Data-driven high‐resolution rainfall reconstruction of 85-year Atlas: A two-tier approach integrating occurrence probability and magnitude

Precipitation gridded products derived from station-based records are crucial for hydrological modelling, uncertainty quantification in reanalysis/remote sensing products, and applications such as flood forecasting. Numerous methodologies exist for precipitation reconstruction, ranging from neighbouring station weighting and geostatistical approaches to advanced machine learning techniques (Serrano‐Notivoli & Tejedor, 2021). However, their performance depends on factors such as regional heterogeneity, network density, and target resolution. To account for rainfall’s dual nature (i.e., occurrence and magnitude), our study draws inspiration from previous studies on dry-event frequency classification (Barancourt et al., 1992) and magnitude variability (Berndt & Haberlandt, 2018). We propose a two-step, data-driven approach that independently models occurrence and magnitude. The resulting product offers high spatiotemporal resolution (i.e., daily maps of precipitation at 2 km resolution) over a long temporal period (from 1940 to present), addressing critical needs in hydrological applications.