs from IMD e-atlas. An
analysis of the MD lifespans and tracks indicate that the average time interval between two consecutive MDs is 19
days, and about 25% of MDs formed within 10 days. Thus,
the choice of averaging period of 1 week prior to MLPS
formation ensures that the antecedent ground wetness conditions are almost unique for each MLPS with minimal
overlap among them. Additionally, in a recent study, Chandrasekar et al. [2008] showed that at weekly time scales,
rainfall and soil moisture are significantly correlated over
the monsoon-trough region during all phases of monsoon
season.
2.4. Relationship of Rainfall With Soil Wetness
[9] To verify the use of antecedent rainfall as surrogate
for soil-wetness condition, we compared satellite-observed
soil wetness with IMD rainfall over central India. Special
Sensor Microwave/Imager based weekly surface wetness
index (BWI) derived by Bassist et al. [1998] was available
for a limited period [1988–2002] and was used for comparison. This index is based on the passive microwave radiometer data at four frequencies (19.35, 22.235, 37.0, and 85.5
GHz) and dual polarization (except at 22.235 GHz). The
weekly composite BWI data were obtained at 0.3 spatial
resolution from the National Climatic Data Center. For
comparison, both the rainfall and the BWI data were
sampled to the same time period (1 week) over a common
spatial domain over central India (72 E–82 E : : 21 N–
26 N). Figure 2b shows a high correlation (r ¼ 0.72)
between rainfall and soil wetness on a weekly time scale.
The highest correlation (r ¼ 0.78) between soil wetness and
rainfall was observed during the initial phase of monsoon
(June-July), somewhat weaker during August (r ¼ 0.67),
when rainfall activity is at its peak and the soil tends to
3591
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
property, the SOMs are effective tools for identifying not
only different regimes within data but also the patterns of
transition states from one regime to another. This is especially suited for the ISM rainfall as it is characterized by
strong intraseasonal variability such as active, break, and
transition states. SOMs have been previously used for the
identification of atmospheric precursors to extreme rainfall
[Cavazos, 1999], analysis of preferred states of monsoon
intraseasonal oscillations (ISOs) [Sahai and Chattopadhyay,
2006], and analysis of synoptic climatic patterns [Hewitson
and Crane, 2002]. Sheridan and Lee [2011] have provided a
review of SOMs in climatological research.
[11] We used SOM-based classification of monsoon rainfall over the monsoon-trough region (major pathway of monsoon lows and MDs) into distinct subgroups with an
objective to analyze the patterns of formation and postlandfall behavior of monsoon lows and MDs during different
phases of monsoon. The 1 week (or 7 day) mean rainfall
composites preceding the formation of each MLPS (n ¼ 408)
during 1951–2007 period was used for SOM analysis. One
important step in this procedure is to select the number of
nodes of SOM. A rule of thumb
pffiffiffi used by many investigators
in cluster analysis is Cmax < n where Cmax is the maximum
number of clusters and n denotes the number of observations.
Jian and Quiansheng [2001] demonstrated the validity of this
rule through a detailed theoretical analysis. For our sample
size of n ¼ 408 the SOM nodes should be less than (5 5).
However, we used a (3 3) square topology to train the
SOM to keep the analyses and interpretations as simple as
possible. It is to be noted that SOM analysis mentioned earlier pertains only to rainfall prior to the formation of MLPSs.
The analysis of the association of these low-pressure systems
with different clusters of rainfall patterns obtained after SOM
analysis is carried out in the next step.
Figure 2. (a) Mean antecedent rainfall (averaging period¼ 7 days) for 196 MDs. (b) Scatterplot showing weekly
averaged IMD rainfall and satellite-observed wetness index
over central India for monsoon months (1988–2002).
saturate and slight increase during the withdrawal phase of
the monsoon (September, r ¼ 0.70). A weaker relation
between soil wetness and rainfall during the peak and withdrawal phase could be indicative of the fact that after saturation the soil exerts little control on evaporative processes.
The comparison from two independent data sources further
corroborates our basis for using 1 week average antecedent
rainfall as a good proxy for soil wetness. It is to be noted
that even during the monsoon season, the soil wetness
shows considerable variability on both interannual and
intraseasonal time scales.
2.5. SOM Analysis of Antecedent Rainfall
[10] SOMs are one of the ‘‘unsupervised’’ cluster analysis
tools that ‘‘map’’ the input data onto a limited number of
nodes [Kohonen 1988, 1990; Hewitson and Crane, 2002].
We prefer SOM to a standard clustering algorithm because it
can inform the apparent mutual relationship among different
clusters. SOMs use a neighborhood function to preserve topological characteristics of the input data [Lampinen and Oja,
1989]. In that on a self-organized map the degree of similarity between two clusters is proportional to their topological
distance from each other. Due to this topology preserving
2.6. Empirical Orthogonal Function Analysis of Lower
Tropospheric Circulation
[12] We used stream functions of 850 hPa vector wind
fields from NCEP/NCAR reanalysis fields as representatives
of lower atmospheric circulation. Stream function, which is a
scalar representation of the vector velocity fields of a 2-D
incompressible flow, is easier to analyze and interpret compared to vector velocity fields. These stream function fields
were subjected to empirical orthogonal function (EOF) analysis for extraction of dominant modes of variability in atmospheric circulation. The EOF analysis decomposes spatialtemporal variations of a variable into combination of orthogonal spatial patterns with corresponding principal components [Lorenz, 1956]. The primary purpose of this analysis is
to isolate the modes of atmospheric circulation linked with
the genesis of MLPSs, particularly, the MDs and quantitatively compare the roles of atmospheric circulation and antecedent ground wetness in postlandfall lifespan of MDs.
[13] It would be worthwhile to explain why we used two
separate pattern extraction methods in the present study.
Both SOM and EOF analyses are efficient pattern extraction tools that are used in geophysical sciences. The conventional EOF analysis relies on the linear relationship
among variables within one pattern and orthogonal relation
among different patterns. On the other hand, SOM can
extract the patterns where the variables are related nonlinearly in a pattern, and different patterns are not orthogonal
3592
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 3. Patterns of antecedent rainfall (mm/d) over the monsoon-trough region projected on SOM
nodes. N denotes the number of MLPS for each node.
to each other. Patterns of rainfall are characterized by large
spatial discontinuity and high spatiotemporal variability. It
is difficult to obtain realistic patterns of rainfall variability
using a linear method like EOF analysis. Dommenget and
Latif [2002] mentioned that principal components of the
dominant EOF mode are often superposition of many different modes, and the center of action derived by EOF analysis can be different from the centers of action of real
physical modes. Considering these pitfalls, they recommended to analyze climate data with different statistical
methods, not only by EOF analysis. We hence chose to use
SOM analysis for analyzing a discontinuous variable like
rainfall and EOF analysis for a continuous variable like
atmospheric circulation. It can be noted that the EOF technique has a unique advantage over SOM. The patterns
extracted by EOF analysis can be numerically compared
with each other through the principal components (which
are analogous to the ‘‘amplitudes’’ of the EOF modes).
3.
Results
3.1. Spatial Patterns of Monsoon Rainfall Variability
Identified by SOM Analysis
[14] SOM analysis resulted in identification of nine distinct states (projected on 3 3 topology) of antecedent
rainfall over monsoon-trough region. These distinct rainfall
patterns, projected on nine ‘‘nodes’’ of SOM, are shown in
Figure 3.The number of MLPS that formed in the Bay of
Bengal after the appearance of these rainfall patterns in the
trough region is shown in each panel. The top right node
(denoted as TR) of SOM shows the typical situation of a
well-developed active phase of the monsoon with tropical
convergence zone (TCZ or ‘‘monsoon trough’’ as commonly referred in monsoon literature) located over the core
monsoon region. In this node which appears during the
months of July and August (period when monsoon activity
is at its peak), rain rates typically exceed 20 mm/d over
central and west-central parts of India, and relatively dry
situations prevail to the north and to the south of the TCZ
(see corresponding panel in Figure 4). The diagonally opposite node at the bottom left (denoted as BL) indicates dry
land conditions that may occur during the early phase of
the monsoon during June, periodic monsoon ‘‘breaks,’’ and
during the withdrawal of the monsoon after second half of
September (see corresponding panel in Figure 4). Area
averaged rain rates during this state of monsoon seldom
exceed 3 mm/d. Normally, monsoon rains first arrive at the
southern tip of Indian peninsula during first week of June
and gradually progress northward [Ramage, 1971]. It is not
before first week of July that most parts of the Indian
3593
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 4.
Monthwise distribution of MLPS genesis for different SOM nodes.
subcontinent are covered by monsoon rain. On the other
hand, withdrawal of monsoon begins from northwest part
of India during second half of September and gradually
progresses in southeast direction. More than 80% of the
time, the dry node BL appears during the months of June
and September (see corresponding panel in Figure 4), while
the remaining 20% appears during July and August associated with prolonged break conditions during these months.
Two nodes adjacent to BL, i.e., the nodes in the middle left
(ML) and bottom center (BC), also denote dry situations
over the core monsoon region and over NW India, but with
relatively wetter conditions in the northeastern and southern regions, respectively (Figure 3). Node (BR) shows the
transition of monsoon state from ‘‘active’’ to ‘‘break’’ when
TCZ migrates northward and lies along the foothills of the
Himalayas, while the node at top left (TL) denotes the
‘‘break-to-active’’ transition with TCZ in the southpeninsular India and a dry situation to the north (Figure 3),
also consistent with Sikka and Gadgil [1980]. Other nodes
indicate ‘‘normal’’ monsoon conditions with widespread
rainfall extending from the southeast to northwestern parts
of India.
3.2. Transition of Monsoon Rainfall States Across the
SOM Nodes
[15] ISM is characterized by strong ISOs that manifest
themselves as spells, or phases of active and weak (or
‘‘break’’) rainfall [Ramage, 1971]. These phases are associ-
ated with distinct synoptic features, e.g., wind circulation,
surface pressure, and precipitation. Furthermore, these
phases are distinctly identified by SOM analysis discussed
in the previous section. Due to the unique neighborhood
function used by SOMs, the patterns projected on neighboring SOM nodes (Figure 3) indicate a gradual, distinct, and
systematic transition of state vectors (rainfall in this case)
from one state to another. In order to understand the patterns of state transition during monsoon, we computed 7
day running mean rainfall during the monsoon season for
the period 1951–2007 and projected them onto the SOM
nodes identified in Figure 3. The projection will assign
each 7 day rainfall to a particular node, and consequently,
the probabilities of monsoon staying in different states (or
nodes) and transiting from one state to another were determined. Our analysis suggested that 80% of the time the
monsoon remains in one of the nine states (stationary
mode), while 20% of the times it transits from one state to
another (transit mode). Furthermore, the probability of
monsoon staying in the driest state (node BL in Figure 3) is
the highest at 38%. As mentioned earlier, the node BL
mostly represents conditions during onset (early June) and
withdrawal (late September) in addition to prolonged monsoon breaks during July and August. On the other hand,
probability of highly active monsoon state (node TR) is just
5%. The probability of other nodes associated with good
monsoon conditions, i.e., nodes TL, TC, MR, and BR are
5%, 5%, 8%, and 11%, respectively (Figure 5a). Thus,
3594
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
higher probability of MLPS genesis during active phase of
monsoon to enhancement of horizontal wind shear of lowlevel winds and positive anomalies of cyclonic vorticity
during this phase. Figure 6 shows the average 850 hPa circulation for wet nodes (Figure 6a), dry nodes (Figure 6b),
and their difference (Figure 6c). Wet state is characterized
by higher wind shear, cyclonic vorticity, and 1–2 hPa lower
surface pressure compared to dry nodes. The negative
anomaly of surface pressure (shown by shaded area) prevails over a wide part of the Bay of Bengal and monsoontrough region.
[16] It is interesting to note the transition of monsoon
from one state to another in Figure 5b. The predominantly
clockwise and asymmetric transition (in SOM space) of
monsoon can be a manifestation of strong ISO within the
monsoon season. Monsoon ISOs are characterized by
repeated northward propagation and fluctuation of TCZ
between two favored locations, one over the monsoontrough region including the northern Bay of Bengal, and
Figure 5. (a) Probability of occurrence of different states
(SOM nodes) during monsoon season (probabilities scaled
by area of circle) and (b) probability of transition of monsoon from one state to another (probabilities scaled by
length of the arrows). Shaded area indicates a general ‘‘wet
state.’’
during the 4 month monsoon season, active rainfall states
denoted by nodes TL, TC, TR, MR, and MC (henceforth
referred as ‘‘wet nodes’’) prevail only 35% of the time,
while dry conditions (nodes ML, MC, BL, and BC, together
referred as ‘‘dry nodes’’) prevail 65% of the time. This
uneven distribution of climatological probabilities of monsoon states can explain the large number of MLPS formations during some of dry nodes. A more realistic
assessment of the monsoon states favorable to the generation of MLPS can be obtained by normalizing the probability of MLPS formation in each state by climatological
probability of that state. These normalized scores are
denoted by symbol in Figure 5a. This analysis shows that
a MLPS is four times more likely to form on a day during
active monsoon state (node TR, ¼ 3.02) than on a day
during dry monsoon state (node BL, ¼ 0.75). Interestingly, is larger than 1.0 for all wet nodes and smaller than
1.0 for all dry nodes. Goswami et al. [2003] attributed the
Figure 6. The 850 hPa circulation for (a) wet nodes, (b)
dry nodes, and (c) difference of circulation between wet
and dry nodes. Shaded region shows negative differences
in surface pressure.
3595
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 7. (a) Average monthly occurrences of MDs, (b) average monthly inland penetration by MDs,
and (c) histograms of IPL (in 100 km bins) for wet (hatched bars) and dry (solid bars) SOM nodes.
the other over the southern Indian Ocean between equator
and 10 S [Sikka and Gadgil, 1980; Goswami, 1994]. During active (break) monsoon conditions, the northern (southern) TCZ is stronger. Northward propagation of TCZ from
these two locations is associated with the transition of monsoon from one phase to the other. The state-transition map
shown in Figure 5b indicates that some monsoon states are
more predictable than others. For example, transitions to
the four corner nodes (TL, TR, BL, and BR) result from
fewer neighboring nodes, leading to least uncertainty and
most predictability for these nodes. Largest uncertainty of
transitions is observed for the nodes MC, MR, and BC.
3.3. Inland Penetration of MDs: Association With
Monsoon States
[17] It is evident from the discussion in the previous section that active monsoon conditions associated with ‘‘wet
SOM nodes’’ are more conducive for genesis of MLPS
over the Bay of Bengal. After the genesis, these low-
pressure systems travel in west-northwest direction along
the monsoon-trough region and bring copious rains to Indian landmass. Thus, the ‘‘inland penetration length’’
(henceforth IPL, the distance traveled after landfall before
these systems dissipate) of these low-pressure systems is of
importance for skillful prediction of extreme rainfall and
flood, consequently enabling effective emergency response.
We hypothesize that the wet conditions over land induced
by antecedent precipitation plays a major role in postlandfall lifespan and inland penetration of monsoon lowpressure systems. In that, presence of wet surface conditions ahead of MD propagation sustains the rainfall within
MD and support the east-west circulation, which in turn,
help the MD survive longer over the land. This hypothesis
builds on a mechanism for westward propagation of MDs
proposed by Chen et al. [2005]. According to this mechanism, the latent heat released by anomalously intense rainfall in west-southwest sector of MD forms east-west
differential heating across the depression in developing an
3596
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 8. Four dominant orthogonal modes of 850 hPa stream function fields based on NCEP/NCAR
reanalysis winds for 57 monsoon seasons (1951–2007). (a) EOF-1, (b) EOF-2, (c) EOF-3, and (d) EOF4. Negative regions are shaded.
east-west asymmetric circulation. The upward branch of
east-west circulation coupled with moisture convergence in
the lower troposphere generates a negative stream function
tendency. Thus, a dynamic interaction between rainfall and
MD circulation leads to westward propagation of MDs
through this negative stream function tendency. This mechanism highlights the importance of MD’s rainfall in its
westward propagation. This rainfall is supported by water
vapor flux convergence and surface evaporation.
[18] For all the 196 MDs we computed, the monthly
genesis and average IPL for each month (shown in Figures 7a and 7b). During the period 1951–2007, 101 MDs
formed during wet state (corresponding to SOM nodes
TL, TC, TR, MR, and BR) and 95 during dry state (nodes
ML, MC, BL, BC) of monsoon. The IPL of wet state MDs
was in the range 5–2250 km (mean ¼ 820 km), and that of
dry state MDs was in the range 33–1800 km (mean ¼ 541
km). Figure 7c shows the frequency distribution of IPL for
both the states. It can be seen that the IPL of dry state (wet
state) MDs is dominated in the shorter (longer) range. The
IPL distribution between the two states was found to be stat-
istically significant at 99% confidence level using a 2 test.
Thus, MDs more often penetrate further inland when the
prestorm surface is wet rather than dry.
4. Inland Penetration of MDs: Association With
Atmospheric Circulation and Antecedent
Precipitation
[19] It can be argued that the general atmospheric conditions (e.g., low-level wind circulation, water vapor loading,
and moisture convergence) during the wet state are responsible for higher inland penetration of MDs. We showed earlier that low-level circulation, relative vorticity, and surface
pressure during wet states is conducive for the genesis of
low-pressure systems. The same atmospheric conditions
are also associated with higher precipitation (and hence,
the ground wetness) along the monsoon-trough region. In
order to separately assess the influences of atmospheric
conditions and antecedent precipitation on IPL of MDs, we
3597
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Figure 9. Regression of antecedent rainfall onto four standardarized EOFs of 850 hPa stream function
fields, (a) EOF-1, (b) EOF-2, (c) EOF-3, and (d) EOF-4. Shaded regions indicate correlations significant
at 0.01% level.
carried out further analysis on the atmospheric circulation
fields described later.
[20] Since the low-level vorticity plays crucial role in
defining active/break monsoon states as well as the genesis
and movement of MDs, we used 850 hPa stream function
fields as a representative of atmospheric conditions. The
Figure 10.
relationship between stream function ( ) and vorticity ()
is given as follows:
r2 ¼ ¼ r v
ð1Þ
where v denotes the vector wind. Decomposition of basic
fields into orthogonal components using EOF analysis is a
(a) First and (b) fourth orthogonal modes of antecedent rainfall.
3598
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
commonly used practice in earth science studies for the
ease of the analysis and interpretation of the results. We
used 850 hPa winds from NCEP/NCAR reanalysis during
57 monsoon seasons (1951–2007) and computed 850 fields
with the earlier relationship and subsequently their dominant orthogonal components using EOF analysis. Figure 8
shows the first four dominant EOFs of 850, of which the
second mode that explains 26% of the variance appears to
be closely linked to monsoon TCZ and rainfall processes
along the monsoon trough. Second part of our analysis
aims to find out how much variability of antecedent precipitation can be attributed to atmospheric conditions represented by 850. For this we regressed the antecedent
rainfall time series with the 850 corresponding to each of
the dominant EOF (shown in Figure 9). We note that the
antecedent rainfall is the average 7 day rainfall prior a day ;
thus, daily 850 lags the rainfall. It is interesting to note
that second EOF of 850 has a strong association with rainfall over monsoon-trough region, peninsular India, and
west coast. Atmospheric conditions that lag rainfall in this
analysis have obviously not caused the rainfall and significant association between the two can be a manifestation of
the persistence of favorable atmospheric conditions. It is
obvious from the earlier analysis that antecedent rainfall
fields are indeed associated with atmospheric conditions.
[21] In order to assess the impact of antecedent rainfall
on IPL, independent of atmospheric conditions, we decomposed the antecedent rainfall also into orthogonal components using covariance matrix based on 57 years (1951–
2007) of monsoon season data from IMD analysis. Composite average of 850 and rainfall corresponding to each of
the four EOFs was computed for the set of short (MDs with
the shortest 25% of IPLs: 49 samples) and long (MDs with
longest 25% of IPLs: 49 samples) IPL MDs separately.
Two-sample t test was performed to compute the significance of the difference in mean 850 and rainfall between
the short and long IPL years. The results of this analysis
indicated that none of the four EOFs of 850 were statistically significant. In other words, the atmospheric conditions
for the two samples of MDs were not much different from
each other. On the other hand the difference of mean of first
EOF of antecedent rainfall was significant at 95% confidence level. Similarly, the fourth EOF of antecedent rainfall was significant at 90% confidence level. The structure
of these two significant rainfall EOFs is shown in Figure
10. The first EOF pattern suggests that the antecedent rainfall along the monsoon trough (higher loadings in the
trough region) is probably responsible for longer inland
penetration of MDs. Similarly, the significance of the
fourth EOFs suggests that dry surface conditions (indicated
by shaded negative contours of EOF) in northwest India
may be associated with short-lived MDs.
5.
Conclusions
[22] Using the data of 408 MLPSs including 196 MDs
formed in the Bay of Bengal during the period 1951–2007
and gridded daily rainfall fields for the same period, we
explored the association between the prestorm rainfall, genesis of MLPSs, and the length of postlandfall inland penetration by MDs. The SOM technique was used to group the
prestorm rainfall into nine clusters that can be broadly clas-
sified into wet and dry states of monsoon. SOM analysis
showed that dry states are more likely to occur during monsoon season compared to wet states; however, the probability of genesis of MLPS during a wet-state day is about four
times higher compared to a dry state day. These maps
showed that in general, the mean inland penetration of wetstate MDs is longer than the MDs that form during dry
state. Larger inland penetration of MDs for wetter prestorm
conditions supports the results from earlier modeling case
studies [Dastoor and Krishnamurti, 1991; Yoon and Chen,
2005; Chang et al., 2009] that a regular supply of lower
tropospheric water vapor is essential for the maintenance of
the structure of a MD. Interestingly, for 49 samples of MDs
with shortest inland penetration and another 49 MDs with
longest inland penetration, the mean atmospheric circulations were not significantly different. On the other hand,
antecedent rainfall over monsoon-trough region exhibited
significance between the two samples. Overall, our results
provide evidence that there is a significant link between antecedent ground wetness and deeper inland penetration of
MDs. These findings have immense implications for
improving short-term forecasting of extreme rainfall, flood
forecasting, and consequently, societal benefits via efficient
emergency response.
[23] Acknowledgments. This study benefited from NSF CAREER
ATM-0847472 grant (Liming Zhou, Anjuli Bamzai, Eric DeWeaver, Jay
Fein). We are thankful to the anonymous reviewers for their valuable comments that were extremely helpful in improving the quality of this paper.
References
Alapaty, K., S. Raman, D. S. Niyogi (1997), Uncertainty in specification of
surface characteristics: A study of prediction errors in the boundary
layer, Boundary-Layer Meteorol., 82, 473–500.
Anthes, R. A. (1984), Enhancement of convective precipitation by mesoscale variations in vegetative covering in semiarid regions, J. Clim. Appl.
Meteorol., 23, 865–889.
Arakawa, A., and S. Moorthi (1988), Baroclinic instability in vertically discrete systems, J. Atmos. Sci., 45, 1688–1707.
Aravequia, J. A., V. B. Rao, and J. P. Bonatti (1995), The role of moist baroclinic instability in the growth and structure of monsoon depressions, J.
Atmos. Sci., 52, 4393–4409.
Avissar, R., and Y. Liu (1996), Three-dimensional numerical study of shallow convective clouds and precipitation induced by land surface forcing,
J. Geophys. Res., 101(D3), 7499–7518.
Bassist, A., N. C. Grody, T. C. Peterson, and C. N. Williams (1998), Using
the Special Sensor Microwave/Imager to monitor land surface temperatures, wetness, and snow cover, J. Appl. Meteorol., 37, 888–911.
Cavazos, T. (1999), Large scale circulation anomalies conducive to extreme
precipitation events and derivation of daily rainfall in northern eastern
Mexico and southeastern Texas, J. Clim., 12, 1506–1523.
Chandrasekar, K., M. V. R. SeshaSai, R. S. Dwivedi, and P. S. Roy (2008),
Surface soil moisture changes during 2007 summer monsoon season
derived from AMSR-E Land3 product, Curr. Sci., 95, 1731–1738.
Chang, H. I., D. Niyogi, A. Kumar, C. M. Kishtawal, J. Dudhia, F. Chen, U.
C. Mohanty, and M. Shepherd (2009), Possible relation between land
surface feedback and the post-landfall structure of monsoon depressions,
Geophys. Res. Lett., 36, L15826, doi:10.1029/2009GL037781.
Chen, T. C., J. H. Yoon, and S. Y. Wang (2005), Westward propagation of
the Indian monsoon depression, Tellus, 57A, 758–769.
Dash, S. K., M. A. Kulkarni, U. C. Mohanty, and K. Prasad (2009), Changes
in the characteristics of rain events in India, J. Geophys. Res., 114,
D10109, doi:10.1029/2008JD010572.
Dastoor, A., and T. N. Krishnamurti (1991), The landfall and structure of a
tropical cyclone: The sensitivity of model predictions to soil moisture
parameterizations. Boundary-Layer Meteorol., 55, 345–380.
Dommenget, D., and M. Latif, (2002) A cautionary note on the interpretation of EOFs, J. Clim., 15(2), 216–225.
3599
KISHTAWAL ET AL.: LANDFALLING MONSOON DEPRESSIONS
Goswami, B. N. (1987), A mechanism for the west-north-west movement
of monsoon depressions, Nature, 326, 376–378, doi:10.1038/326376a0.
Goswami, B. N. (1994), Dynamical predictability of seasonal monsoon
rainfall: Problems and prospects, Proc. Indian Natl. Sci. Acad., 60A,
101–120.
Goswami, B. N., R. S. Ajayamohan, P. K. Xavier, and D. Sengupta (2003),
Clustering of low pressure systems during the Indian summer monsoon
by intraseasonal oscillations, Geophys. Res. Lett., 30(8), 1431,
doi:10.1029/2002GL016734, 2003.
Goswami, B. N., V. Veugopal, D. Sengupta, M. Madhusoodan, and P. Xavier (2006), Increasing trend of extreme rain events over India in a warming environment, Science, 314, 1442–1445.
Hewitson, B. C., and R. C. Crane (2002), Self organizing maps: Application to synoptic climatology, Clim. Res., 22, 13–26.
Holt, T., and S. Sethuraman (1986), A comparison of the significant features of the marine boundary layers over the Arabian Sea and the Bay of
Bengal during MONEX-79, in Proceedings of National Conference on
FGGE, January 14–17.
IMD-e-atlas (2008), Tracks of Cyclones and Depressions (1891–2007)
Electronic Version 1.0/2008, India Meteorol. Dep., New Delhi, India.
Jian, Y. U., and C. Quinsheng (2001), The upper bound of the optimal number of clusters in fuzzy clustering, Sci. China, 44, 119–125.
Kalnay, E., et al. (1996), The NCEP/NCAR 40-Year Reanalysis Project,
Bull. Am. Meteorol. Soc., 77, 437–472.
Kaplan, J., and M. DeMaria (1995), A simple empirical model for predicting the decay of tropical cyclone winds after landfall, J. Appl. Meteorol.,
34, 2499–2512.
Kishtawal, C. M., D. Niyogi, M. Tewari, R. A. Pielke, and J. M. Shepherd
(2010), Urbanization signature in the observed heavy rainfall climatology over India, Int. J. Climatol., 30, 1908–1916, doi:10.1002/joc.2044,
2010.
Kohonen, T. (1988), Self-Organization and Associative Memory, Ser. in
Inform. Sci., vol. 8, 2nd ed., Springer-Verlag Berlin Heidelberg.
Kohonen, T. (1990), The self organizing map, Proc. IEEE, 78(9), 1464–
1480.
Krishnakumar, V., R. N. Keshavamurty, and S. V. Kasture (1992), Moist
baroclinic instability and the growth of monsoon depressions—Linear
and nonlinear studies, J. Earth Syst. Sci., 101, 123–152.
Krishnamurti, T. N. (1979), Tropical meteorology, in Compendium of
Meteorology II, edited by A. Wiin-Nielsen, WMO 364, 428 pp., World
Meteorol. Organ, Geneva.
Krishnamurti, T. N., Y. Ramanathan, P. Ardanuy, R. Pasch, and P. Grieman
(1980), Quick look summer MONEX atlas. Part III: Monsoon depression
phase, Rep. 80-8, 135 pp., FSU.
Krishnamurti, T. N., M. C. Sinha, B. Jha, and U. C. Mohanty (1998), A
study of South Asian monsoon energetics, J. Atmos. Sci., 55, 2530–2548.
Lampinen, J., and E. Oja (1989), Fast self-organization by the probing algorithm, in Proceedings of the International Joint Conference on Neural
Networks, IJCNN, pp. II 503–II 507.
Lorenz, E. N. (1956), Empirical orthogonal functions and statistical
weather prediction, Sci. Rep. 1, 49 pp., Mass. Inst. of Technol.,
Cambridge.
Nitta, T., and K. Masuda (1981), Observational study of a monsoon depression developed over the Bay of Bengal during summer MONEX, J.
Meteorol. Soc. Jpn., 59, 672–682.
Niyogi, D., C. Kishtawal, S. Tripathi, and R. S. Govindaraju (2010), Observational evidence that agricultural intensification and land use change
may be reducing the Indian summer monsoon rainfall, Water Resour.
Res., 46, W03533, doi:10.1029/2008WR007082.
Pielke, R. A., Sr. (2001), Influence of the spatial distribution of vegetation
and soils on the prediction of cumulus convective rainfall, Rev. Geophys.,
39, 151–177.
Pielke, R. A., Sr., and D. Niyogi (2010), The role of landscape processes
within the climate system, in Landform—Structure, Evolution, Process
Control: Proceedings of the International Symposium on Landforms
Organised by the Research Training Group 437, Lecture Notes in Earth
Sciences, vol. 115, edited by J. C. Otto and R. Dikaum, 258 pp.,
Springer-Verlag Berlin Heidelberg.
Rajeevan, M., J. Bhate, J. D. Kale, and B. Lal (2006), High resolution daily
gridded rainfall data for Indian region: Analysis of break and active
monsoon spells, Curr. Sci., 91(3), 296–306.
Ramage, C. S. (1971), Monsoon Meteorology, Academic, 296 pp., New
York.
Sahai, A. K., and R. Chattopadhyay (2006), An objective study of Indian
summer monsoon variability using the self organizing map algorithm,
IITM Res. Rep. RR-113, 41 pp., Indian Inst. of Trop. Meteorol., Pune,
India.
Segal, M., and R. W. Arritt (1992), Nonclassicalmesoscale circulations
caused by surface sensible heat-flux gradients, Bull. Am. Meteorol. Soc.,
73, 1593–1604.
Shepard, D. (1968), A two-dimensional interpolation function for irregularly spaced data, in Proceedings of the 23rd National Conference ACM,
p. 517–524.
Sheridan, S. C., and C. C. Lee (2011), The self-organizing map in synoptic
climatological research, Prog. Phys. Geogr., 35, 109–119.
Shukla, J. (1978), CISK barotropic instability and the growth of monsoon
depressions, J. Atmos. Sci., 35, 495–508.
Sikka, D. R. (1977), Some aspects of the life history, structure and movement of monsoon depressions, Pure Appl. Geophys., 115, 1501–1529.
Sikka, D. R., and S. Gadgil (1980), On the maximum cloud zone and the
ITCZ over Indian longitudes during the southwest monsoon, Mon.
Weather Rev., 108, 1840–1853.
World Meteorological Organization Technical Document (2012), Tropical
cyclone operational plan for the Bay of Bengal and the Arabian Sea,
Document, Edition 2012, WMO/TD 84, pp. 106.
Yoon, J.-H., and T.-C. Chen (2005), Water vapor budget of the Indian monsoon depression, Tellus, 57A, 770–782.
3600