Minute-scale disdrometer signatures of extreme rainfall in Prague (2011–2018): microphysical context for regime-adaptive radar QPE

High-intensity rainfall often occurs on minute time scales and in rapidly changing precipitation regimes, which challenges both real-time monitoring and radar-based quantitative precipitation estimation (QPE). Radar retrieval relations implicitly rely on assumptions about drop size distributions (DSDs), yet DSDs can change substantially between convective rain, embedded convection within larger-scale precipitation, and stratiform rain affected by melting-layer processes. Here we use high-temporal-resolution disdrometer data as an independent microphysical reference to document these regime-dependent differences during intense events and to explore their potential for regime-adaptive QPE and event-based characterization.

We analyse minute-resolution observations from Prague, Czech Republic, measured by a 2D video disdrometer operated by the Institute of Atmospheric Physics of the Czech Academy of Sciences. The dataset covers 2011–2018 and comprises 710 minutes classified as strong convection, 162 minutes as embedded convection, and 418 minutes as stratiform rain with a bright band. Microphysical variability is summarized in the (D_m,log₁₀N_w) phase space, where D_m​ is the mass-weighted mean diameter and N_w​ the normalized intercept parameter, and complemented by metrics sensitive to the large-drop tail.

The three regimes exhibit distinct and physically interpretable “fingerprints”. Strong convection is characterized by larger D_m​, elevated N_w​, and the highest contribution of large drops (typically above ~2–3 mm), consistent with intense coalescence and breakup dynamics during convective peaks. Stratiform rain with a bright band shows smaller characteristic sizes and markedly lower N_w​, reflecting a different balance of microphysical processes linked to melting-layer precipitation. Embedded convection occupies an intermediate region and frequently alternates between convective-like and stratiform-like DSD states, making it the main source of ambiguity at minute scales.

To translate these differences into a practical regime indicator, we employ a linear Support Vector Machine trained on canonical end-members (strong convection vs bright-band stratiform) and interpret the signed distance to the boundary as a probabilistic “convective likelihood” with a transition band. This microphysical perspective provides added value for (i) regime-aware radar QPE assumptions during mixed events, and (ii) event-based indices that communicate not only how extreme an event is, but also what kind of extreme rainfall is occurring—an aspect relevant for urban hydrological impacts and preparedness in a changing European climate.