Uncertainty-informed regionalization of sub-daily rainfall extremes in northwestern Italy using Bilinear Surface Smoothing with credible-interval-based data screening

Study region: This study focuses on the northwestern part of Italy, where accurate spatial characterization of sub-daily rainfall extremes is essential for hydrological design, flood risk assessment, and infrastructure planning. Observational data are drawn from the I2-RED dataset, comprising approximately 820 rain gauges with records spanning 1927-2020. The region is characterized by complex terrain and substantial spatial variability in rainfall extremes. Study focus: The Bilinear Surface Smoothing with Explanatory variable (BSSE) framework is applied to regionalize mean annual maximum rainfall depths at 1, 3, 6, 12, and 24 h timescales. Elevation is incorporated as an explanatory variable. The stochastic formulation of BSSE was exploited to identify stations whose observations fall outside the credible intervals of the fitted spatial field. These stations were removed, and the BSSE parameters were re-optimized to assess the effect of data screening on in-sample and leave-one-out cross-validation performance. New hydrological insights for the region: The identified outliers predominantly correspond to legacy gauges with short or temporally unrepresentative records, whose anomalously low mean annual maxima likely reflect sampling limitations and, in some cases, documented intensification of rainfall extremes in the region. Their removal leads to systematic performance improvements across all timescales. The results show that heterogeneous rain gauge networks can introduce significant uncertainty in regionalization analyses and that BSSE offers a reproducible framework for spatial interpolation and model-based data quality assessment.