The GRB lies between the longitudes 73°21^′–83°09^′ E and latitudes 16°07^′–22°50^′ N, as shown in Fig. 1. GRB is the largest peninsular river, and the catchment accounts for 9.5 % of India's geographical area (i.e., 312 812 km²). The catchment area extends over the states of Maharashtra, Madhya Pradesh, Andhra Pradesh, Chhattisgarh, Orrisa, and Karnataka. The basin receives 80 % of its annual rainfall from ISMR, with an average annual rainfall of about 1097 mm (1971–2005). However, GRB has a tropical climate with temperatures varying between 26 and 44 °C (1969–2004) (Mishra et al., 2020). Also, GRB has more than 15 major water infrastructures built across the river with a combined storage capacity of more than 21 BCM to serve irrigation, water supply, and hydropower production. Moreover, the installed hydropower generation capacity is more than 2.7 GW, and the total irrigated area is around 6000 km². Similar to many large river basins which experience the impacts of extreme climate events, GRB is also vulnerable to extreme calamities. For instance, the downstream regions of GRB has been experiencing the floods while severe droughts were reported in the significant portion of the upstream of the basin.

A wide range of inputs, such as meteorological forcing and physiographic features (i.e., elevation, soil, land cover and vegetation information), are necessary to setup and run the VIC model. Meteorological forcing variables such as rainfall, minimum temperature (Tmin⁡), and maximum temperature (Tmax⁡) were obtained from India Meteorological Department (IMD). The gridded rainfall datasets with a spatial resolution of 0.25°×0.25° (Pai et al., 2014), daily minimum and maximum air temperature datasets of 1°×1° resolution (Srivastava et al., 2009) were used as the observed forcing to simulate VIC model.

Digital Elevation Model (DEM) of 30 m spatial resolution developed by Shuttle Radar Topographic Mission (SRTM) was used to delineate and extract terrain-related information of GRB. The Decadal Land Use Land Cover (LULC) map of 100 m spatial resolution for 2005 was obtained from NASA's ORNL DAAC portal. The soil information was extracted from a 30 arcsec Harmonized World Soil Database (HWSD), obtained from the Food and Agriculture Organization (FAO). HWSD raster comprising of soil attributes up to a depth of one meter, bifurcating the soil layer into topsoil (0–30 cm) and subsoil (30–100 cm) was used. The observed daily streamflow data for GRB was obtained from 1980 to 2008 from the Central Water Commission (CWC), Government of India, to validate the VIC model simulations. ENSO conditions were defined based on the mean monthly ONI values from National Oceanic and Atmospheric Administration (NOAA) (https://origin.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ONI_v5.php, last access: 15 April 2022).

VIC model is a macroscale and physically-based semi-distributed model which simulates water and energy fluxes independently at each grid cell (Liang et al., 1994). Prominent features of the model include (a) simulation of each grid cell independently, (b) explicit representation of sub-grid vegetation and elevation heterogeneity, and (c) consideration of non-linear baseflow. The water balance at each time step in the VIC model is based on the continuity equation (Liang et al., 1994) as described below: 1∂S/∂t=P-E-R where ∂S/∂t is the change in water storage, P is precipitation, E is evapotranspiration and R is runoff. The units of all the above-mentioned variables are mm within chosen timestep (daily scale was adopted in this paper).

VIC curve accounts for the spatial heterogeneity of runoff generation in the model, whereas baseflow generation is based on the Arno Model conceptualization (Franchini and Pacciani, 1991). The Penman-Monteith equation estimates the evapotranspiration (ET) for each vegetation type defined in the model. The simulated runoff and baseflow fluxes from each grid cell are utilized by a separate routing model based on the linearized Saint–Venant equation to get the total discharge at the selected gauge stations (Lohmann et al., 1996, 1998).

The flowchart (Fig. 2) represents the methodology adopted in this study. The VIC model was employed in the study for hydrological modelling of GRB. The total area of the basin is 312 812 km² and discretized into uniform grids of resolution 0.15°×0.15° (about 277.22 km², a total of 1325 grids) and all input parameters were established at each grid level. Genetic algorithm (GA) optimization was employed for finding the optimal parameter set of VIC model (Whitley, 1994). Parameterization at each grid level was derived from thematic layers such as LULC map, FAO-HWSD global soil map, and DEM to obtain model parameters in the prescribed format (ASCII file format). Grid-wise flux outputs computed from VIC-3L were fed to a standalone routing model to simulate daily streamflow at the outlet (Polavaram). Different performance metrics like the coefficient of determination (R2) and Nash–Sutcliffe Efficiency (NSE), with a threshold limit of above 0.75, were adopted to analyse the model behaviour for calibration and validation. After calibration of the VIC model, grid-wise flux outputs, namely surface runoff, baseflow, soil moisture, and evapotranspiration, were analysed for various El Niño episodes. The present study applied the ONI, which represents the running 3-month mean of sea surface temperature (SST) anomalies for the Niño 3.4 region to identify El Niño years. El Niño events were defined as five consecutive overlapping 3-month periods at or above the +0.5 anomaly, and the events were further subclassified into weak (with a 0.5 to 0.9 SST anomaly), moderate (1.0 to 1.4), strong (1.5 to 1.9) and very strong (≥2.0) events (Table 1). Finally, the percentage change in water balance components of GRB, namely precipitation, soil moisture and abstractions, were compared with normal years for various classes of El Niño events.

An inverse relationship between the average precipitation and El Niño intensity from moderate to very strong was noticed for GRB during 1980–2008 (Fig. 4). The positive correlation in average precipitation for weak El-Niño years could be due to spatial variations in impacts of the events in different basin regions, resulting in an overall increase of precipitation in the basin. POC in precipitation was observed with the index of -0.74 for weak El-Niño years, and a significant increase in POC was noticed with increasing ONI. This implies that the amount of precipitation received during El Niño years was lower than normal years, and the total precipitation would have been even less with the increasing magnitude of El Niño events. These findings very well match with previous studies (Ihara et al., 2007; Rajeevan and Pai, 2007).

Spatial variation of precipitation over the basin revealed that the upstream region of GRB was less influenced during weak El Niño episodes than the middle and downstream regions. In contrast, the POC in precipitation was noticeably higher for upstream and middle Godavari regions for moderate El Niño episodes. Furthermore, the impact of El Niño events was becoming more prominent when there was an increasing magnitude of the event. However, POC was observed lower for strong El Niño than the very strong El Niño episodes. It may be noted that the study did not quantify the relative impacts of tropical climate patterns such as the Indian Ocean Dipole (IOD), and the tropical Atlantic variability (ATL) on the ISMR. Therefore, a possible reason for lower POC in the case of a strong ENSO than the very strong ENSO for precipitation may be due to the co-occurrence of the El Niño and IOD events in the Indo-Pacific Ocean sector or over India. Ashok et al. (2004) showed that IOD significantly reduces the impact of ENSO on the ISMR whenever IOD and ENSO events co-occur. Cherchi and Navarra (2013) also showed large anomalies in the characteristics of ISMR due to the occurrence of ENSO and IOD events.

An apparent increase in POC for abstractions was observed with the increasing intensity of El Niño from moderate to very strong (Fig. 5). In contrast, POC in abstractions was observed to be -6.10 for weak El-Niño years due to increased precipitation indicating more water available for losses. A significant increase in POC for abstractions was noticed with increasing ONI from moderate to very strong El Niño. This can be attributed to the fact that there was less soil moisture availability during the El Niño years due to the reduction in precipitation, which resulted in decreased evapotranspiration and other losses.

POC in soil moisture was highest for very strong El Niño, which means soil moisture availability is reduced during El Niño years (Fig. 6). This can be explained relating the low quantum of precipitation for the El-Niño years, which resulted in the reduction of soil water recharge, less percolation, and hence less soil moisture availability in soil layers. A similar finding in the spatial variability of soil moisture was observed in various regions of GRB. Overall, it was found that El Niño-induced impacts were high for intense El Niño episodes throughout the basin.

The analysis of summary statistics computed from the model simulations corresponding to the normal as well as ENSO years (Table 2) revealed that the average annual precipitation for El Niño years (1050.82 mm) is lesser than normal years (1131.65 mm). Similar to our findings, Bothale and Katpatal (2016) showed a precipitation deficit during El Niño years over the Pranhita catchment (a sub-catchment of GRB). Our analysis also highlighted a significant reduction in annual average streamflow from 2845.06 m³ s⁻¹ in normal years to 2422.13 m³ s⁻¹ during El-Niño years (Table 2). Additionally, we compared the rainfall intensity obtained from 1 and 3 d rainfall depths for El Niño and normal years from 1980–2008. For normal years, the mean and standard deviation of maximum daily rainfall intensity was found to be 121.56 and 86.10 mm d⁻¹, respectively, while 87.21 and 26.47 mm d⁻¹ were reported for El-Niño years. Similarly, the mean and standard deviation of maximum 3 d rainfall intensity was found to be 61.87 and 51.72 mm d⁻¹ (normal years), respectively, while 46.59 and 18.82 mm d⁻¹ were reported for El-Niño years. In addition, the number of rainy days were computed for El Niño and normal years from 1980–2008. The mean and standard deviation of rainy days was found to be 81 and 4.83 d for normal years, respectively, while 70 and 13.01 d were observed during El-Niño years. The results showed the reduction in rainfall intensity and the associated number of rainy days for El-Niño years than normal years. Furthermore, the analysis revealed that rainfall deficits during El Niño years could be the major reason for the reduction in soil water recharge, less percolation, and hence less soil moisture availability in soil layers. Consequently, insufficient soil moisture during the crop season has a high risk of leading to agricultural drought and crop failure (Sai et al., 2016). Numerous researchers have shown the occurrence of droughts in GRB due to El Niño events, particularly in the Marathwada region of Maharashtra (Dixit and Jayakumar, 2021; Chavadekar and Kashid, 2021). Additionally, Kulkarni et al. (2016) demonstrated a dramatic 52 % decline in pulse (Tur, Moong and Urad Dal) output in Maharashtra during the drought of 2015 caused by the El-Niño event. Hence, El Niño events have adverse impacts on the water resources and agricultural productivity in GRB.