An analogue-conditioned multi-satellite framework for daily precipitation reconstruction

Adrian Huerta , Roberto Serrano-Notivoli, Stefan Brönnimann

Precipitation is a key component of the Earth system, yet remains difficult to reconstruct accurately due to its stochastic nature, physical drivers, and sparse observational coverage. Existing gridded datasets rely on trade-offs between station-based observations, satellite retrievals, and multi-source blending, often introducing inconsistencies and temporal limitations. Here, we present an analogue-conditioned multi-source daily precipitation reconstruction framework that integrates physically consistent analogue-pattern selection with spatial data fusion. By conditioning reconstruction on precipitation analogues, the framework enables the transfer of information across time, allowing satellite-era observations to inform historical reconstructions while preserving large-scale structure. Applied over Tropical South America, the method demonstrates robust performance across observational variants, climate regimes, and spatial scales, accurately reproducing precipitation magnitude and occurrence. The framework provides a unified approach from station to grid scale and reduces dependencies on blended datasets. It generates spatially coherent fields suitable for the analysis of extremes and long-term trends. This approach offers a flexible and scalable pathway for reconstructing high-resolution precipitation datasets in regions with limited observational coverage.