1 Introduction

Recent increases in extreme events are manifestations of climate change. A warming climate not only alters the frequency, intensity, spatial extent, duration, and timing of these events [1], but also augments the adverse impacts on almost all facets of society and natural ecosystems [2]. Notably, the intensification of the hydrologic cycle, driven by global warming has been linked to frequent and intensified hydroclimatic extremes [3], such as heat waves, flash floods, landslides, and droughts [4,5,6]. Thus, the changing characteristics of hydroclimatic extremes have triggered widespread global concern, due to their impacts on global economy. Recently published Global Climate Risk Index shows that among the major world economies, only India and Japan are ranked among the 10 most climatologically vulnerable countries out of 180 countries [7]. India, the most populated (more than 1.4 billion) country in the world [8], is heavily reliant on agriculture (https://pib.gov.in/newsite/PrintRelease.aspx?relid=186413, accessed on April 20, 2024) and is highly vulnerable to changing characteristics of precipitation extremes due to inadequate disaster mitigation policies and infrastructure. During the past few decades, different regions of the country have experienced devastating heavy precipitation events. Recent studies also reported that dry extremes, even during the monsoon season, have become more frequent across India, and associated soil moisture depletion primarily affects crop production [9]. In addition, dry extremes are compounded with the increasing hot conditions leading to further depletion of the soil moisture due to positive radiation feedback [10]. The crop production losses may increase up to 10–40% by the end of the twenty-first century [11, 12].

A recent study revealed a southwards shift in precipitation extremes over South Asia utilizing past observations and indicated a greater increase in precipitation extremes over South India than over North and Central India [13]. At present, India ranks as the fifth largest economy worldwide and the southern part of India (also known as peninsular India), accounts for approximately 44% of the country's GDP (https://mospi.gov.in/data, accessed in April 2024). The peninsular India encompasses coastal regions, which frequently experiences heavy precipitation events under the combined influence of the southwest/northeast monsoon and tropical cyclones. These extreme precipitation events result in storm surges, flooding and landslides affecting many regions of southern India [14]. Rao et al. [15] showed that peninsular India is experiencing more frequent extreme precipitation events as a consequence of climate change, which can potentially hinder the growing economic aspirations of India. For instance, Kerala was substantially affected by severe flood events in 2018 and 2019 owing to unusually high precipitation events during the monsoon season [16]. Other examples include, heavy precipitation induced flooding in Tamil Nadu and Puducherry (in November and early December 2015); frequent extreme precipitation events over Mumbai [17, 18]; and frequent flooding in Bengaluru [19]. Such extreme events in peninsular India has caused significant economic losses. For example, the Bengaluru flood in 2022 alone caused a loss of USD 26 million in a single day due to disruption in the service-based industry (https://indianexpress.com/article/cities/bangalore/floods-caused-rs-225-crore-loss-outer-ring-road-companies-associations-8126614/, accessed on April 2024), while the Kerala flood of 2019 caused damage of USD 750 million primarily due to the losses in agricultural production (https://timesofindia.indiatimes.com/city/kochi/kerala-agriculture-flood-damage-touches-rs-6281-crore/articleshow/69352122.cms, accessed on April 2024).

To this end, this study will focus on the changing characteristics of precipitation extremes across peninsular India, particularly in terms of intensity, frequency, and duration. Understanding these changes is crucial for developing effective adaptation strategies to mitigate the impacts of extreme weather events on various sectors such as agriculture, water resources, and infrastructure. To make the characterization of the precipitation extremes more consistent and comparable, many studies have utilized the ETCCDI (Expert Team on Climate Change Detection and Indices) [20,21,22] extreme precipitation indices. Utilizing these indices, Dash and Maity [18] found that both wet and dry extremes exhibited significant changes over the southern parts of India. Further, significant increase in extreme precipitation frequency and intensity is observed across the coastal regions [23]. Besides, studies conducted at basin-scale, have also reported evidences of changing precipitation extreme characteristics. For instance, considering the major river basins across India, the maximum increases in extreme precipitation events is noted across the western flowing rivers in the peninsular India [24]. Similarly, long-term increasing trends were found in extreme precipitation magnitudes across the lower Godavari and Krishna basins [25]. Further, the Nagavali and Vamsadhara basins have also experienced increased heavy precipitation occurrences [26]. Conversely, the upper Tapi basin showed a decreasing trend in annual total precipitation, yet an increase in the precipitation intensity [27].

While these studies mostly evaluated the extreme precipitation characteristics at national or basin scale, consideration of only entire peninsular India as the study region could offer a broader perspective on the regional changes in these characteristics. Moreover, many of the existing studies focused on a singular time frame, while those considered multiple time frames are primarily centered around the climate regime shifts in the 1970s and 1980s [28, 29]. However, the effects of the climate regime shift in the 1990s [30], is not addressed, which could provide insights on the recent trends in the precipitation pattern and the potential abrupt shifts. In the peninsular India, a noteworthy distinction in precipitation characteristics exists between its eastern and western sides. Consequently, the evolving nature of extreme precipitation features may manifest divergent patterns. Thus, it is essential to examine the spatial and temporal patterns of precipitation variability [31]. This raises the following crucial question: how are the spatial patterns of extreme precipitation characteristics altered over time? We undertook this study to investigate the temporal evolution in the spatial variability of extreme precipitation characteristics across the peninsular India, particularly comparing the last two decades (post-2001) with respect to the preceding 50 years (1951–2000). This will offer a clearer understanding of how the regional precipitation characteristics have evolved in the recent times.

2 Study area and data

The southern part of India (southwards from 23.5° N latitude, i.e., the Tropic of Cancer), known as peninsular India, is the study area for this analysis and mostly lies within 8.25° N to 23.5° N and 69.00° E to 93.25° E. The region is surrounded by the Arabian Sea to the west, the Bay of Bengal to the east and the Indian Ocean to the south. The majority of the region is characterized by a tropical climate and experiences two distinct monsoon periods: the southwest monsoon from June to September and the northeast monsoon (also known as the return monsoon) from October to December. Most of the region receives the majority of the annual precipitation during the southwest monsoon. However, during the northeast monsoon, precipitation is primarily concentrated over the southeastern part of the study region.

Daily gridded precipitation data with a 0.25° × 0.25° (latitude × longitude) spatial resolution for the period from 1951 to 2021 were obtained from the India Meteorological Department (IMD). To develop this gridded precipitation dataset, daily precipitation recordings from 6955 stations were used. The inverse distance weighted method was used for interpolating the station measurements into gridded estimates. This method is based on the assumption that stations closer to grid locations tend to have more similar rainfall characteristics than do distant stations [32, 33].

3 Methodology

An overall methodological outline is shown in Fig. 1, which starts with the computation of extreme precipitation indices at each grid point. A total of eleven such indices are used. These are defined by the Climate Variability and Predictability (CLIVAR) project of UNWMO’s World Climate Research Programme jointly with the Expert Team on Climate Change Detection and Indices (ETCCDI) (https://www.wcrp-climate.org/data-etccdi, accessed in December 2023). The indices can be grouped into three categories portraying different characteristics of extreme precipitation, namely, intensity-based, frequency-based and duration-based, as shown in Table 1. Detailed descriptions of their computations are also provided in Table 1 along with their respective units.

Fig. 1
figure 1

Overall methodological outline showing major steps. Details are provided in Data and Methodology section and Appendix A (refer to the supplementary document)

Full size image
Table 1 Details of the indices characterizing the extreme precipitation events
Full size table

3.1 Identification of change points

Change point analysis is a statistical technique used to identify abrupt changes or shifts in a time series. Change points (here onwards, CPs) associated with different precipitation-based extreme indices were assessed from 1951 to 2021 at each grid point in the study region. It detects the year of significant change, if it exists, in the underlying statistical properties of an extreme index. As mentioned before, three different statistics of the extreme indices are considered, i.e., trend, mean and variability (in terms of standard deviation), as illustrated in Fig. 1, to comprehensively evaluate the abrupt shifts in different extreme precipitation characteristics. CPs considering linear trends in the extreme precipitation indices are detected employing the sequential Mann‒Kendall test [34]. CPs in the mean and standard deviation are identified following [35, 36]. The CPs are evaluated at a significance level of 0.05. The mathematical details of these methods can be found elsewhere. However, these are also presented in Appendix A.1 (refer to the supplementary document).

The CPs (year) may exhibit variability across regions and even between individual grids. There may be a common time frame when the majority of the grids in the study region experience CPs. Nevertheless, the analysis of CPs proves instrumental in identifying specific overall time epochs that merit further investigation. Thus, the purpose of the change point analysis is to determine the time epochs for the subsequent epochwise analysis to identify the changes in the spatial patterns of extreme indices.

3.2 Epochwise analysis of extreme indices

3.2.1 Trend analysis

The trend in each of the eleven extreme precipitation indices is computed at all the grid points across the study area using the Mann–Kendall (M–K) test [37]. This test examines monotonic increasing or decreasing patterns in the extreme indices over different time periods. The statistical significance of the identified trends was assessed at a significance level of 0.05. The M–K test is a nonparametric approach that considers the rank of the extreme precipitation time series rather than the actual values. The mathematical details of the M–K test are provided in Appendix A.2 (refer to the supplementary document).

3.2.2 EOF analysis

EOF analysis is a statistical technique used to capture the dominant modes of variability [38,39,40,41]. EOF analysis is carried out to identify coherent spatial and temporal patterns of variability in extreme precipitation indices. The changes in spatial and temporal variability patterns across the entire study region and for each epoch are evaluated. Whereas the basic details of EOF analysis can be found elsewhere [37, 42, 43], the steps based on the EOF analysis to identify the changes in the spatial pattern of extreme indices over time are explained as follows:

  1. Step 1:

    Let \({M}_{mn}\) be the data matrix of an extreme precipitation index, where \(m\) is the number of time steps in the time series and \(n\) is the number of locations (grid points) in the study region. Each row is a spatial map, and each column is a time series. The data matrix \(M\) is processed (the temporal mean is removed from each grid point) to form an anomaly matrix \({F}_{mn}\).

  2. Step 2:

    The covariance matrix \({R}_{nn}\) of the extreme precipitation index is obtained as

    $$\begin{array}{c}{R}_{nn}={ {{ F}_{mn}}^{T}\times F}_{mn}\end{array}$$
    (1)
  3. Step 3:

    The eigenvalues (\({\lambda }_{1,}{\lambda }_{2}{,\dots .\lambda }_{n}\)) and the eigenvectors \({V}_{nn}\) of the covariance matrix \({R}_{nn}\) are obtained. These eigenvectors (\({V}_{nn}\)) are the EOFs.

  4. Step 4:

    The time coefficient matrix \({T}_{mn}\) is calculated by:

    $$\begin{array}{c}{T}_{mn}={F}_{mn}{\times V}_{nn}\end{array}$$
    (2)

These time coefficients are also known as principal components (PCs). The variance contribution rate of an eigenvector is calculated as follows:

$$\begin{array}{c}{R}_{k}=\frac{{\lambda }_{k}}{{\sum }_{i=1}^{m}{\lambda }_{i}}\times 100\% \end{array}$$
(3)

Thus, the outcomes of this second part of the analysis are (i) spatial variability patterns as depicted through EOFs, (ii) temporal coefficients as identified through the PCs, and (iii) the variance explained by each PC. These spatial patterns and temporal coefficients reveal important features of changes in the spatiotemporal patterns of extreme precipitation characteristics.

4 Results and discussion

4.1 Change points in the precipitation extreme indices

The change points (CPs) at each grid point for all eleven extreme precipitation indices are identified at the 0.05 level of statistical significance. Considering the entire study period from 1951 to 2021, the most significant CPs based on trend, mean and standard deviation are identified separately and shown in Fig. 2 (trend-based), Fig. 3 (mean-based) and Fig. 4 (standard deviation-based). Grids with no significant CPs are shown in white, and if there is a CP, the grid is shown in colour depending on the decade when the change had occurred. There are several observations to note from Figs. 2, 3 and 4. These categories are discussed in Table 1.

Fig. 2
figure 2

Spatial pattern of grids with a change point in some year during 1951–2021 for different extreme precipitation indices based on its trend, at 0.05 significance level

Full size image
Fig. 3
figure 3

Same as Fig. 2 but the change point is based on mean

Full size image
Fig. 4
figure 4

Same as Fig. 2 but the change point is based on standard deviation

Full size image

First, out of all the intensity-based indices, the most extreme, i.e., R99p, which represents the annual total extreme precipitation magnitude, is observed, with the maximum areal coverage exhibiting significant CPs in the trend. Most of the CPs in the case of R99p occurred during 1961–1981. Additionally, the probability density function (PDF) and cumulative distribution function (CDF) of R99p are illustrated in Figure S1, for two distinct time periods, i.e., before and after the identified change point year. Differences in the underlying PDFs and CDFs across these periods are clearly discernible, indicating significant changes in R99p, post the change point. In case of the SDII, the CPs in the trend are found to occur during different time windows at different locations. In general, it occurred earlier (1951–1981) across the eastern region than across the western region (1971–2011). This contrast can be associated to regional differences in the rate of warming [44]; changes in the cyclonic activities in the Bay of Bengal and the Arabian sea [45]; extent of urbanization and corresponding land use changes, and the changes in anthropogenic aerosols [46, 47]. In contrast, CPs in the mean SDII are prominent over the eastern regions, mostly during the recent past, i.e., 2001–2011. Apart from this, the southernmost regions and some patches along the western coast also exhibited CPs in the mean SDII along with PRCPTOT and R95p during 2001–2011. A contrast is observed in the case of the indices expressing extreme intensity, namely, R95p, R99p, Rx1day and Rx5day, when comparing the eastern and western coastal regions. Specifically, CPs in the mean predominantly occurred during the post-1980s for the eastern coastal areas, contrasting with the pre-1980s timeframe noted across the western coastal regions. The same pattern is noted for the CPs in the standard deviation of R95p. Considering most of the intensity-based indices, changes in the standard deviation occurred predominantly in the post-1980s period.

Second, among the frequency-based indices, the CPs in the trend are observed with maximum spatial coverage in the case of R50mm compared to other frequency-based indices, i.e., R10mm and R20mm. This may indicate that the frequency of moderate events (i.e., 10mm < precipitation < 50mm) has reduced over time and are compensated through increased frequency of the heavy precipitation (i.e., precipitation ≥ 50mm) events. Furthermore, the CPs in the trend of R50mm are more widespread than those in the mean and standard deviation. In the case of the mean and standard deviation, more or less similar spatial patterns of CPs are noted for R10mm and R20mm considering the respective statistics. Furthermore, a contrast is noted between the coastal and central regions for the CPs in the mean. The coastal regions exhibited CPs during recent times, i.e., post-1980s, while the central regions mostly exhibited CPs during the period 1961–1981. Over the southernmost regions, changes in both the mean and standard deviation of the frequency-based indices occurred during the recent decade, i.e., 2001–2011.

Third, between the two duration-based extreme indices, i.e., CWD and CDD, a higher areal coverage of grids showing CPs is observed in the case of CWD for all three statistics, i.e., trend, mean and standard deviation. The mean of the CWD exhibits CPs across nearly all grid points, predominantly during the pre-1980s. This suggests that there is an abrupt increase/decrease in the maximum wet spell duration across most of the southern India. However, the CWD after/before the identified change point in mean has not possibly experienced consistent increase/decrease over time, imparting insignificant change points in trend over the region. Conversely, for CDD, CPs in the mean are primarily evident in the post-1980s period. Furthermore, regions near the western coast exhibited CPs during the recent past, i.e., 2001–2011. Further, CPs in its trend over the western coastal region is noted during 1980s, which can be attributed to the global climatic regime shifts [29] that might have influenced the trend more notably than the mean CDD. Considering the trend in the CWD, most of the changes occurred during 1971–1991, and the Western Ghats exhibited fewer CPs than the other regions. In the case of CDD, most of the CPs in the trend are mostly concentrated over the eastern coastal regions, occurring during 1961–1981. Similar to the trend, most of the CPs in the standard deviation of CDD were obtained during 1961–1981.

Overall, the extreme wet indices R99p and R50mm exhibited more extensive spatial extents that experienced a change in trend, mostly during 1961–1981. This is likely because higher threshold values correspond to more prominent CPs, or more explicitly, extreme events with high magnitude/frequency exhibit a change more explicitly. This can be attributed to the global climatic regime shifts around 1960s and 1970s, characterized by abrupt and persistent changes in various climate variables including precipitation pattern [28, 48,49,50]. For instance, various characteristics of El Niño-Southern Oscillation (ENSO) and its relationship with the Indian Ocean Dipole (IOD) changed since the 1970s, consequently affecting the Indian summer monsoon precipitation regime [51, 52]. Besides, significant increase in urbanization was evidenced during 1971–1981 in India, which has notably influenced the extreme precipitation occurrences [53]. Furthermore, out of the two duration-based indices, CWD (expressing extreme wet spells) changed more extensively, considering all three statistics (trend, mean and standard deviation). The CPs in the mean are spatially more extensive than those in the other two statistics for most indices.

The opposite feature between the eastern and western coastal regions is also observed. The CPs in the mean predominantly emerged in the post-1980s for the eastern coastal areas, in contrast to the pre-1980s points across the western coastal regions. Furthermore, over the southernmost regions, CPs in all frequency-related indices and many intensity-related indices have occurred mostly during the recent decade (2001–2011). Based on the identified CPs considering the majority of indices, statistics, and regions, the entire time period is divided into three epochs for further analysis: epoch 1: 1951–1975 (T1); epoch 2: 1976–2000 (T2); and epoch 3: 2001–2021 (T3).

4.2 Epochwise spatial pattern of extreme indices

First, the mean and standard deviation of all the extreme indices are computed at each grid location for three epochs, i.e., T1, T2 and T3. The spatial distributions of these statistics for all three epochs are shown in Figs. 5 and 6. Considering the wet indices (i.e., all indices except CDD), a higher mean is observed across the western coastal regions apart from Gujarat. This is expected because the western coastal parts are the regions that receive the maximum annual precipitation compared to the other regions. In contrast, the areas besides this region had considerably lower average values, as this region lies in the rain shadow/leeward region. The northeastern part of the study region, which includes Chhattisgarh, Odisha, Jharkhand, West Bengal and some parts of Andhra Pradesh, has a medium range of average values. This finding does not hold much resemblance in the case of R99p during T1. For CWD and PRCPTOT, the extent of areas with lower mean values increased over the epochs. On the other hand, for R99p, the area with higher mean values increased across the epochs. In the case of CDD, the western part of the study region, mainly Gujarat and some parts of Madhya Pradesh and Maharashtra, exhibited very high mean values, and the southern tip of India (parts of Kerala and Tamil Nadu) and the northeast part of the study region (West Bengal and Odisha) exhibited low values.

Fig. 5
figure 5

Spatial distribution of mean of the intensity-based (PRCPTOT, SDII, R95p, R99p, Rx1day and Rx5day), frequency-based (R10mm, R20mm and R50mm) and duration-based (CWD and CDD) indices, for three epochs

Full size image
Fig. 6
figure 6

Same as Fig. 5 but for the standard deviation of the indices

Full size image

Like the mean values, most of the wet indices exhibited higher standard deviations in the western coastal regions apart from those in Gujarat. However, there are a few exceptions. For instance, the areal extent with a higher standard deviation for RX1day gradually increased across the three epochs. More prominent increases are captured over the eastern coastal and northwestern coastal and Gujarat regions. Notably, the more impactful extremes, i.e., R99p and RX1day, exhibited more pronounced increases over time considering both the mean and standard deviation across the eastern coastal regions. In the case of R10mm and R20mm, southern tip regions are observed with higher values in T3, which is not observed during prior epochs. During T3, the standard deviation of the SDII in some regions of Gujarat and R99p across the eastern regions of the study area (West Bengal and Odisha) increased. The CDD gradually increased in standard deviation across the northwestern parts (Maharashtra and Madhya Pradesh) of the study region.

Overall, summarizing the findings, the spatial pattern of these statistics for all three epochs reveals the following specific features: (i) there is a clear variation in the extreme precipitation pattern from epoch to epoch, (ii) extremes possessing higher potential impacts showed more prominent increase in their mean and standard deviation over the eastern coastal regions, and (iii) the southern tip is noted with more frequent precipitation occurrences during 2001–2021.

4.3 Spatial patterns of trends in extreme indices

In this section, epochwise trends in the extreme indices, if they exist, are identified to investigate their nature. For instance, have the recent epochs experienced significant increases/decreases, and does the spatial extent exhibiting significant trends encompass a larger area than earlier? At the 5% significance level, more spatial coverage exhibiting a significant trend (either increasing or decreasing) is noted in the cases of SDII, R20mm, PRCPTOT, CWD, CDD and R95p. Spatial maps for these indices are shown in Fig. 7 (right panel). On the other hand, less spatial coverage with a significant trend is observed for Rx1day, Rx5day, R10mm, R50mm and R99p (Fig. 8, right panel). Notably, these trends occur between two consecutive CPs. Thus, the absence of a trend does not imply no change in the extreme indices over the entire study period (1951–2021). It might have changed relatively abruptly at the CPs and then exhibited no trend during an epoch, as noted with R99p, for example. To further explore this, the overall trend during 1951–2021 is also assessed, as shown in the left panels of Figs. 7 and 8. For most of the indices, spatially more extensive changes are noted considering the entire time period, than that computed within the different epochs. This observation supports the discussion above. It is worthwhile to note that the extreme indices, which signify greater potential severity, namely R20mm, R50mm, R95p, R99p, Rx1day, exhibited spatially more extensive increasing trends across the country. Further, this observation is more prominent over the eastern coastal regions than the western coastal regions. Among all the indices, SDII has shown the most substantial increase, more specifically, in the eastern coastal areas. In contrast, the most widespread decrease has been observed in CWD. This indicates an intensification of precipitation characterized by an increase in short-duration, heavy precipitation events, especially in the eastern coastal regions.

Fig. 7
figure 7

Spatial pattern of grids with increasing (red) and decreasing (blue) trends in extreme indices during the entire time period (left panel) and the three different epochs (right panel). Only statistically significant trends are shown in colour. The spatial coverage of these indices with significant trends is relatively greater than that of the indices shown in Fig. 8

Full size image
Fig. 8
figure 8

Same as Fig. 7 but for indices exhibiting less spatial coverage as compared to those in Fig. 7

Full size image

Considering the three different epochs, for most of the wet indices, the areal extent with significant trends increased over time. R10mm, R20mm, PRCPTOT, SDII and R95p showed patches of decreasing trends across the regions of Odisha and Chhattisgarh in the first epoch. However, over the epochs, the areal coverage with decreasing (increasing) trends decreased (expanded). The maximum wet spell (CWD) significantly decreased across most of southern India during T2. However, during T3, it increased across a few regions, including the Western Ghats, and the decreasing trends were concentrated in the eastern regions. Considering R95p and R99p, the increasing trends are mostly distributed across the southern parts of the study area. Contrasting spatial patterns of increasing/decreasing trends are observed for CDD over the epochs. With the exception of a few patches on the eastern coast, most of the regions exhibited increasing trends during T1 and T2; however, mostly decreasing CDDs were observed during T3. This indicates a reduced duration of dry spells during the recent period from 2001–2021. During T3, the spatial coverage with increasing trends for PRCPTOT, R10mm, R20mm, and R95p increased prominently. The total precipitation magnitude (PRCPTOT) increased in the western region, while the intensity (SDII) increased mostly in the eastern region. Furthermore, decreased wet spell durations was noted across the eastern parts (Odisha, West Bengal, and Chhattisgarh). This indicates that precipitation in these regions has become more intense in recent years, which supports the inferences drawn for the overall trends during 1951–2021.

To summarize the findings, the recent epoch (2001–2021) have seen an expansion in the spatial extent with increasing trends in most of the wet extreme precipitation indices in the southwestern regions of India. Conversely, the occurrence of dry extremes (CDD) has decreased across most of southern India during. Additionally, precipitation events have intensified during 1951–2021, throughout most of the southern India, with a more pronounced increase observed along the eastern coastal regions compared to the western regions. Out of the three epochs, this pattern is more evident for the recent epoch, 2001–2021. This can be majorly attributed to the increased frequency of tropical cyclones in the Bay of Bengal along the eastern coast in recent decades [45]. A study by Patel and Kuttippurath [54] indicates that atmospheric water vapor levels were elevated over the northern Indian Ocean, the Bay of Bengal, and the peninsular India from 2003 to 2020, thereby enhancing the potential for heavy precipitation. Gupta and Jain [55] found that global warming leads to increases in the extreme precipitation intensity at a higher rate over the eastern coastal regions. Furthermore, urbanization influences precipitation patterns in various ways, e.g., urban heat islands that intensify precipitation by increasing atmospheric instability and enhancing moisture transport [56]; higher emissions from urban areas augments aerosol production, consequently increasing the number of cloud condensation nuclei [57]. In this context, the recent LULC changes owing to the rapid urbanization during the last two decades, particularly in the coastal regions [58, 59], has resulted in higher surface temperatures and sensible heat fluxes, causing a relatively higher convective available potential energy and, consequently, intensified precipitation events [60]. The observed changes in seasonality has also influenced the regional precipitation patterns [61].

Additionally, changes in large-scale circulations can influence localized convection through modulations of synoptic circulation patterns [62,63,64]. Across peninsular India, the changing extreme precipitation characteristics have shown a significant positive relationship with various large-scale indices, primarily the Arctic Oscillation (AO) [65]. In coastal regions, extreme precipitation events have a notable association with the Niño 3.4 and Indian Ocean Dipole (IOD) indices, as compared to the other regions [66]. Increased precipitation variability along the eastern coast is primarily driven by the Southern Oscillation Index (SOI), Pacific Decadal Oscillation (PDO), Dipole Mode Index (DMI), Niño 3, and Niño 3.4 [67]. The duration of extreme precipitation events is strongly linked to the ENSO at an inter-annual scale, while the magnitude of these events is significantly associated with ENSO, Equatorial Indian Ocean Oscillation (EQUINOO), and PDO at an inter-decadal scale [68]. Furthermore, La Niña and the active phases of the Madden–Julian Oscillation (MJO) are connected to enhanced convection across the peninsular region [69]. Thus, these factors together have altered the extreme precipitation characteristics, emphasizing the need for continuous monitoring and adaptation strategies in response to evolving climatic conditions.

4.4 Spatial and temporal variability patterns of the extreme indices

The spatial variability of the extreme indices is investigated through EOF analysis. The spatial pattern of variability corresponding to each mode and the corresponding principal components (PCs), showing temporal variability, are derived for all the indices. Changes in the spatial and temporal variability patterns are evaluated over three epochs, T1, T2, and T3. EOF analysis produces many modes of variability, which together capture the total variance. The first few modes explain most of the total variance. To illustrate the differences in the spatial and temporal variability explained through different modes, we selected two indices expressing dry (CDD) and wet (R20mm) conditions. The leading three EOF modes of CDD and R20mm and the corresponding PCs for the three epochs are presented in Fig. 9. In all instances, the first three modes together captured approximately 60–70% of the total variance, with the first mode alone capturing approximately 30–40% of the total variability. Considerable differences in the spatial and temporal patterns are noted for the three modes in the case of both indices. For example, during T1, the first mode (EOF 1) showed positive CDD values across the entire region, with higher values over the southeastern and central parts of peninsular India. The regions with higher loadings contribute more to the total variability and indicate above-normal CDD across those regions during T1. In the case of EOF 2, most of the positive values are observed only over the eastern parts. Similarly, EOF 3 revealed a different spatial pattern, where the southern tip and some north-central parts of the study area exhibited high negative loadings, while the rest showed positive loadings. Considering EOF 1 for T2 and T3, a contrasting spatial pattern compared to that of T1 is noted. During T2 and T3, most of the region had negative loadings, with higher values distributed across the northern part of the study region, indicating below-average CDD, i.e., a shorter duration of extremely dry spells.

Fig. 9
figure 9

Spatial variability pattern and principal components of CDD (top) and R20mm (bottom) considering the first three EOF modes, i.e., EOF 1, EOF 2 and EOF 3 for three epochs: T1 (1951–1975), T2 (1976–2000), T3 (2001–2021)

Full size image

The temporal pattern of EOF1 indicates more frequent above-normal CDD years towards the end of T1 and the beginning of T2 for the entire study region. During T3, a greater magnitude of above-normal CDD was observed during 2013–2015 and 2020–2021. For R20mm as well, different spatial and temporal patterns are observed across the three epochs. EOF1 accounted for 42.59%, 32.96%, and 31.08% of the total variability during T1, T2, and T3, respectively. During T1, higher positive loadings are observed over the entire Western Ghats region, with a contrasting pattern observed during T2. In T3, along with the Western Ghats, the southeastern coastal regions exhibited high positive loadings. The above-normal R20mm over the Western Ghats indicated by the high positive loadings is expected, as this region is the high precipitation region of India, receiving the maximum annual precipitation. However, the observed higher positive loading across the southeastern coast during recent times indicates that this region has become wetter in terms of the above-normal frequency of daily precipitation > 20 mm. Furthermore, from the temporal patterns of CDD and R20mm during T3, it is evident that recent years have experienced above-normal dry and wet precipitation conditions simultaneously over the southern Indian region.

For the remaining indices, only the first mode is shown for discussion (Figs. 10 and 11). The Western Ghats consistently exhibited high variability in most cases. However, a contrast in the phase of variability is observed in this region over time for some indices. For example, considering R95p, a positive phase with high loadings is observed during T1 and T3, while a negative phase is obtained during T2. RX5day, PRCPTOT, R10mm, and R20mm showed similar spatial patterns of variability over time. During the first two epochs, higher loadings in the negative phase are observed mainly across the Western Ghats, and in T3, this region is observed to have high positive loadings, along with the southern tip of India. For all the wet indices except R99p and SDII, most of the study region exhibited positive loadings during the recent epoch, albeit with variable magnitudes across the region.

Fig. 10
figure 10

Spatial variability pattern and principal components of RX1day, RX5day, PRCPTOT, SDII, and CWD across South India considering the first EOF for three epochs: T1 (1951–1975), T2 (1976–2000), T3 (2001–2021)

Full size image
Fig. 11
figure 11

Same as Fig. 10 but for the extreme precipitation indices, R10mm, R50mm, R95p, and R99p

Full size image

CWD, which represents the maximum wet spell, exhibited more consistent spatial variability patterns over time than did the other indices. Strong positive loadings are observed over the Western Ghats region across the three epochs, with weak loadings observed elsewhere. Higher positive loadings are observed over the eastern coastal regions during T3 in the cases of R95p, RX1day, and RX5day. This indicates above-normal extreme precipitation conditions across the eastern coastal regions, specifically over the state of Odisha, during the last two decades. This can be attributed to the increasing frequency of tropical cyclones in the Bay of Bengal over the eastern coast during recent decades, with Odisha experiencing the greatest number of cyclone strikes [70]. Further, warming of the Indian Ocean is significantly associated with the increasing extreme events across coastal Odisha [71]. During T3, the indices with relatively greater spatial extent with strong (positive/negative) loadings are R95p, R20mm, and CDD. Therefore, more areas contributed to the higher spatial variability obtained for these indices. Considering all of southern India, different temporal patterns over time are observed for different indices. Consequently, above-normal PRCPTOT, SDII, R10mm, and R50mm were observed during the most recent years (2019–2021). In contrast, below-normal CWD, RX1day, and RX5day have been observed in recent years. The consecutive years 2005–2008 are observed with below-normal SDIIs. For R99p, most of the years exhibited above-normal conditions during the post-1975 period (i.e., during T2 and T3).

5 Conclusions

This study investigated the spatial and temporal variability in extreme precipitation characteristics and associated long-term changes across southern India for the period of 1951–2021. Eleven indices expressing different precipitation characteristics, i.e., frequency, duration, and intensity, are utilized. Abrupt changes in the precipitation indices are identified through change point analysis considering three characteristics: trend, mean and variability (in terms of standard deviation). Extreme precipitation indices with higher thresholds exhibited more spatially extensive change points (CPs). For example, R99p and R50mm, which express the intensity and frequency of extreme precipitation, respectively, exhibited the maximum spatial extent with CPs in trend. The time of abrupt change and associated spatial pattern varied across the indices. A pronounced disparity in the extreme intensity indices, i.e., R95p, R99p, Rx1day, and Rx5day, is evident when comparing the eastern and western coastal regions: CPs in the mean predominantly manifested post-1980s across the eastern coastal areas, in contrast to the pre-1980s shifts observed across the western coastal regions. Additionally, over the southernmost regions, discernible shifts in all frequency-based indices and many of the intensity-related indices have occurred primarily during the recent decade, i.e., 2001–2011.

Consequently, the entire temporal span was segmented into three distinct epochs (1951–1975, 1976–2000, and 2001–2021) for subsequent analysis. Contrasts in the changing characteristics of dry and wet extremes have been found in recent years, with wet (dry) extremes increasing (decreasing) significantly across most of southern India during 2001–2021. A significant increase in the extreme precipitation intensity across the eastern coastal regions and a significant decrease in the wet spell together indicate the intensification of precipitation events over these regions during the recent epoch (2001–2021). The observed changes are likely attributable to the increased occurrence of tropical cyclones in the Bay of Bengal over the eastern coast, increased urbanization, regional warming, elevated atmospheric water vapor levels and changes in the large-scale circulation, during the recent decades.

Further analysis explored the spatiotemporal variability patterns of precipitation indices within three distinct epochs using empirical orthogonal functions (EOFs). The first three leading EOF modes collectively explained approximately 60–70% of the total variance for most of the indices. Most of the indices consistently exhibited greater spatial variability over the Western Ghats. However, a contrast in the phase of the variability is noted over time for some of the indices. The maximum wet spell exhibits a persistent coherent spatial variability pattern over time. In terms of temporal variability, above-normal R99p values are indicated for a majority of the years post-1975. Finally, the southernmost tip regions have become wetter in terms of precipitation frequency during recent decades.

These findings have important implications for climate change adaptation and disaster risk management in the region. For example, the increased occurrence of extreme precipitation events on the eastern coast and the southernmost tip of southern India suggests that these regions are at greater risk of flooding and other related hazards. As a result, there is a need to develop and implement appropriate adaptation and disaster risk management strategies in these regions.