Polarimetric QPE with an X-band radar on the northwest coast of Peru: a ray-coherence diagnostic and a reliability-weighted specific-attenuation estimator

Abstract. The northwest coast of Peru is an arid region punctuated by extreme rainfall events, making accurate quantitative precipitation estimation (QPE) critical for hydro-meteorological risk management. This study develops, validates, and selects locally calibrated polarimetric QPE algorithms for a new dual-polarization X-band radar in Piura, Peru, using raindrop size distribution (DSD) data from the 2024 rainy season applied to radar data from the February–April 2025 season and evaluated against a network of rain gauges.

The manufacturer's default algorithm severely overestimated total rainfall (percentage bias, PBIAS, of +25.6 %), an error removed by the local calibration. Among the locally calibrated candidates, four variants of a specific-attenuation (A_H) algorithm were formulated around different data-selection thresholds for the total differential-phase span, ΔΦ_DP, that constrains the ZPHI retrieval. At the gauges these variants proved nearly indistinguishable, yet noise in the span estimate perturbs each ray by a nearly multiplicative factor, imprinting radial streaks that daily point accumulations cannot register. To measure this artifact directly from the radar data, we introduce a gauge-independent, scan-level diagnostic based on the azimuthal anomaly of ln R, whose primary metric is the ray-coherent fraction of anomaly variance (coh). Applied to more than 26 000 scans, it reveals contamination in all precipitation regimes and under every threshold configuration, whereas a span-independent R(Z_H, K_DP) control sits at the noise floor. This motivates a reliability-weighted estimator that blends R(A_H) with the local composite through a per-ray inverse-variance weight driven by the estimated noise of the ΔΦ_DP constraint itself. The weighted estimator matches the best threshold variant at the gauges while reducing the ray-coherent contamination by factors of 2.4–2.7 in the moderate and heavy regimes, approaching the control floor without blurring the field.