Reference precipitation (RP) serves as a benchmark for evaluating the accuracy of precipitation products; thus, the selection of RP considerably affects the evaluation. In order to quantify this impact and provide guidance for RP selection, three interpolation methods, namely inverse distance weighting (IDW), ordinary kriging (OK), and geographical weighted regression (GWR), along with six groups of station densities, were adopted to generate different RPs, based on the super-high-density rainfall observations as true values, and we analyzed the errors of different RPs and the impacts of RP selection on the assessment of GPM precipitation products. Results indicate that the RPs from IDW and GWR both approached the true values as the station density increased (CC > 0.90); while the RP from OK showed some differences (CC < 0.80), it was similar to GWR when the station density was low, but the accuracy improved at first and then worsened as the station density continued to increase; the evaluation results based on different RPs showed remarkable differences even under the same conditions; when the average distance between rainfall gauges that were utilized to generate RPs was below the medium value (i.e., d < 20 km), the evaluation based on RP derived from IDW and GWR was close enough to that based on the true precipitation, which indicates its feasibility in evaluating satellite precipitation products.

Multi-source fusion techniques have emerged as cutting-edge approaches for spatial precipitation estimation, yet they face persistent accuracy limitations, particularly under extreme conditions. Machine learning offers new opportunities to improve the precision of these estimates. To bridge this gap, we propose a hybrid artificial neural network–geographically weighted regression (ANN–GWR) model that synergizes event recognition and quantitative estimation. The ANN module dynamically identifies precipitation events through nonlinear pattern learning, while the GWR module captures location-specific relationships between multi-source data for calibrated rainfall quantification. Validated against 60-year historical data (1960–2020) from China’s Yongding River Basin, the model demonstrates superior performance through multi-criteria evaluation. Key results reveal the following: (1) the ANN-driven event detection achieves 10% higher accuracy than GWR, with a 15% enhancement for heavy precipitation events (>50 mm/day) during summer monsoons; (2) the integrated framework improves overall fusion accuracy by more than 10% compared to conventional GWR. This study advances precipitation estimation by introducing an artificial neural network into the event recognition period.