Spatial varying and co-occurring future climate changes over a heterogeneous river basin: a multivariate approach

Recent observations reinforce the substantial influence of global warming on changing climate, leading to widespread impacts on various ecosystems (Easterling et al. 2000; Hao et al. 2018; IPCC 2022a). Raftery et al. (2017) concerned a low likelihood, i.e., 1% (5%), of global warming being below 1.5 °C (2.0 °C) with an unprecedented temperature rise of 2.0–4.9 °C by the end of the 21st century. Subsequently, this anthropogenic warming exacerbates climate-related risks to natural and human ecosystems through intensified and frequent extreme weather events, sea-level rise, disrupted global and regional precipitation patterns, etc. (IPCC 2022a). These changes also influence the daily lives of living beings by challenging their health, agriculture, environment, economic status, and availability of water resources (Carleton & Hsiang 2016; Ebi et al. 2021). In recent times, the frequent catastrophic losses during floods in India and Bangladesh (in 2022), heatwaves in Europe and America (in 2021), Australian wildfire (in 2020), Cyclone Idai (in 2019), etc., have provoked public alertness about the daunting prospects of a changing climate. Therefore, a sustainable and climate-resilient world eventually calls for coordinated mitigation measures and decisions at various scales of governance, i.e., from local to global scale (IPCC 2022b), with active public participation. Thus, understanding and quantifying the potential impacts of future climate changes on various spatial and temporal scales plays a pivotal role in effective mitigation and adaptation measures for enhanced climate and socio-economic resiliency.

The global/regional climate models (GCMs/RCMs) are credible tools for understanding future climate changes and anticipating risks (Raju & Nagesh Kumar 2020). The GCM datasets from different phases of the Coupled Model Intercomparison Project (CMIP) have provided a basis for predicting impacts, guiding adaptations and mitigations under varying representative concentration pathways (RCPs). These climate projections are subject to uncertainties related to the parameterisation of physical processes, numerical formulations, initial and boundary conditions, and internal variability. However, the multi-model ensemble of climate models is proven to improve model projections' reliability, consistency, and skills, including climate-time scale simulations, by sampling initial conditions and model uncertainties (Gholami et al. 2023). Mishra et al. (2014) assessed the reliability of GCM/RCM datasets to simulate precipitation extremes over India and found that the ensemble mean of CMIP-5 GCMs outperformed their corresponding RCMs. Knowing the spatial disparity and diverse impacts of climate change, the coarser spatial resolution of GCMs limits their direct applicability for regional estimations. To overcome this limitation, statistical or dynamic downscaling transforms large-scale GCM outputs into local/regional-scale hydro-meteorological variables. Statistical downscaling approaches (SDA), i.e., artificial neural network (Schoof & Pryor 2001), support vector machine (Tripathi et al. 2006), change factor method (Anandhi et al. 2011), non-parametric kernel regression (KRSD; Salvi et al. 2013), etc., are computationally viable compared with dynamic downscaling. The KRSD approach, later modified by Kannan & Ghosh (2013), overcomes the limitations of conventional SDAs by capturing the spatial dependence of rainfall magnitude and hydro-meteorological extremes in predicting multisite daily rainfall magnitude. The bias, if any, in the downscaled outputs, mainly rainfall, precludes their direct relevance with field observations (Nanditha & Mishra 2022) and hence needs to be corrected. Chen et al. (2013) analysed various bias-correction methods and showed that distribution-based methods, such as the quantile mapping approach (QMA), perform consistently better than mean-based methods. QMA also served as a reliable bias-correction option in topographically diverse regions (Enayati et al. 2021).

The Expert Team on Climate Change Detection Indices (ETCCDI) defined 27 climate indices to assist spatiotemporal hydroclimatic investigations, particularly in rainfall and temperature. Typically, these indices characterise moderate/extreme events that occur within a year or less, striking a balance between data availability and the observed changes' reliability (Zhang et al. 2011). Studies across the globe used these indices; however, they are not limited to Donat et al. (2013), Sillmann et al. (2013), Alexander (2016), etc., to assess the spatiotemporal variability of individual variables and indices. Such univariate analyses of hydroclimatic variables fell short of describing their mutually inclusive geographical and temporal patterns, including their complicated dependence structure, like storm duration and intensity, drought duration and severity, rainfall and soil moisture, etc. (Sadegh et al. 2017). As a result, multivariate analysis techniques emerged as a powerful tool to overcome these shortcomings (Kao & Govindaraju 2007; Kumar et al. 2021). Multivariate regression analysis, factor analyses, copula theory, various machine learning algorithms, etc., are popularly used multivariate analysis techniques in hydrology, water resources, and climate science. The principal component analysis (PCA), an unsupervised machine learning algorithm, is a multivariate statistical technique to identify patterns and correlations among hydroclimatic variables and has numerous applications in the atmospheric sciences and hydroclimatic studies (Bethere et al. 2017; Raju & Nagesh Kumar 2020).

Past studies have used observed and projected climate variables (primarily rainfall and temperature) and ETCCDI climate indices in univariate and multivariate frameworks for hydroclimatic assessments at global/regional scales. Donat et al. (2013) reported significant warming during winter months and more spatially heterogeneous increasing trends in total rainfall, intensity, and frequency over most parts of the world during 1901–2010. Sillmann et al. (2013) showed pronounced global changes in minimum temperature extremes with an intensification of dry conditions over Australia, Central America, South Africa, and the Mediterranean region during 2006–2100. Bethere et al. (2017) showed increasing mean monthly rainfall and temperature over the Baltic states during 2071–2100 using PCA. During the late 21st century, Western Europe tends to have a faster rise in extreme temperature than the mean (Sherwood 2020). In the Indian context, a study by Goswami et al. (2018) discussed increased exposure of the Northern Sikkim Himalayas to co-occurring heavy and weak precipitation during 2006–2100. Kumar et al. (2021) reported increasing rainfall extremes, wet spell duration, and maximum and minimum temperatures during 2025–2095 over India. Using bivariate copula assessment, they also demonstrated that most of the region in India would be vulnerable to precipitation extremes along with hot days and cold nights. Sharma et al. (2018), Chandole et al. (2019), and Kale (2020) studied the rainfall and/or temperature characteristics over the Tapi River basin using the univariate approach. They reported spatially heterogeneous increasing and decreasing temporal patterns. Gehlot et al. (2021) and Jibhakate et al. (2023a) explored the joint space–time variability of rainfall and temperature variables over physioclimatically heterogeneous TRB using PCA and cluster analysis. These past studies efficiently used ETCCDI indices and univariate/multivariate analysis techniques to explore the spatiotemporal variability of rainfall and temperature variables. However, the literature lacks studies exploring the joint hydroclimatic variability of multiple rainfall and temperature indices. Further, co-occurring or concurrent climate-scale changes in the hydroclimatic regime of precipitation and temperature have remained unexplored.

Water availability in peninsular India primarily relies on the Indian Summer Monsoon Rainfall (ISMR). It administers the agriculture-based Indian economy but has been challenged by anthroposphere changes, seasonal and spatially heterogeneous monsoons (Ghosh et al. 2016). These changes would compromise the availability of water resources and food security of 1.4 billion people. The Tapi River, the second-largest west-flowing river in peninsular India, has multi-faceted socio-economic importance, from sustaining livelihoods and agriculture to supplying water to numerous industries and fostering regional growth and cultural heritage. The increasing area under chronic water shortage, from 34.0% (in 1981) to 63.2% (in 2011), is alarming (Sharma et al. 2019a). The TRB has witnessed various catastrophes in past years, including the 2006 flood, which inundated nearly 80% of Surat City and caused the death of 150 people and losses of 210 billion rupees (Jibhakate et al. 2023b), which is expected to be 0.26% of the city's gross domestic product in the year 2050 (Hallegatte et al. 2013). Notably, Surat is one of the fastest-growing cities in the world. Considering the agricultural dependency of the region, heavy rainfall events damaged standing crops over 40 and 5.7 lakh hectares (4,000,000 and 570,000 ha) in the years 2021 and 2022, respectively, in Maharashtra state (Dasgupta 2022). The basin is also found to be sensitive to climate hazards and risks, as evident from the recent worldwide aggregate damage ratio ranking of Maharashtra (38th), Gujarat (44th), and Madhya Pradesh (52nd) states, respectively, sharing 76.73%, 9.07%, and 14.2% of the area of TRB (2023 XDI Gross Domestic Climate Risk dataset; https://xdi.systems/xdi-benchmark-gdcr/, accessed on March 4, 2023). Thus, considering the socio-economic importance, climatic sensitivity, and vulnerability of the region, the shortcomings of previous studies and the knowledge gap, the present study proposes a methodological framework for comprehensive assessment and compact interpretation of hydroclimatic variability in TRB for the baseline and future periods using PCA, a multivariate unsupervised machine learning algorithm. The study also considers the co-occurring climate-scale changes and their plausible implications on the health and well-being of the region's people, agriculture, and water resources, which would help devise policies for regional sustainability and mitigation measures to enhance climate resiliency.

The current paper is organised into five sections. Section 1 includes the significance of the problem, a detailed review of the literature on the theme of the manuscript, the research gap, and the formulation of the objectives of the problem. Section 2 describes materials and methods, including the study area's physiographic characteristics, the data used, and the methodological framework adopted. Section 3 presents the univariate evaluation of the rainfall and temperature indices, followed by a multivariate aspect of joint climatic variability for future and co-occurring climate-scale changes. The anticipated repercussions of the findings on human health, agriculture sectors, urban centres, and plausible adaptation measures are discussed in Section 4. Finally, Section 5 concludes with a summary of the work, major findings, key limitations, and future directions of the current investigation.

The regional climate projections in India with statistical downscaling have been undertaken by the Indian National Committee on Climate Change (INCCC), Department of Water Resources, River Development and Ganga Rejuvenation, Ministry of Jal Shakti, Government of India, through a research project entitled ‘Statistical Downscaling for Hydro-climatic Projections with CMIP5 Simulations to Assess the Impact of Climate Change’. The five GCM models (specified in Table 1) were selected as a part of the INCCC project as they reliably represent Indian climate conditions (Mishra et al. 2014; Shashikanth et al. 2014; Madhusoodhanan et al. 2018). These models also consistently simulated rainfall over TRB (Singh et al. 2016). The current study used the datasets developed in the INCCC project, i.e., KRSD rainfall (Kannan & Ghosh 2013; Salvi et al. 2013) and bias-corrected temperature for the historical period (1951–2005) and future period (2021–2100) for RCP 4.5 and RCP 8.5 (denoted as R4.5 and R8.5). The former is a ‘medium’ emission scenario, while the latter represents a ‘business as usual’ scenario. Notably, the cumulative greenhouse gas concentration in the atmosphere up to 2020 is traced closely to RCP 8.5 (Schwalm et al. 2020). The high-resolution daily scale gridded rainfall (0.25° × 0.25°) and temperature data (1.0° × 1.0°) collected from the India Meteorological Department (IMD), Pune, is used as a reference dataset for the period 1951–2020. These gridded rainfall and temperature datasets were developed by Pai et al. (2014) and Srivastava et al. (2009), respectively, using daily observations of 6,955 and 395 quality-controlled gauging stations across the country. These are the most reliable among other gridded datasets in capturing climate seasonality over India and are used by various researchers (Mukherjee et al. 2018; Kumar et al. 2021; Nanditha & Mishra 2022). The daily temperature data were bilinearly interpolated to 0.25° resolution. The spatially consistent gridded rainfall and temperature data over TRB (149 grids, see Figure 1) was used in the present study to unravel their joint hydroclimatic patterns.

The unsupervised machine learning approach, PCA, unravels the multivariate hydroclimatic variability and spatial patterns. Using an attribute similarity structure, the PCA reduces the intricate association among variables to mutually exclusive indices known as principal components (PC) (Jolliffe 2002). These PCs efficiently and compactly represent the spatial patterns among the hydroclimatic attributes. The annual average values of the 12 climate indices at 149 grids of TRB, calculated using IMD data for the baseline period 1991–2020, formed a matrix with 149 rows and 12 attributes. Since these indices have different units, they were standardised using the mean (⁠⁠) and standard deviation (⁠⁠) of the baseline period. The climate indices for the future scenarios were also standardised using the same mean (⁠⁠) and standard deviation (⁠⁠) as the baseline period. This standardisation scheme scales the departures of future climate considering the present scenario and opts for comprehensive climate-scale assessments (Bethere et al. 2017). The PCA used the covariance matrix of these standardised anomalies of the baseline period as a similarity structure to calculate the eigenvalues and eigenvectors (loadings). The loadings calculated using reference data were used as a coefficient for the linear transformation of standardised anomalies for baseline and future periods. Since each PC represents the climate-scale phenomenon and describes the various characteristics of rainfall and temperature regime over TRB, the co-occurring changes in these PCs and their plausible implications on living beings, agriculture and water resources are discussed in section 4. The standard methods used in the present study are not detailed, for brevity. The cited literature studies may be referred to for comprehensive information on all the key components of this methodology.

Most regions in the TRB depict mixed signatures of increasing and decreasing AnnualTOT during the baseline period, except LTB. The MTB, having the least annual rainfall among the four heterogeneous regions of TRB during the baseline period, shows a statistically significant rise of 21.2% and 48.3%, respectively, during R4.5-DF and R8.5-DF (see Figure 4(a) and Figure SF2). AnnualTOT in Purna and Burhanpur sub-catchments shows a rise of 16.1% and 14.6% under R4.5-DF compared with 8.2% and 9.5% under R8.5-DF. The LTB shows an increase of 23.2% (26.4%) in AnnualTOT under R4.5-NF (R8.5-NF), which decreases marginally in the DF. The transitions in AnnualTOT during R8.5-DF would alter the rainfall regime of UTB to wet and MTB to wetter from the existing baseline scenario. The RD in Purna (Burhanpur) sub-catchment and LTB showed an increasing (decreasing) trend during the baseline period. The Purna and Burhanpur sub-catchments would experience a significant rise of 20.2% and 10.2%, respectively, in RD by R4.5-DF. On the contrary, RD in MTB (LTB) during R8.5-DF would increase by 23.1% (29.4%) (see Figure 4(b) and Figure SF2). Notably, AnnualTOT and RD are likely to rise over TRB in future scenarios (see Figure 4(a) and 4(b)); however, the disproportionate rise in RD and AnnualTOT would cause SDII to upsurge over MTB and LTB (except R8.5-DF) and reduce SDII over the Purna and Burhanpur sub-catchments (except R4.5-DF), as shown in Figure 4(c). The temporal trends depict a remarkable rise over the MTB region under both RCP scenarios. The greatest length of dry spells (CDD) during the period 1944–2013 has increased significantly over more than 92% of the basin stations (Gehlot et al. 2021), while the signatures of the future climate are diametrically opposite and statistically significant, particularly in UTB and MTB (see Figure 4(d) and Figure SF2). The CDD in Purna sub-catchment would reduce to 98 (86) days in R4.5-DF (R8.5-DF) from 155 days during the baseline period. Similarly, the MTB shows a reduction of CDD to 146 days, on average, in DF from 175 days during the baseline period. In contrast, the LTB would likely experience nominally longer CDD in the future (223 days, on average) compared with the present (214 days). During the baseline period, the length of wet spells (CWD) showed decreasing characteristics over the Burhanpur sub-catchment and MTB, which would increase diametrically over TRB under both RCP scenarios. The future estimates for CWD are a disproportionate rise over TRB, as shown in Figure 4(e). The percentage increase of CWD in the future ranges from 17.2% (in LTB in R4.5-DF) to 82.6% (in MTB in R8.5-DF); however, the statistically significant increase is concentrated in MTB and LTB during the RCP 8.5 scenario. Nearly 39.5%–44.5% of AnnualTOT over TRB is contributed by very wet days (see definition of R95pTOT in Table 2), showing the non-uniform temporal distribution. The humid/sub-humid agro-climatic regimes of TRB, i.e., Burhanpur sub-catchment and LTB, show a dichotomy in R95pTOT in the future (see Figure 4(f)). The former shows rising non-uniformity (R95pTOT ranging from 2.8% to 14.2%). In contrast, the latter shows a uniform annual rainfall distribution, with R95pTOT ranging from −2.3% to −34.7%. The increased share of heavy rainfall events over MTB during DF under both RCP scenarios is statistically significant. It is a cause of concern with reference to soil erosion as the region relies heavily on agriculture.

The assessment of temperature-based climate indices shows that the average maximum-TXm (average minimum-TNm) temperature over TRB during the baseline period ranges between 32.9 and 33.5 °C (19.5–21.1 °C), with mean temperature-Tm ranging from 26.2 to 27.3 °C (see Figure 4(g)–4(i)). The temporal trends in these variables also characterise increased warming during the baseline period. The TXm in TRB in NF and DF under R4.5 (R8.5) scenario would rise to 34.3–34.6 °C (34.8–35.0 °C) and 35.0–35.2 °C (37.1–37.3 °C), respectively. Similarly, the corresponding TNm in TRB would be 21.0–21.3 °C (21.4–21.7 °C) and 21.7–22.1 °C (24.0–24.3 °C), respectively. The temporal variations strongly adhere to statistically significant warming in TXm and TNm over the entire TRB. A remarkable escalation of minimum temperature vis-à-vis maximum temperature is also expected in the MTB and UTB regions. As a result, the diurnal temperature variation, denoted by DTR, would reduce. In contrast, the LTB would likely experience higher DTR variation, ranging from 5.0% to 8.1%, in the future. The DTR in the baseline period showed decreasing trends over TRB, except UTB. In agreement with the baseline trends, the temporal variations in the future period show a decreasing trend in DTR over TRB. The frequency of daily maximum temperatures above the 95th percentile, i.e., very hot days (TX95), shows large deviations in the future as the threshold values at individual grids were based on the entire future period, i.e., 2021–2100. Similar observations are there for the frequency of daily minimum temperatures below the 5th percentile, i.e., very cold nights (TN5). The discrepancy in the reduced frequency of TX95 and exaggeration of TN5 in the NF period under both scenarios could be due to underlying overestimation/underestimation in the maximum/minimum temperature. Despite this, the TX95 (TN5) shows a remarkable rise (decline) of 41.8%–49.3% (73.9–80.5%) across TRB in the R8.5-DF as compared with the present scenario. Notably, these indices have also shown increasing warming trends over TRB during the baseline period, except TN5 over MTB and LTB showing decreasing trends. Overall, the temperature indices unanimously indicate the dramatic expansion of the warming regime in the future, which is statistically significant at a 5% significance level (see Figure SF2).

Considering the TRB as a whole, the KPDE of AnnualTOT (see Figure SF3(a)) for future periods shows a shift towards a wet regime with a pronounced increase in its variability. On the other hand, the changes in the location parameters of RD and SDII are not significant, except for R8.5-DF. However, it shows a substantial rise in variability over the TRB (see Figure SF3 (b,c)). Similar expansions in the spatial variability are observed for the CDD, CWD, and R95pTOT. The CDD (CWD) shows substantial reduction (increment) in their longest spell durations, as shown in Figure SF3(d,e). Conversely, the temperature indices exemplify sizable shifts towards warming, which are more pronounced in night-time temperature than day-time (see Figure SF3(g–l)). However, they do not indicate significant changes in the variability.

A substantial difference in the mutual spatial patterns of climate indices is observed for the future. The PC1 is positively (negatively) correlated with DTR and TN5 (AnnualTOT, SDII, and CWD) and shows increasing (decreasing) magnitude of indices with an increase in PC1 score over space. Thus, the regions with relatively lower PC1 scores can be characterised as the humid/wetter exposed to higher intensity and longer wet spells with lower diurnal temperature ranges and colder nights. For instance, the said characteristics are represented by the relatively lower PC1 score in LTB compared with the other regions of TRB for the baseline period (see Figure 6(a)). These values of PC1 clearly show the tendency of LTB to have aggravated rainfall magnitude, intensity, and wet spells, particularly in DF. During the baseline period, the PC1 score ranged from −14.1 to 6.1. The corresponding ranges during the R4.5-NF (R8.5-NF) and R4.5-DF (R8.5-DF) periods are −12.3 to 5.6 (−11.3 to 8.8) and −21.8 to −3.5 (−32.9 to −9.4), respectively (see Figure 6(b)–6(e)). It also describes the joint shifting of the TRB towards a warm and wet regime with longer wet spells compared with the baseline scenario; however, there are spatial discrepancies at the grid level. In contrast, PC2 does not show substantial differences in spatial patterns (see Figure 6(f)–6(j)). The increasing magnitude of PC2 signifies increasing (decreasing) RD (R95pTOT) due to its positive (negative) correlation with these indices. Therefore, the high PC2 values have a relatively more uniform annual rainfall distribution during the monsoon period. The magnitude of PC2 scores over TRB during the baseline period and all four future periods, i.e., R4.5-NF (R4.5-DF) and R8.5-NF (R8.5-DF), respectively, ranged from −7.0 to 8.4, −8.6 to 12.6 (−8.9 to 11.5) and −8.5 to 11.2 (−10.8 to 18.9). These values signify the spatial disparities as both ends of PC2 expand in the future, indicating the regions with pronounced non-uniformity of rainfall during baseline will get further aggravation, like certain regions in MTB. On the other hand, the LTB region shows a transition towards a wet regime but with a more uniform temporal distribution. The PC3 primarily describes the common spatial patterns of the dry spell durations (CDD) and mean (Tm), minimum (TNm) and maximum (TXm) temperature over TRB, with all being negatively correlated with PC3 scores. Figure 6(k) clearly shows that LTB has the persistence of dry spells compared with MTB and LTB, demonstrated by lower PC3 values. The more consistent and significant changes in temperature indices overshadow the changes in the rainfall index, i.e., CDD. Thus, a dramatic shift in the warming regime is reflected over the TRB in PC3 (see Figure 6(k)–6(o)) and can be seen from the PC3 scores during baseline, R4.5-NF (R4.5-DF) and R8.5-NF (R8.5-DF) periods ranging from −9.2 to 8.6, −13.2 to −1.4 (−19.0 to −6.9), and −15.7 to −4.0 (−33.7 to −21.7), respectively. The scores of PC4 primarily describe the spatial pattern of very hot days (TX95) with a negative correlation over space. The corresponding PC4 score during the baseline, R4.5-NF (R4.5-DF) and R8.5-NF (R8.5-DF) periods varied from −3.0 to 4.9, 2.8 to 6.5 (1.8 to 5.9), and 8.5 to 11.6 (−7.4 to −0.8), respectively. It can be seen from Figure 6(p)–6(t) that along the direction of flow of the Tapi River, the frequency of very hot days gradually reduces under the baseline scenario; however, in the future, the entire basin is anticipated to experience very hot days, particularly in the DF. The analysis of the results obtained from PCA showed that the multivariate assessment comprehensively presented the spatial variability of hydroclimatic variables in the future period.

The methodology presented in this study primarily focuses on the comprehensive assessment and compact interpretation of hydroclimatic variability. At first, a univariate analysis of 12 climate indices for the baseline period (1991–2020) and the future periods (NF: 2021–2050, and DF: 2071–2100) under RCP 4.5 and RCP 8.5 scenarios was performed. The joint spatial variability of these climate indices, under the influence of climate change, was explored using PCA. The 12 climate indices were transformed into four components that described the primary spatial aspects of climate across TRB more proficiently and compactly. These components were used to identify the regions likely to be critically affected by the disproportionate and co-occurring climate-scale changes in the future. The following paragraphs establish the corroboration of current findings with existing literature and suggest the plausible implications considering the human health, socio-economic, and environmental aspects of the TRB.

TRB's geographical settings and relief features are the primary cause of spatial heterogeneity in rainfall patterns. As a result, the LTB and Burhanpur sub-catchment have better annual rainfall than the MTB and Purna sub-catchment. Notably, the latter two regions have more than 60% of their land use patterns as agriculture (see Figure 1), and low rainfall augments the overexploitation of groundwater to meet the water requirement of around 64% of the net irrigated area in TRB (Sharma et al. 2019a). Therefore, annual extractable groundwater resources are marked semi-critical/critical/overexploited, especially in the Purna sub-catchment (CGWB 2022). Sharma et al. (2019a) were also concerned about the increased chronic water stress conditions in TRB from 34.0% (in 1981) to 63.2% (in 2011) due to reduced surface water availability under a rapidly growing population. In line with the present, TRB is anticipated to experience more spatially heterogeneous rainfall patterns, especially in NF projected scenarios compared with DF. As a result, coherent rainfall patterns are observed in the baseline period and NF projections. On the other hand, the DF future would experience wetter rainfall patterns. Satoh et al. (2017) reckoned additional stress on the per capita water availability, leading to intensified water scarcity amid increasing surface water availability in the region by the 2050s. Notwithstanding high water-challenges in agricultural, domestic, and industrial sectors, the basin also demonstrated high coping capacity with improved basin-scale economic–institutional capacity (Satoh et al. 2017). The policy frameworks of the Government of India, like Atal Bhujal Yojana, Jal Kranti Abhiyan, National Water Mission, National Action Plan on Climate Change, etc., are functional in providing the right ground for the Indian river basins to cope with future water challenges and meet the sustainable development goals (Sarkar & Bharat 2021). Apart from the challenges posed by water scarcity, the TRB is equally sensitive to floods, as evidenced by the history of catastrophic floods in Surat City (LTB) (Jibhakate et al. 2023b). Further, the climatological transition of rainfall patterns to wet regimes would trigger more wet weather extremes in the DF. Thus, under the prevailing rise in very wet days in UTB and MTB, the significant increase of CWD over LTB in DF would augment pluvial floods in the city in conjunction with riverine floods, like the floods of 2006 or 2013 in Surat City. The city would additionally be susceptible to coastal urban flooding under the warming-induced sea-level rises (Hallegatte et al. 2013). The findings of the present study for rainfall indices during the baseline period closely follow the findings of Gehlot et al. (2021) and Sharma et al. (2018). The possible discrepancies in the statistical results could be due to differences in the datasets (station vs gridded) and the study period (1944–2013 vs 1991–2020). Kumar et al. (2021) used CWD and CDD indices under R4.5 and R8.5 with two different subperiods, i.e., 2025–2055 and 2065–2095, demonstrating similar climatological patterns over TRB.

The baseline period's minimum and maximum temperature indices showed regionally heterogeneous warming patterns. In contrast, the temperature indices give more homogeneous indications of dramatic warming supported by the mean temperature rise of 1.1 °C (R4.5-NF), 1.8 °C (R4.5-DF), 1.5 °C (R8.5-NF), and 4.0 °C (R8.5-DF), which is close to the postulated warming of 2.0–4.9 °C (Raftery et al. 2017). The results also show profound night-time warming vis-à-vis day-time warming in the future. The differential rates of temperature change in the future lead to a significant reduction of DTR in UTB and MTB, diminishing the heat dissipation cushion of night-time temperature. This would exacerbate heatwaves and heat stress, heat-related morbidity and mortality rates, and crop stress, leading to ill effects on crop yield, marginal labours and farmers, their work performance, and family income (Kjellstrom et al. 2016; Mazdiyasni et al. 2017; Sharma et al. 2020). Notably, the Maharashtra, Madhya Pradesh, and Gujarat states, respectively, reported approximately five, nine, and four fatalities per heatwave event from 1978 to 2014, with Maharashtra having the highest frequency of heatwave events in the country (Malik et al. 2021). Also, the thermal heat stress would lead to a loss of 5.8% of India's working hours, equivalent to 34 million full-time jobs (60% in the agriculture sector) by 2030 (ILO 2019). Moreover, the increasing DTR in sub-humid LTB and concurrent intensification of atmospheric temperature and humidity would adversely impact human health by causing an upsurge in cardiovascular and respiratory diseases (Cheng et al. 2014) and increasing outdoor thermal stress and the overall energy consumption and power requirement for indoor cooling (Sherwood 2020). The warming tendency reported in the current study closely corroborates with past studies on TRB (Sharma et al. 2018; Chandole et al. 2019; Jibhakate et al. 2023a), Indian regions (Mall et al. 2021; Rehana et al. 2022), and regions across the globe (Donat et al. 2013; Alexander 2016). The studies by Sherwood (2020), Bethere et al. (2017), and Sillmann et al. (2013) on a global scale and Kumar et al. (2020, 2021) on the Indian region affirm the similar remarkable rise in the temperature mean and extremes during the future periods.

The multivariate assessment of climate indices' variability using PCA comprehensively represents the potential changes in the future climate. The assessment resulted in three plausible states of climate transition scenarios in future periods. For instance, scenario C1 would prevail in the TRB under R4.5-NF and R8.5-NF, respectively, at 22.8% and 41.6% of its area. In these regions, warming and rainfall deficit can cause dry–warm compound phenomena. Conversely, wet–warm compounding represented by scenario C2 is likely at 16.8% (47.0%) and 1.3% (30.2%) area of TRB under R4.5-NF (R4.5-DF) and R8.5-NF (R8.5-DF), respectively. Mishra et al. (2020) demonstrated that the warm phase of the sea surface temperature in the Pacific (El Niño events) triggered dry–warm compound events over west-central India in the recent decade. Gehlot et al. (2021) also associated the moderate to strong El Niño and weak La Niña (moderate to strong La Niña and weak El Niño) with drier (wetter) ISMR conditions over TRB. Further, the climate-scale transitions to wet–warm (dry–warm) climatology would exacerbate the concurrent or isolated occurrence of extreme weather events, such as heavy rainfall, prolonged wet spells, flood, heatwave/coldwave etc. (droughts, heatwaves, wildfires, etc.) in the basin. Raymond et al. (2020) discussed the severe consequences of dry–warm/wet–warm compound events on water resources, standing crops and agricultural productivity, wildlife, human health, and the socio-economic infrastructure and environment. The co-occurrence of wet–warm and warm-air quality events intensifies the human mortality risk and healthcare hazards (Zscheischler et al. 2018). Apart from the detrimental circumstances discussed above, the uniform temporal distribution of the annual rainfall (scenario C3) with skilful management of the water resources can be proven beneficial for TRB. Notably, scenario C3 is most probable at 21.5% (8.7%) and 53.0% (69.8%) of the area of TRB under R4.5-NF (R8.5-NF) and R4.5-DF (R8.5-DF), respectively. The uniformity of temporal distribution of annual rainfall and increased water availability can improve agricultural productivity and groundwater reserves, lead to more significant biodiversity and reduce the potential of flood risk and soil erosion, particularly under distant future scenarios. TRB shows mean climatological transitions to wetter and warmer regimes by the end of the 21st century. However, it is subjected to severe weather-scale events, as inferred from the increased variability. Further, the atmospheric temperature rise interacts with and exacerbates risks, such as drought, extreme rainfall, or floods, which makes the adaptability of existing schemes/policies questionable.

Kumar & Parikh (2001) showed the adverse effects of temperature outweigh the positive impacts of rainfall, particularly in agriculture. Thus, micro-irrigation, soil nutrient management, multiple stress-tolerant crop varieties, improved seed quality, etc., enhance crop resiliency while rainwater harvesting and drainage, artificial groundwater recharging, crop insurance, renewable energy-based technologies, strengthening and developing more reliable weather forecasting and early warning systems, etc., reduces the overall vulnerability of the ecosystems to multiple climate hazards and mitigates the risks (Pathak 2023). The subset of these measures has increased the yields of wheat, rice, and cereals, respectively, at 43, 27, and 15.5 kg/ha despite declining rainfall under rising temperatures in India during 1951–2016 (Mishra et al. 2020). To meet the urban-scale sustainable development goals, Surat City, being projected to be the fastest growing city in the world during 2019–2035, has planned to increase its green cover from 18% to 25%, management of solid and liquid waste, implement green infrastructure codes and standards, etc. (https://www.shiftcities.org/post/surat-smart-city-leading-indias-sustainable-urban-development-movement, accessed on July 25, 2023). Kleerekoper et al. (2012) also emphasised increasing vegetation cover, use of water infrastructure, recycling and reuse of treated wastewater, built form, and use of absorbent materials based on their thermal admittance, along with the planning of urban developments to combat the urban heat island effects in major urban centres. In addition, improved drainage systems, land use and land cover planning, and suitable structural and non-structural measures, i.e., flood forecasting systems and flood hazard–vulnerability–risk mapping, would enhance its resiliency to pluvial and riverine flooding. Implementing such good practices throughout the basin would set back the potential damages of global warming and climate change. Considering the implications of climate change and variability outlined in the present study, evaluating hydroclimatic risk for TRB is an emerging concern that may be an extension of the current study. The agricultural susceptibility/risk to hydroclimatic extremes would also be a research priority in future.

The recent evidence signifies the detrimental impacts of climate change on hydroclimatic regimes, globally and regionally. Regardless of the importance of global investigations, regional studies establish repercussions at the lowest administrative level and are practicable for successful execution at the basin level. A comprehensive assessment and compact interpretation of hydroclimatic variability over the TRB for the future have been performed using PCA, a multivariate unsupervised machine learning algorithm. The principal components have been used to identify the regions likely to be critically affected by the disproportionate and co-occurring climate-scale changes in the future.

The current analysis showed that the mean climatology of TRB during the distant future would be wetter and warmer compared with the baseline period. However, owing to more spatial heterogeneity, coherent rainfall patterns are observed in the baseline and near-future periods. Amongst the heterogeneous regions of TRB, MTB showed a profound statement towards the wet regime. In contrast, the temperature assessment gives more homogeneous indications of dramatic warming supported by the mean temperature rise of 1.1–1.8 °C and 1.5–4.0 °C under RCP 4.5 and RCP 8.5 scenarios. This gives rise to various plausible scenarios, like, dry–warm and wet–warm climatology with varying rainfall and temperature characteristics. Considering the former, 22.8% (R4.5-NF) and 41.6% (R8.5-NF) of TRB would experience lower annual rainfall with pronounced non-uniform distribution. In contrast, the latter would expose 47.0% (R4.5-DF) and 30.2% (R8.5-DF) of TRB to higher annual rainfall with high rainfall intensity, long wet spells, and pronounced non-uniform distribution. The assessment also reflected a plausibly better rainfall scenario in the distant future wherein increasing annual rainfall with more uniform temporal distribution is viable over more than 50% of TRB; however, it is exposed to the consequences of severe warming.

The climate variability under substantial warming makes evaluating hydroclimatic risk and agricultural susceptibility/risk to hydroclimatic extremes for TRB an emergent priority that may be an extension of the current study. This study contributes critical information regarding the regional exposure of TRB to projected climate scenarios, which can help devise mitigation strategies. This study considered the two most widely used RCP scenarios; however, including more climate models considering multiple strategies can lead to a more robust inference. The impact of formulating multi-model ensembles on their performance would also require consideration with the inclusion of several climate models. The uncertainties in the climate projections were not considered during the present assessment. The mutual information among the used climate indices is assumed stationary while projecting future components. Also, the present study explicitly does not consider the changes in the concurrent weather events. These are the key limitations of the methodology adopted; however, it is generic in nature and unanimously applicable to other regions.

The authors thank the Indian National Committee on Climate Change (INCCC), under the Department of Water Resources, River Development & Ganga Rejuvenation, Ministry of Jal Shakti, Government of India (GoI), for providing funding and data support for this research [16/22/2016-R&D/3059-3076 dated November 7, 2016]. The authors also appreciate the infrastructural support provided by the Centre of Excellence on ‘Water Resources and Flood Management’, TEQIP-II, Ministry of Education, GoI. The data disseminating agency, i.e., India Meteorological Department, Ministry of Earth Sciences, GoI is also duly acknowledged. Acknowledgements are due to the esteemed editor and reviewers for providing valuable and thoughtful suggestions in improving the quality of this manuscript.

LKG conceptualised the whole article, developed the methodology, analysed the data, and wrote the original draft; PLP rendered support in funding acquisition and data curation, supervised the process, reviewed and edited the article; PVT rendered support in funding acquisition and data curation, supervised the process, reviewed and edited the article.

This work was supported by Indian National Committee on Climate Change (INCCC), Department of Water Resources, River Development & Ganga Rejuvenation, Ministry of Jal Shakti, Government of India [16/22/2016-R&D/3059-3076 dated November 7, 2016].

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.