Interpretable machine learning reveals contrasting drivers of streamflow droughts in regulated and natural river systems

Abstract

This study investigates the spatial, temporal, and process-based dynamics of streamflow droughts across the Godavari River basin using daily discharge records from multiple gauging stations. A multivariate framework integrating drought duration, intensity, and deficit was employed to characterize event severity through a joint probability approach. Two composite indices, the Short-Duration Drought Index (SDDI) and Long-Duration Drought Index (LDDI), were derived to distinguish between rapid, high-intensity droughts and prolonged, cumulative-deficit events. Random Forest models and one- and two-dimensional partial dependence analyses were applied to identify dominant drought characteristics and nonlinear interactions among hydrological characteristics across high human-influence (HHI) and low human-influence (LHI) regions. Results reveal a pronounced hydro-anthropogenic dichotomy: HHI sub-basins, heavily regulated by major reservoirs, exhibit long-duration but moderate droughts controlled primarily by operational persistence that alters the natural flows, while LHI catchments display shorter yet more intense events governed by climatic forcing. The analyses further show that in HHI regions, flow intensity dominates short-term drought dynamics, whereas in LHI regions, duration and cumulative deficit are the key controls on long-term drought propagation. Extreme event analysis (>80th percentile) confirms this contrast, highlighting regulation-induced moderation in HHI basins and climatic amplification in natural systems. The findings emphasize that drought management in the Godavari basin must adopt region-specific strategies, improving adaptive dam operations to maintain critical environmental flows in regulated areas and enhancing climatic and ecological resilience in natural basins. The framework offers a transferable approach for diagnosing drought behaviour and associated environmental risks in other large, human-influenced river systems.