Developing a radar-rain gauge hourly blended precipitation dataset for Great Britain using the Gauss Blending Method

Blended rainfall datasets take advantage of the strengths of multiple sources of rainfall data, producing higher accuracy and broader coverage. However, current blending methods either suffer from low accuracy or are overly complex. Some focus solely on improving statistical performance, but sacrifice the spatial structure of rainfall. Here we develop a Gauss Blending Method (GBM), based on an adaptation of the Gauss-Seidel method, to merge radar and gauge rainfall (970 gauges) in Great Britain to produce a high-resolution (1 km) hourly blended precipitation product. A comparison to 194 independent gauges (2006–2018) demonstrates that the blended product reduces the RMSE, MAE and MRE, compared with the only radar data, by 14.5%, 14.5% and 22.1% respectively, while improving its CC and NSE by 7.8% and 23.2%. GBM also enhances rainfall detectability without introducing new systematic bias. GBM is compared with five standard approaches. The Multiplicative Adjustment Method and Mean Field Bias Adjustment Methods perform poorly and fail to improve radar rainfall. The Additive and Mixed Adjustment Methods show slightly worse performance and less stable rainfall detection skill, producing rainfall fields with artificial discontinuities and wedge-shaped artefacts. The only comparable method across most metrics is Kriging with External Drift. However, the GBM generates more complete data coverage, shows a better performance for extreme rainfall of ≥ 10 mm h −1 , better preserves rainfall structure and local variability, and is easy to apply. This is shown for several example extreme rainfall events. The resulting radar-gauge blended dataset will facilitate the analysis of spatial rainfall variability and extreme rainfall over GB, with significant advantages for extremes analysis, hydrological modeling, and flood risk assessment.