Hydrological Drought Propagation is Memory-driven but Recovery is Event-driven

Drought propagation and recovery connect meteorological and hydrological deficits in opposing directions, yet are generally treated as symmetric processes. Here we show they are governed by fundamentally distinct mechanisms. Analyzing over 75,000 matched meteorological–hydrological drought pairs across 3,606 U.S. basins, we found that propagation is driven by antecedent hydrological condition while recovery is predominantly event-driven, governed by characteristics that vary from event to event. Antecedent streamflow conditions explained 79% of propagation-lag variance at the basin scale. Recovery timing was most strongly predicted by false recoveries, the number of times streamflow temporarily exceeded the drought threshold before final termination, with importance distributed across five distinct drivers (R²cv = 0.56). Independently initiated droughts remained unexplained by catchment-scale variables (R²cv ≤ 0). We identified four drought lifecycle phases (Gate, Pathway, Brake, and Regime), each with a distinct dominant predictor. A storage duality emerged at both ecoregion and event scales: the same groundwater reserves associated with faster onset were linked to slower termination, consistent with baseflow deficits persisting after precipitation resumes. These findings indicate that drought recovery cannot be predicted by the same indicators used for propagation, with