Disdrometer-based microphysical contrasts between convective and stratiform rainfall to improve radar rainfall retrievals

Rainfall retrieval algorithms for weather radars are linked to assumptions about drop size distributions (DSDs), but DSD properties vary strongly across rainfall regimes. To reduce regime-dependent biases in radar-based quantitative rainfall estimation, we use high-temporal-resolution disdrometer observations to quantify microphysical differences between strong convection, embedded convection, and stratiform rainfall with a bright-band, and to test how well these regimes can be separated in the (D_m, log₁₀N_w) phase space, where D_m is the mass-weighted mean diameter and N_w the normalized intercept parameter.

Our analysis shows a systematic convective–stratiform contrast. Strong convection has larger characteristic drop sizes and higher normalized concentrations (mean D_m ≈ 1.07 mm; mean N_w ≈ 2.93 × 10⁴ m⁻³ mm⁻¹). Embedded convection has slightly smaller D_m but N_w remains comparably high (mean D_m ≈ 1.02 mm; mean N_w ≈ 2.00 × 10⁴ m⁻³ mm⁻¹). Stratiform rainfall with a bright-band has smaller D_m and markedly lower N_w (mean D_m ≈ 0.92 mm; mean N_w ≈ 6.38 × 10³ m⁻³ mm⁻¹).

Cumulative DSD curves indicate that regime separation is driven primarily by the large-drop tail: strong convection shows the highest contribution of drops above ~2–3 mm, embedded convection is intermediate, and stratiform rainfall declines steeply at large diameters. To translate these findings into an objective regime indicator, we train a linear SVM (Support Vector Machine) on canonical samples (strong convection vs stratiform rainfall with a bright-band) and apply it to all events. Convective and stratiform rainfall are largely separable, while embedded convection occurs on both sides of the boundary, supporting a probabilistic classification with a transition band. These results provide microphysical insights that can be used to refine regime-dependent radar retrieval parameterizations and improve radar-based rainfall estimates at hydrologically relevant scales.