This study evaluates the relationship between flow variability of unregulated and regulated streamflow stations and global climate indicators. Mann–Kendall and change-point analysis is applied to investigate the gradual and abrupt changes in streamflow data, followed by the investigation of multi-scale fluctuations in streamflow data using Continuous Wavelet Analysis. Linkages between streamflow and global climate indicators are examined using Cross-Wavelet and Wavelet Coherence Analysis. Results showed contrasting trend values for unregulated and regulated streamflow stations. Surprisingly, all unregulated stations experienced a significant abrupt shift in change point contrary to the regulated streamflow. Further, for unregulated stations, streamflow variability and hydroclimatic teleconnections were observed at a lower scale, indicating that variations in streamflow are more frequent and generally occur on an intra-annual to inter-annual scale. Contrary to this, regulated stations observed the streamflow variability and hydroclimatic teleconnections at a larger scale (8–10 years), indicating that all the fluctuations are smoothed out. Thus, unregulated stations cannot be used as a proxy for regulated stations in any given basin. Indeed, for better water resource planning and management, both regulated and unregulated streamflow should be investigated.
Distinct characteristics of unregulated and regulated streamflow variability and hydroclimatic teleconnections.
Intra-annual and annual variability, i.e., 0.5 and 1 year, are prominent in unregulated in contrast with the appearance of intra-decadal scale in the regulated streamflow stations.
Hydroclimatic teleconnections of unregulated stations are prevalent up to inter-annual scale, whereas they existed up to intra-decadal scale for regulated stations.
Gradual or abrupt variation in streamflow directly affects the socio-economic condition of the particular region, and it is essential to understand its variability for appropriate management and planning of the water resources (Ehsanzadeh et al. 2011). The streamflow variability can be analyzed with the help of trend, stationarity, homogeneity, periodicity, and noise present in the hydrologic times series (Drissia et al. 2019). However, many studies showed trend analysis as one of the most widely used techniques for identifying variability in hydrological time series (Rathnayake 2019). Gudmundsson et al. (2019) evaluated the global change in mean and extreme streamflow for more than 30,000 streamflow sites and found the contrasting hydrological change. Sahoo & Jha (2020) examined low flow long-term trends for 14 stations of the Mahanadi river basin in India and observed an increasing trend in the upper Mahanadi River basin. Further, Birsan et al. (2014) found increasing winter and autumn streamflow trends and decreasing summer flow trends in Romanian streamflow. Long-term streamflow variability is attributed to anthropogenic climate change and the influence of large-scale climate drivers (Jiang et al. 2013). Many researchers from India (Sahu et al. 2012; Panda et al. 2013; Arfan et al. 2019) and other parts of the world (Rochdane et al. 2012; Petpongpan et al. 2020) have also established from their study that the changing trends of hydrological variables such as streamflow or rainfall are the causal effects of varying climatic conditions. Therefore, it is essential to study flow variability and its connections with regional climate and atmospheric circulation patterns to develop probabilistic flood and drought forecasting models in assessing hydrologic risk. This would further help in decision-making for disaster management across any particular basin.
Evaluating the impact of large-scale climate patterns on streamflow has gained attention over the years. For example, Lauro et al. (2019) investigated hydroclimatic variability of the Colombian Andes region and associated it with El Niño Southern Oscillation (ENSO) and observed extreme discharge in the study area during El Niño. Based on the study conducted by Burn & Hag Elnur (2002), various drivers, including the Pacific Decadal Oscillation (PDO), North Atlantic Oscillation (NAO), and North Pacific index (NP), affect runoff timing in the Mackenzie River basin in northern Canada. The impact of Indian Ocean Dipole (IOD) and ENSO on extreme streamflow in the Citarum River in Indonesia was studied by Sahu et al. (2012) and established the strong relationship between IOD and seasonal streamflow.
Although large-scale drivers could influence local streamflow at multiple temporal scales, only a few studies are dedicated to quantifying multi-scale flow variability. Wavelet is an emerging technique extending its application to examine the variability in any time series at different scales or frequencies. Wavelet transform has been extensively used worldwide in other research to analyze streamflow variability at multiple temporal scales (Coulibaly & Burn 2004; Labat 2010; Briciu & Mihăilă 2014). Further, this method is also used to understand the mechanisms that drive flow variability and to evaluate the teleconnections of climate drivers (Joshi et al. 2016; Schulte et al. 2016; Rathinasamy et al. 2019). Climate change would have detrimental effects on Asia's densely populated river basins (Immerzeel et al. 2010). Thus, a country like India needs to understand streamflow dynamics since major rivers serve as the primary source of water supply for irrigation, power production, industrial use, and other purposes.
The river basins of the Indian subcontinent region are very sensitive to changing climate scenarios (Panda et al. 2013). Previously, many studies have carried out the analysis for Indian basins in the context of global climate indices influence (like ENSO & IOD) on streamflow variability (Krishna Kumar 1999; Ashok et al. 2001; Gadgil et al. 2003; Chowdhury & Ward 2004; Maity & Nagesh Kumar 2008; Maity & Kashid 2011). It is worth noting that all of the studies mentioned above are based on streamflow variability across a whole basin. However, there are numerous regulated and unregulated streamflow stations in each basin. To understand streamflow variability at a regional or local scale, it is essential to examine the responses of different stations in any basin. Beaupré et al. (2020) and Growns et al. (2000) compared regulated and unregulated streamflow stations. However, the studies for Indian basins demonstrating the response of regulated and unregulated streamflow variability and hydroclimatic teleconnections are scarce (Yeditha et al. 2021). Thus, trend and wavelet analysis methods are used in this study to investigate the streamflow variability of six different unregulated streamflow stations in India and quantify the influence of precipitation and global climate indices at different time scales. Furthermore, the study had expanded to the 40 regulated streamflow stations of the same chosen basins, primarily in the context of hydroclimatic teleconnections, intending to distinguish these relationships for regulated and unregulated streamflow. Thus, the current study deals with the following objectives:
To investigate and quantify the influence of global climate indices on streamflow variability at six unregulated and 40 regulated streamflow stations at different time scales.
To identify the distinction between regulated and unregulated stations regarding streamflow variability and hydroclimatic teleconnections.
STUDY AREA AND DATA
Indian river basins are considered in this study to evaluate the streamflow variability at multiple time scales and to assess its link with the climate drivers. Central Water Commission has divided India into 20 river basins, including 12 primary and eight composite basins. Rivers belonging to the peninsular region of India are primarily non-perennial and mainly fed by rainwater. Because of the peninsular region's stiff and rigid lithology, there is a much lower chance of adding underground water into the surface water. So, understanding the streamflow variability in this region gives more accuracy than the Himalayan River basin, in which streamflow primarily depends on snow, precipitation, and underground water. So, by considering these facts, the present study assessed six major river basins of the peninsular region of India, including Brahmani, Cauvery, Mahanadi, Narmada, Godavari, and Subarnarekha. Six unregulated stations and 40 regulated stations are selected from the selected river basins (Figure 1). Refer to the supplementary file (Table S1 and S2) for detailed information on the streamflow stations. Streamflow data sets for the selected stations are obtained from the Water Resources Information System (WRIS) India website http://indiawris.gov.in/wris/ for 41 years from 1975 to 2015. The daily streamflow data are converted to monthly time series using the downsample toolbox in MATLAB 2019a.
Climate indices data, namely, IOD, NAO, ENSO (Niño3.4), and PDO are used to investigate the hydroclimatic teleconnections on the regulated and unregulated streamflow station of the six Indian peninsular basins. The IOD Mode Index is calculated by subtracting the SST anomalies from the western (10°S–10°N, 50°E–70°E) and eastern (10°S–0, 90°E to 110°E) tropical Indian Oceans (Saji et al. 1999). The NAO is calculated based on the sea surface temperature between the North Atlantic Ocean subpolar low and high subtropical regions. The Niño3.4 index, which represents the ENSO and has better correspondence with the Indian monsoon rainfall, is computed by averaging the sea surface temperatures (SSTs) anomalies over the Niño 3.4 region (5°N–5°S, 120°W–170°W). Pacific Decadal Oscillations (PDO) is a climate index that occurs on a larger scale, with patterns primarily observed in the North Pacific Ocean. Its phases are classified as warm or cool based on temperature changes. It was often observed that large-scale patterns of atmospheric circulation like IOD (Guan & Yamagata 2003; Behera et al. 2006), NAO (Hurrell 2001; Visbeck et al. 2001), ENSO (Neelin et al. 1998; McLaughlin et al. 2014), and PDO (Mantua & Hare 2002; D'Arrigo & Wilson 2006) determine the spatio-temporal distributions of crucial atmospheric variables like surface temperature and precipitation over the land surface, which in turn control the hydrologic cycle. For this study, global climate indices data are downloaded from the website https://www.esrl.noaa.gov/psd/data/climateindices/list/
METHODS AND METHODOLOGY
Streamflow and global climate indices are the variables investigated in this study. Streamflow provides information on the overall response of the river basin, and climate indices influence streamflow variability. The temporal analysis of streamflow will be more concentrated in this study; however, it is necessary to understand the gradual and abrupt changes before understanding the temporal variation. So before proceeding straight to the temporal analysis trend, step-change analysis was performed to evaluate the significant gradual and abrupt changes in streamflow. Wavelet transforms due to its potential competence in evaluating the temporal variability at a varying frequency of time series signal. It is employed in this study for temporal analysis at multiple time scale and identifying hydroclimatic teleconnections. The flowchart in Figure 2 depicts the step-by-step methodology used in this study to examine streamflow variability and quantify the influence of hydroclimatic teleconnections across six river basins in India. Here, in this section, the brief methodology adopted in the study is discussed, and then the different methods used for research are explained.
Daily streamflow data of 41 years from 1975 to 2015 of six unregulated and 40 regulated stations in India are obtained from the Center Water Commission (CWC) Board of India. For the same period, data of four selected climate indices (IOD, NAO, Niño3.4, & PDO) are also collected.
Variability in both unregulated and regulated streamflow is checked in terms of gradual and abrupt change. For finding gradual or monotonic change in streamflow data, the non-parametric Mann–Kendall trend test is used, while for abrupt or step change in streamflow data, change-point analysis is performed. MATLAB has been used for trend and step-change analysis.
For further investigation, the multi-scale variability in unregulated and regulated streamflow, wavelet analysis is used. Continuous wavelet analysis is performed to identify the multi-scale variability in streamflow data.
Cross-wavelet and wavelet coherence analyses are employed to identify the coherency and correlation of streamflow with precipitation and global climate indices.
Mann–Kendall trend test
To determine the monotonic or the gradual trend in unregulated streamflow series for each gauging station, the Mann–Kendall (MK) test is used. The MK test capabilities were robust in detecting trends. Many researchers worldwide in their study used the MK test for finding the trend in many domains, including the trend in hydrologic data (Burn & Hag Elnur 2002; Hamed 2009), temperature data (Alhaji et al. 2018), water quality parameter (Kisi & Ay 2014), meteorological variables (Gocic & Trajkovic 2013), evapotranspiration data (Shadmani et al. 2012), and precipitation data (Ahmad et al. 2015; Guntu et al. 2020). The Mann–Kendall test is a statistical test that is frequently applied for climate and hydrology trend analysis. The use of this test has two benefits. First, it is a non-parametric test and does not require the normal distribution of the data. Secondly, due to inhomogeneous time series, this test is sensitive to abrupt breaks. The data are evaluated as a sequence of times. The following data values are compared with each data value in time series (Alhaji et al. 2018).
A very high positive value of S indicates an upward trend, while a low negative value indicates a downward trend. To statistically quantify the significance of the trend, the probability associated with S and the sample size, n, must be computed.
Change points are abrupt variations in time series data. Such abrupt changes could represent transitions between states. The detection of change points is helpful in time series modelling and prediction. Time series data are time series measurements that describe the behaviour of systems. These behaviours may change over time due to external influences, events, or systematic internal distribution changes (Kawahara & Sugiyama 2012). Change-point detection (CPD) is the problem of finding abrupt changes in data when a property of the time series changes (Reeves et al. 2007). Along with hydrologic data, the CPD method has been studied for a variety of purposes over the last several decades, for example, medical condition monitoring (Yang et al. 2006), speech recognition (Rybach et al. 2009), image analysis (Radke et al. 2005), human activity analysis (Tran 2019), and climate change detection (Ducré-Robitaille et al. 2003; Reeves et al. 2007; Itoh & Kurths 2010). In this study, a Changepoint analyzer provided by Taylor (2000) and software MATLAB2019a were used to find the abrupt changes in unregulated streamflow data.
Continuous wavelet transform
Because hydrologic data are non-stationary, their frequency content varies over time. Time-frequency decomposition is one of the methods for obtaining changing frequency information in any time series. Various methods for time-frequency analysis of non-stationary signals exist. The popular short-time Fourier transform (STFT) method generates a time-frequency spectrum by performing the Fourier transform over a specified time window (Cohen 1995). But, in STFT, the time-frequency resolution is constant, which is not suitable for non-stationary data, whose parameters are changing over time. Over the past two decades, the wavelet transform has been applied in many branches of science and engineering. Unlike the STFT, which has a constant resolution at all times and frequencies, the wavelet transform has a good time and poor frequency resolution at high frequencies and good frequency and insufficient time resolution at low frequencies (Daubechies 1990).
The continuous wavelet transform (CWT) is a part of the wavelet analysis method, primarily used to display and analyze signal characteristics that vary with time and scale. As a result, it can be a valuable tool for detecting and identifying signals with unusual spectral characteristics, transient information content, or other non-stationary properties (Sadowsky 1996). At both time and frequency resolutions, the CWT provides helpful information about any time series.
The function (t), which can be natural or complex, plays the role of a convolution kernel and is called a wavelet. The parameter a can be interpreted as a dilation (a>1) or contraction (a<1) factor of the wavelet function corresponding to different scales of observation. The parameter can be interpreted as a temporal translation or shift of the function which allows the study of the signal f (t) locally around the time .
The wavelet function has the following properties.
Here is the CWT of the time series X considered. The complex conjugate of the CWT of time series Y. Cross-wavelet constructs the relationship between two CWTs with high standard power (covariance) of the two time series (Grinsted et al. 2004). The significant covariance of these time series can be observed as red noise in the wavelet scalograms.
Wavelet coherence analysis
In the above equation,
R (y, x) – wavelet coherence between y and x; (y, x) – squared wavelet coherence between y and x; W (y, x) denotes the corresponding cross-wavelet transforms; denotes smoothing factor.
Wavelet coherence ranges from 0 to 1. If the correlation coefficient value is closer to 1, there is more correlation between the two time series and vice versa.
The time-frequency analysis of streamflow data from both types of stations was used to identify streamflow variability of unregulated and regulated stations in terms of monotonic and abrupt change. Finally, four climate indicators are used to explore hydroclimatic teleconnections of streamflow. Using the Mann–Kendall trend test, only two unregulated stations from the Cauvery and Narmada basins exhibited a significant long-term negative trend. In contrast, most regulated stations in the Brahmani, Cauvery, and Godavari river basins showed a significant negative trend, while Mahanadi, Narmada, and Subarnarekha river basins observed a positive trend. Significant abrupt variations in regulated streamflow are rare; nonetheless, all unregulated stations observed significant abrupt changes in their streamflow data.
All unregulated streamflow stations observed strong features at the intra-annual (0.5 years) and annual (1 year) time scales. However, most regulated stations are dominated by oscillations from the annual to the intra-decadal scale rather than intra-annual variability. The cross-wavelet and wavelet coherence plots between streamflow and several climate indices demonstrate that the Niño3.4 index has the strongest feature at lower and higher frequencies. The association between unregulated streamflow and all global climate indices was observed up to inter-annual scale for the unregulated station, but it was observed up to intra-decadal scale for the regulated station.
The following sections provide a complete description of all of the findings.
Long-term trend and CPD for unregulated and regulated stations
The gradual and abrupt changes in streamflow data for six unregulated and 40 regulated stations from 1975 to 2015 are estimated using MK trend analysis and step-change analysis. Table 1 presents the Z value, p-value obtained from the MK test, and step-change year from the step-change detection technique. From the values tabulated, only two unregulated stations belonging to the Cauvery and Narmada basins show a significant long-term negative trend at the 5% significance level. Three out of four stations in Brahmani and Cauvery basins and 12 out of 16 stations in the Godavari river basin show a significant negative trend in regulated streamflow stations. Further, three and two stations out of eight in the Mahanadi basin show positive and negative trends, respectively. In addition, all the regulated streamflow stations in Narmada and Subarnarekha basins display positive trends.
|River .||Station .||Indicated as .||Z value .||P-value .||Significance .||Step change .|
|Cauvery||Kudige||CK||−3.95||7.69 × 10−5||**||1985−|
|Brahmani||Altuma||BAL||−4.95||7.05 × 10−7||**||1994−|
|Cauvery||Biligundulu||CBI||−5.06||4.14 × 10−7||**||−|
|Kollegal||CKO||−6.64||3.08 × 10−11||**||−|
|Polavaram||GPO||−8.07||6.66 × 10−16||**||−|
|Purna||GPU||6.86||6.52 × 10−12||*||−|
|Saigaun||GSI||−7.07||1.48 × 10−12||**||−|
|River .||Station .||Indicated as .||Z value .||P-value .||Significance .||Step change .|
|Cauvery||Kudige||CK||−3.95||7.69 × 10−5||**||1985−|
|Brahmani||Altuma||BAL||−4.95||7.05 × 10−7||**||1994−|
|Cauvery||Biligundulu||CBI||−5.06||4.14 × 10−7||**||−|
|Kollegal||CKO||−6.64||3.08 × 10−11||**||−|
|Polavaram||GPO||−8.07||6.66 × 10−16||**||−|
|Purna||GPU||6.86||6.52 × 10−12||*||−|
|Saigaun||GSI||−7.07||1.48 × 10−12||**||−|
Note: ‘–’represents no significant trend, ‘*’ represents a significant positive trend, and ‘**’ represents a significant negative trend. Step change denotes the year in which change has occurred, and ‘−’ indicates a negative, and ‘+’ indicates a positive shift.
Step-change analysis reveals the significant negative abrupt change for the unregulated stations of Brahmani, Cauvery, Mahanadi, and Narmada basins in 2015, 1985, 2007, and 2006, respectively. In addition, a positive abrupt shift is observed for the Godavari and Subarnarekha river basins in 1992 and 1985. Surprisingly, significant abrupt changes in regulated streamflow stations are much less common. Even the shifts observed are all negative except the regulated streamflow stations in the Subarnarekha basin. Interestingly, an abrupt shift is observed in the mid-1990s for most of the stations in Brahmani and Godavari basins. One out of four stations in Cauvery (CMU) and one out of eight stations in the Mahanadi (MTI) basin experienced an abrupt negative shift. Finally, no regulated stations in the Narmada basin show any significant shifts.
Multi-scale temporal analysis
Daily streamflow data from six unregulated stations and 40 regulated stations are decomposed using continuous wavelet transforms (CWT) to determine the scale-specific dominant parameters which might cause the trend and step change in streamflow data. This is followed by cross-wavelet and wavelet coherence analysis determining the hydroclimatic teleconnections of four global climate indices.
The CWT was used to decompose unregulated streamflow data from six stations. Figure 3 depicts the scalograms obtained from the continuous wavelet analysis. All unregulated stations studied (BA, CK, GT, NG, MB, and SA) had high intra-annual, annual, and inter-annual variability (0.5, 1, 2, and 4 years). Significant intra-annual oscillations of 0.5 years were observed primarily for stations BA, GT, NG, and SA. However, at this scale, such oscillations were absent at stations CK and MB. Except for stations CK and MB, which had some discontinuity from 1985 to 2000, annual fluctuations were active for all stations during the study period. Significant oscillations on an inter-annual scale (2 and 4 years) were observed at stations BA, MB, NG, and SA, primarily between 1990 and 2000.
Similar to unregulated stations, scalograms are developed for 40 regulated streamflow stations. All the scalograms of regulated stations are included in the supplementary section (Fig. S1) in the same order as listed in Table 1 for brevity purposes. All four stations except station BTI, from the Brahmani basin, observed high variability in streamflow from 0.5 to 8 years of scale. These oscillations are mostly predominant from 1985 to 2000 in stations BAL and BJE. Whereas, for the station, BJA streamflow variation can be observed throughout the study period. In the case of the Cauvery basin, all stations showed significant oscillation at the annual (1 year) scale. These oscillations have appeared in three ranges from 1975 to 1985, 1990 to 2000, and 2005 to 2010. Both intra-annual and inter-annual fluctuations are absent in the case of the Cauvery basin; however, some stations observed intra-decadal oscillation of 8 years of scale in the year 2000–2005. In the case of the Godavari basin, all 16 stations showed annual oscillation in streamflow; however, some stations also observed inter-annual and intra-decadal oscillation. Significant oscillation at intra-annual and annual scales can be observed for the Mahanadi basin. However, most of the stations also observed significant oscillation at inter-annual scale (up to 8 years). The station from Narmada and Subarnarekha basin observed these oscillations up to inter-annual scale, in which the Subarnarekha basin observed it up to 8 years of scale. In contrast, oscillation for the Narmada basin is restricted at 4 years of scale.
Hydroclimatic teleconnections with unregulated and regulated streamflow
Indian ocean dipole
The XWT plot (Fig. S2 and Figure 4) revealed a strong relationship between IOD and streamflow. From 1980 to 2010, all unregulated stations (BA, CK, GT, MB, NG, and SA) showed a strong yearly connection. Surprisingly, the annual correlation appears to be vanishing from 2000 to 2005 and has been nonexistent in recent years. But along with smaller scales, regulated stations from Brahmani, Cauvery, Godavari, Mahanadi, and Subarnarekha observed some linkage with IOD at the intra-decadal scale (8 years). Mostly, the significant correlations are intermittent, but they seem to be continuous from 1985 to 2000 for most of the regulated stations. The direction of arrows in the cross-wavelet plot is not in one direction. So for a better understanding of coherence between streamflow and IOD, wavelet coherence analysis (WTC) (Fig. S3 and Figure 5) was performed. Significant areas are transient with concise duration and mainly occurred at 0.5–1, 2–4, and 4–8 years of scale. Although the IOD is an essential modulator of Indian monsoon rainfall (Ashok et al. 2001), its impact on streamflow variability at the regional scale is different. This research also revealed a complex link between IOD and streamflow of distinct basins at different scales and periods. From the direction of arrows, it can be observed that streamflow of the Brahmani basin has an in-phase relationship with IOD at the annual scale. However, still, it is in an anti-phase relationship at an inter-annual and intra-decadal scale. Most significant areas from Cauvery and Godavari showed an anti-phase relationship, whereas stations from Narmada and Subarnarekha observed an in-phase relationship with IOD. Mahanadi basin kept an anti-phase relationship up to 4 years of scale and an in-phase relationship with IOD at 4–8 years of scale. For the majority of the station, an anti-phase relationship can be observed from 1985 to 2000.
North Atlantic Oscillation
In-phase associations were observed between unregulated with NAO at the intra-annual (0.5 years) and annual (1 year) scales in XWT analysis (Fig. S4) from 1980 to 2010, with breaks in 1985, 1995, 2000, and 2005. The WTC analysis (Fig S5) produced similar results; however, stations such as GT, MB, and NG showed significant correlation at 2- and 4-year time scales.
In the case of the regulated station, except stations from the Cauvery basin, there was a significant but intermittent intra-annual correlation for all stations observed in the XWT plot (Fig. S6). Annual correlations are present for all stations of each basin, but the significant areas of high correlation are absent between the year from 1980 to 1990 at this scale. Inter-annual and intra-decadal correlation can be observed for all stations of the Brahmani and Subarnarekha basin. In contrast, only a few stations from other basins observed a significant correlation area at these particular scales, in the case of the WTC plot (Fig. S7), in which most of the stations from the Brahmani basin showed significant coherence with NAO up to 4 years of scale except for station BAL, which observed this only at the annual scale. All stations from the Cauvery basin showed almost the same coherency at intra-annual, annual, and inter-annual time scales. In contrast, stations from the Godavari basin observed distinct coherency at different time scales; for example, a station like GAS, GJA, GNO, GPA, and GPU observed significant coherency at inter decadal scale whereas, for another station, coherency can be seen only up to 4 years of scale. These similar observations can also be seen for the Mahanadi basin. The station from the Narmada basin observed the significant coherency up to 8 years of scale, whereas Subarnarekha observed this only up to 4 years of scale. In the case of the phase relationship between streamflow data and NAO, there is an in-phase relationship between Brahmani and NAO. Still, an anti-phase relationship can also be observed for these stations in 1980. The most significant area showed the upward movement of arrows for the Cauvery basin. The majority of the Godavari, Mahanadi, and Narmada stations showed an in-phase relationship, whereas Subarnarekha observed a mixed (in-phase, anti-phase, and lagged) relationship with NAO.
The study found that Niño3.4 has the strongest association with unregulated streamflow on an annual scale compared to other indices. Based on the XWT plot (Figure 6), unregulated stations such as BA, CK, MB, and SA are highly correlated with Niño3.4 throughout the period (1975–2010), whereas GT and NG have some vanishing features from 1995 to 2005. The WTC plot (Figure 7) observed periodicities at intra-annual, inter-annual, and intra-decadal scales along with the annual scale.
The XWT plot (Figure 8) shows a significantly high correlation between Niño 3.4 and streamflow stations from an annual to an intra-decadal scale for regulated stations. Annual correlations are intermittent and mostly absent in the years 1982, 1988, and 1997. Inter-annual and intra-decadal correlations are continuous for the period from 1980 to 2005. Interestingly none of the stations from any basins observe any intra-annual correlation with Niño 3.4 except the appearance of some covariance between 0.5 and 1-year scale can be observed for the entire basin in the WTC plot (Figure 9). For the Brahmani basin, a significant area with high coherency can be seen at 2–4 years of scale; however, station BJA was also observed at 4 to 8 years of scale from 1995 to 2005. Cauvery basin is not influenced by Niño 3.4 at annual, inter-annual, and decadal scale, but some intermittent significance coherency can be observed for inter-annual and intra-decadal scales. The station from the Mahanadi basin also observed some small significant areas from 0.5 to 8 years of scale.
Similarly, the Narmada basin showed coherency up to 4 years of scale, whereas Subarnarekha observed a highly significant area with high covariance at 4–8 years of scale. Overall, Niño 3.4 influences the streamflow of the chosen basin with a very short-lived significant region. These regions can be found in the years 1978, 1982, 1985, 1992, and 2006 for smaller scale, and for larger scale, these can be found in 1990–2000 and 2005. The direction of arrows is not in one direction in both XWT and WTC plots. Still, most insignificant regions showed an anti-phase relationship with Brahmani and Narmada, Leading with Cauvery, and in-phase relationship with Godavari, Mahanadi, and Subarnarekha basins.
Pacific decadal oscillation
PDO indices showed a leading association with unregulated streamflow at a 1-year scale in the XWT plot (Fig. S8). However, stations GT, MB, and SA observed these periodicities at 2, 4, and 8 years of scale in the WTC plot (Figure 9), in addition to 1 year of scale. These associations can be seen primarily between 1985 and 2000.
In the case of regulated stations, the XWT plot (Fig. S10) between PDO and streamflow give a similar result as of Niño 3.4 and observe the significant correlation from annual to intra-decadal scale; however, the WTC plot (Fig. S11) provides some different features of covariance with PDO, for example, Brahmani and Cauvery basin observe significant coherence at 0.5, 1, and 2–4 years of scale. All the considerable areas are short-lived and intermittent. Each station from the Cauvery basin observed similar coherence with PDO. Stations from Godavari basin and Mahanadi observed these coherencies mainly at inter-annual and intra-decadal scale, mostly from 1985 to 2000 and small coherency at 0.5 and 1 year of scale. Narmada and Subarnarekha basins also observed coherency at 0.5, 1, 2, and 4 years of scale. These covariances are of short duration up to the annual scale, but from 2 to 4 years of scale, these significant regions ranged mainly from 1985 to 2000. From the direction of the arrow, it can be observed that the station belonging to Brahmani, Cauvery, Narmada, and Subarnarekha showed an in-phase relationship with PDO. In contrast, the Godavari basin observed an anti-phase relationship. The station from the Mahanadi basin showed an in-phase relationship at the inter-annual scale, whereas there was an anti-phase relationship at the intra-decadal scale.
Variation in streamflow can profoundly impact the national or possibly global spatial-temporal structure of water availability (Pahl-Wostl 2019). Frequent severe droughts or flood conditions directly impact water availability and significantly impact India's ecology and socio-economic growth (Rajamani 2009). This study identified streamflow variability of unregulated and regulated stations in terms of monotonic and abrupt change, followed by the time-frequency analysis of streamflow data of both types of stations. Finally, hydroclimatic teleconnections of streamflow with four climate indices are investigated.
Based on the Mann–Kendall trend test, at a 5% significance level, only two unregulated stations from the Cauvery and Narmada basins revealed a significant long-term negative trend. In contrast, most regulated stations in the Brahmani, Cauvery, and Godavari river basins showed a significant negative trend. In contrast, a significant positive trend was observed in Mahanadi, Narmada, and Subarnarekha river basins. The above results related to the Godavari and Subarnarekha basin are concurrent to the findings of various studies. Das (2019) and Kuriqi et al. (2020) studied the streamflow of various stations in the Godavari basin and concluded that upstream gauging stations show no significant trend. Despite this, the majority of the downstream stations recorded a significant downward trend in streamflow data. Similarly, upstream gauging sites in the Subarnarekha basin showed no trend, but downward gauging sites showed an increasingly significant trend. In the case of the Mahanadi basin, the results are in congruence with the findings of Murthy et al. (2017) and Panda et al. (2013), in which authors concluded a significant decreasing trend in the streamflow data.
Several major and minor storage projects have existed in all the six basins considered as characterized by Jain et al. (2017). For example, in the Brahmani basin, some major and medium projects were developed in the 1960s, 1980s, and 1990s. Several medium-sized storage projects were constructed in the Cauvery basin throughout the 1930s and 1950s and a few in the 1970s. The Godavari basin covers more than 0.3 million square kilometres and consists of several medium-sized storage projects built during the 1980s and 1990s. Still, no significant storage projects have been done since 2001. In 1958, the Hirakud dam, a major project in the Mahanadi basin with a live storage capacity of 5.8 BCM, was completed. Following that, plenty of smaller projects proliferated. During the 1980s and 1990s, many medium-sized storage projects were established in the Narmada basin. Subarnarekha basin also contained several major and minor projects. Despite this, significant abrupt changes in regulated streamflow are much less common; on the other hand, all unregulated stations experienced a significant abrupt change in their streamflow data. Although unregulated catchments are not supposed to be influenced by hydropower plants, dams, or other reservoirs, such drastic changes in unregulated streamflow suggest that there may be other factors influencing variations in streamflow for these basins. According to Das et al. (2018), the main reason for the step-change in streamflow for the Subarnarekha River could be the strong influence of other nearby basins. Das et al. (2018) reported that urbanization and changes in land use land cover have a significant impact on streamflow alteration in addition to anthropogenic activity. Furthermore, they discussed increasing population density over these basins, causing cropland to be converted into built-up areas, causing surface water infiltration to be hampered, and increasing runoff.
CWT was used to understand better streamflow variability for the temporal analysis of streamflow data. All unregulated streamflow stations observed strong features at the intra-annual (0.5 years) and annual (1 year) time scales. Throughout the study period, the majority of the stations displayed a prominent feature at these scales. However, rather than intra-annual variability, oscillations from the annual to the intra-decadal scale dominate most regulated stations. Stations in regulated catchments observed a tiny region on a 0.5-year scale.
The IOD is essential for determining hydrologic factors such as streamflow and precipitation and acting as a modulator of the Indian monsoon (Ashok et al. 2001). This paper revealed that IOD had a minimal effect on streamflow in the basins studied, showing the spatial diversity of Indian streamflow teleconnections. This may be because IOD does not influence the hydrologic variables of the Indian basin compared with ENSO, which is also established by many studies (Ashok et al. 2003; Saji & Yamagata 2003; Marchant et al. 2007). In the absence of ENSO, Crétat et al. (2017) and Rathinasamy et al. (2019) investigated the impact of IOD on the Indian Summer Monsoon. They discovered that it did not push the monsoon circulation. Pokhrel et al. (2012) also found that the combined influence of ENSO and IOD improved forecasting of the Indian summer monsoon. It highlighted the necessity to investigate further these two indices’ combined effect on streamflow variability in the Indian peninsular basin.
The essential areas of the vital feature in XWT and WTC are less with NAO than other indices, indicating less NAO dominance in these basins. A prior study by Yadav et al. (2009) supports this claim. The pressure in the polar region drops during the positive phase of NAO. In contrast, the pressure in the mid-latitudes (Indian subcontinent) rises. This positive phase of the NAO strengthens the western disturbance by intensifying the westerly jet stream over the Middle East (Nageswararao et al. 2016). Heavy rainfall in the winter season amplifies these disturbances, primarily impacting the northwest portion of the Indian subcontinent. Our research area is centred on the peninsular region, which has experienced less western disturbance and is unaffected by the NAO. Kurths et al. (2019) also found a weaker relationship between NAO and hydrologic factors in India's southern area.
Hydroclimatic teleconnections between streamflow and four climate indices unravel maximum significant correlation with Niño 3.4, followed by the PDO index. PDO influences on six basins were similar to Niño 3.4, highlighting the possibility of proxy connections (Rathinasamy et al. 2019). The relationship between PDO indices and Indian River basins is still fully explored, except for a study by Krishnamurthy & Krishnamurthy (2014), who found that the warm and cold phases of PDO influence PDO ENSO conditions, affecting streamflow and precipitation in Indian basins. This could be related to the warm phase of the PDO, which raises the Pacific Ocean's sea surface temperature, altering trade winds and boosting the Walker Circulation, amplifying the Niño state (Agarwal 2019). Usually, moisture-driven currents move towards land due to pressure and temperature differences between the oceanic surface and the landmass of the Indian subcontinent. Still, during the Niño3.4 condition, the warm phase of ENSO, SSTs rise. As a result, the temperature difference between landmass and sea decreases, further hindering adequate moisture transport. A number of studies (Maity et al. 2007; Panda et al. 2013; Jena et al. 2014) established a positive relationship between streamflow and precipitation, implying that indices like PDO (positive phase) and Niño3.4 both have a negative impact on summer monsoon rainfall and thus reduce streamflow in the study area.
Comparison between regulated and unregulated streamflow variability and their hydroclimatic teleconnections
In unregulated catchments, all considered stations showed high variability at intra-annual, annual, and inter-annual scales. Oscillations at 0.5 and 1-year scales are more dominant and appeared throughout the study for almost all the stations. After the 1990s, the significant areas with oscillations at 2–4 years of scale can also be observed. But interestingly, for regulated stations rather than intra-annual variability, oscillations from annual to intra-decadal scale are dominant for most stations. Stations that belong to regulated catchments exhibit a less significant region at 0.5 years of scale. At annual and inter-annual scales, oscillations observed a similar pattern as unregulated, but regulated stations also observed some additional oscillation at the intra-decadal scale (8 years).
In hydroclimatic teleconnections, streamflow of unregulated stations is influenced mainly by climate indices at intra-annual, annual, and inter-annual scales. Mostly annual correlations are active between IOD and unregulated streamflow of Brahmani, Cauvery, Godavari, and Subarnarekha basin, but for Mahanadi and Narmada, inter-annual correlations can also be observed. With NAO, intra-annual and annual correlations are predominant, but some significant areas with high inter-annual correlations are also present. Similarly, both Niño 3.4 and PDO influence the streamflow from intra-annual to inter-annual scale. However, by looking at cross-wavelet and wavelet coherence plots between climatic indices and regulated streamflow, along with smaller scales, significant dominant connections at the intra-decadal scale (8 years) can also be observed. The oscillations at a larger scale can be observed for almost all basins except Mahanadi and Narmada, which do not observe a significant correlation with IOD and Niño 3.4 at the intra-decadal scale; however, the same station is showing correlation with NAO and PDO at this particular scale.
Previously, many basin-wise studies for the Indian peninsular regions were conducted (Panda et al. 2013; Jain et al. 2017; Murthy et al. 2017; Das 2019). However, the majority of the studies focus solely on regulated stations. Some works are available for unregulated stations as well, but none provide detailed information. In this study, streamflow variability and its response to climate indices for regulated and unregulated stations were identified separately.
Interestingly, in the context of streamflow variability and hydroclimatic teleconnections, this study observed the distinct behaviour of unregulated and regulated streamflow stations. Overall, streamflow variability and hydroclimatic teleconnections can be observed at a lower scale or a higher frequency for unregulated stations, indicating that variations in streamflow are more frequent and generally occur from intra-annual to inter-annual scale. The global climate indices also showed a correlation with unregulated streamflow at these scales, since climate drivers, such as IOD, NAO, and Niño 3.4, are periodic and usually appear every 2–3 years. This conveys the impression that climatic phenomena primarily influence streamflow at unregulated stations.
On the other hand, streamflow variability at the regulated station does not observe the significant oscillations at 0.5 years of scale, rather than the variation at a larger scale or lower frequencies (8–10 years). The hydroclimatic teleconnections also showed the correlation up to intra-decadal and decadal scale for most stations, which indicates that the variation in streamflow in regulated stations is not as frequent as unregulated ones. It also gives an idea that along with climatic patterns, some other factors, such as land use, land cover, population density, and other anthropogenic activity, are also predominately influenced by the streamflow at regulated stations, as suggested by Das et al. (2018).
By considering all these results obtained from both unregulated and regulated stations, the study concludes that both types of stations respond differently in any particular basin. So, the regulated station cannot be considered as a proxy of unregulated stations. Thus, for better planning and management of water resources, both regulated and unregulated streamflow should be examined. However, understanding the streamflow variability at any unregulated station provides a comparatively better understanding because, in the regulated station, regulated hydropower plants, dams, or any different structure alter the natural flow regime and affect the streamflow variability (Milner et al. 2019). The unregulated catchment, on the other hand, is unaffected by such changes. These water resource structures, such as dams or reservoirs, are typically built near the unregulated catchment's downstream end. Thus, unregulated stations, once again, provide a genuine image of any catchment and are useful for various practical applications like drainage structure design, runoff forecasts, and water distribution at the downstream end (Swain & Patra 2017).
This study attempted to investigate the nature of runoff variability among six different catchments of India and analyzed the influence of climate teleconnections on runoff variability. The study was carried out for six unregulated stations and 40 regulated stations from these six chosen basins. Mann–Kendall and step-change analysis was carried out to detect the gradual long-term and abrupt changes in the streamflow time series. Wavelet analysis was used to understand the multi-scale association of basin streamflow with the precipitation and climatic indices. The summary of significant findings from the study is as follows:
Based on the Mann–Kendall test and step-change analysis results, only two unregulated Narmada and Cauvery stations showed significant adverse changes in trend. In contrast, most regulated stations in the Brahmani, Cauvery, and Godavari river basins showed a significant negative trend. In contrast, a significant positive trend was observed in Mahanadi, Narmada, and Subarnarekha river basins. All unregulated stations experienced a significant abrupt change in their streamflow data. On the other hand, significant abrupt changes in regulated streamflow are much less common.
All CWT spectra of unregulated streamflow consistently show dominant periodicities at 0.5 and 1 year of band. In contrast, regulated stations are also observed streamflow variability at intra-decadal (8 years of scale), which suggests that higher (lower) frequency components might affect the streamflow variability of unregulated (regulated) streamflow stations.
The cross-wavelet and wavelet coherence plot between streamflow and different climate indices reveal that maximum strong feature at lower and higher frequencies is observed with NIÑO 3.4 index. These plots also show an in-phase relationship at the starting year and a lagged relationship at the end year with streamflow. Similarly, IOD and NAO show an in-phase relationship, and PDO indices showed a leading relationship with streamflow. All the global climate indices observed a relationship with unregulated streamflow up to inter-annual scale, whereas it can be observed up to intra-decadal scale for the regulated station.
Finally, the findings showed that both unregulated and regulated stations act differently in the context of streamflow variability and hydroclimatic teleconnections in any given basin. As a result, regulated stations cannot be used as a substitute for unregulated stations. Thus, both stations should be studied for improved water resource planning and management of any basin.
AA, AG, and TP acknowledge the funding support provided by the Indian Institute of Technology Roorkee through Faculty Initiation Grant number IITR/SRIC/1808/F.I.G and COPREPARE project funded by UGC and DAAD under the IGP 2020–2024.
The authors declare that they have no conflict of interest.
DATA AVAILABILITY STATEMENT
All relevant data are included in the paper or its Supplementary Information.