The research team, led by scientists at the UNESCO-Madanjeet Singh Centre for South Asia Water Management at the University of Moratuwa in Sri Lanka, together with a colleague at Tohoku University in Japan, focused on the Narayani River Basin, also known as the Gandaki Basin. Covering roughly 36,492 square kilometers and shared by China, Nepal, and India, the basin climbs from the lowland Terai plains at about 60 meters above sea level to peaks exceeding 8,000 meters. This enormous elevation range produces climates spanning humid tropical to polar alpine, with the summer monsoon delivering nearly 80 percent of annual rainfall. The team selected six sub-catchments to represent this diversity: the highly snow-dominated Budhi Gandaki, the moderately snow-influenced Myagdi Khola and Madi Khola, the rainfed reservoir-regulated Tadi Khola and Rapti, and the purely rainfed Manahari Khola.

To detect drought-to-flood transitions objectively, the researchers built their analysis on the Standardized Precipitation Index, or SPI, a widely used metric that converts monthly precipitation into standard deviations from the long-term norm. They computed the month-to-month change in SPI, denoted ΔSPI, for every calendar transition between 1980 and 2014, using daily rainfall records from 17 stations maintained by Nepal’s Department of Hydrology and Meteorology. Historical drought and flood events were compiled from the UNDRR DesInventar database and cross-checked against ReliefWeb, national reports, and other archives. By testing three candidate thresholds, the team found that a ΔSPI of 1.25 best separated meaningful transitions from minor fluctuations. A threshold of 1.0 flagged too many small swings, detecting 15 to 45 events per station, while 1.5 was overly restrictive, typically catching only 1 to 16. The intermediate value of 1.25 closely matched the empirical gap between the distributions of wet and dry anomalies in the historical record, where the lower quartile of positive anomalies sat at 0.98 and the median drought anomaly at −0.23.

With the threshold calibrated, the team examined when these transitions occur. The results revealed striking spatial contrasts. The Manahari Khola recorded only 8 transitions in 35 years, all clustered tightly in the pre-monsoon to early monsoon window, with a circular-statistics mean timing in May–June and a very strong directional concentration of 0.91. The Tadi Khola recorded 17 events, all confined to April through July, and the Rapti recorded 28, mostly between April and July. In these low-snow catchments, the monsoon onset is clearly the master control. By contrast, the Budhi Gandaki exhibited 53 transitions spread across a wide seasonal window whose timing depended strongly on elevation. At the high-elevation Larke Samdo station, events appeared from late autumn through early monsoon, tracking persistent snow cover that exceeded 0.80 in midwinter, while the lowest-elevation station saw events only from March to July.

Detecting the transitions was only the first step. The central question was why similar precipitation jumps produce such different hydrological outcomes. To answer it, the researchers paired ΔSPI with two companion indices: the Standardized Streamflow Index difference, ΔSSI, derived from six river gauges, and the Soil Water Deficit Index difference, ΔSWDI, computed from ERA5-Land near-surface soil moisture. Regression of ΔSPI against ΔSSI quantified how efficiently precipitation anomalies translate into streamflow, while Principal Component Analysis of the three variables together revealed the dominant multivariate patterns. In the snow-dominated Budhi Gandaki, the two approaches told a coherent story of seasonal decoupling and re-coupling. During winter and early spring, when snow cover peaked at 0.86 to 0.88, ΔSPI–ΔSSI relationships were weak or negative: precipitation fell as snow and was stored in the pack rather than reaching the river. Principal components showed opposing loadings for precipitation and streamflow, with PC1 explaining up to 100 percent of the variance in some transitions.

The picture changed dramatically with the melt. In May–June, as snow cover at Larke Samdo dropped sharply, the ΔSPI–ΔSSI slope turned strongly positive, with a regression R² of 0.77, and the PCA loadings for precipitation and streamflow aligned at 0.71 and 0.70 respectively. Rainfall and snowmelt together were now transferring water efficiently into soil moisture and discharge. At the mid-elevation Jagat station, coupling appeared even earlier, with a perfect R² of 1.0 in January–February, while at the low-elevation Aru Ghat station, strong coupling emerged in May–June with an R² of 0.96. This elevation-dependent gradient shows that snowpack acts not merely as a delay but as a reshaper of the entire drought-to-flood transition, pushing the flood window later and higher up the basin.

In the regulated catchments, a different mechanism dominated. The Rapti sub-catchment sits downstream of the Kulekhani-I and II hydropower cascade, with a combined 92 megawatts of capacity and 85.3 million cubic meters of reservoir storage. During the April–May transition, DtF events showed weak or decreasing ΔSPI–ΔSSI relationships, and PCA loadings pointed in opposing directions, consistent with early monsoon rainfall being captured in the refilling reservoir rather than released downstream. By June–July, however, the system flipped: regressions steepened to R² values of 0.75 to 1.00, and PC1 explained nearly all the variance with fully aligned precipitation and streamflow loadings. Once reservoir storage filled, additional rainfall was rapidly routed downstream through controlled releases, amplifying the flood response. The smaller run-of-river scheme on the Tadi Khola buffered less strongly but still subdued pre-monsoon responses, confirming that the type and size of regulation matter.

Soil moisture emerged as a mediator threading through all regimes. During winter and early spring, the ΔSWDI loading typically aligned with ΔSPI, indicating that precipitation anomalies were replenishing depleted soil and subsurface storage. During late pre-monsoon and monsoon transitions, the sign reversed, marking saturated soils with no remaining capacity, where further rainfall could only become runoff. The authors interpret these reversals as evidence that soil moisture regulates the transition pathway itself, distinguishing storage-dominated from transmission-dominated regimes, rather than merely modulating the size of the eventual flood.

The practical implications are significant for a basin where floods between 1990 and 2022 affected more than 3.2 billion people globally and where 1.81 billion people worldwide are exposed to 1-in-100-year flood events, the majority of them in South and East Asia. The framework offers a transferable tool for early-warning systems in data-scarce mountain basins: snow-informed inflow forecasting for snow-fed tributaries, pre-monsoon soil-moisture monitoring for rainfed ones, and adaptive reservoir release strategies, including pre-monsoon drawdown, to reduce compound risk at monsoon onset. The authors caution, however, that the ΔSPI threshold of 1.25 is basin-specific and would need recalibration elsewhere, and that reservoir effects were inferred from hydrological signatures because operational release data were unavailable.

Limitations also temper the findings. The monthly resolution of the indices may miss rapid compound sequences, such as a warm, dry winter followed by intense short-duration rainfall, and ERA5-Land soil moisture reflects only near-surface conditions rather than the root zone. Snow cover fraction served as a proxy for cryospheric storage, and incorporating snow water equivalent would sharpen attribution of melt-driven processes. Most importantly, drought-to-flood events are rare, with samples of only 2 to 7 events per transition, so some very high regression values should be read as indicators of process consistency rather than strict statistical significance. Even so, the study marks a clear advance: by jointly analyzing atmospheric, hydrological, and soil-moisture anomalies, it demonstrates that drought-to-flood whiplash in the Himalayas is not a single phenomenon but a family of regime-specific processes, each demanding its own monitoring and management response.

Subject of Research: Drought-to-flood transition dynamics and compound hydroclimatic extremes in the snow-fed sub-catchments of Nepal's Narayani River Basin

Article Title: Multivariate analysis of compound hydroclimatic extremes in snow-fed sub-catchments: an application for climate-resilient water management of Narayani Basin, Nepal

Article References: Sedai, S., Gunawardhana, L., Bamunawala, J., & Rajapakse, L. (2026). Multivariate analysis of compound hydroclimatic extremes in snow-fed sub-catchments: an application for climate-resilient water management of Narayani Basin, Nepal. Frontiers in Environmental Science, 14, Article 1789376. https://doi.org/10.3389/fenvs.2026.1789376

Image Credits: AI Generated

DOI: 10.3389/fenvs.2026.1789376

Keywords: drought-to-flood transition, compound hydroclimatic extremes, Narayani River Basin, Nepal, Standardized Precipitation Index, snowmelt, soil moisture, reservoir regulation, principal component analysis, Himalayan hydrology, early warning systems, water management