Relative influence of ENSO, IOD, and AMO over spatiotemporal variability of hydroclimatic extremes in Narmada basin, India

Floods and droughts have been becoming more frequent in changing patterns of climate (Hirabayashi et al. 2008). Studies on climate model projections suggest that there will be more frequent and intense climate extremes in the projected warming future with increasing greenhouse gases emission (Canadell et al. 2022). Therefore, an improved understanding of the global climate system and associated extremes at the regional scale may help in reducing risk (Dai 2011; Field et al. 2012; Rummukainen 2012; Alfieri et al. 2017; Dosio et al. 2018; Naumann et al. 2018). The implications of climate change may vary from local to global scale. Therefore, it becomes important to study the spatiotemporal variability of major hydroclimatic extremes at the basin scale for a better understanding of the impact of climate change on the local water resource system and to take consequent adaptation and mitigation measures (Panda et al. 2013; Garner et al. 2015). These hydroclimatic extremes (droughts, heatwaves, floods, etc.) significantly affect the productivity of the agriculture-dependent economy of Madhya Pradesh with rainfall as a major source of irrigation (Thomas et al. 2015). Apart from local factors (urbanization, orography, etc.), regional precipitation/temperature patterns have also been observed linked with large-scale climate oscillations (LSCOs) (Krishna et al. 1999; Saji et al. 1999; Rose & LaRow 2016). The LSCOs or teleconnections are recognized as a known significant connection between local hydrometeorology and large-scale oceanic-atmospheric circulation patterns, spread across the globe (Maity et al. 2007; Bracken et al. 2018; Jha et al. 2021). Therefore, it is important to study the relative influence of major LSCOs (EL Nino-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD), and Atlantic Multidecadal Oscillation (AMO)) over hydroclimatic extremes at the basin scale to ascertain the probable variations in them.

The ENSO, coupled oceanic-atmospheric mode of the tropical Pacific Ocean, has been studied extensively across the globe for its strong influence over climate drivers (Sahu et al. 2012; Gao et al. 2017; Agilan & Umamahesh 2018; Khaki et al. 2018; Naveendrakumar et al. 2019; Song et al. 2019). The positive (El Nino) and negative (La-Nina) phases of ENSO cause deficit and excess rainfall over a major part of India. The extreme phases of ENSO have been found linked with extreme flooding and droughts in various regions including India across the globe (Trenberth & Guillemot 1996). Likewise, El Nino and La Nina conditions in the tropical Pacific Ocean, and the Indian Ocean also experience a physical phenomenon known as IOD. IOD is the measure of the difference in sea surface temperature (SST) between the tropical western Indian Ocean and the south-eastern Indian Ocean (Saji et al. 1999). The positive and negative phases of IOD represent the warming and cooling of the surface water of the Indian Ocean. The warming (cooling) phases of the tropical western Indian Ocean signify the excess (deficit) precipitation over India (Ashok et al. 2001; Vinayachandran et al. 2002). It has also been found that cool SST over the south equatorial Indian Ocean had modulated the precipitation pattern over Central India during the period 1982–2003, also the continuous warming of the tropical Indian Ocean may lead to extreme precipitation over India in the future (AjayaMohan & Rao 2008). In reference to the Indian continent, another important teleconnection is AMO which is related to North Atlantic decadal variability. The Indian summer monsoon rainfall has been found positively modulated by AMO (Ratna et al. 2021). For a detailed study of AMO influence over Indian summer monsoon rainfall, several other pieces of literature can be considered (Vittal et al. 2020; Ratna et al. 2021).

Climate extremes pose a fundamental threat to places, species, and people's livelihoods. Some standard methods to define extreme events are: consideration of record-breaking events (Coumou et al. 2013; Beniston 2015), identification of extremes from percentile-based approaches (Russo et al. 2015; Schär et al. 2016), extreme value theory (EVT). EVT has been used extensively to interpret climate extremes at regional and global scales on the basis of various standardized indices (Trenberth & Fasullo 2012; Katz 2013; Goswami et al. 2018; Singh et al. 2023). There are 27 indices available from ‘climdex.org’, all of which are derived from data on daily precipitation and temperature (https://www.climdex.org/learn/indices/). The expert team on climate change detection and indices (ETCCDI) of the World Climate Research Program (WCRP) has recommended these standardized indices to compare results across time periods, regions, and source datasets. Here, in this study, to analyse instantaneous and as well as cumulative responses of hydroclimatic extremes in the Narmada basin, we have taken three standard indices for precipitation (Table 1). Similar to precipitation, we have taken three indices for analysing the spatiotemporal variability of streamflow (discharge) extremes (Table 1). Central India faces extreme heat in summer that adversely affects the agricultural productivity of the region (Sharma & Mujumdar 2017; Singh et al. 2021). To analyse the spatiotemporal variability of extreme heat in the Narmada basin, we have also analysed three extreme indices for temperature (Table 1).

The descriptions of all nine extremes analysed in the study are mentioned in Table 1.

The spatiotemporal assessment of these indices can provide a better understanding of the risk associated with extremes (floods, droughts, heat waves, etc.) over the Narmada basin. Analysis of this link between prominent teleconnections and hydroclimatic extremes can give us a better understanding of floods, droughts, and heatwaves associated hazards (Ward et al. 2014; Seo et al. 2019). Several studies have been conducted using EVT under the stationary assumption to analyse risk due to extreme events. Under changing climate, this stationary assumption over extreme value probability distribution induces inaccuracies in the analysis of extreme events. This assumption may lead to underestimating or overestimating the probability distribution of extremes for the specific design period (Coles 2001). Therefore, a nonstationary framework would be more appropriate to assess hydroclimatic extremes accurately (Parey et al. 2010; Salas & Obeysekera 2014). Since ENSO, IOD, and AMO are major LSCOs that affect the patterns of hydroclimatic extremes over India, we have investigated the relative influence of individual and collective response of these teleconnections at the basin scale for better assessment of risk associated with hazards. In this study, we analysed the nonstationary behaviour of extreme precipitation, temperature, and streamflow over the Narmada basin considering ENSO, IOD, and AMO as covariates in the nonstationary framework of generalized extreme value (GEV) distribution.

The specific objectives of the study are: (i) to understand the instantaneous and short-term spatiotemporal variability of hydroclimatic extremes in the Narmada basin; (ii) to identify the region of the basin more prone to different hydroclimate extremes; (iii) to understand the relative influence of ENSO, IOD, and AMO over hydro-climatology of Narmada basin.

Streamflow extreme indices were calculated using daily observed discharge data from 10 discharge stations (Mohggaon, Manot, Patan, Gadarwara, Sandia, Hoshangabad, Chhidgaon, Handia, Mandleshwar, Garudeswar) for the period of 37 years from 1979 to 2015 (Table 2). Rest stations at Narmada basin sites were discarded because of the unavailability of data for the concerned period. The discharge data have been obtained from Central Water Commission (India, WRIS). The daily rainfall data have been obtained from Indian Meteorological Department (IMD) at 0.25° resolution. The daily temperature dataset for the basin has been obtained from the Global Meteorological forcing dataset (GMFD) at 0.25° resolution.

SST anomalies over the Nino 3.4 region (17° E-120° W, 5° S-5° N) of the Central Pacific are considered as one of the potential indicators for ENSO. The monthly SST-Nino 3.4 index data were obtained from National Oceanic and Atmospheric Administration (NOAA) Centre for Weather and Climate Prediction (https://www.cpc.ncep.noaa.gov/data/indices/). IOD also referred to as Indian Nino, is an irregular oscillation of SST of the western and the eastern Indian Ocean. IOD, represented as Dipole mode Index (DMI), is the difference of SST anomalies between the tropical western Indian Ocean and tropical south-eastern Indian Ocean. The monthly DMI data were obtained from Physical Science Laboratory, NOAA (https://psl.noaa.gov/gcos_wgsp/Timeseries/DMI/). Similarly, AMO data were taken from NOAA Physical Science Laboratory (https://psl.noaa.gov/data/timeseries/AMO/). These indices data have been obtained for the same period of 37 years from 1979 to 2015. As the Narmada basin receives the majority of its rainfall in the summer monsoon season, June–July–August–September (JJAS). Therefore, the mean values for JJAS season of LSCOs (ENSO, IOD, and AMO) were taken to analyse various extremes in the monsoon season.

As discussed above, changing climate invalidates stationary assumptions and thus, affects the traditional methods of frequency analysis as well as the definitions of the return period, risks, and reliability (Pechlivanidis et al. 2017). Trend analysis has been carried out to understand the spatiotemporal variability of hydroclimatic extremes in the Narmada basin (Ouarda & Charron 2019). A nonparametric modified Mann Kendall (MK) trend test has been carried out using ‘modifiedmk’ package of R studio to analyse monotonic trends in the data. The null hypothesis (H0) of the test is that there is no trend in the data and the alternative hypothesis (H1) is that the data follow a monotonic trend. The outputs from MK test are MK Z-statistic, Sen's slope, MK S statistics, Variance of S statistics, MK p-value, and MK's tau. The MK Z-statistics represent the level of confidence in the results of the test and Sen's slope represents the average rate of trend in the data, which have been discussed more elaborately. The statistics of the test were analysed at 5% level of significance.

Results from modified MK tests (z-statistics and Sen's slope) along with mean and standard deviation (SD) of precipitation, streamflow, and temperature extremes were analysed for the trend analysis. The Z-statistics of MK test indicate the significance in the monotonic trend of the time series of data whereas Sen's slope gives the rate of change of that time series of data. We have interpreted the results of Z-statistics in four categories at 5% level of significance. It was considered significantly decreasing at Z < −1.96, decreasing at −1.96 < Z < 0, increasing at 0 < Z < 1.96 and significantly increasing at Z > 1.96. The rest of the three parameters (mean, SD, and Sen's slope) have been analysed into three categories based on the observation.

A total of 10 stations (Mohggoan, Manot, Patan, Gadarwara, Sandia, Hoshangabad, Chhidgaon, Handia, Mandleshwar, Garudeswar) were taken into consideration for the streamflow and were represented as S1, S2, S3, S4, S5, S6, S7, S8, S9, and S10, respectively (Table 1). The stations have spread along the Narmada basin starting from the upper part to the tail of the river basin (Figure 1). Streamflow extremes, Qx1day, Qx5day, and Q95p for any of the stations were not found with any significant trend. Due to the contribution from various tributaries in the downstream, the streamflow of downstream stations was comparatively higher (Table 4). Q90p is the cumulative streamflow in a year from days when streamflow was more than the 90th percentile (cumulative contribution from very high streamflow) of daily streamflow data at a station, therefore as per the definition, the magnitude of Rx5day will always be more than Rx1day and Q90p will be comparatively too high than Rx1day and Rx5day and hence the mean and Sen's slope was comparatively higher in case of Q90p (Table 4). The streamflow magnitude and consequent extremes (Qx1day, Qx5day, and Q95p) from the station (S7) installed in the tributary of the main river have also been noted with a lesser magnitude of mean and MK test statistics.

The temperature indices (TXx, TNx, and TX90p) were analysed in a similar way. There was also not any significant trend in the past in temperature extreme, TXx and TNx in the basin, i.e., the daily maximum temperature in the day and night were not significantly increasing or decreasing in the past. Another temperature extreme, TX90p, a potential indicator of heatwaves, has shown increasing throughout the basin whereas significantly increasing trend at the uppermost and lowermost parts of the basin (Figure 3(b)).

The distribution of extremes at all the grids/stations were checked for the nonstationarity caused by ENSO, IOD, and AMO in the extremes. LR test revealed that there was a significant amount of nonstationarity caused by LSCOs present in the extremes. Statistics of the LR test with minimum p-value were used to find the best combination of LSCOs influencing most of the distribution of the extremes. Rx5day shows significantly more nonstationary behaviour in majority of the basin due to LSCOs in comparison with PRCPTOT and R95p (see Supplementary Material, Figure S1) The results from the best combination of LSCO(s) (nonstationary model(s)) and stationary model were discussed in terms of return levels (RLs) of extremes at 10, 20, 50, and 100 years.

The streamflow extreme indices (Qx1day, Qx5day, and Q95p) for the basin have also been increasing with increasing return periods. The downstream stations are more sensitive because they have been influenced by the upstream tributaries (Figure 5(ii)). Also, the variance in the instantaneous streamflow extremes (Qx1day and Qx5day) of the basin was observed high in comparison with Q95p (Figure 5(ii)). The variance in Qx1day extreme of discharge for station S7, located at one of the tributaries of the main river was observed maximum (Figure 5(ii)(a)). For given streamflow extremes in almost all stations, we have found nonstationarity due to LSCOs (see Supplementary Material, Table S1).

The temperature extremes, TXx, TNx, and TX90p were noticed with significant nonstationary behaviour due to LSCOs throughout the basin (see Supplementary Material, Figure S1). The central upper part of the basin experiences a comparatively higher magnitude of maximum and minimum daily temperature (Figure 5(iii)(a),(b)). The lower part of the basin experiences more number of days with temperatures more than the 90th percentile of daily maximum temperature (T90p) (Figure 5(iii)c).

Observing the spatiotemporal variation in PRCPTOT, if RLs in the stationary scenario were 100 mm or more amount of deficit in precipitation amount, we have considered as highly underestimated whereas a surplus of 100 or more would be highly overestimated. In case the surplus or deficit of PRCPTOT was lesser than 100 mm, as overestimated and underestimated. From plots (Figure 6(i)(a)), we found that the spatial trend of overestimated or underestimated magnitude of PRCPTOT is similar but increasing with the return period. In the case of Rx5day and R95p, the limit of overestimation and underestimation is taken as 50 and 25 mm, respectively. For all the three precipitation indices, the central part of the basin experiences underestimated the magnitude of RLs whereas the upper and lower part of the basin experience overestimated RLs.

The difference in RLs of the stationary and nonstationary was plotted for all 10 discharge gauging stations at 10, 20, 50, and 100 return periods (Figure 6(ii)). Qx1day and Qx5day are comparatively instantaneous responses of a river basin whereas Q95p is the comparatively long-term behaviour of a river basin. Therefore, the limits for underestimation/overestimation, in this case, are 1,000, 5,000, and 10,000 m³/s for Qx1day, Qx5day, and Q95p, respectively. In case of Q95p, for discharge gauging stations, S6 and S10, we got overestimated discharge values, whereas for S6 and S8 RLs were highly overestimated, and at S9, streamflow was highly underestimated. For station S7, Qx1day and Qx5day were highly underestimated for higher return periods whereas, for station S6 and station S5, Qx1day and Qx5day were highly overestimated, respectively, in the stationary framework.

TXx was considered underestimated/overestimated with a decrease/increase by 1 °C/0.5 °C whereas in case of TNx and TX90p a decrease/increase by 0.5 °C/0.5 °C and by 0.5 °C/1 °C, respectively. In case of TXx, RLs were consistently underestimated for some regions and overestimated for others at different return periods. The TNx and TX90p have been found overestimated and highly overestimated in the stationary framework (Figure 6(iii)(b),(c)). This magnitude of overestimated RLs has been found to increase with the return periods (see Supplementary Material, Figure S3).

Hydroclimatic extremes have a significant impact on the hydrological cycle of the basin, such as floods, droughts, or intense precipitation. These events are often caused by complex interactions between atmospheric and hydrological processes and can have severe consequences for water resources and human activities in the affected area. For a major part of the basin, extreme precipitation indices (PRCPTOT, Rx5day, and R95p) were observed to increase but not with any significant trend in the past. The mean values of these precipitation extremes were of higher magnitude at the upper part of the basin whereas the lower part of the basin was showing an increasing trend. In the case of temperature extremes also, the daily maximum temperature was decreasing whereas the daily minimum temperature at annual scale was increasing at the uppermost and lowermost parts of the basin. The temperature extreme, TNx was increasing through the basin whereas in the regions where the daily minimum temperature in a day was increasing, TX90p has been found significantly increasing trend in the past. For streamflow extreme indices (Qx1, Qx5, and Q95p), Z-statistics and Sen's slope from the MK test have been found to decrease for all the 10 discharge gauging stations.

It becomes essential in changing climate to assess the role of local and global factors associated with spatiotemporal variability of hydroclimatic extremes. In this study, we have included major climate oscillations (ENSO, IOD, AMO) in reference to Indian climatology to understand their association with the hydroclimatic extremes at the Narmada basin. The results have been interpreted in two ways: (i) Stationary framework (no influence of LSCOs) and (ii) nonstationary framework (significant influence of LSCOs). The results from the GEV distribution of hydroclimatic extreme indices based on the likelihood ratio test show a significant amount of nonstationarity caused by LSCOs. The precipitation extremes, PRCPTOT and R95p were observed significantly influenced by ENSO, IOD, and AMO individually whereas extreme Rx5day was relatively more influenced by ENSO, and AMO individually and collectively. In terms of temperature extremes, it was shown that TXx in the majority of the Narmada basin (26.47% of the region) was significantly more influenced by ENSO alone, while TNx was observed to be substantially more influenced by ENSO and AMO. Combined with IOD and AMO, ENSO had a considerably greater influence on the temperature extreme TX90p, a potential indicator of heatwaves in a region, and has been noticed with an increasing trend. This linkage may give rise to more frequent and intense heatwaves in the regions which consequently may affect production adversely. By ignoring spatiotemporal variability of extremes caused by ENSO, IOD, and AMO may lead to underestimating and overestimating the magnitude of hydroclimatic extremes of the Narmada basin. The overestimation of RLs will increase the cost of hydraulic projects whereas underestimation may lead to disaster (failure of hydraulic structure) in long term. Some other global and regional factors can also be included in a linear and nonlinear combination of covariates of GEV distribution to better understand the spatiotemporal variability of hydroclimatic extremes. The upper Narmada basin was experiencing a high magnitude of precipitation extremes, hence could be vulnerable to flooding and whereas the basin was projected to experience more frequent and intense heatwave-associated disasters at high return periods. Appropriate adaption and mitigation measures can also be taken to reduce the climate risk in the vulnerable region of the basin.

The authors are thankful to the Indian Meteorological Department (IMD) for providing us daily precipitation dataset at 0.25° resolution and Terrestrial Hydrology Research Group of Princeton University for the product global meteorological forcing dataset (GMFD) to provide us daily temperature (maximum, minimum and average) dataset at 0.25° resolution from 1979 to 2015 and also, we are thankful to the Physical Science Laboratory of National Oceanic and Atmospheric Administration (NOAA) to provide us ENSO (SST (Nino 3.4), IOD, and AMO) indices datasets.

No funding has been received to assist with the preparation of the submitted manuscript. Observed Daily gridded Precipitation has been obtained from Indian Meteorological Department (IMD) at 0.250 spatial resolution from 1979 to 2015 (https://www.imdpune.gov.in/cmpg/Griddata/Rainfall_25_Bin.html). Observed Daily gridded temperature has been obtained from Global Meteorological forcing dataset (GMFD) by Terrestrial hydrology research group of Princeton University (https://hydrology.princeton.edu/data.pgf.php). The monthly SST-Nino 3.4 index data has been obtained from NOAA Centre for Weather and Climate Prediction (https://www.cpc.ncep.noaa.gov/data/indices/). The monthly DMI data has been obtained from NOAA Physical Science Laboratory (https://psl.noaa.gov/gcos_wgsp/Timeseries/DMI/). Similarly, AMO index data has also been taken from NOAA Physical Science Laboratory (https://psl.noaa.gov/data/timeseries/AMO/).

All relevant data are available from an online repository or repositories.

The authors declare there is no conflict.