A localisation strategy for scalable and efficient multi-source precipitation blending at high resolutions over large domains

Driven by growing demands for higher resolution and quality in hydrometeorological applications, precipitation products are increasingly being developed at sub-daily and kilometre-scale resolutions by blending information from multiple data sources. However, these developments face substantial modelling complexity, as the integration of multiple heterogeneous sources expands the inter- and intra-source error relationships that need to be represented, particularly when applied across large spatial domains. To address this challenge, this study presents the theoretical formulation and scalable implementation strategies for generalised multi-source precipitation blending within a statistical interpolation framework. We use an error-informed localisation concept, which formalises the insight that, in most blending scenarios, only observations that are both spatially representative and characterised by relatively low errors exert meaningful influence on blended precipitation field. Motivated by this concept, we develop a highly efficient implementation that approximates source influence by selecting only the nearest grid cell from each additional spatial input, enabling a substantial reduction in covariance modelling while preserving the essential blending characteristics. The framework is demonstrated in a three-source blending experiment integrating satellite, radar, and gauge data to produce 2-km and hourly rainfall fields over two radar domains in Australia. Qualitative assessment of precipitation and error spatial structures during extreme events, together with quantitative evaluation using the fractions skill score (FSS), show that the simplified implementation retains blending performance comparable to a comprehensive scheme, while requiring only a fraction of the computational cost. This work provides both conceptual understanding and practical implementation strategies to support large-scale, high-resolution, multi-source precipitation product development.