Influence of large-scale teleconnections on annual and seasonal floods in Godavari and Narmada River basins

In India, flooding is one of the three prominent climate extremes, the other two being droughts and cyclones (Bhattacharya & Das 2007). The majority of flooding in Indian watersheds occurs during summer monsoon months due to uneven distribution of rainfall. For example, the recent devastating floods in Kerala were in response to the abnormally high rainfall received within a short period of 3 days, i.e. from 15th to 17th August 2018 (e.g. Mishra et al. 2018). Approximately 80% of rainfall over the Indian subcontinent is received during the summer monsoon. Summer monsoon rainfall being the major source of water input to the Indian subcontinent, optimal design and operation of water resources infrastructure (e.g. major dams) is very much essential.

The Indian summer monsoon is substantially influenced by several low-frequency atmosphere-ocean oscillations including Pacific Decadal Oscillation (PDO), El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD), Atlantic Multidecadal Oscillation (AMO), etc. (e.g. Walker 1933; Saji et al. 1999; Roy et al. 2003; Sajani et al. 2007; Krishnamurthy & Krishnamurthy 2013a, 2013b; Li et al. 2017; Saini et al. 2022). For example, Krishnamurthy & Krishnamurthy (2013a) identified that the warm phase of PDO is associated with the rainfall deficit over the Indian subcontinent, whereas the cool phase of PDO is associated with the rainfall excess. A similar relationship is identified between El Niño and La Niña episodes of ENSO pattern (e.g. Krishnamurthy & Krishnamurthy 2013a; Saini et al. 2022). Despite these studies on teleconnections and Indian summer monsoon rainfall (ISMR), there has been little work done on the influence of these teleconnections on annual mean and/or peak streamflow in the watersheds of the Indian subcontinent. Henceforth, this study explores the influence of such teleconnections, i.e. PDO, ENSO and IOD specifically, on the characteristics of annual floods (or annual peak flows) in the Godavari and Narmada River basins. Such information plays an important role in the optimal planning and design of various infrastructures, and informs adaptive management policy for better reservoir operations.

The existing knowledge of flood characteristics in the Indian watersheds is primarily based on the assumption of stationarity, i.e. the annual floods are independent and identically distributed (i.i.d.) or the system fluctuates within a fixed envelope of variability (e.g. Milly et al. 2008). However, this assumption is widely questioned across the globe and several studies indicate that this assumption is inadequate (e.g. Milly et al. 2008; Stedinger & Griffis 2008; Gurrapu et al. 2016; 2022). For example, Franks (2002) and Kiem et al. (2003) demonstrated that the frequency of floods in New South Wales, Australia is impacted by Inter-Decadal Pacific Oscillation (IPO) and ENSO. They suggest that annual peak flow data should be assessed as a function of the causal climatological factors that affect regional climates. In a similar study, Ward et al. (2014) determined that La Niña episodes produce higher annual floods compared with El Niño episodes in the majority of the river basins across the globe, whereas few basins show the opposite relation. In another study, Andrews et al. (2004) determine that El Niño episodes produced higher annual floods along the California coast, USA. In a more recent study, Gurrapu et al. (2016) demonstrated that the frequency of floods in the watersheds of western Canada is substantially impacted by a large-scale low-frequency PDO, where the negative phase of PDO produced higher magnitude floods and the positive phase of PDO produced relatively lesser magnitude floods. So, the existing know-how on the flood characteristics of the Indian watersheds becomes immaterial unless proven. Therefore, this study explores the relationships between annual floods in the selected watersheds and the large-scale low-frequency atmospheric oscillations including ENSO, PDO and IOD.

This study is motivated by the observation that such teleconnections are not yet a key ingredient in the planning and design of regional water resources and/or transportation infrastructure. This study is the first of its kind to evaluate the impact of these low frequency oscillations, which are known to substantially control the magnitude and frequency of ISMR, on the annual and seasonal, i.e. summer monsoon (southwest or SW) and winter (northeast or NE) monsoon peak flows in Godavari and Narmada River basins.

Annual peak flow in Godavari and Narmada Rivers typically occurs during the southwest (i.e. during June–September) and/or northeast (i.e. during October–December) monsoon seasons, from the high magnitude and/or high intensity rainfall. The time series of annual peak flows, the maximum of the daily streamflow recorded during a water year starting from 1st June and ending on 31st May of the following year, at each station are extracted from the daily averaged streamflow records. In addition, the time series of seasonal peak flows, i.e. from 1st June to 30th September for SW monsoon and from 1st October to 31st December for NW monsoon, are extracted for each station. At each gauging site, peak flows for any given year or season are extracted from all the available data, regardless of the missing data. Any year or season with missing daily streamflow data or with a ‘0’ mean streamflow is discarded and is not used in the analysis. Details of the selected streamflow gauging stations and the recorded daily streamflow data for all the selected gauging stations are downloaded freely from the India-WRIS website (https://indiawris.gov.in/wris/#/DataDownload).

In addition to these two indices, the influence of the IOD is also evaluated. Sustained changes in the difference between sea surface temperatures of the tropical western and eastern Indian Ocean are known as IOD; more details of this oscillation are available at www.bom.gov.au/climate/iod/. IOD is quantified by Dipole Mode Index (DMI; Saji et al. 1999) and is freely available from Earth Systems Research Laboratory (ESRL), National Oceanic and Atmospheric Administration (NOAA), USA. DMI is computed at a monthly timescale and in this study, June to December monthly averaged index (DMI_{Jun_Dec}) is used for annual floods, June to September monthly averaged index (DMI_JJAS) for SW monsoon and October to December monthly averaged index (DMI_OND) for NE monsoon seasonal peak flows. Using DMI_{Jun_Dec}, the annual peak flow series at each of the selected gauging stations are stratified as responses to positive (DMI_{Jun_Dec} ≥ 0.5) and negative (DMI_{Jun_Dec} ≤ −0.5) episodes of IOD.

The effect of large-scale teleconnections on maximum (annual and/or seasonal) flows in the Godavari and Narmada River basins is first analyzed by measuring the strength of correlation between them using non-parametric Spearman's correlation coefficient (ρ). Rank-based Spearman's correlation is a robust method with no prior assumption of a distribution-fit to the hydrological or meteorological data (Wilks 2006; Gurrapu et al. 2016). In this study, the correlation between the peak flows and each of the teleconnections, namely, PDO, ENSO (ONI) and IOD (DMI), at all the 64 gauging sites were computed and the correlation coefficient (Spearman's ρ) with p-value less than or equal to 0.1 is considered significant at a 90% confidence level (i.e. based on the two-tailed significance test with α = 0.05). The longest length of streamflow data enables the detection of the impact of teleconnections and so the full period of record was used in the analysis.

Flood frequency curves fitted to the stratified peak flows help in investigating the impact of the teleconnections (e.g. Gurrapu et al. 2016). In this study, stratified peak flows were fit to a 3-parameter lognormal distribution (LN3) and 90% confidence intervals were constructed (USGS 1982). A clear separation between the flood frequency curves and non-overlapping confidence intervals indicate that the peak flows are not identically distributed (e.g. Franks & Kuczera 2002; Gurrapu et al. 2016). If there is a substantial overlap of the 90% confidence intervals, it may be assumed that the peak flows are i.i.d., however, the ratio of flood quantiles may indicate otherwise. Therefore, to further evaluate, the ratio of flood quantile of the negative episode to the flood quantile of the positive episode, termed as flood ratio (FR), is computed for selected return periods, 2-, 5-, 10-, 25-, 50- and 100-years (Franks & Kuczera 2002; Gurrapu et al. 2016). If the flood ratio is greater than 1 (FR > 1), it may be assumed that the higher magnitude floods are more common during the negative episodes of the teleconnection and vice-versa. While the flood ratio is computed in the same way for evaluating the influence of the ENSO, it is computed as the ratio of flood quantile from La Niña episodes to the quantiles from El Niño episodes to evaluate the effect of ENSO.

The correlations between annual peak flows and June to December averaged PDO indices (PDO_{Jun_Dec}) and between SW monsoon seasonal peak flows and June to September averaged PDO indices (PDO_JJAS) were also computed. Table 2 lists the strength of correlation (Spearman's ρ) between the standardized annual (WY) and seasonal (SW and NE monsoon) peak flows in the selected gauges and the indices of PDO, ENSO and IOD. The SW monsoon seasonal peak flows in seven gauging stations of the Godavari basin show statistically significant negative correlations with PDO_JJAS, whereas no station from the Narmada basin shows a statistically significant correlation. The annual peak flows in nine gauging stations of the Godavari basin show statistically significant negative correlations with PDO_{Jun_Dec}, whereas only one station from the Narmada basin shows a statistically significant positive correlation with PDO. Annual peak flows at Bamni Banjar (Stn ID: NBAM04) show a positive correlation indicating that the positive episodes of PDO produced higher magnitude peak flows. Although it is generally agreed that the positive phase of PDO produces dry years and the negative phase produces wet years, these relationships deviate moderately along parts of central India, northeastern states, and western and eastern Ghats (Krishnamurthy & Krishnamurthy 2013a). Hence, the contrasting correlation at Bamni Banjar and a few other gauging stations showed a positive correlation, although statistically not significant.

Table 2

Rank-based Spearman correlations between standardized annual (WY), SW monsoon seasonal (SW), NE monsoon seasonal (NE) peak flows and the low-frequency atmosphere-ocean oscillations including Pacific Decadal Oscillation (PDO), El Niño-Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD)

The statistically significant (α = 0.1) correlations are bolded, and the negative (positive) statistically significant correlations are shaded in blue (red).

View Large

In a similar manner, NE monsoon seasonal streamflow in 20 out of 46 gauging stations (≈43%) in the Godavari River basin shows statistically significant negative correlations, indicating that higher magnitude flows are more common during the La Niña episodes of ENSO (ONI_OND < −0.5) (Table 2). The strength of Spearman's ρ correlation ranged between −0.30 and −0.56, indicating that as much as 56% of the variability in NE monsoon seasonal peak flows at these gauging stations can be explained by the ENSO pattern. Four out of 18 gauging stations (≈22%) from the Narmada River Basin show statistically significant correlations. In comparison, only a few gauging stations showed statistically significant correlations with annual and SW monsoon seasonal peak flows, i.e. 7 each for the Godavari River basin and none for the Narmada River basin. All these correlations are negative, indicating that higher magnitude floods have occurred during the La Niña episodes of ENSO.

The correlations between IOD and annual peak flows are statistically significant and negative in four gauging stations in the Godavari basin and none in the Narmada basin. Whereas the correlations between IOD and SW monsoon seasonal peak flow are statistically significant in five (four negative and one positive) gauging stations of Godavari and one (negative) in the Narmada basin. The signal of IOD is strongly seen in the NE monsoon seasonal peak flows, similar to the other teleconnections, 13 out of 46 gauging stations (28%) in the Godavari basin and 4 out of 18 stations (22%) in the Narmada basin show statistically significant negative correlations, indicating that higher magnitude floods are more common during the negative episodes of IOD. Overall, the NE monsoon seasonal peak flows indicate a strong signal of PDO, ENSO and IOD, and all these correlations indicate that higher magnitude floods are common during the negative episodes of these teleconnections (Table 2).

The Q-Q plots are also constructed to evaluate the relationships between the ENSO and IOD and annual and seasonal peak flows in all the selected gauging stations. These Q-Q plots were tested for statistical significance at a 90% confidence interval (α = 0.1), two-sided. Table 3 lists the negative (–ve) and positive (+ve) episodes of a teleconnection, i.e. PDO, ENSO and IOD, in which higher magnitude annual (WY) and/or seasonal (SW and NE) floods may be expected, based on the statistically significant Q-Q plots. In summary, 17 stations indicate higher magnitude annual floods, and 14 stations indicate higher magnitude SW monsoon seasonal floods in response to negative PDO episodes. On the contrary, one station indicates higher magnitude annual floods, and three stations indicate higher magnitude SW monsoon seasonal floods in response to the positive episodes of PDO. The ENSO signal indicates that 16 stations show higher magnitude annual floods, 25 stations show higher magnitude SW monsoon seasonal floods and 25 stations show higher magnitude NE monsoon seasonal floods in response to La Niña episodes. Similarly, four stations show higher magnitude annual floods, five stations show higher magnitude SW monsoon seasonal floods and 29 stations show higher magnitude floods in response to the negative episodes of the IOD. Only a few stations (≤7.5%) indicate that higher magnitude floods occur in response to positive episodes of these teleconnections. Overall, the NE monsoon seasonal floods show a clear signal of PDO (21 out of 64 or ≍33%), ENSO (25 out of 64 or ≍39%) and IOD (29 out of 64 or ≍45%), all indicating that higher magnitude floods occur in response to the negative episodes of these teleconnections (Table 3). These results are in agreement with the earlier observations made, i.e. negative episodes of these teleconnections produce wetter years (e.g. Krishnamurthy & Krishnamurthy 2013a; Saini et al. 2022).

The regional influence of PDO and ENSO on annual and seasonal peak flows is also analyzed using the flood ratio approach, results not presented. The annual floods fail to show the signal of either of the teleconnections, i.e. PDO, ENSO or IOD, because the flood ratio is less than or equal to 1 in nearly 50% of the stations and is more than 1 in the other half, for most of the return periods. However, floods with higher frequency (i.e. RP = 2 or 5) indicate that higher magnitude floods are marginally higher during the negative episodes (FR > 1 in nearly 60% of the stations) of PDO and IOD. The regional influence of ENSO on annual peak flows in the gauges of the Narmada River basin is not distinctly seen using the flood ratio approach. In contrast, the influence of all these teleconnections is clearly seen on NE monsoon seasonal floods, indicating that higher magnitude floods are more common during the negative episodes of PDO, ENSO and IOD.

The return period or frequency of a flood event is one key piece of information required for adequate planning and design of water resources infrastructure and for effective management of available water. Traditionally, such information is obtained from FFA, which assumes that the annual peak flow series is independent and identically distributed (i.i.d). Almost all the FFA done in India invokes the assumption of i.i.d. and this study aimed to evaluate its competency in making estimates of the design flood. The results indicate that the annual and seasonal peak flows in the gauges spread across the Godavari and Narmada River basins indicate that their magnitude and frequency are substantially influenced by the phases of PDO, ENSO or IOD. In particular, the influence of these teleconnections is clearly seen on the seasonal, namely, northeast (NE) and southwest (SW) monsoon, peak flows. In the majority of the gauges, higher magnitude floods seem to be more common during the negative episodes of PDO, ENSO (La Niña) or IOD. In addition, the regional influence of these teleconnections is seen in the magnitude and frequency of seasonal peak flows using the flood ratio approach. The signal of these teleconnections is clearly seen in the seasonal floods of higher frequency (i.e. RP = 2, 5 and 10 years), where almost all the gauging stations indicate that the higher magnitude floods are common during negative episodes. Overall, the results from this study highlight the potential inadequacy of the i.i.d. assumption and are not tenable where the hydroclimatology is strongly influenced by the low-frequency atmosphere-ocean oscillations. These results are in agreement with the observations made by other researchers across the globe (e.g. Kwon et al. 2008; Stedinger & Griffis 2008, 2011; Lόpez & Francès 2013; Barros et al. 2014; Gurrapu et al. 2016). This is manifest in the Godavari and Narmada River basins in India and other parts across the globe including western Canada, California and eastern Australia (e.g. this study; Franks & Kuczera 2002; Ward et al. 2014; Gurrapu et al. 2016). The extent of this problem in other Indian watersheds remains to be explored. Any region with a strong teleconnection with such large-scale atmosphere-ocean oscillations may be subject to under- or over-estimation of the design flood. Therefore, the knowledge of the regional hydroclimate with regards to phases of the large-scale low-frequency atmosphere-ocean oscillations should be considered prior to estimating the design flood. Furthermore, the effect of other atmosphere-ocean oscillations, e.g. North Atlantic Oscillation (NAO), Arctic Oscillation (AO) and Atlantic Multi-Decadal Oscillation (AMO), on extreme hydrology of Indian watersheds needs to be explored.

All relevant data are available from an online repository or repositories.Streamflow: (https://indiawris.gov.in/wris/#/DataDownload). PDO: (http://jisao.washington.edu/pdo/) ENSO: (https://psl.noaa.gov/data/correlation/oni.data). IOD: (www.bom.gov.au/climate/iod/)

The authors declare there is no conflict.