Enhanced Modulation of Streamflow Flash Droughts by Reservoir Operations in India

Streamflow flash droughts (SFDs) occur due to a rapid decline in streamflow and cause major challenges associated with water availability for downstream ecosystems, hydropower generation, and irrigation water demand. Human interventions such as reservoir operations and reservoir storage can considerably influence streamflow variability. However, the crucial role of dams/reservoirs on the occurrence of SFDs in India remains unexamined. Using a combination of hydrological and hydrodynamical models, we examined the role of reservoirs on SFDs in India during 1971–2020. Reservoirs play a considerable role in the occurrence of SFDs during the summer monsoon season (June‐September) in India. The frequency and severity of SFDs in the presence of reservoirs are substantially lesser than in the absence of them. In addition, high and low antecedent reservoir storage conditions before meteorological flash droughts (MFDs) do not support the rapid decline of streamflow (i.e., SFDs) downstream of reservoirs, while the medium reservoir storage conditions prior to MFDs favor the development of SFDs. Similar to conventional (or long‐term) streamflow droughts (SDs), SFDs are more frequent in the El Nino phase than in the La Nina phase. Unlike the long‐term streamflow droughts, the implementation of dams considerably reduces the frequency of SFDs during both the negative and positive phases of El Nino Southern Oscillation. Overall, our findings demonstrate the considerable role of human interventions in flash drought occurrence, as SFDs are triggered by MFDs but modulated by reservoir storage and operations.


Introduction
Flash drought is characterized by a rapid depletion of soil moisture or streamflow associated with high evaporative demand and large precipitation deficit (Beguería et al., 2014;Vicente-Serrano et al., 2010).Flash droughts cause severe impacts on water resources, agriculture, and ecological systems (Christian et al., 2021;Mahto & Mishra, 2023a, 2023b;Pang, 2023).Flash droughts have been identified using hydrometeorological variables such as soil moisture, precipitation, temperature, and evapotranspiration (Mahto & Mishra, 2023b;Mo & Lettenmaier, 2015;L. Wang et al., 2016;Yuan et al., 2023).For instance, low rainfall and abnormally high-temperature cause sudden increases in evapotranspiration rates, contributing to the rapid intensification of water-stressed conditions, which results in flash droughts (Sreeparvathy & Srinivas, 2022).The frequency of flash droughts has increased in India over the past decades, which is likely to rise under the warming climate (Christian et al., 2021;Mahto & Mishra, 2020, 2023a, 2023b).A broad category of flash drought, meteorological flash drought (MFD), is primarily caused by the lack of rainfall (Mo & Lettenmaier, 2016;Sreeparvathy & Srinivas, 2022;Zhang & Yuan, 2020).Significant advancements have been made in identifying MFDs and their changes at regional and local scales (Mo & Lettenmaier, 2016;Noguera et al., 2021;Sreeparvathy & Srinivas, 2022;Yuan et al., 2023).For instance, Sreeparvathy and Srinivas (2022) reported that the frequency of MFDs is projected to rise by 34% in the future, with the highest occurrence probability and exposure risk in the Indian sub-continent during the summer monsoon season (Mahto & Mishra, 2020).
The occurrence and propagation of MFDs can adversely impact hydrological systems (Kim et al., 2021;Quansah et al., 2021).The transition from meteorological drought to hydrological drought is referred to as drought propagation (Eltahir & Yeh, 1999).Conventional meteorological droughts may lead to agricultural and hydrological droughts (Bhardwaj et al., 2020).The long-term hydrological droughts and the role of climate conditions in their occurrence are relatively well-understood (Bhardwaj et al., 2020;Huang et al., 2015;Van Loon & Laaha, 2015).However, hydrological flash droughts that occur due to the rapid decline of streamflow have not been examined in the monsoonal climate of India.Similar to conventional long-term hydrological droughts, hydrological flash droughts can be characterized by abnormally low flow conditions in rivers (Tallaksen & Van Lanen, 2004;Wilhite & Glantz, 1985) that occur and get intensified rapidly compared to conventional (or longterm) droughts.Thus, streamflow flash drought (SFD) can be defined by the rapid decline of streamflow below a certain threshold in a short duration.
SFDs can be affected by climatic conditions as well as human interventions associated with reservoir storage and release, especially in the downstream regions.Human interventions related to reservoir operations substantially alter streamflow in downstream regions (Dai et al., 2008;López-Moreno et al., 2004;Magilligan & Nislow, 2005).Reservoirs play a critical role in the hydrological system by altering streamflow regimes through attenuation, delay, and controlled releases (Zajac et al., 2017).Large reservoirs reduce streamflow during the wet season and increase streamflow during the dry season (Graf, 2006;Vicente-Serrano et al., 2017;Yang et al., 2017).In addition, water storage in reservoirs can mitigate the low flow condition in the downstream region during droughts (Lazin et al., 2023;Raczyński, 2020).Dams/reservoirs considerably influenced streamflow across the globe during the twentieth century (Chao et al., 2008;Lettenmaier & Milly, 2009).More than 2.8 million dams with 7,000-10,000 billion m 3 capacity (Grill et al., 2019) have been built globally to reduce flood risk and increase water availability during dry periods, which represent over one-sixth of the annual continental discharge to global oceans (Abbott et al., 2019;Chao et al., 2008).India is currently the third largest dam-building country in the world, with more than five thousand large dams (Pradhan & Srinivasan, 2022), which provide water for irrigation and hydropower, and store water to control flooding in the downstream regions.All the major rivers in India are influenced by reservoirs (Lehner et al., 2011).Therefore, understanding the role of reservoirs on SFDs is crucial.
The role of reservoirs in conventional hydrological droughts has been examined (Dai et al., 2008;López-Moreno et al., 2004;Magilligan & Nislow, 2005).Conventional hydrological droughts have a longer persistence time and may not be considerably affected by reservoirs due to their long durations (Brunner, 2021;Chang et al., 2019;Luo et al., 2023;Sun et al., 2023).However, the duration and severity of conventional streamflow drought can be reduced by reservoir operations (Wu et al., 2019).In contrast, SFDs occur for a relatively shorter duration; therefore, these can be mitigated through the quick release of water from reservoir storage.The interannual variability of Indian summer monsoon rainfall due to oceanic and atmospheric circulations often leads to abnormal droughts and floods in India (Mishra et al., 2012;Roxy et al., 2015), which have a significant impact on the Indian economy.El Nino Southern Oscillation (ENSO) is a well-known phenomenon related to monsoon variability through large-scale Walker and Hadley circulations (Ashok et al., 2001(Ashok et al., , 2004)).Sea surface temperature (SST) over the central Pacific region considerably affects the Indian summer monsoon rainfall (Kumar et al., 2013), which remains the most crucial driver of droughts in India.Consequently, the occurrence of SFDs during the summer monsoon season can be affected by the warm and cold phases of ENSO.
As streamflow and reservoir storage observations in the Indian subcontinental river basins are limited, simulations from the hydrological models can fill this critical gap (Asoka & Mishra, 2020;Chuphal & Mishra, 2023a;Tiwari & Mishra, 2019;Vegad & Mishra, 2022).Mahto and Mishra (2020) reported that India faces frequent flash droughts during the summer monsoon.Moreover, hydrological drought development strongly depends on hydro-meteorological factors during the summer monsoon season in India (Ganguli et al., 2022).Regardless of the potential implications of reservoir storage on flash droughts, their impacts remain unexplored in the monsoonal climate in India.In addition, the role of reservoirs in modulating SFDs Water Resources Research 10.1029/2024WR037036 during the positive and negative phases of ENSO has not been examined.Considering the large number of dams and reservoirs in India, we aim to address the following questions: (a) how do dams affect the frequency of the SFDs during the summer monsoon season in India?(b) what is the role of reservoirs in the propagation of meteorological to SFDs? and (c) How does the ENSO variability affect the SFDs and conventional droughts during the summer monsoon season considering with and without dams scenarios?To address these questions, we used the H08 and CAMA-Flood model to simulate streamflow and reservoir storage for subcontinent river basins (Dai Yamazaki et al., 2011;Hanasaki et al., 2018).We examined the SFD events in the presence and absence of dams during the summer monsoon (June-September) in the major river basins of India during 1971-2020.

Data Sets
Daily gridded precipitation data were obtained from the India Meteorological Department (IMD) at 0.25°spatial and daily temporal resolutions for the period of 1971-2020.The gridded precipitation is developed using stationbased rainfall observations from more than 6,900 gauge stations distributed over India (Pai et al., 2014).Distribution of rain gauge stations effectively captures the key features of the summer monsoon (June-September) variability and orographic rainfall over the western Ghats and foothills of the Himalayas.The gridded rainfall from IMD has been widely used for hydrological assessments in India (Kushwaha et al., 2021;Mahto & Mishra, 2020;Mishra et al., 2018;Shah & Mishra, 2016).We used daily maximum and minimum temperatures from IMD, which are developed at 1°spatial and daily temporal resolutions from more than 395 gauge stations across India by employing Shepard's (1968) inverse distance-weighted interpolation technique (Srivastava et al., 2009).Gridded maximum and minimum temperatures were re-gridded at 0.25°resolution using bilinear interpolation to make the temperature data consistent with the gridded precipitation.
We collected daily streamflow (m 3 /s) at gauge stations and reservoir storage across the major river basins of India from the Water Resources Information System of India (India-WRIS), which is available at https:// indiawris.gov.in/wris/.The Central Water Commission (CWC, Government of India) monitors and provides daily streamflow and reservoir storage for Indian river basins through the India-WRIS platform.Previous studies used streamflow and reservoir storage data from India-WRIS to examine the climate change impact on water availability in India (Asoka & Mishra, 2020;Chuphal & Mishra, 2023a;Mishra, 2020;Tiwari & Mishra, 2022).We selected 36 dams (height ≥100 m and capacity ≥1 km 3 ) from the National Register of Large Dams (NRLD) (Figure 1) based on the availability of observed streamflow and reservoir storage.We obtained the observed streamflow data for 21 gauge stations, which consider the influence of the selected 36 dams.The detailed information for gauge stations and dams is mentioned in supplemental Tables S1 and  S2 in Supporting Information S1.
Sea surface temperature anomaly over the central Pacific region strongly correlates with the summer monsoon rainfall and air temperature in India (Mishra et al., 2012;Mishra, Thirumalai, et al., 2021;Mishra, 2020).Using SST data, we estimated the Oceanic Nino Index (ONI), which is a 3-month running mean SST anomaly.The National Oceanic and Atmospheric Administration uses the ONI index for classifying El Nino (positive/warm phase) and La Nina (negative/cool phase) events in the eastern tropical Pacific.We considered the El Nino/La Nina years if the ONI index for the three consecutive pairs (May-July, June-August, and July-September) remains above/below 0.5 (Figure S1 in Supporting Information S1).Thus, we selected six El Nino years and six La Nina years during 1971-2020 (Table S3 in Supporting Information S1).We also estimated the effectiveness of dams/ reservoirs for reducing SFDs and SDs, which is calculated as where (N wo ) i and (N w ) i are the number of events (SFDs or SDs) without dams and with dams during the 1971-2020 period, respectively, for ith catchment and "n" is the total number of catchments.

Hydrological Model Simulations of Streamflow and Reservoir Storage
We used H08 and Catchment-based Macro-scale Floodplain (CaMa-Flood) models to simulate daily streamflow and reservoir storage to examine their role in the SFD occurrence.The H08 is a large-scale grid-based global hydrological model, which is used for the simulation of hydrological fluxes such as runoff and baseflow (Hanasaki et al., 2018).We selected the CaMa-Flood model for streamflow routing at the selected locations as the CaMa-Flood model considers the role of reservoirs in streamflow routing, unlike most other routing models.
The H08 model consists of six distinct modules: land surface hydrology, river routing, reservoir operation, crop growth, environmental flow, and water abstraction.We considered the land surface module of the H08 model.The H08 model uses the leaky bucket method to estimate subsurface runoff, while surface runoff is calculated based on saturation excess non-linear flow within individual grid cells (Hanasaki et al., 2008(Hanasaki et al., , 2018)).The model solves water balance to estimate soil moisture and snowmelt.The soil and vegetation parameters for the H08 model were obtained from the Harmonized World Soil Database and ISIMIP3a database (Volkholz & Ostberg, 2022), respectively.We used the H08 model with input meteorological forcing from IMD at 0.25°spatial and daily temporal resolutions for 1971-2020.We integrated the H08 model with CaMa-Flood (H08-CaMa-Flood combined model) to simulate streamflow and reservoir storage.
The CaMa-Flood is a large-scale distributed model that computes hydrodynamics at the continental scale using a shallow water flow equation (Dai Yamazaki et al., 2011, 2013).The model estimates flow velocity and discharge  S2 in Supporting Information S1.

Water Resources Research
10.1029/2024WR037036 using a local inertial equation (Bates et al., 2010) along with an automatically generated river network using the Flexible Location of Waterways method (D Yamazaki et al., 2009).The CaMa-Flood model has been benchmarked against the other streamflow routing models for better performance in simulating discharge and reproducing observed streamflow (Zhao et al., 2017).For instance, He et al. (2020) developed the global drought and flood catalog from 1950 to 2016 using river discharge from CaMa-Flood model simulation.We used H08simulated daily runoff at 0.25°spatial resolution as input in the CaMa-Flood model and simulated the daily streamflow at a finer grid size (0.1°) by using the high-resolution stream network available in the CaMa-Flood model.
To simulate the effects of water regulation due to human interventions on downstream streamflow, we combined the H08 land surface module with the CaMa-Flood reservoir module.The CaMa-Flood (version 4.1) incorporates the role of dams and reservoirs for simulating streamflow (Hanazaki et al., 2022).We used the basic information for each dam including total storage capacity, drainage area, and location from NRLD to incorporate the dams in the CaMa-Flood river network.Reservoir operations related decisions can also influence the timing and amount of released water that can alter the occurrence of SFDs in the downstream regions.However, reservoir rule curve data sets are not available in the public domain for most reservoirs in India.Moreover, reservoir operations-related decisions may not always follow the standard rule curves (Jain et al., 2023).Therefore, we could not examine the role of reservoir operation policy on SFD occurrence in the downstream region due to the unavailability of rule curves.The CaMa-Flood model estimates the reservoir storage by implementing the target storage-and-release operation scheme, which is the most suitable for multipurpose dams (for example, flood control, irrigation, hydropower, and water supply) (Hanasaki et al., 2022;Yassin et al., 2019).We performed CaMa-Flood routing for two scenarios: (a) the influence of a dam/reservoir was not considered, and, (b) the effect of a dam/ reservoir on downstream streamflow was considered.The daily gridded runoff generated by the H08 model was forced into CaMa-Flood to simulate the streamflow in the presence/absence of dams.Model simulated streamflow with consideration of dams includes the influence of upstream reservoirs.Previous studies used the H08-CaMa-Flood combined model to simulate streamflow and investigate the influence of the reservoir on streamflow in Indian river basins (Boulange et al., 2021;Chuphal & Mishra, 2023a, 2023b;Vegad et al., 2023).

Calibration and Evaluation of H08-CaMa-Flood Model
We manually calibrated the H08-CaMa-Flood combined model against observed daily streamflow and evaluated the model's performance against observed daily streamflow and reservoir live storage using the coefficient of determination (R 2 ), Nash-Sutcliffe Efficiency (NSE), and refined index of agreement (d r ) (Figure 2).R 2 is a statistical measure to test the goodness of fit of the model.In addition, NSE and d r are widely used to check the predictive skill of hydrological models.R 2 , NSE, and d r can be calculated as: Water Resources Research 10.1029/2024WR037036 SINGH AND MISHRA where Q i and S i are observed and simulated daily streamflow for the day "i," respectively.Q and S represent the mean of observed and simulated daily streamflow, respectively, and "n" is the total number of days.
We calibrated the H08 model for each basin by adjusting four parameters (single-layer soil depth, gamma, bulk transfer coefficient, and tau) (Hanasaki et al., 2008) within their defined range (Tables S4-S5 in Supporting Information S1).These four critical model parameters considerably influence the runoff generation (Hanasaki et al., 2008).A detailed discussion of the calibration parameters of the H08 model can be obtained from Dangar and Mishra (2021).We selected 21 gauge stations at the farthest downstream of 14 major river basins with observed streamflow availability for at least 10 years to evaluate the model performance (Table S1 in Supporting Information S1).For transboundary river basins (Ganga and Indus), daily observations are not available in the public domain, therefore, we used monthly observed streamflow for calibration and evaluation of the model  S1 and S2 in Supporting Information S1.
Next, we evaluated the model performance with the remaining half of the streamflow observations (Figures 2d-2f).The model exhibits R 2 (median: 0.75), NSE (median: 0.71), and d r (median: 0.74) greater than 0.6 for all the basins (except Sabarmati; R 2 = 0.57, NSE = 0.51, and d r = 0.57) during the evaluation period (Table S1, Figure S3 in Supporting Information S1).In addition, the model exhibits R 2 , NSE, and d r greater than or near 0.7 for nine river basins and comparatively good skill (NSE > 0.75) for the Ganga, Narmada, and Subarnarekha basins during the evaluation period.The H08 and CaMa-Flood combined model performs well for the low flow during calibration (Figures S4 in Supporting Information S1) and evaluation periods (Figures S5 in Supporting Information S1).In this study, we used streamflow pentads (5-day mean streamflow) instead of daily streamflow; therefore, we evaluated the model performance using streamflow pentads (Figure S6a-S6c in Supporting Information S1).The model shows comparatively higher R 2 and NSE coefficients for all river basins using streamflow pentads than daily streamflow (Figures S3 and S9 in Supporting Information S1).
After calibrating the model against daily streamflow, we evaluated the model performance against the observed daily live storage of 36 reservoirs (Table S2 (Chuphal & Mishra, 2023a, 2023b;Vegad et al., 2023).In addition, Yamazaki et al. (2011) evaluated the CaMa-Flood model against gauge observations and reproduced streamflow for 30 major river basins across the globe.Compared to other global hydrological models, the CaMa-Flood captures peak discharge well, favouring the more accurate reproduction of observations (Zhao et al., 2017).After evaluating the model performance, we simulated the daily streamflow and reservoir storage for the 1971-2020 period at selected 36 catchments to investigate the influence of dams on the occurrence of SFDs at downstream of catchments.

Identification of Flash Drought
We delineated the upstream basins draining to reservoirs (Figure 1) using a digital elevation model (30s spatial resolution) obtained from HydroSHEDS (Lehner et al., 2008).Basin-averaged precipitation and temperature were estimated for each catchment using daily gridded data sets.We estimated the pentads (5-day averages using a fixed window) for rainfall, temperature, streamflow, and reservoir storage.Pentads reduce the high-frequency fluctuations in daily observations and provide a more accurate characterization of drought conditions (Christian et al., 2019;Nguyen et al., 2019;Singh & Mishra, 2024a).In a calendar year, pentads from the 31st to the 55th represent the summer monsoon season (June-September).We estimated streamflow percentiles for each pentad using the empirical Weibull distribution, considering 1971-2000 as the reference period (Mahto & Mishra, 2020;Mishra, Aadhar, & Mahto, 2021;Singh & Mishra, 2024a).We employed the rapid intensification characteristics of flash drought to identify SFDs using streamflow percentiles (Otkin et al., 2018;Singh & Mishra, 2024a) (Figure 3).Specifically, an SFD event was identified when streamflow percentiles dropped from above 40th percentile to below 20th percentile within three pentads (Figure 3).A SFD event terminates when streamflow percentile exceeds 40th percentile.In addition, we considered only those SFD events in which the mean streamflow percentile during the event (after onset and before termination) is below 25th percentile (Figure 3).To ensure the normal streamflow conditions before the occurrence of an SFD event, we applied a threshold for the

Water Resources Research
10.1029/2024WR037036 antecedent streamflow such that the average streamflow over the preceding three pentads (including onset pentad) exceeds the median (50th percentile) streamflow (Figure 3).Furthermore, we included only those SFDs in our analysis that persisted for a duration of three pentads or more (excluding onset and termination pentads).
We identified the MFDs in the upstream catchments using basin mean precipitation percentiles and employing the same percentile threshold as in SFDs.The occurrences of SFDs and MFDs were identified for the summer monsoon season (June-September) during 1971-2020.Next, we identified SFDs that occurred after MFDs, which refers to the propagation of MFDs to SFDs.To do so, we enforced two criteria: (a) the commencement of an MFD should precede the onset of an SFD, and (b) there must be an overlap of at least one pentad between the durations of MFDs and SFDs.

Meteorological and SFDs
First, we estimated the frequency of MFDs over the 36 catchments (Figure 1) upstream of reservoirs across India during the summer monsoon season (June-September) for 1971-2020 (Figure 4).Each catchment exhibited at least 25 MFDs over the 50 years (1971-2020), with most catchments (29 out of 36) showing more than 35 MFDs and almost half of the catchments showing more than 40 MFDs (Figure 4).Moreover, the catchments in transboundary river basins (Ganga and Indus) exhibit comparatively lower frequency of MFDs than the other river basins.The Ganga basin experienced a decline in the summer monsoon rainfall and streamflow (Mishra et al., 2012;Shah & Mishra, 2016), which can be associated with a relatively lesser occurrence of MFDs as most of them might have been converted to conventional long-term meteorological droughts due to persistent rainfall deficit.
We identified SFD events at downstream of reservoirs during the summer monsoon season for 1971-2020.We estimated the frequency of SFDs in the absence and presence of dams (Figures 5a and 5b).The frequency of SFDs in the presence of dams is lower than that in the absence of dams in each catchment.The majority of the catchments (29 out of 36 catchments) exhibit 10 or more SFDs in the absence of dams, with more than half of the catchments showing relatively higher frequency (15 or more) of SFDs (Figure 5a).In contrast, almost all the catchments show less than 10 SFDs in the presence of dams, with more than half of the catchments exhibiting only up to five SFDs during the observed record (Figure 5b).Catchments in southwest India (longitude: 73°-78°and latitude: 10°-18°) show a relatively higher frequency of SFDs in the absence of dams (Figure 5a).Mahto and Mishra (2020) reported a relatively higher frequency of MFD events in western ghat, northeast, and northwest regions of India during the summer monsoon season, which could trigger SFDs in these regions.
We estimated the severity of SFDs in the absence and presence of dams (Figures 5c and 5d), which is defined as the minimum of mean streamflow percentile for two continuous pentads after onset and before termination of the SFD event.A lower value of the mean streamflow percentile indicates a higher severity of the SFD event.The severity of SFDs is reduced for most of the catchments after the implementation of dams at upstream.In the absence of dams, almost all the catchments (33 out of 36 catchments) exhibit SFDs with mean streamflow falling below 14th percentiles (Figure 5c).In the presence of dams, 10 catchments did not exhibit any SFD event.Therefore, severity of SFDs was estimated only for 26 catchments (Figure 5d).Almost half of the catchments (11 out of 26) exhibited SFDs with mean streamflow above the 14th percentile.Overall, MFDs are more frequent than SFDs during the summer monsoon season.Moreover, the implementation of dams decreases the frequency and severity of SFDs in the downstream regions.Our results highlight that the human interventions play a major role in modulating the severity and frequency of SFDs in the downstream of reservoirs.

Influence of Dams on the Propagation of Meteorological to SFDs
We examined the influence of dams on the propagation of MFDs to SFDs in the downstream of each reservoir during the summer monsoon season for 1971-2020.We identified the MFDs that propagate into SFDs in the absence and presence of dams (Figure 6).The frequency of propagation of MFDs to SFDs considerably decreases after the implementation of dams (Figures 6a and 6b).In the absence of dams, half of the catchments show more than nine MFDs, which propagated into SFDs (Figure 6a).Whereas, in the presence of dams, most catchments show only up to three MFDs that propagate to SFDs (Figure 6b).The SFDs that do occur even after the implementation of dams were relatively more severe (low streamflow percentile) than those SFDs that did not occur after the implementation of dams (Figure S12 in Supporting Information S1).Moreover, a relatively higher frequency of propagation of MFDs to SFDs was observed in southwest India (Figure 6a), which also experiences a high frequency of SFDs (Figure 5a).Southwest India receives considerable rainfall during the summer monsoon (Guhathakurta et al., 2015;Rajeevan et al., 2006).A long dry spell due to summer monsoon break causes a rapid decline in rainfall and a rise in air temperature, which leads to the development of MFDs (Mahto & Mishra, 2020;Mishra, Aadhar, & Mahto, 2021).Substantial precipitation deficit and anomalous high temperature due to land We also examined the influence of dams on the severity of SFDs that occurred during propagation.We calculated the severity of SFDs in the absence and presence of dams (Figures 6c and 6d).During propagation, the SFDs show comparatively low severity in the presence of dams.In the absence of dams, the mean streamflow percentile for SFDs (during propagation) moves from the 8th to 14th percentile range in almost all the catchments (Figure 6c).In the presence of dams, the propagation of MFDs to SFDs occurred only in 24 catchments.Out of 24 catchments, half of the catchments show mean streamflow percentiles between the 14th and 20th percentiles (Figure 6d),

Role of Antecedent Reservoir Storage on SFDs
Reservoirs play a crucial role in the hydrological system since they affect the downstream flow regime by changing the magnitude and timing of streamflow through storage and subsequent release of water (Ayalew et al., 2013;Hawkins, 1969;Zajac et al., 2017).Therefore, we further examined the role of reservoir storage on the occurrence of SFD in downstream.We estimated the antecedent reservoir storage, considering mean reservoir storage in the previous three pentads from the onset of MFD, for all the MFDs for each catchment to understand the role of reservoirs on the propagation of MFDs to SFDs.We divided MFDs into two sets for each catchment: (a) the MFDs that did not propagate SFDs (i.e., SFDs do not occur), and (b) the MFDs that propagated SFDs.We calculated the antecedent storage conditions (in percentiles) for both sets of MFDs for each catchment upstream of reservoirs (Figure 7).
In the first set of MFDs, we identified two categories of MFDs based on their antecedent storage conditions.A few MFDs have antecedent storage much higher than the median (50th percentile), while others have antecedent storage much lower than the median.Consequently, we calculated the mean antecedent storage for both the categories of the first set of MFDs for each catchment upstream of reservoirs (Figures 7a and 7b).For the first category of MFDs, the mean antecedent storage varies from ∼70th to ∼85th percentiles for all the catchments, with an average of ∼78th percentiles, indicating high reservoir storage conditions (Figure 7a).The high reservoir storage before the onset of an MFD event indicates sufficient water availability in the reservoir.As a result, even after an MFD occurs in the upstream catchment, a gradual release of the water from the reservoir restricts the rapid decline in streamflow below the 20th percentile (threshold used to define SFDs), preventing the propagation of MFDs into SFDs.
For the second category of MFDs, the mean antecedent storage varies from ∼17th to 30th percentiles for all the catchments, with an average of ∼22nd percentile, exhibiting the low reservoir storage conditions (Figure 7b).The

Water Resources Research
10.1029/2024WR037036 low reservoir storage before the onset of MFDs indicates less reservoir storage, resulting in low flow and conventional long-term streamflow drought.In this case, the onset of MFDs cannot exacerbate the SFDs, that is, the rapid decline of streamflow (from above 40th to below 20th percentile) is not possible since streamflow is already below the threshold.
We estimated mean antecedent reservoir storage for the second set of MFDs (the MFDs that propagated SFDs) for each catchment upstream of reservoirs.In addition, we calculated mean antecedent reservoir storage for the catchments with at least three MFDs that propagate into SFDs.Mean antecedent reservoir storage varies from ∼44th to ∼58th percentile for all the catchments, with an average of ∼51st percentile, which indicates medium storage conditions (Figure 7c).The availability of medium reservoir storage before the onset of an MFD event reflects the normal flow regime.Moreover, the reservoirs do not release water at downstream during the onset of MFDs potentially because of medium storage conditions.As a result, streamflow declines rapidly from normal conditions to below the 20th percentile, leading to the development of SFDs in the downstream region.Overall, both high and low antecedent reservoir storage conditions before the onset of MFDs prevent the propagation of MFDs into SFDs.High antecedent storage maintains streamflow above the threshold, while low antecedent reservoir storage favours the low flow (below the threshold) conditions in the downstream restricting rapid decline of streamflow.Therefore, the medium reservoir storage condition before the onset of MFDs is favourable for the propagation of MFDs to SFDs.

Role of ENSO Variability on SFDs
The ENSO exerts an external forcing on Indian summer monsoon rainfall, directly impacting streamflow in India (Ihara et al., 2007).To examine the role of climate variability on the occurrence of SFDs, we examined the propagation of MFDs to SFDs during the summer monsoon season for the El Nino (warm) and La Nina (cool) phases in the absence/presence of the dams.We find a higher frequency of SFDs in the El Nino phase than in the La Nina phase in the absence of dams (Figure 8a).The warm phase of ENSO is associated with deficient and weak summer monsoon circulation, which delays the onset of the Indian summer monsoon (Athira et al., 2023;Ramesh Kumar et al., 2005;Sahastrabuddhe et al., 2023), contributing to monsoon breaks and the development of MFDs Water Resources Research 10.1029/2024WR037036 (Mishra, Aadhar, & Mahto, 2021).The development of MFDs in the catchments upstream of dams/reservoirs causes a rapid decline of streamflow, leading to SFDs.In contrast, surplus summer monsoon rainfall during the La Nina phase ensures a higher inflow to dams (Kripalani & Kulkarni, 1997) and a relatively lesser frequency of SFDs.In the presence of dams, the frequency of SFDs is reduced considerably during both phases of the ENSO, which is modulated by antecedent reservoir storage.Low and high antecedent reservoir storage restricts the propagation of MFDs into SFDs, while medium antecedent reservoir storage favors MFDs propagation into SFDs (Figure 8c).The effectiveness of dams for SFDs was ∼0.87 and ∼0.89 during the El Nino and La Nina phases, respectively.Furthermore, we estimated the frequency of conventional/long-term streamflow droughts (SDs) in the absence and presence of dams during the El Lino and La Nina phases.We considered the occurrence of conventional streamflow drought when the 4-month streamflow standardized index (SSI) for the summer monsoon period (June-September) falls below 1.Similar to SFDs, we find a higher frequency of SDs during the El Nino phase than La Nina in the absence of dams and in the presence of dams (Figure 8b).The summer monsoon rainfall deficit during the El Nino phase leads to conventional SDs.Almost all the catchments upstream of reservoirs show negative/positive rainfall and streamflow anomalies for the summer monsoon season during the El Nino/La Nina phases (Figures 8d and 8e).However, a few catchments located in peninsular India show negative anomalies of rainfall and streamflow during the La Nina phase.South India receives relatively less rain in the summer monsoon season during the cool phase (La Nina) of ENSO (Hrudya et al., 2021;Mishra, Aadhar, & Mahto, 2021).
In the presence of dams, the frequency of SDs does not decrease considerably (Effectiveness ≈ 0.35) during the El Nino phase, while there is no reduction (Effectiveness = 0) in the frequency of SDs during the La Nina phase.The prolonged conventional droughts can impact the effectiveness of dams as they may not be able to supply water downstream (M.Wang, Jiang, et al., 2022).We find 12 SDs during the La Nina phase in the absence of the dam (Figure 8b), which are caused by negative rainfall anomaly (Figure S13 in Supporting Information S1).Most drought events occur in the summer monsoon during the El Nino phase in India (Mishra, 2020).Further, the presence of a dam fails to reduce the occurrence of conventional SDs, as the dam becomes ineffective due to declined storage.Overall, similar to SDs, SFDs show a higher frequency of occurrence during the El Nino phase than in the La Nina phase in the absence of dams.Moreover, dams are more effective in reducing the frequency of SFDs than SDs during both positive and negative phases of ENSO.

Flash Drought Episodes
To further establish the role of reservoir storage on the occurrence of SFD, we examined a few selected events using the simulated streamflow and reservoir storage demonstrating the occurrence of MFDs in the catchment upstream of Bansagar dam/reservoir (Ganga river basin) and their propagation into SFDs during the summer monsoon seasons of 2016, 1982, and 1981 (Figure 9) (Figure 9).We calculated the standardized anomalies of average precipitation and air temperature for the catchment (Figures 9a-9c).The large precipitation deficit (standardized anomalies < 1) and anomalously high temperature (standardized anomalies >1) caused the development of MFDs (Figures 9d-9f).In 2016, the precipitation dropped from ∼67th percentile to ∼19th percentile within one pentad duration (18th-19th pentad) and remained below 40th percentile for the three continuous pentads (19th-21st pentads) with mean precipitation as ∼18th percentile (<25th percentiles), leading to MFD (Figure 9d).Further, precipitation increased to the 58th percentile (>40th percentile) in the 22nd pentad, indicating the termination of the MFD.In addition, the mean precipitation for the previous three pentads (16th-18th pentad) from the onset of MFD remains more than the 50th percentile, showing the normal conditions prior to MFD (Figure 9d).Similarly, the development of MFDs occurred in the summer monsoon season of 1981 and 1982 (Figures 9e and 9f).
We examined the propagation of MFDs into SFDs in the absence and presence of the Bansagar dam.In the absence of dams, the high temperature anomalies and precipitation deficit in the upstream catchment cause the development of MFDs, which later propagate to SFDs (Figures 9g-9i).In 2016, after the onset of MFDs, the streamflow dropped from ∼55th to ∼16th percentile in three pentads (19th-22nd pentad) and remained below 40th percentile for four pentads (20th-23rd pentad), with 22.6th percentile as the mean (<25th percentile), indicating the occurrence of SFD after the onset of MFD (Figure 9g).The SFD terminated at 24th pentad when streamflow reached to ∼68th percentile (>40th percentile).During the MFDs, the positive temperature anomalies enhanced the evaporative water demands and depleted soil moisture (Christian et al., 2021).As a result, the reduced moisture flux and high evaporative demand resulted in a decline in inflow (Christian et al., 2021).

10.1029/2024WR037036
Similarly, the development of the SFDs occurred after the onset of MFDs in 1982 and 1981 (Figures 9h and 9i).The onset of SFDs occurred one pentad later than the onset of MFDs in all the 3 years.Moreover, the normal streamflow conditions (mean antecedent streamflow >50th percentile) persisted before the occurrence of SFDs (Figures 9g-9i).
In the presence of dam, MFDs did not propagate into SFDs in 2016 and1982 (Figures 9j and9k), while MFD propagated into SFD in 1981 (Figure 9l).We investigated the role of reservoir storage in the occurrence of SFDs.In 2016, the antecedent reservoir storage (average of 16th to 18th pentad) was ∼80th percentile before the onset of MFD, indicating high storage conditions and sufficient water storage availability before the onset of the MFD.During the MFD, the release of water from reservoir maintained the streamflow percentile above the threshold (20th percentile) (Figure 9j).Thus, the MFD did not propagate to SFD.In 1982, the antecedent storage (average of 4th-6th pentad) was ∼22nd percentile (low storage condition) (Figure 9k), indicating relatively lesser reservoir storage before the occurrence of an MFD.As a result, the downstream streamflow remained below ∼20th percentile throughout the MFD event (Figure 9k).In 1981, the MFD event commenced from the 17th pentad and lasted till the 23rd pentad.The antecedent reservoir storage (average of 15th-17th pentad) was above the 40th Figure 9.Time series of precipitation, air temperature, streamflow (simulated), and reservoir storage (simulated) with MFD and streamflow flash drought (SFD) events in the presence and absence of Bansagar dam during the summer monsoon season of 2016, 1982, and 1981. (a-c) standardized anomalies of precipitation (blue) and air temperature (red) for the catchment upstream of the dam with a gray area highlighting the MFD period, (d-f) precipitation percentiles (blue) during MFD highlighted by blue shaded area, (g-i) streamflow percentiles during SFD highlighted in green shade in absence of dam, (j-l) streamflow (green) in the presence of dam and reservoir storage percentiles (orange) in the downstream of the dam.

Water Resources Research
10.1029/2024WR037036 percentile (Figure 9l) indicating medium reservoir storage condition, leading to SFD (Figure 9l).Overall, the MFDs propagated to SFDs in the absence of dam, while in the presence of dam, the high and low antecedent reservoir storage prevented the propagation of MFDs to SFDs.Moreover, in the presence of dam, the medium antecedent reservoir storage favored the propagation of the MFD to SFD.

Observed and Model Simulated SFDs
We compared a few simulated events from the H08-CaMa-Flood combined model with observed events during the summer monsoon season of 2008, 2010, and 2002 (Figure 10) (Figure 10) for five reservoirs (Rengali, Tungabhadra, Bango, Bargi, and Indira Sagar) for 2000-2018 (Table S6 in Supporting Information S1).We compared the propagation of MFDs into SFDs at Rengali dam using simulated and observed reservoir storage and streamflow at downstream.In 2008, an MFD event occurred from the 17th to 22nd pentad (Figure 10a), which did not propagate to SFD in the model simulations (Figure 10b) as well as in the observations (Figure 10c).The antecedent storage (average of 15th 17th pentad) prior to MFD is 95th and 90th percentile in the simulations and observations,  , (d-f) the model simulated streamflow (green), reservoir storage (orange), and streamflow flash drought (SFD) event (green region) with gray patch showing the MFD period, (g-i) observed streamflow (red) at Jenapur gauge station, reservoir storage (black), and SFD event (red region), (j, k) observed mean antecedent storage for MFDs that do not propagate SFDs, and (l) observed mean antecedent storage for MFDs that propagate SFDs for the five dams (Rengali, Tungabhadra, Bango, Bargi, and Indirasagar).

Water Resources Research
10.1029/2024WR037036 respectively (Figures 10b and 10c).As a result, both simulations and observations indicate that the high antecedent storage conditions maintain the streamflow above the threshold (20th percentile) by releasing the discharge at downstream throughout the MFD, which does not lead to the SFD.Similarly, in 2010, both simulations and observations showed low antecedent storage before the onset of the MFD event (Figures 10d-10f), which did not lead to an SFD event.In 2002, the propagation of MFD to SFD occurred in the model simulations and observations (Figures 10g-10i).The antecedent storage before the onset of the MFD was estimated as ∼57th and 65th percentile in simulations and observations, respectively (Figures 10h and 10i).Consequently, the medium storage conditions before the onset of MFDs in both simulations and observations propagate the MFDs into SFDs.Using observations for the five dams, we estimated the mean antecedent storage for MFDs that propagate/ do not propagate SFDs from observations (Figures 10j-10l).Overall, the mechanism and characteristics of SFDs and their propagation are consistent in both model simulations and observations.

Discussion and Conclusions
Reservoirs play a crucial role in downstream flow regime by attenuation and delay in flow and anthropogenic operating rules (Ayalew et al., 2013;Hawkins, 1969;Mateo et al., 2014;Zajac et al., 2017).Previous studies examined the role of reservoirs on conventional streamflow drought and reported that reservoir regulations can decrease the duration and severity of conventional drought (Wu et al., 2019).We examined the propagation of MFDs to SFDs during the summer monsoon season in the absence and presence of dams.We find that MFDs propagate into SFDs more often in the absence of dams.The frequency and severity of SFDs are reduced in the presence of dams.The development of MFDs in the catchments draining to the dams is characterized by a large precipitation deficit and anomalous high temperatures during the summer monsoon (Mo & Lettenmaier, 2016;Sreeparvathy & Srinivas, 2022).During the summer monsoon, the precipitation deficit due to monsoon breaks increases the evaporative demand, resulting in a drier air column and reduced precipitation formation (Christian et al., 2021), which directly impacts streamflow.Further, the land-atmospheric feedback intensifies the flash drought occurrence (Mahto & Mishra, 2023a), which contributes to the decline in streamflow due to increased evapotranspiration and reduced soil moisture (Kim et al., 2021;Quansah et al., 2021).
Conventional streamflow drought develops slowly and persists for a longer duration (Bhardwaj et al., 2020;Haslinger et al., 2014;Tallaksen & Van Lanen, 2004); therefore, the complete mitigation of conventional drought by reservoir release can be more challenging due to the restricted temporal efficacy of reservoirs (Brunner, 2021;Chang et al., 2019;Luo et al., 2023;Sun et al., 2023).We examined the role of reservoir release on short-span streamflow droughts.High antecedent storage prior to the occurrence of MFDs prevents the propagation of MFDs to SFDs.Because of high storage availability, the reservoirs release discharge during the MFD event to maintain the streamflow above the threshold and prevent the rapid decline of streamflow.In addition, low antecedent storage before the occurrence of MFDs causes low flow conditions in downstream regions, which do not support the occurrence of SFDs.Ehsani et al. (2017) reported that the effectiveness of existing dams in creating drought resilience is limited under the impact of low reservoir storage.In contrast, the medium antecedent storage (near to median storage) favors the propagation of MFDs to SFDs.
We further examined the occurrence of SFDs during the summer monsoon season in the positive and negative phases of ENSO.The warm (EL Nino) phase of ENSO is associated with the weak summer monsoon circulation and low moisture transport during the Indian summer monsoon season, leading to a higher frequency of streamflow droughts (Ashok & Tejavath, 2019;Athira et al., 2023).The summer monsoon during the El Nino phase contributes to the increased occurrence of summer monsoon breaks (Athira et al., 2023;Qadimi et al., 2021;Ramesh Kumar et al., 2005), which favor the development of MFDs that lead to SFDs.Therefore, we found a higher frequency of SFDs during the El Nino phase than the La Nina phase.Above-average rainfall during the La Nina phases contributes to higher streamflow.The increased moisture transport during the negative phase of ENSO contributes to above-average streamflow conditions in the Indian subcontinent region, which do not favour the long-term persistent low flow conditions in the river regime.Therefore, SFDs and SDs are less frequent during the La Nina phase than during the El Nino phase.In the presence of dams, the lack of monsoon rainfall may lead to a decrease in the reservoir's level, which can impact the effectiveness of dams in modulating the drought occurrence.However, quick reservoir release can modulate SFDs (i.e., SFDs).Therefore, the implementation of dams has more influence on SFDs than conventional SDs during the positive and negative phases of the ENSO cycle.
We note the limitations associated with simulated streamflow and reservoir storage.For instance, we did not account for the withdrawal of water from rivers for irrigation, residential, and industrial demands, which can also influence the occurrence of flash droughts.In addition, reservoir operations-related decisions can also influence the release and occurrence of SFDs.However, the standard rule curves are not available for most reservoirs, therefore, we did not consider their role on SFDs.The consideration of groundwater pumping may influence the streamflow droughts in river basins of India.For instance, the increased irrigation water demand has led to extensive groundwater pumping, causing a reduction in groundwater levels (Dangar & Mishra, 2021) and baseflow (Mukherjee et al., 2018).Further, irrigation may alter the surface energy balance and local circulations, affecting water and heat transport from the surface to the atmosphere (Lawston-Parker et al., 2023).The coupling of atmospheric and hydrological models with proper representation of human interventions can further improve the accuracy of hydrological processes (J.Wang, Jiang, et al., 2022).
Notwithstanding the limitations, reservoirs significantly influence the propagation of meteorological to SFDs during the summer monsoon season in India.The reservoirs reduce the frequency and severity of SFDs.The development of SFDs can significantly impact hydropower productivity as a rapid decline in streamflow reduces hydropower generation (Wan et al., 2021).Our findings have implications for mitigating SFDs as the frequency and severity of flash droughts are projected to increase in the twenty-first century (Mishra, Aadhar, & Mahto, 2021).The construction of dams in humid regions where MFDs occur more frequently (Zhang et al., 2017) has a higher tendency to mitigate the SFDs.
Based on the findings, the following conclusions can be made: • The implementation of dams decreases the frequency and severity of SFDs during the summer monsoon season, with more than half of the study catchments experiencing a three-fold decline in SFDs.Moreover, the catchments in southwest India show a relatively higher frequency of SFDs in the absence of dams.• The MFDs propagate into SFDs more frequently in the absence of dams; however, in the presence of dams, almost all catchments exhibit comparatively low frequency (up to three) of MFDs that propagate into SFDs.Moreover, in the presence of dam, half of the catchments show relatively less severity of SFDs (i.e., mean streamflow percentiles ranging from 14th to 20th) compared to the severity in the absence of dams (8th-14th percentiles).• The low (17th-30th percentile) and high (70th-85th percentile) antecedent reservoir storage conditions prior to the onset of MFDs prevent the propagation of MFDs to SFDs, while medium (44th-58th percentile) antecedent storage conditions prior to the onset of MFDs favor the propagation of MFDs to SFDs.• Similar to SDs, SFDs occurred more frequently in the El Nino phase than in the La Nina phase in both scenarios (absence/presence of dams).Moreover, the implementation of dams decreases the frequency of SFDs considerably than SDs in both phases of ENSO.

Figure 1 .
Figure 1.Delineation of upstream catchments of 36 large dams using digital elevation model (DEM).Detailed information for dams is mentioned in TableS2in Supporting Information S1.

Figure 2 .
Figure 2. Coefficient of determination (R 2 ), Nash-Sutcliffe efficiency (NSE), and refined index of agreement (d r ) for calibration and evaluation of the H08-CaMa-Flood combined model.(a-c) performance measures during calibration for daily streamflow at the selected gauge stations,(d-f) performance measures during evaluation for daily streamflow, and (g-i) performance measures during evaluation for daily live storage of reservoirs.More detailed information on the calibration and evaluation period is presented in supplemental TablesS1 and S2in Supporting Information S1.

Figure 3 .
Figure 3.A schematic representation of streamflow flash drought (SFD) identification method.Streamflow drops from above 40th percentile (orange dashed line) to below twentieth percentile (red dashed line) within the three pentads during a SFD.Once the declined streamflow increases back to above the 40th percentile, SFD terminates.

Figure 4 .
Figure 4. Spatial distribution of frequency of meteorological flash droughts (MFDs) occurred in the upstream catchments of dams during the summer monsoon season for 1971-2020.A histogram plot illustrates the distribution of MFDs across all the catchments.

Figure 5 .
Figure 5. Spatial distribution of the frequency and severity of streamflow flash droughts (SFDs) occurred at reservoirs downstream in the absence/presence of dams during the summer monsoon season for 1971-2020.(a, b) the frequency of SFDs in the absence and presence of dams, and (c, d) the mean severity of SFDs in the absence and presence of dams; grayfilled circles represent the dam with no SFD event.Histogram plots illustrate the distribution of frequency and severity of SFDs across all the catchments.

Figure 6 .
Figure 6.Spatial distribution of the frequency of propagation of MFDs to streamflow flash droughts (SFDs) and severity of SFDs (SFDs occurred after MFDs) downstream of reservoirs in the absence/presence of dams during the summer monsoon season for 1971-2020.(a, b) the frequency of propagation of MFDs to SFDs in the absence and presence of dams, and (c, d) mean severity of SFDs (SFDs that occur after MFDs) in the absence and presence of dams; gray-filled circles represent the dam with no SFD event.Histogram plots illustrate the distribution of the frequency of propagation and severity of SFDs across all the catchments.

Figure 7 .
Figure 7. Mean and standard deviation of antecedent reservoir storage prior to meteorological flash drought (MFDs) events when MFDs propagate/do not propagate to streamflow flash drought (SFD) events in each catchment upstream of reservoirs.(a, b) antecedent storage when propagation of MFDs to SFDs does not occur (i.e., SFDs do not exist), and (c) antecedent storage when propagation of MFDs to SFDs occurs.

Figure 8 .
Figure 8. Streamflow flash droughts (SFDs) and conventional streamflow droughts (SDs) with rainfall and streamflow conditions during the summer monsoon season in the El Nino and La Nina years.(a) the frequency of SFDs in the absence/presence of dams for all the 36 catchments upstream of reservoirs, (b) the frequency of SDs in the absence/presence of the dams, (c) antecedent reservoir storage conditions before MFDs when SFDs occur/do not occur, (d) rainfall anomalies during the summer monsoon season, and (e) streamflow anomalies during the summer monsoon season in El Nino and La Nina phases of El Nino Southern Oscillation.

Figure 10 .
Figure 10.Comparison of observed and simulated time series of dam outflow and reservoir storage for Rengali dam in 2008, 2010, and 2002.(a-c) precipitation percentile (blue) time series with MFD event (blue region), (d-f) the model simulated streamflow (green), reservoir storage (orange), and streamflow flash drought (SFD) event (green region) with gray patch showing the MFD period, (g-i) observed streamflow (red) at Jenapur gauge station, reservoir storage (black), and SFD event (red region), (j, k) observed mean antecedent storage for MFDs that do not propagate SFDs, and (l) observed mean antecedent storage for MFDs that propagate SFDs for the five dams (Rengali, Tungabhadra, Bango, Bargi, and Indirasagar).