Merging dual-polarization X-band radar network intelligence for improved microscale observation of summer rainfall in south Sweden

Abstract

Compact dual-polarization doppler X-band weather radars (X-WRs) have recently gained attention in Scandinavia for sub-km and minute scale rainfall observations. This study develops a method for merging data from two X-WRs in Dalby and Helsingborg, southern Sweden (operated at five and one elevation angle levels, respectively) to improve the accuracy of rainfall observations. In total, 87 rainfall events from May-September 2021, observed by 38 tipping bucket gauges in the overlapping coverage of the X-WRs, were used for ground truth. The gauges were classified into four zones. An artificial neural network using doppler and dual-polarization variables (ANN) and a regression-based hybrid of RATEs (single-level rainfall products built-in to the X-WRs) based on the Marshall-Palmer equation (RMP) were calibrated for each zone. The calibrated models at 5-min scale significantly outperformed RATEs for all zones verified by Gilbert skill score (GSS), relative bias (rBIAS), mean absolute error (MAE), and Nash-Sutcliffe efficiency (NSE) not using the calibration data. Quantile-quantile plots confirmed a considerable improvement of the statistical distribution of the merged rainfall estimates for Zone I (closest to Dalby), II (mid-way between Dalby and Helsingborg), and IV (similar range as II for Dalby but farthest to Helsingborg) especially using ANN. Zone III (farthest to Dalby and closest to Helsingborg) was problematic for all RATEs, ANN, and RMP. The lowest-level elevation angle for both X-WRs showed the most erroneous RATEs. Consequently, the problems with Zone III can be solved if multiple levels of Helsingborg X-WR at higher levels are available.