Developing a Gauge–Radar Merged Precipitation Dataset (1 hour and 1 km) for Great Britain: GRaD-GB (1H1K)

High-quality gridded precipitation datasets are essential for climate analysis and flood-risk assessment in Great Britain (GB); however, such datasets remain limited, and existing products suffer from important limitations. Rain gauge measurements provide highly accurate point-scale observations, but sparse gauge networks limit their applicability. Radar quantitative precipitation estimates (QPEs) offer useful spatial information on rainfall fields at national scale, but suffer from multiple artefacts and errors. Blended rainfall datasets therefore represent a promising approach, as they capitalise on the complementary strengths of radar and gauge observations. Accordingly, this study aims to develop a high-resolution blended precipitation dataset for GB, focusing on two key components: quality control (QC) of radar QPEs and the merging of radar and gauge rainfall.

First, radar QPEs are shown to contain substantial and spatially variable errors even after standard reflectivity-based QC. We assess the Met Office composite radar QPE for GB (hourly, 1 km resolution; 2006–2018) against approximately 1300 hourly rain gauges, demonstrating that errors increase with elevation, distance from radar, and rainfall intensity. Radar QPEs frequently underestimate high-intensity hourly rainfall and fail to detect many extreme events (≥40 mm h⁻¹), with underestimation occurring approximately 1.7 times more often than overestimation (for rainfall ≥0.2 mm h⁻¹). To address these issues, we develop a holistic, rule-based QC framework that exploits spatial–temporal continuity and rainfall-field uniqueness to further quality-control radar QPEs already processed by the Met Office. The framework (i) detects and recovers beam-blocked regions, (ii) classifies normal versus suspect rainfall fields, and (iii) identifies and replaces bad rainfall pixels associated with radar malfunction, ground clutter, and electronic noise. Application of this framework reduces the Root Mean Squared Error (RMSE) relative to gauges from 0.546 to 0.386 (−29%) and increases the correlation coefficient from 0.552 to 0.725 (+31%), while preserving genuine extreme rainfall.

Second, building on the quality-controlled radar product, we introduce a Gauss Blending Method (GBM), adapting the Gauss–Seidel method to merge radar rainfall with gauge constraints (970 gauges) and generate a spatially complete, structure-preserving hourly precipitation field at 1-km resolution. Independent evaluation using 194 gauges (2006–2018) shows that the blended product improves RMSE and mean absolute error by ~14.5% and reduces mean relative error by ~22% compared with radar-only data. The GBM also enhances rainfall detectability and outperforms commonly used adjustment approaches, including the Additive Adjustment, Multiplicative Adjustment, Mixed Adjustment, and Mean Field Bias Adjustment methods. Its overall performance is comparable to Kriging with External Drift; however, GBM shows superior performance for higher rainfall intensities (≥10 mm h⁻¹), provides substantially greater spatial data coverage, better preserves local rainfall variability, and is easier to implement in practice.

Together, the proposed QC framework and GBM enable the production of GRaD-GB (1H1K), an hourly 1-km gauge–radar merged precipitation dataset for Great Britain covering the period 2006–2023. The dataset combines hourly quality-controlled radar QPEs with hourly rainfall observations from approximately 1500 quality-controlled rain gauges. GRaD-GB (1H1K) is well suited for analysing precipitation variability, storm life cycles, and extreme rainfall, thereby providing a robust basis for hydrological applications, flood risk estimation, and extreme rainfall analysis.