1 Introduction

Urbanization and regional climate change are major global environmental concerns (Solecki et al., 2013) and the two are interlinked (Pielke Sr et al., 2011). Urbanization manifests through the increase in the urban population and the expansion of impervious surface area. Since 1900, the proportion of the global urban population has increased from 13% to more than 50% (Seto & David, 2010). Urban sprawl alters local and regional meteorology- humidity, wind speed, temperature, and planetary boundary layer height (Arnfield, 2003). Urbanization also modifies the exchange of heat, moisture, aerosols, and momentum between the land surface and the overlying atmosphere. Consequently, urbanization directly influences the changing weather patterns over and around cities and impacts the community and economy (National Research Council, 2010).

In a recent meta-analysis review of urbanization studies, Liu and Niyogi (2019) conclude that there is an ongoing need for observational assessments related to the impact of urban landscapes on different weather parameters. Using a combination of modeling and observational datasets, studies have concluded that the changes in rainfall over a city are associated with an increase in urbanization of the city (e.g. Buishand, 1979; Shepherd et al., 2002; Burian & Shepherd, 2005; Chen et al., 2007; Liang et al., 2011; Zhang et al., 2019; Lorenz et al., 2019; Liu & Niyogi, 2019; Niyogi et al., 2020). The urban-rural heterogeneity strengthens mesoscale convection. Urbanization may suppress rainfall over the city and create conditions favorable for downward enhancement (Niyogi et al., 2011; Zhang et al., 2019; Wen et al., 2020). With the increase in built-up areas, urbanization generally modifies the regional atmospheric circulations, which results in more intense rainfall (Kumar et al. 2013). Studies such as Yang et al. (2017) report that short-lived extreme rainfall events are strongly associated with urbanization. Studies (Siswanto et al., 2016; Niyogi et al., 2020) also suggest a possible delay in late afternoon rainfall due to urbanization. Analysis of spatio-temporal variation of rainfall and urbanization reveals that urban areas with the highest population areas are dominated by the slightly longer rainfall duration and higher rainfall intensity (Zhu et al., 2019). There is also a significant correlation between the spatial distribution of heavy rainfall amount and an increase in urbanization over India (Kishtawal et al., 2010), the United States (Niyogi et al., 2017; Singh et al., 2020), and China (Zhi-Hong et al., 2016, and Shi et al., 2017).

Over the Indian monsoon region, Bisht et al. (2018) concluded that there is an increasing trend in pre-monsoon season rainfall over most parts of India during the post-urbanization era (1971–2015) compared to during the pre-urbanization era (1901–1970). Changes in urbanization are likely to significantly alter Indian Monsoon Rainfall (IMR) and cause extreme rainfall over the various urban regions (Kishtawal et al., 2010; Dimri and Niyogi 2013; and Singh et al., 2016). This study is focused on the rapidly urbanizing Indian city of Bhubaneswar, the capital of Odisha state along the Bay of Bengal. The city is designated as the first ‘smart city’ being developed in India (Bhattacharya et al., 2015). The remarkable urbanization rate experienced for Bhubaneswar has been studied in Pathy and Panda (2012), and the impact on temperature has been reported in Swain et al. (2017), Barik et al. (2019), and Gogoi et al. (2019). The city has also experienced several heavy rainfall events that, according to long-term residents, appear to be on the rise.

The study’s main goal is to determine whether the rainfall in Bhubaneswar, which is quickly urbanizing, has changed and, if so, what the characteristics of that change are. The overarching goal is to determine whether urbanization change and rainfall variation are connected. Pre-monsoon (March–April–May: MAM), monsoon (Jun–Jul–Aug–Sept: JJAS), post-monsoon (Oct–Nov–Dec: OND), and winter (Jan – Feb: JF) are some of the seasons with interest in terms of rainfall characteristics. The investigation is driven by its possible effects on managing flash floods, mudflows, and waterlogging in metropolitan areas.

2 Data and methods

This section describes the datasets and methodology used to carry out the present research work.

2.1 Study area

Bhubaneswar (centered at 85.84° E and 20.27° N and 45 m above sea level, is a prominent city on the eastern coast of India Fig. 1(a)). Bhubaneswar is located in Khordha district, surrounded by Cuttack on its north, Nayagarh on the west, Puri on its south, and Jagatsinghpur on the east. The extended map of the city with surface relief is shown in Fig. 1(b). The color maps show that there are about 21 different land use land cover (LULC) classes. Bhubaneswar is one of the fastest-growing urban areas in the country, and a designated ‘smart city’. The city has also experienced frequent natural hazards such as heatwaves, tropical cyclones, and flooding (Mohanty et al., 2013; Panda et al., 2014; Gupta, 2020). The city experiences a humid sub-tropical climate with World Meteorological Organization (WMO) code 42971. Climatologically, the average annual temperature is 27.4 °C, and annual rainfall is about 1505 mm. The maximum temperature is observed during pre-monsoon season (MAM), and maximum precipitation is observed during the July & August month of the year over the city (Pandey et al., 2014). The city has grown from a population of 8170 in 1921 to 881,988 in 2011 (Census of India, Odisha State, 2011). It has been projected for 2021 year and which is about 2150, 000 (Thandar, 2012). Bhubaneswar city area is characterized by undulating upland topography in western and central part while the eastern part is relatively flat with gentle slope. The main drainage channel is the Kuakhai River which is a distributary of the Mahanadi River. The soils are mainly Alfisols and Ultisols (Choudhury et al., 2011).

Fig. 1
figure 1

Study area showing (a) the east coast state of Odisha and the location of the capital city Bhubaneswar; b municipal boundary of the city, with the numbers indicating the landmarks across Bhubaneswar city (names are in the table shown left side to the image) discussed in the text

Full size image

2.2 Data used

The spatio-temporal characteristics of rainfall during four different seasons over Bhubaneswar station are analyzed using quality-controlled hourly rainfall data obtained from the National Data Centre (NDC) of India Meteorological Department (IMD), Pune. For the 39 years (1980–2018) of the study period, there are 49 days of missing data for Bhubaneswar. The gridded daily rainfall analysis obtained from the IMD (Pai et al., 2014) at a horizontal resolution of 0.25° × 0.25° is used in addition to the station data. The study period is confined to the recent era of 39 years (1980–2018) since more automated weather station data was used to generate the gridded rainfall analysis using the WMO standard quality control method. To supplement the IMD observations, three-hourly rainfall data at a horizontal resolution of 0.25° × 0.25° for the 1999–2018 year (depending on the availability of data) from Tropical Rainfall Measuring Mission (TRMM) has been used (Huffman et al., 2015). To assess the urbanization changes, cloud-free images from United States Geological Survey (USGS) Landsat satellite for 1980, 1990, 2000, 2010, and 2019 were acquired (https://landsat.gsfc.nasa.gov). These were used for supervised classification, and a summary is shown in Table 1 (Barsi et al., 2007; Dewan & Yamaguchi, 2009). Population density data for Bhubaneswar city has been taken from Socio-Economic Data and Applications Center (SEDAC) for 2000, 2010, and 2020 year (Center for International Earth Science Information Network (CIESIN), Columbia University, 2018).

Table 1 Details of Landsat Satellite data processed for the urban land use/cover analysis
Full size table

2.3 Methodology used

Rainfall for different seasons over Bhubaneswar has been studied by using both hourly and yearly trend analysis. In addition, both parametric and non-parametric statistical methods have been used to detect trends and other changes in climate variables and are discussed in the following section (Suhaila & Jemain, 2009, and Jha & Singh, 2013). Methodology used in this recent study has been shown in Fig. 2. To investigate trend yearly and hourly rainfall over Bhubaneswar both Continuous Wavelet Transform (CWT) technique and the Mann-Kendall trend tests have been carried out. First, the possible existence of significant autocorrelations in the used datasets have been checked.

Fig. 2
figure 2

Flowchart of methodology used

Full size image

2.3.1 Mann-Kendall significant test

The non-parametric Mann-Kendall (MK) method has been widely used in recent years for climatic time series analysis (Mann, 1945; Kendall, 1948; Yue et al., 2003; Singh et al., 2008). In this study, the MK test has been used to verify the confidence in the trend analysis. This was calculated using MATLAB 2017a. The advantages of using the MK test are that (i) it is distribution independent and suitable for data with missing or censoring values, (ii) it has high asymptotic efficiency (Berryman et al., 1988; Fu et al., 2004; Modarres & Sarhadi, 2009), and (iii) the test is not affected by gross data errors and outliers. The p-value of the test generally defines the significance of the trend. Lesser than threshold p-values suggest rejection of the null hypothesis.

The significant test has been applied to an annual maximum rainfall time series xi ranked from i = 1…., n – 1, and xj is ranked from j = i + 1, …., n. Each data point xi is used as a reference point and is compared with all other data point xj such that

$$\operatorname{sgn}\ \left(\textrm{x}\right)=\left\{\begin{array}{cc}1,& \textrm{when}\ {\textrm{x}}_{\textrm{j}}>{\textrm{x}}_{\textrm{i}}\\ {}0,& \textrm{when}\ {\textrm{x}}_{\textrm{j}}={\textrm{x}}_{\textrm{i}}\\ {}-1& \textrm{when}\ {\textrm{x}}_{\textrm{j}}<{\textrm{x}}_{\textrm{i}}\end{array}\right.$$
(1)

Kendall statistics (S) is defined as follows,

$$\textrm{S}=\sum\nolimits_{\textrm{i}=1}^{\textrm{N}-1}\sum\nolimits_{\textrm{j}=\textrm{i}+1}^{\textrm{N}}\operatorname{sgn}\left({\textrm{x}}_{\textrm{j}}-{\textrm{x}}_{\textrm{i}}\right)$$
(2)

Where “N” represents the length of the data series.

The variance is defined as,

$$\textrm{Var}\left(\textrm{S}\right)=\frac{1}{18\ }\ \left\{\ \textrm{n}\left(\textrm{n}-1\right)\left(2\textrm{n}+5\right)-\sum\nolimits_{\textrm{i}=1}^{\textrm{g}}{\textrm{t}}_{\textrm{i}}\ \left({\textrm{t}}_{\textrm{i}}-1\right)\left(2{\textrm{t}}_{\textrm{i}}+5\right)\right\}$$
(3)

Where ‘ti’ indicates the extend of ith tied groups (a set of sample data with similar values) number and ‘g’ denotes the tied groups number.

The value of (S) and Var (S) has been calculated to develop standardized test measurement of ‘Z’ and which is as follows.

$$Z=\left\{\begin{array}{cc}\frac{S-1}{\sqrt{\mathit{\operatorname{var}}(S)}},& when\ S>0\\ {}0,& when\ S=0\\ {}\frac{S+1}{\sqrt{\mathit{\operatorname{var}}(S)}},& when\ S<0\end{array}\right.$$
(4)

Positive values of Z-statistics depict increasing, while the negative values indicate a decreasing trend in the time series. Statistics of MK test is based on the ‘+’ or ‘-’ signs; so the trends obtained from MK test are less affected by the outliers and the non-parametric MK test is beneficial for such data (Helsel & Hirsch, 1992, and Birsan et al., 2005).

2.3.2 Wavelet analysis

The wavelet transform (WT) is the other statistical approach used in this study. It is a mathematical technique for analyzing the data in a broad spectrum of time-frequency. A WT mathematically decompose a signal into multiple lower resolution levels. Among several mathematical transformations, Fourier transforms are the most popular (Santos et al., 2003). This analysis simultaneously provides both a time and frequency view of one-dimensional time series. The amplitude and time variation of such amplitude of any time series data can be obtained (Torrence & Compo, 1998). Wavelet analysis is a widely used method to detect different parameters periodic variation on different time scales (Weng & Lau, 1994, and Ling et al., 2012). The wavelet transforms analysis is effective in analyzing non-stationary time series and capturing useful information from it (Wei et al., 2012, and Adarsh & Reddy, 2015). The continuous wavelet transform shows a time-scale representation of the signal and helps analyze long-period datasets (Meyers et al., 1993; Wang & Wang, 1996).

Early applications of wavelet analysis include ocean waves, El Nino-Southern Oscillation, and turbulence (Wang & Wang, 1996). However, wavelet aided methods have been widely used in precipitation data to detect trends (Partal & Küçük, 2006; Sang, 2013; Shiau & Chiu, 2019; Pal et al., 2019). In recent years, wavelet analysis has been an effective approach for studying extreme precipitation (Kumar, 1996; Wang et al., 2013; Tan et al., 2016; Rathinasamy et al., 2019).

For a wavelet function ‘ φ ’, the time-scale continuous wavelet transform (CWT) is defined as (Daubechies, 1990),

$$\textrm{W}\left(\textrm{a},\uptau \right)=\frac{1}{\surd \mid \textrm{a}\mid }{\int}_{-\infty}^{\infty}\textrm{f}\left(\textrm{t}\right)\upvarphi \left(\frac{\textrm{t}-\uptau}{\textrm{a}}\right)\textrm{dt}$$
(5)

Where ‘t’ is the time, ‘a’ and ‘ τ ’are the scale and location parameters. The CWT equation represents the signal (t); where the original signal can be reconstructed with the minimum of W(a, τ). The cone of influence (COI) is the area of the wavelet power spectrum where edge effects cannot be ignored (Gu et al., 2020).

2.4 Land cover classification

For developing the land cover change assessment, cloud – free imageries for Bhubaneswar city have been acquired from United States Geological Survey (USGS) Landsat satellite datasets. The sampling corresponding to five different years: 1980, 1990, 2000, 2010, and 2019 (Table 1) was analyzed. ArcGIS 10.2 has been used to digitalize the image and create a GIS database. Initially, Landsat imageries were formatted to the World Geodetic System (WGS84) reference datum and zone 43 north projection of Universal Transverse Mercator (UTM) system. Three follow up steps are involved for satellite image interpretation such as image pre-processing, processing, and post-processing, and were undertaken next.

The flowchart of the methodology followed for the generation of built-up area maps is presented in the Fig. 3. Preprocessing of image includes geometric corrections, radiometric calibration of the image data and filtering of noise present in the data. As a part of image processing, false color composite (FCC) images have been created using red, green, and near infrared channels.

Fig. 3
figure 3

Flowchart of methodology used for LULC generation

Full size image

Red areas present in the FCC have been interpreted as vegetation, black color areas as deep water, white color as barren land, and cyan color as urban built-up area, respectively. Variation of color tones for urban area vary with density and material use. Multispectral images were captured by stacking several bands of the image in FCC for each year separately. For 1990, 2000, 2010, and 2019, two Landsat scenes from paths 139 and 140 have been mosaiced to visualize the image as per landuse using ArcGIS. Row refers to the latitudinal center line of a frame of the imagery taken by the satellite. Satellite moves along its path, and the instruments present in the satellite scans the earth surface.

A supervised classification methodology (Jain et al., 2016; Swain et al., 2017; Barik et al., 2019) is used to produce LULC maps for Bhubaneswar city. The raster image is then classified into eight classes (urban built up, water, wetland, barren land, broadleaf deciduous forest, mixed forest, shrub/grass, and irrigated cropland). In supervised classification, the user develops the spectral signatures for different land use categories and then the ArcGIS software help assign the other pixels of the same signature into one class (Eastman, 2003). This process is frequently used in remote sensing for quantile analysis of datasets. In supervised classification, the first step is to define the training site i.e., area of interest. After the selection of training site, several pixels have been assigned for different land use categories. Then the supervised classification has been applied over the assigned land use categories.

3 Results

The following section analyzes the spatiotemporal characteristics of rainfall for different seasons from 1980 to 2018 over Bhubaneswar city and its surroundings.

3.1 Temporal analysis of rainfall over Bhubaneswar

The wavelet coefficients obtained from CWT for precipitation at Bhubaneswar for the 1980–2018 period, seasonally. White contour lines enclose statistically significant wavelet power regions at a 95% confidence level in the time-frequency space (Fig. 4(a (i-iv))). The calculated wavelet power presented below the solid black line (COI) in Fig. 4(a) is considered unreliable due to its edge effects.

Fig. 4
figure 4

A Wavelet transforms of monthly rainfall for (i) Jan – Feb, (ii) Mar-Apr-May, (iii) Jun-Jul-Aug-Sept, and (iv) Oct-Nov-Dec. The black line represents the cone-of-influence and white contours represent the 95% significance level; B Annual trend of seasonal maximum rainfall over Bhubaneswar from 1980 to 2018 for (v) Jan – Feb, (vi) Mar-Apr-May, (vii) Jun-Jul-Aug-Sept, and (viii) Oct-Nov-Dec.

Full size image

For the winter and post-monsoon season, significant inter-annual oscillations with periodicity as 1–130 years occurred from 1997 to 2003 over Bhubaneswar. Whereas, in pre-monsoon period significant inter-annual oscillations are within a more constrained 1–12 years’ period, and more notably in the 1990s to 2000s period. For the monsoon season, inter-annual oscillations are more pronounced at a 2–6 years’ period. In other words, the monsoonal changes are most notable, followed by the pre-monsoon period. The winter and post-monsoon rainfall features do not show any notable change. Analysis of annual variation of seasonal maximum rainfall is shown in Fig. 2(b), which indicates an increasing trend in maximum seasonal rainfall for all seasons except for post-monsoon season across Bhubaneswar. The maximum seasonal rainfall showed an increase of 15.6 mm over 39 years for pre-monsoon season (0.4 mm per day in a season or year). In other studies, it is also analyzed that there is an increase in rainfall from last decade over the city (Li et al., 2020; Jiang et al., 2020). The rainfall in winter showed nearly half the increase as the pre-monsoon rains, while the monsoon rainfall showed even more muted change with only about 1 mm per decade. In post-monsoon season, the maximum seasonal rainfall showed a decrease of about 0.5 mm per decade. The MK significant test has been carried out for all the trend analysis. The p-value for these trends is 0.19, 0.15, 0.2, and 0.16 for winter, pre-monsoon, monsoon, and post-monsoon season are statically significant at 81%, 85%, 80%, and 84% significant level respectively.

Hourly rainfall across Bhubaneswar is analyzed next. The study period has been divided into two parts to examine how the peak rainfall hour changes with time. It is also observed from previous studies for different urbanized cities (Zhu et al., 2019; Yuan et al., 2021). Wavelet transforms for 1980–1999 (first section) and 2000–2018 (second section) for different seasonal rainfall has been shown in Fig. 5(A) and (B), respectively. The analysis reveals a gradual but notable shift in the timing of the maximum rainfall. In the initial data period, maximum rainfall generally occurs in the morning, i.e., around 8 am to 10 am. Whereas, during the second, it occurs during the afternoon, i.e., around 4 pm to 8 pm. This is more evident for pre-monsoon season. The initial rainfall data shows a noisy frequency, while the variability was more organized with a 1–3 h periodicity. For winter season, maximum rainfall occurs during early morning, i.e. from 4 am to 10 am in the first section and in the second section it is around 3 pm to 5 pm (it is statistically insignificant). For the monsoon season, there is no detectable change in rainfall peak timing but the intensity showed increase in the second half of the data period. In the post-monsoon season, maximum rainfall occurs around or after 5 pm in the early half of the data period, and the rainfall occurs earlier i.e., around 11 am to 4 pm in the second section.

Fig. 5
figure 5

A Wavelet transform of hourly rainfall during 1980–1999 time period for (i) Jan – Feb, (ii) Mar-Apr-May, (iii) Jun-Jul-Aug-Sept, and (iv) Oct-Nov-Dec; B is same as A but during 2000–2018 time period. The black line represents the cone-of-influence and white contours represent the 95% significance level; C Variation of hourly rainfall anomaly for 1980–2018 for (ix) Jan – Feb, (x) Mar-Apr-May, (xi) Jun-Jul-Aug-Sept, and (xii) Oct-Nov-Dec over Bhubaneswar

Full size image

The hourly anomaly trend for 1980–2018 seasonal rainfall, and the analysis suggests an overall increase in rainfall intensity from midnight to late-evening over Bhubaneswar (Fig. 5(C)). It is also observed that maximum rainfall occurs from 5 pm to 8 pm during pre-monsoon season, and from 5 pm to 7 pm, from 6 pm to 8 pm, and around 6 pm during winter, monsoon, and post-monsoon seasons, respectively. This change in the rainfall timing is consistent for results from another urban center (Hyderabad, which is about 1000 km southwest) with similar urban sprawl in recent decades (Niyogi et al., 2020). The similarity of findings for Bhubaneswar, which has much more rapid urbanization and a dramatic change in the timing, is instructive. In that, the difference (or change) in rainfall timing over the city appears to be another signature of urban modification of rainfall – the first being the spatial change reported in the literature (Shepherd, 2005; Ntelekos et al., 2010; Routray et al, 2010; Liu & Niyogi, 2019).

3.2 Spatial analysis of rainfall over Bhubaneswar

Climatological mean rainfall during different seasons indicates that Bhubaneswar receives about 40–45 mm rains during pre-monsoon season which is maximum compared to other seasonal rainfall (Fig. 6(A)). Maximum rainfall occurs typically along the east side of the urbanized city for all seasons. The thunderstorm-centric east side of the city gets more rainfall (about 60 mm) than the west side of the city (about 30 mm) for pre-monsoon season. The distance of the rainfall maxima is approximately 40–50 km from the center of the city. Thus, there is a gradation in rainfall intensity from the west to the east side of the city, and climatologically, the rainfall difference is about 30 mm. That is, a factor of two increase in the rainfall east side of the city as compared to the west side is found. The rainfall maxima occur at ~ 50 km from the city center, which conforms to the summary of urban rainfall features reported in Liu and Niyogi (2019). Huff, 1975 did experiments in the St. Louis area, and they found that highest rain cell occurs over the urban region which is east side of the city. Further, the same has been confirmed in the recent study (Liu & Niyogi, 2019), that the east side of the storm is more impacted due to the anti-cyclonic circulation of low-pressure system over the city.

Fig. 6
figure 6

Spatial distribution of A seasonal mean rainfall (black colour figure represents Bhubaneswar city boundary) for (i) Jan – Feb, (ii) Mar-Apr-May, (iii) Jun-Jul-Aug-Sept, and (iv) Oct-Nov-Dec; B is same as A but for trend of seasonal rainfall (black colour figure represents Bhubaneswar city boundary and black “+” shows the grid points having stastically significant rainfall)

Full size image

Trend analysis of pre-monsoonal rainfall over Bhubaneswar has been carried out with IMD gridded data and presented in Fig. 6(B). Results show that the rain over the city and along the east side of the city shows an increasing trend compared to the rainfall trend along the city’s west side. But, during winter and post-monsoon season, rainfall along the west side of the city is showing an increasing trend compared to east side of the city. It is highlighted that the trends are relatively small, and the results should be considered as tentative.

For analyzing the spatiotemporal rainfall variation, 3-hourly rainfall data for 20-year period (1999–2018) using TRMM fields is presented in the supplementary Figs. 1–4. Considering the rapid urbanization rate noted for Bhubaneswar, the study period is divided into four epochs of 5-year intervals: (i) 1999–2003, (ii) 2004–2008, (iii) 2009–2013, and (iv) 2014–2018. During winter, the maximum rainfall is during 17.30 IST in the first interval, whereas more rainfall is occurring during 14.30 IST in recent times, i.e., the fourth interval. During pre-monsoon season, maximum rainfall is received around afternoon 14:30 IST in the first interval and during evening 17:30 IST in the second. The rainfall from past midnight (01.30 IST) to early morning (08.30 IST) has decreased from 1999 to 2018. In monsoon season, afternoon rainfall typically around 14:30 IST has decreased while the early evening 17:30 IST rainfall has increased over the time period. During post-monsoon season, maximum rainfall has occurred around 11:30 IST in the first interval and around 17:30 IST in second interval. This is summarized from the 3-hourly trend analysis of rainfall over Bhubaneswar (Fig. 7 (I-IV)). This hourly changes of rainfall in a day has also been found from Siswanto et al., 2016.

Fig. 7
figure 7

Trend of 3-hourly rainfall during 1999–2018 year for (a) 00 UTC, (b) 03 UTC, (c) 06 UTC, (d) 09 UTC, (e) 12 UTC, (f) 15 UTC, (g)18 UTC, and (h) 21 UTC for (i) Jan – Feb, (ii) Mar-Apr-May, (iii) Jun-Jul-Aug-Sept, and (iv)Oct-Nov-Dec from TRMM gridded data (black colour figure represents Bhubaneswar city boundary and black “+” shows the grid points having stastically significant rainfall)

Full size image

3.3 Changes in LULC over Bhubaneswar

The changes in the urbanized area from 1980 to 2019 over Bhubaneswar are examined (Fig. 8). The aim is to broadly assess the relation between urban sprawl and rainfall changes. Supervised classification of LULC map using ArcGIS leads to the quantification of change in LULC (Fig. 8(a)). The changes in the urban built-up area are shown in Fig. 8(b), which shows a dramatic increase in urban built-up area from 1980 to 2019. The urbanized area is concentrated over the central part of the city in 1980s, compared to the significant sprawl noted in the 2019 imagery. It is important to note that the urban growth noted over Bhubaneswar is possibly emblematic of the changes occurring across the different emerging cities of the world (Wang et al., 2006, Li et al., 2013, and Bechtel et al. 2017).

Fig. 8
figure 8

a Classified LULC map for the study area, b spatial distribution of urban built-up area over the study area for 1980, 1990, 2000, 2010, and 2019 year, c Decadal variation of urban area (km2) and population density across Bhubaneswar, and d Spatial distribution of population density over Bhubaneswar for 2000, 2010, and 2020 year

Full size image

Then the population density for Bhubaneswar city has been plotted from 2000 (Fig. 8(c)). It shows an increasing trend in the urban built-up area, and population density over the city. Similar results have also been observed for Bhubaneswar city from 2003 to 2017 in Barik et al. (2019). The urban area is about 22.5 km2 in 1980 and shows a more than five-fold increase to 122 km2 in 2019. Thus, there is a change in about 100 km2 in the last 39 years of the period. Spatial distribution of population density over Bhubaneswar suggests that the population density is increasing over the city and more is over east side of the city (Fig. 8(d)). It is concluded that the urban area, rainfall, and population density are more along east side i.e. Rasulgarh (20.29 N, 85.86 E), Master Canteen (20.27 N, 85.84 E), and Rajmahal (20.26 N, 85.83 E) station points of Bhubaneswar city.

3.4 Change in rainfall over urban and peri-urban grid points

To clarify the association of rainfall, change with the increase in urbanization, the rainfall over urban and peri-urban grid points has been analyzed for different seasons (Fig. 9 (i-iv)). The difference in urban and peri-urban rainfall shows a modest but detectable increase from 1980 to 2019 period for all seasons. The increase in rainfall is more in monsoon season. For 1980s to 2000s the difference is negative or constant and does not show any notable impact of urbanization; whereas in recent period, the rainfall over the urban area is more compared to peri-urban regions; which is often associated with the urban heat island and rainfall impacts. The same has been observed in Li et al., 2020. Increase in urbanization increases surface temperature; which causes more surface evaporation, more local updraft over the region and intensify the rainfall intensity (Houston and Niyogi 2007).

Fig. 9
figure 9

Difference in rainfall over urban and per-urban grid-points during 1980–2019 for (i) Jan – Feb, (ii) Mar-Apr-May, (iii) Jun-Jul-Aug-Sept, and (iv) Oct-Nov-Dec

Full size image

3.5 Possible mechanism

A framework for the changes in rainfall due to urban intensification is presented here. The changes in the landscape characteristics due to urbanization lead to an increase in surface air temperature over the urban area relative to the peri-urban regions. This urban heat island (UHI) also leads to spatial gradients in the net radiation flux, as identified for the study region in Gogoi et al. (2019). The presence of UHI, especially in the pre-monsoon period, indicates a potential for enhanced vertical mixing and mesoscale circulation across the urban-rural interface. This favors a setup for moisture advection over the city. Surface winds upwind of the city are frequently altered (reduced) by urban roughness, and they may even accelerate downwind. This results in modified divergence and convergence patterns that are reflective of the microscale lateral pressure differential throughout the city transect and are visible along the city boundaries (Liu & Niyogi, 2019). The urban area likely provides sufficient energy to develop convective systems, elevating the clouds and rainfall water mixing ratio over the urban areas and thus causes more intensified rainfall (Simpson et al., 2008; Shem and Shepherd, 2009; Niyogi et al., 2011). The change in the timing of the rains is further supported by the UHI feedback on the regional circulation and convection potential during the peak UHI conditions – late afternoon and evening. These features will be need to be evaluated in a future modeling study.

4 Conclusions

Spatiotemporal variation in four different seasonal rainfalls and the concurrent changes in LULC over Bhubaneswar city are analyzed. Study results are as follows.

Both wavelet and MK-trend analysis indicate a notable increase in rainfall intensity for different seasons except for post-monsoon season over Bhubaneswar. The maximum increasing amount is about 0.4 mm averaged over the city per season which occurs during pre-monsoon season. The analysis of hourly rainfall data indicates a change in both duration and magnitude of rainfall occurrence over the city for all seasons. Interestingly, there is a change in the timing of the rainfall with a decrease in the midnight to early-morning rains and a corresponding increase in the late-afternoon to midnight rainfall from 1980 to 2018. In addition to the changes in the rainfall timing, there is an increasing trend of rainfall over the city and along the city’s thunderstorm-relative east side compared to the west side of the city (which is at > 95% significant level).

A five-fold increase in urban built-up area over the study area is found over the last four decades. Thus, urbanization in the rapidly expanding city along the east coast of India has likely caused a notable change in the timing, magnitude, and rainfall location. The data trend is still weak but there are indicators that the transition will likely continue as the city expands even further into the coming decades. This research will shed light on how heavy precipitation in the urbanized metropolis of Odisha is affected by changes in urban infrastructure. The civil engineers, city planners, and urban water managers will benefit from this product in order to plan the city in consideration of future environmental effects of urbanization.