Multidimensional drought-propagation modes across U.S. catchments under methodological uncertainty

Abstract. Meteorological drought is transformed as it propagates through catchments, but estimated propagation behaviour depends on how drought events are defined and linked. We analysed meteorological-to-hydrological drought propagation across CAMELS-US using four non-redundant response dimensions: propagation probability, onset lag, severity transmission, and recovery extension. Susceptibility was evaluated across 162 combinations of 27 event definitions, three criterion-weighting methods, and two multicriteria ranking methods. Of 488 event-reliable catchments, 438 had sufficient coverage for the methodological ensemble. Catchment classifications derived from a single baseline event definition differed substantially from those obtained after integrating across event definitions (adjusted Rand index = 0.256). The integrated analysis identified two broad response modes: a rapid transmission–attenuation mode and a delayed persistence mode. These modes were comparatively stable under bootstrap resampling (median adjusted Rand index = 0.937), although they overlapped within a continuous multidimensional response space. Continuous susceptibility was largely insensitive to including duration transmission as a fifth criterion (Spearman ρ = 0.990), while approximately half of rank uncertainty arose from interactions among methodological choices. Susceptibility remained moderately predictable under spatially blocked validation, with random-forest R² decreasing from 0.424 under random cross-validation to 0.245 under leave-one-HUC2-out validation. Climate provided most of the transferable predictive information. The transmission–persistence association was weak and became uncertain after adjustment for hydroclimate. Overall, integrating plausible methodological choices produced more reproducible drought-propagation structure than reliance on a single event definition.