A detailed investigation on the potentially drought-prone regions over India
is presented in this study based on the balance between precipitation
and potential evapotranspiration (PET) during the southwest Asian
mid-monsoon season. We introduce a parameter named dry day frequency (DDF)
which is found suitable to present the drought index (DI) in mid-monsoon
season, hence strongly associated with the possibility of drought
occurrences. The present study investigates the probable aspects which
influence the DDF over these regions, revealing that the abundance of
anthropogenic aerosols especially over urbanized locations has a prevailing
role in the growth of DDF during the last few decades. The prominent increasing
trend in DDF over Lucknow (26.84
Drought is a natural and recurrent phenomenon which occurs in all forms of climate. Although similar to aridity in many ways, droughts are mainly temporary in nature and thus they should not be confused with water scarcity due to excess of water demand over available supply. On the other hand, these weather extremes are more reasonably linked with the distribution and frequency of rainfall over any region. Although there are no generally accepted definitions for drought, the American Meteorological Society has categorized it into four types: meteorological or climatological, agricultural, hydrological, and socioeconomic (Heim, 2002). A prolonged drought lasts several months or even years, while the absence or reduction of precipitation creates meteorological droughts. On the other hand, short-term (few weeks) dryness in the surface layer could result in an agricultural drought (Heim, 2002). However, when prolonged meteorological droughts reduce the ground water level severely then hydrological droughts occur. Finally, all of the first three droughts with a deficit in water availability are named as socioeconomic drought. Among these four, the agricultural drought might be a serious issue when farming or crop producing in humid or subhumid zones are concerned. The situation has, however, become more serious in the present due to rapid population growth across all continents, thereby also producing an increase in their global demand (Sivakumar, 2011).
India is a country where agriculture and its allied activities act as major
sources of livelihood and hence it is expected to be deeply affected by
drought occurrences, especially if they occur in the mid-monsoon period (as it
experiences
India happens to be one of the most vulnerable drought-prone countries, as severe droughts occur at least once in a 3-year time span since the past few decades. In addition, there are numerous instances of severe drought conditions during monsoon as reported in the recent past (Pai et al., 2011). Consequently, several studies have been carried out in recent years in order to understand the drought occurrences during the Indian summer monsoon period (Gore and Sinha Ray, 2002). Bhalme and Mooley (1980) defined the Drought Area Index for drought intensity assessment using monthly rainfall distribution. Raman and Rao (1981) suggested a possible relation between summer droughts and prolonged brake phase of the southwest monsoon over the Indian subcontinent. Parthasarathy et al. (1987) identified the extreme drought years by analyzing the decade-long anomalies in the Indian summer monsoon rainfall. Tyalagadi et al. (2015) analyzed more than 100 years of rainfall and identified 21 drought years, half of which were associated with El Niño. Gadgil et al. (2003) explained the excess rainfall or drought in terms of the Equatorial Indian Ocean Oscillation (EQUINOO) during 1972–2002, especially during monsoon season. Francis and Gadgil (2010) also suggested the role of the El Niño Southern Oscillation (ENSO) and EQUINOO behind the 48 % deficit of June rainfall over India. Apart from these oscillations, like ENSO or IOD (Indian Ocean Dipole), there are also lots of other parameters which may have prominent influences on drought occurrence: Himalayan ice cover, Eurasian snow cover, the passage of intraseasonal waves, effects of accumulated pollution etc., e.g., Krishnamurti et al. (2010) reported the intrusion of desert air mass to be responsible for the drought occurrences over the central Indian region.
In general, most of the previous studies on monsoon droughts are discussed on the basis of rainfall accumulation, and there are very few which quantify its relation with the direct or indirect radiative effects of aerosols (Twomey, 1977) while considering both rainfall and PET. Absorbing aerosols such as black carbon (BC) or dust have the capability of atmospheric heating by absorbing solar radiation, while nonabsorbing aerosols (e.g., sulfates) scatter the solar radiation and have less of an effect over the same time span (Lau and Kim, 2006). Additionally, they have the capability of modulating the cloud characteristics by altering cloud radiative properties (Wencai et al., 2015). Previous studies have shown the presence of the aerosols (mainly dust and BC) and their ability to impact rainfall (depending upon their sizes) during Indian summer monsoon as described by the elevated heat pump hypothesis (Solmon et al., 2015). During late premonsoon or early monsoon season, the aerosol loading over India is nearly 3 times higher than the average due to the dust abundance, which is partly dependent upon the winds, precipitation and surface temperature (Dey, 2004). However, the opposite can also be true (e.g., Moorthy et al., 2007). Very recently some new attempts were undertaken to study the long- and short-term implications of both natural and anthropogenic components in producing several atmospheric processes in the boundary layer which produce a hindrance to convective rainfall, especially over urbanized coastal locations, which may also lead to subsequent drought occurrences (Chakraborty et al., 2017; Talukdar et al., 2017). Keeping all these assertions in mind, an effort is made to establish a possible relationship between aerosol loading and summer monsoon rainfall, and consequently over drought occurrences during this period in the past few decades.
A detailed investigation is presented on the evolution of the dry phase leading to drought conditions during mid-monsoon over three Indian regions based on the balance between precipitation and PET during the monsoon season. A new parameter called dry day frequency is used to understand the trends of drought potential over the mentioned Indian regions. This is followed by a three-pronged investigation to identify the most dominant factor behind these trends after which future projections of DDF is observed and explained for these locations during the mid-monsoon period.
Most of the research attempts in the recent past have employed the SPEI as an
indicator of drought occurrence over the Indian region (Beguería et
al., 2010). SPEI, which is precipitation minus PET, mainly represents the
climatic monthly water budget. Interestingly, this parameter is found to be
the most reliable identifier of drought occurrences as it can be expressed
in terms of standardized Gaussian variance with 0 mean and 1 standard
deviation. Another advantage of using SPEI over any other multiscalar
drought indicators (e.g., SPI) is that it not only includes the effect of the
evaporative demand in its calculation but also can be calculated for
different timescales (Beguería et al., 2010), unlike the Palmer Drought Severity Index (PDSI) which
relies on a water balance of a particular system. In this study the SPEI is
calculated using monthly precipitation and PET from the CRU TS3 dataset
(
Datasets of sunspot numbers are considered here as a reliable representative
of solar activity, which in turn may modulate the earth's hydrological
balance. There have been several scientific mentions in the past underlying
the effect of solar intensity on tropical rain and monsoon strengths both
over India and abroad (Agnihotri et al., 2002). Monthly averaged sunspot
numbers are obtained from the Solar Influences Data analysis Center (SIDC)
in the Royal Observatory of Belgium from the year 1749 till present (Cliver
et al., 2013). This study also considered the ENSO index, obtained from the
Oceanic Niño Index (ONI), which is calculated using a 3-month running mean
of Extended Reconstructed Sea Surface Temperature, Version 5 (ERSST.v5). Sea surface temperature (SST) anomalies in Niño 3.4 region (5
In addition to all general data quality issues, there is another potential issue regarding the discrepancy in data quality before and after the year 1999. To solve this issue, AOT datasets were taken in two clusters before and after 1999 and a detailed statistical analysis on the data revealed widespread overlapping which supports the continuity in data quality standards of aerosols for this analysis. Further details are provided in the Supplement Text 1 and Table S3.
In addition, the ERA-Interim reanalysis low-cloud-cover data are utilized
(
This study uses gridded population density (as a proxy of urbanization),
obtained from the Gridded Population of the World (GPWv4) and provided by the
CIESIN-SEDAC database from Columbia University for the years 2000, 2005, 2010
and 2015. This dataset is constructed by extrapolating the population data
from national or subnational administrative units all around the world. The
resolution of the product is 30 arcsec, or approximately 1 km at the
Equator; further details about the data can be obtained from the following link:
Considerable conditions for drought occurrences are identified on the basis
of the balance between monthly PET and rainfall accumulation during
June–September as depicted in Fig. 1. It is seen that due to arid
climates, northwestern India experiences higher values of PET, particularly
up to July, which may happen due to late arrival of monsoons at that location,
and hence this region may be considered for the analysis. On the other hand,
the southeastern peninsula of India experiences higher PET values, hence it
has been considered for further analysis. However, the rest of the country
experiences much lower values of PET. In contrast, precipitation values are
consistently lower both in northwestern India as well as the southeastern peninsula, so both these regions may face more probability to
experience negative DI, hence are selected for analysis. Another highlight
from the figure is that, the mid-section of the Indo-Gangetic Plain (IGP) depicts a sharp gradient of
precipitation. This diversity becomes more prominent during the months of
July–August as during this period, the entire IGP experiences very heavy
rain accumulation (> 300 mm on average) but the mid-IGP
experiences much lower rainfall
Monthly averaged maps of potential evapotranspiration rate and precipitation during June–September.
After the identification of the drought-prone regions, the main objective is
to determine a suitable parameter which best represents the probability of
droughts and which also can be related to other natural and anthropogenic
factors in all regions. Hence an assumption is made that if the temporal
distribution of rainfall is considered constant each month, then a drought
is only possible when both PET is high and precipitation is low. Now low
precipitation and high PET mainly arise from multiple dry day occurrences
in a month leading to droughts. For simplicity, during each of the four months
in the three seasons, the difference between precipitation and PET is calculated
over 115 years and the obtained data are normalized with respect to mean and
The correlation analysis depicts a set of reasonable correlation coefficients in both regions 1 and 3 over 115 years. Better correlation values are observed typically over July in region 3 and August in region 1, while it is lower in all other cases. It may be noted that regions situated in the western and northwestern parts of the country experience delayed monsoon (supported from many independent sources) which have led to high correlation values during June and July over region 3. However, region 1 especially shows a good correlation in August, a mid-monsoon month, which needs more attention in the following sections. Here, it may be noted from Fig. S4 that the agreement between DDF and DI is not strong in most of the cases other than three to four instances only. The reason is that the DI is dependent on the monthly accumulated difference between precipitations and PET, while DDF depends upon the erratic distribution of daily rainfall accumulation, hence the temporal scales of these two parameters are different from each other. The second reason is the presence of an independent factor called PET which depends upon various components (location, season, vegetation and soil type, temperature, moisture content, wind speed, surface pressure, and net radiation flux) but not on precipitation. Hence this explains the disagreement between these two parameters on a climatic scale.
Considering the last 60 years, correlation coefficients are improved in all regions and months as expected. Region 3 shows high correlations in July followed by August, while region 1 depicts comparatively much higher values during July and August. Thus the consideration of delayed monsoon onset may bring out more dry days in regions 1 and 3. But on the other hand, region 1 shows a high association between DI and dry days in August which demands investigations. Region 2 is mainly influenced by precipitation occurring during the late monsoon months, i.e., September, and not by the mature monsoon stage which is evident from the higher correlation values at that time. Hence this region may not fit with the scope of the present study. Additionally, higher correlation values are obtained during the 60 years span compared to the 115-year scale, hence DDF trends will be studied over the last 60 years span in the following sections.
It can be seen from the preceding sections that the correlation between DI
and dry days for region 1 is noticeable, but it is not highly prominent due
to the presence of many outliers in the scatter plots (Fig. S4). This is
because region 1 encompasses a total spatial coverage of 5
It has already been discussed that the drought intensity has significant correlation with DDF on a monthly basis. However, it is also necessary to investigate whether the intramonthly distribution of rainfall may also have its own impact in modulating the dry day frequencies especially during the mid-monsoon months which experience maximum precipitation variability. Hence, the monsoon months (JJAS) are now divided into eight equal slots of 15 d each and the 60-year time series for all these regions are obtained. The robust-fit trend analysis at 95 % confidence level is done to find the mean yearly trends, which is multiplied by 60 years and then normalized with respect to mean to generate a percentage-wise change in DDF.
The percentage changes are shown in Fig. 2c, which depicts an overall
increase in DDF for all regions with a few exceptions. Region 2 shows very
weak trends (< 5 %) all throughout monsoon; however, by the end of
September, a reasonable trend of
Region 3 shows minimal but alternating dry day frequency trends (<
On the other hand, region 1 shows very strong increasing trends in dry day frequencies with a similar pattern over 1a and 1b. Both these subregions experience relative wetting in late June, followed by a prolonged dry phase up to September. But the main difference between the two subregions is that the trends are consistently high all throughout region 1a with as much as 60 % and 20 % increases over August which also continues onto September; in region 1b, the trend values are comparatively lower (40 % and 5 %) during August. Thus, it can be inferred that although a clear increase in DDF is obtained all throughout region 1 during July–September, the trends are relatively stronger in region 1a, especially during August, hence it is given primary importance in this study.
In light of the previous sections, the probable influences behind the
increasing trends in dry day occurrences are investigated over region 1.
A number of natural or anthropogenic factors may be responsible for this
phenomenon. While natural factors mainly include the effect of solar
activity, ENSO oscillations or moisture tendencies, the anthropogenic
constituents mainly include aerosols which again encompass a lot of organic
and inorganic pollutants. To quantify the effect of aerosols, the aerosol
extinction coefficient values can be utilized from either satellite
observations (Multi-angle Imaging SpectroRadiometer, MISR) or from dedicated model simulations (MERRA-2). Since
observational datasets from MISR satellites are very sparse during monsoon
season and also the total measurement period is only 16 years, the MERRA-2
datasets are used for further analysis. Keeping the availability of AOD
datasets in mind, further analysis has to be concentrated on the 36-year time span
between 1980 and 2015. Owing to the prominence of DDF trends during the
month of August, further studies are concentrated on this period only. As
already mentioned, natural factors like solar activity and ENSO oscillations
(hereafter referred to as SSN and ENSO) may have some impact on precipitation
variability, which is also supported from previous attempts, and hence they are
considered. Additionally, moisture content also directly controls
precipitation and so their monthly means at 850 hPa (corresponding to
maximum moisture content during monsoon) are also utilized from the MERRA-2
reanalysis database. Some additional factors such as surface meteorological
conditions, circulation pattern and atmospheric thermodynamics also play
significant roles in controlling the occurrence of isolated but intense
convective precipitations, which also indirectly affect the dry day frequency
count. Hence parameters such as 2 m surface temperature, 850 hPa
geopotential and vertical totals index (difference between 850 and 500 hPa temperature) are also considered in this analysis. To understand the
dependence of these factors on DDF, the monthly DDF values during August,
1980–2015, are arranged in descending order and then the sorted dataset is
divided into three equal groups as short dry phase (SDP), corresponding to
normal conditions (8–10 d with average of 9); medium dry phase (MDP),
signifying near drought (10–14 d, with average of 12.5 d); and long
drought period (LDP), which represents full drought conditions (14–18 d,
average
Frequency distribution analysis results of various controlling
factors behind DDF evolution for various types of dry phase lengths over
region 1a, namely surface temperature, geopotential, VT, humidity, SSN,
ENSO, total aerosols, sulfates, sea salt, BC, dust PM
Classification of dry phase conditions according to its length for region 1a, Lucknow and region3.
As the dry phase length distribution fails to identify the dominant factor behind the rise of dry days in region 1a, hence all these four factors are passed through principal component analysis (PCA) test and the results are shown in Fig. S5a. The analysis produced a set of three orthogonal components out of which pc1 and pc2 account for 38 % and 19 %, respectively, of variances so we can neglect the contribution of the third component. Next, the corresponding variance scores of these components are plotted in Fig. S5, which shows that most of the parameters have very little variance according to both pc1 and pc2 axes, except aerosols, thereby indicating its increasing dominance over dry day evolution which can be validated using multilinear regression (MLR) analysis.
Further, multilinear regression analysis is done to see the independent
contribution of these four parameters to DDF. All datasets are normalized so
as to get uniform variability to enable easy identification of the
dominating factors. The MLR concludes that the coefficients for T2m, VT,
aerosol, SHUM,
MLR coefficients for all general factors affecting DDF for region 1a, Lucknow and region 3.
In view of the dominance of total aerosol AOT over DDF, the analysis is
concentrated on the datasets of various aerosol components over region
1a. Total columnar extinction values of five aerosol components, namely black
carbon (BC), dust PM
It was discussed in the previous sections that aerosols have a
dominating influence over dry day occurrences; however, it is yet to be
specified which types of aerosols (natural or anthropogenic, organic or
inorganic) are becoming major influencing factors for this phenomenon over
region 1a. Hence, time series datasets of these five components are again
taken for 36 years and are grouped with respect to the corresponding dry day
ranges as already explained in the previous section. After that, the
corresponding distributions are plotted in box plots in Fig. 3. The
distribution analysis depicts that the sea salts show some overlapping which
reduces the impact on DDF. Sulfates have quite high values all throughout
but their distribution exhibits a prominent overlapping so they cannot be
used here. Dust AOT values are low but its median shows a weak contribution
towards drying, but the overlapping in the distribution makes the association
very weak. Compared to others, BC and OC have shown a better association
with DDF along with reasonably increasing tendencies in medians and
quartiles. But this phenomenon also hints towards a dominant component of
pollution coming from certain highly urbanized sectors of region 1a such as
Lucknow, Allahabad (25.43
In the previous sections, when the magnitudes of BC and sulfate AOTs are compared then a question may arise as to how such small changes in BC have a dominant influence on DDF while sulfates have relatively no effect on it. Hence to have a double check on this fact, a statistical analysis is again performed on the AOT datasets, the details of which can be found in the Supplement Text 2 and Table S4. The analysis revealed that sulfate AOTs experience tremendous overlapping between the clusters which is about 2–3 times the actual increase in its cluster mean; however, the cluster mean increase in the case of BC is 1.5 times its SD, thereby explaining its net effectiveness in controlling the DDF.
The PCA results depicted in Fig. S5b show the contribution of pc1 alone is 60 % followed by pc2 of 25 % to be more prominent, hence there may not be a need to study pc3 here. From the scores, it is found that sulfate and dust behave similarly in their variances with high pc1 and low pc2 values, but OC and BC have both high pc1 and pc2 components so they may be found responsible for the variability in dry day changes. However, sea salt also may have some influence but it is not clearly understood from the figure.
MLR coefficients for aerosol components affecting DDF for region 1a, Lucknow and region 3.
To clarify any remaining misconceptions, the MLR coefficients are computed which gives the values as 0.542, 0.129, 0.263, 0.326 and 0.124 for BC, dust, OC, sea salt and sulfates, respectively (shown in Table 3). It is expected that the dust and sulfate have very low contributions so should be neglected. BC, OC and sea salt have higher values, of which OC and sea salt have comparable magnitudes, but sea salt has much low AOT values with a lower pc1 variance score and also reasonable distribution overlapping, so the effect of OC may be considered better. BC has a very high MLR coefficient with high pc1 score and also a clear variability in distributions. Hence, it may be concluded that owing to urbanization, the effect of BC followed by OC has a strong association with drought intensity and dry day occurrence.
From the previous section, it has surfaced that anthropogenic emission (of
BC and OC) as a result of urbanization may have a significant association
with the increase in DDF during August. To be definite about this, a
reinvestigation has been done over Lucknow (26.8
Frequency distribution analysis results of various controlling
factors behind DDF evolution for various types of dry phase lengths over
Lucknow corresponding to five aerosol components such as BC, dust PM
Figure S6 shows the distribution analysis of these components with PCA tests. The analysis reveals the presence of three strong principal components where pc1 is 60 % and pc2 is 30 %; hence pc3 is not considered further. When the variance scores for these parameters are plotted, then all factors show almost similar values of pc1 score, so pc2 becomes important. While judging the pc2 scores, we see that BC followed by OC have the best variability in this set, hence they may be considered for the dry day variation. To confirm this, multilinear regression is done on the components and the results yield values of 0.864, 0.218, 0.556, 0.0106 and 0.155 for BC, dust, OC, sea salt and sulfate (Table 3). According to previous results, the contribution of BC and OC is much higher than the others, with BC showing a higher correlation in all cases compared to OC, hence the dependence of dry days is found to be primarily associated with urbanization, more evidence of which will be produced in later sections. Dust follows this parameter but its dependence is comparatively much smaller than both BC and OC, which further supports these findings.
In the previous subsection, the effect of aerosols with BC in particular is
found to be strongly associated with low rainfall occurrence. However, the
effect of all meteorological parameters was not isolated in the previous
analysis. Also a time series analysis showing the impact of present AOD on
impending rainfall accumulation was not demonstrated earlier. Hence an
attempt has been made over Lucknow as it is an urbanized location in region
1a. To isolate the effect of various meteorological parameters such as
temperature, pressure, winds, moisture content and rainfall accumulation,
these datasets have been collected and then plotted in Fig. S7 for 16–30 July of 1980–2015. The long-term mean and
It is further required to see the effect of low rainfall periods and AOD on impending DDF for the next few days during these years. Hence a set of years having comparatively lower rainfall accumulation during 16–31 July were identified. A total of 16 years was recorded which had rainfall values between the 50th and 25th percentile of the population. It may be noted that certain years experienced rainfall below the 1st quartile and hence they were neglected to preserve the data uniformity. The average AOD values were accumulated for those years and interestingly two well separated clusters having a set of nonadjacent 8 years in each were observed: one with AOD below 0.3 and the other above 0.4. To study the effect of these two AOD clusters on rainfall, their corresponding DDF values are observed for the next 15 d (1–15 August). This time shift was employed in order to investigate the net effect of changing AOD on impending rainfall distributions. It was observed that DDF values are distinctly higher for high AOD compared to the lower AOD case. This supports the hypothesis that higher AOD necessarily leads to more DDF in the next few days.
In the previous section it was clarified that there is potential growth of aerosols in a particular region which takes part in certain complex atmospheric processes which sequentially leads to dry phase developments. However, this study also needs to be done for a larger region to test whether the same hypothesis is also valid over a widespread area. Hence, this test has now been done over region 1 and region 1a. In this case all similar steps are followed but now the AOD–DDF cluster relationship is shown side by side for regions 1, 1a and Lucknow together to understand whether localized urbanization inputs do really have any influence over DDF growth. The cluster analysis results from Fig. 5 shows almost similar clustering in DDF with respect to aerosols but the effect of AOD is seen to become more diffused as one shifts from a small urban region such as Lucknow (having more localized anthropogenic dominance) to region 1 (having lower urbanization density) and this is also well reflected from slightly higher DDF values over Lucknow.
Sequential association between AOD cluster (16–30 July) and DDF (1–15 August) for regions 1, 1a and Lucknow.
The preceding sections have given an idea of how urbanization is influencing
the evolution of dry day occurrences. To understand quantitatively its
climatic impact, the averaged DDF of the last 60 years are plotted for regions
1, 1a and Lucknow in order to examine the change in DDF patterns as one
downscales from a broad synoptic scale (IGP) to a small localized urban
location. Figure 6 reveals that region 1 has a weak but discernible increase
from 9 to 13 d in last 60 years. When robust-fit analysis was performed,
it was inferred that the net change in dry day frequencies over region 1 is
Statistical comparison of the climatology of all parameters during
August for region 1, region 1a, Lucknow during various time spans
It was reported earlier that an increase in anthropogenic aerosols may lead to
more cloud condensation nuclei, thus causing a reduction in cloud particle radius which may
result in a low occurrence of rain in spite of the increase in cloud cover.
From the previous section it is clear that dry day frequency exhibits a definite
increase in magnitude over region 1a and Lucknow. Since anthropogenic
components have shown the highest possible dominance on dry day occurrences,
an attempt is made to identify how cloud parameters like cloud cover have
changed with time over regions 1, 1a and Lucknow having different
urbanization growth and so on the anthropogenic components. In this study,
the main emphasis is given to low cloud cover only since aerosols have a
tendency to be limited to the lower atmosphere, especially in the monsoon
season. Region 1, which is covering a broad area, does not show a prominent
change in DDF and it is also observed that the change in cloud cover
over region 1 (
The long-term trends of dry day occurrences have exhibited a prominent
growth in dry days but the effect of these trends were found to be subdued to
some extent by several periodicities over the last 60 years in both region 1
and 1a. To understand their role on a quantitative scale, periodicity
analysis is done on the last 60 years using autocorrelation functions and the
results are depicted in Fig. S8. The autocorrelation function (ACF) values show highest value of 1
for a time lag 0, hence it is removed. Also, there is no use in understanding
periodicities greater than half of the period, hence the maximum period is
fixed to 30 years. The
In most of the preceding sections, the variability in DDF has been studied
over region 1 falling in the IGP. However, the northwestern part of the
country also comes under a high drought severity zone as already discussed,
hence this region is studied in detail. Figure 2 shows that the DDF
trend is comparatively higher during the month of July, hence DDF during
that month will be considered hereafter for further analysis over region 3.
It may be noted that the change is not so much prominent here as in region 1
(with a cumulative average of
Frequency distribution analysis results of various controlling
factors behind DDF evolution for various types of dry phase lengths over
region 3, namely surface temperature, geopotential, VT, humidity, SSN,
ENSO, total aerosols, sulfates, sea salt, BC, dust PM
A better insight into the interdependence of all these components is
investigated by the PCA test in Fig. S9a. The analysis reveals six PCA
components out of which two principle components (PCs) are considered to explain the complete range
of variances in dry days. The scores signify no definite pattern with the
total aerosol AOT assuming high pc1 and low pc2, followed by surface
temperature and ENSO having high contributions in only one of the two PCs. SHUM
falls in a completely different quadrant while the other parameters also show
an equally poor variance relationship and hence are neglected. Since aerosols have
a higher pc1 component which is comparatively stronger than other PCs, it
may be a deciding factor. To clarify this, MLR coefficients are calculated
which are around 0.178, 0.101, 0.241,
In view of the previous subsection, analysis is concentrated on the aerosols components over region 3. The distribution analysis of aerosol components is shown in Fig. 7, which depicts that, as usual, sea salt aerosols and sulfates have no role in modulating the DDF. It may be noted that here the magnitudes of sea salts and sulfates are higher than in region 1 or 1a, possibly due to its transport from the nearby seas which has not been washed away by rain in its path owing to the arid climate. However, they experience a very prominent overlapping between the components, which reduces the overall trend. The variation in OC is not clear and hence is removed. BC as usual has a deterministic variance with some overlapping, but still the whiskers and median values indicate its impact on dry days. Another important aspect here is that the range of values for these parameters is much lower here due to lower urbanization which still affects the DDF. But the contribution of dust aerosols emerges as the dominant component here as it not only shows higher values compared to all other regions but it also signifies a clear trend in the medians and distribution values. Thus it can be inferred that both dust and BC may contribute to this phenomenon.
To investigate which parameter has more dominance in dry day formation, PCA analysis is done on the individual components and the results are depicted in Fig. S9b. Here four principle component analyses (PCAs) are obtained, but the first two PCAs contribute 80 % of variability so the 2-D variance is seen. Also, the contribution of pc1 is comparable to pc2 so here both will be important. While analyzing the scores, it is observed that only dust and BC have both high pc1 and pc2, while most of the others have lower pc2 scores so they can be neglected. Further investigation is done on MLR analysis towards the trend contribution, which also gives similar outputs as 0.464, 0.431, 0.120, 0.182 and 0.033 for BC, dust, OC, sea salt and sulfate, respectively (Table 3). Again here both BC and dust emerge as potentially significant for the region 3 to be considered in association with the weak rise in dry days. Both of these components may have local sources but owing to their location there are possibilities of having an added amount of dust aerosol being transported from adjoining deserts, from dust storms, or from fumigation of dust from the ground during intense dryness, which are not found prominent over region 1a (where BC and OC were high due to high urbanization). Further, for more meticulous observation, cloud cover values have also been checked (Fig. 8) which show that the cloud cover has remained almost unchanged over the years unlike region 1a and Lucknow. This is again in good agreement with a less prominent increase in anthropogenic emissions or in short less of an increase in urbanization over region 3 compared to region 1a or Lucknow. This is further discussed in the following sections but a few things are important to mention here: the trend of dry days in region 3, though it is weaker compared to region 1a, may have a serious impact in the future as the region already experiences a high number of dry days itself, so a slight increasing trend is also alarming. Thus the effect of urbanization will still be an important parameter contributing towards the increase in BC and (some of) dust aerosol growth, and thereby leading to stronger trends in DDF over this region.
Statistical comparison of the climatology of all parameters during
July for region 3 during various time spans.
After the unclear dominance of dust followed by anthropogenic components was explained in the previous sections, now it is again necessary to check whether the aerosol growth in region 3 has really any sequential effect on the impending DDF growth, hence the same study is also repeated over region 3. The results from this analysis, shown in Fig. 8, indicate that the DDF values are much higher over this region due to the prevalence of a normal arid climate and not primarily due to aerosols, which is also understood from the widespread overlapping between the two clusters. Hence the definite relationship between aerosol growth and DDF cannot be firmly established and it may need more detailed analysis in future.
From the previous section, a strong association has been observed between
dry day frequency and anthropogenic emissions such as BC and OC, which in
turn is closely related to the urbanization growth. On the other hand,
high population density is also generally associated with the growth of
urbanization and hence it may be taken as a suitable proxy for the latter in
this study. The population density values were taken from the gridded
1
Region-wise population densities and BC AOT values (during August)
for regions 1, 1a, Lucknow and region 3 during (2000–2015); vertical bars
represent the corresponding
The primary distribution of population for the year 2000 is shown in Fig. S10, which depicts more values for region 1a compared to region 1b, and Lucknow is still found as a patch of very high population even during 2000. On the other hand, region 3 had much lower populations at the same time. Next, the long-term variation in population density is again observed over regions 1, 1a and Lucknow from Fig. 9. It may be noted from the figure that all throughout region 1 population density rises from 650 to 800 persons per square kilometer, which is quite a high value. Region 1a shows even higher values than region 1 with a steep rise from 760 to 1000 persons per square kilometer. Thus it follows that region 1a has consistently higher population average and trends leading to higher OC and BC. However, the situation worsens in Lucknow where population density changes drastically from 850 to 1100 persons per square kilometer with most of the change happening in the last 10 years, hence this phenomenon strongly supports the amplified DDF trend over Lucknow compared to 1a. But region 3 shows very little variations in last few years (100 to 140), which may have led to the comparatively lower BC and OC emissions. However, it may also be noted that the relative change over region 3 is higher (40 %) compared to Lucknow (30 %). Hence in future, if urbanization and population persist to grow at the same rate over region 3, then BC, OC and dust will also expectedly grow to alarming limits which can cause a drastic change in DDF over northwestern Indian regions.
The next concern of this study is to investigate the projected change of dry
phase lengths over the foreseeable future. Many attempts in recent years
have employed CMIP5 general circulation model (GCM) simulations to provide future projections for any
urbanization scenario. In accordance with the present study, RCP 8.5
projections of rainfall (and DDF) corresponding to maximum urbanization
levels have been considered over the mentioned regions. It may be noted that
in the last 60 years itself, DDF values have reached
The total variation in dry days are investigated over regions 1 and 3
including both historical and CMIP5 RCP 8.5 projections data to get a 150-year trend of dry day frequencies in Fig. 10a. The DDF for all 29 grid
points in region 1 and 20 grid points over region 3 are averaged yearly and
then depicted in Figs. 5 and 7. The multimodel mean data show that even
when averaged spatially, dry days show a clear increase from
Hence region 1 creates a more alarming situation with dry days increasing by around 5 times compared to the other regions. To further investigate this abrupt change spatially, the model averaged data of DDF for the 50-year span are shown for region 1 in Fig. 10b. As expected, the figure shows a high value around Lucknow for the 50-year periods but its effect diffuses as one goes towards the outskirts of Lucknow facing lesser urbanization. Places adjoining Lucknow show a very drastic change only after 2010. Thus, most of the places adjoining Lucknow show a very high number of dry days (> 45 d) near the end of this century, which will grossly affect the monsoonal rainfall leading to severe droughts and so it needs to be addressed by policymakers.
It is an essential aspect to study the probability of drought occurrences
over India during monsoon as agricultural and economical issues are directly
related to it. In the present study, a detailed analysis on the occurrence
of dry days during monsoon over the Indian region is presented. In this
study, three potentially drought-prone regions in India based on the dearth
of precipitation and abundance of PET is considered. Region 1 mostly belongs
to the State of Uttar Pradesh (UP), region 2 covers major parts of the
states of Andhra Pradesh and Tamil Nadu and a small portion of Karnataka, and
region 3 encompass the arid part of Rajasthan. Detailed investigations
revealed that over the eastern part of region 1, which is referred to as region
1a, urbanization plays a significant role in increasing DDF. Prevailing impact
of anthropogenic emission like BC or OC aerosols becomes more prominent as
the study goes in depth with a downscaling approach from a broad region 1 to
a specific urbanized location like Lucknow, which is one of the urbanized
sectors of IGP. The increase in cloud cover and nonoccurrence of rain
events indicate rain suppression phenomena over region 1, which is yet to be
investigated in detail. This also indicates the scope of the study over
several other point locations having drought occurrence record but could not
be included in the present study. Finally, the long-term projections of DDF
are drawn over regions 1a and 3 using the intense urbanization scenario of RCP
8.5 and an average of 70 % rise in dry days are seen, which may be a very
crucial concern by the year 2100 and hence it needs to be considered by
policymakers in future aspects. However, this study is mainly done from
modeled components of aerosols, so a far more accurate analysis can later
be done over the IGP subject to more availability of aerosol in situ data in the
other major urban locations over India. The main findings of the study are
shown in a schematic presentation in Fig. S12 and are highlighted as
follows:
The DDF (based on the frequency of days having local precipitation
accumulation less than 1 mm) has a significant level of correlation with the
universally accepted monthly SPEI drought index (DI) especially in the last
60 years. Further, the correlation levels between DI and DDF are more
prominent during August in region 1a and during July in region 3. The trends of DDF (within a 15 d window) are more prominent during August
for region 1a. However, region 3 shows a descent trend during July, while
region 2 shows the same during late September (corresponding to the monsoon
retreating phase), hence it has been neglected as it may not completely
reflect a monsoonal drought. Results from region 1a indicate prevailing contribution of aerosols compared
to ENSO, humidity, surface meteorology, circulation instability or SSN. Our
study shows that BC and OC aerosols over urbanized region are more active at
increasing the DDF, which is also supported from distribution, PCA and MLR
analyses. The trend analysis on DDF reveals that the increasing trends become stronger
as the spatial coverage is downscaled from region1 to 1a and followed by a
local urbanized location of Lucknow. About 50 % increase in DDF is found
in Lucknow compared to 17 % all through region 1. Further, a periodicity
of 4 and 8 years is found to be stronger in region 1, which is overpowered by the
randomly varying urbanization component over Lucknow. The sequential association analysis between aerosols and DDF reveals that
aerosol growth in the period of 16–30 July over region 1 has a direct impact on
DDF developments during the next 15 d time span. However, the relationship is
slightly more definite for localized urbanized areas like Lucknow, having a
more anthropogenic aerosol dominance. Population density maps have been taken as a proxy of the urbanization
component owing to its significant agreement with anthropogenic carbonaceous
emissions (BC). A higher population density is observed over Lucknow
(average of 850 persons per square kilometer and trends of In-depth investigation revealed that the increase in urbanization components
like BC or OC exhibits a significant association with increased cloud
lifetime ( Though in region 3 aridity plays a major role to experience a high number of
dry days ( The climatic projections of dry day frequency from CMIP5 simulations of three
GCM models (CNRM CM5, CAN ESM and NorESM 1M) show a sharp increase in dry
days during 15 July to 15 September with DDF reaching up to 50 dry days over
region 1 and 45 dry days over region 3 by 2100.
Daily rainfall data used in the present study were obtained from the
National Climate Centre, India Meteorological Department
(
The supplement related to this article is available online at:
RC performed the main analysis in the study. BKG, ST, MVR and AM provided the initial concept, main guidance, needed data and also contributed to the analysis, discussion and editing.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Interactions between aerosols and the South West Asian monsoon”. It is not associated with a conference.
One of the authors (Rohit Chakraborty) thanks the Science and Engineering Research Board, Department of Science and Technology, for providing fellowship under a National Post Doctoral Fellowship scheme (grant no. PDF/2016/001939). He also acknowledges the National Atmospheric Research Laboratory for providing necessary support and data for this work. The authors also thank Soumyajyoti Jana from Calcutta University for his suggestions.
This paper was edited by B. V. Krishna Murthy and reviewed by three anonymous referees.