Abstract: Objective The gully region of the Loess Plateau features complex terrain and prominent spatial heterogeneity in the canopy structure of Robinia pseudoacacia forests. Traditional sparse random sampling cannot accurately capture spatial variability. This study aimed to precisely characterize its spatial variation and determine the reasonable sampling quantity. Methods From June to October 2025, a 30 m × 30 m pure Robinia pseudoacacia forest plot was established in the Zhifanggou Watershed of Ansai County, Shaanxi Province. Based on monitoring data from 248 throughfall collectors during 25 effective rainfall events, we analyzed the spatial distribution patterns and variation characteristics of throughfall in the vigorous and late growth stages, and optimized the minimum sampling number. Results 1) The total throughfall during the growing season reached 458.1 mm, with an average throughfall ratio of 88.9% and a spatial coefficient of variation (CV) of 13.7%. The average throughfall ratios at the peak and late growing seasons were 88.4% and 89.4%, with corresponding CVs of 13.3% and 14.1%, respectively. The CV of throughfall decreased in a power function pattern with the increase of rainfall amount, duration and intensity. 2) The throughfall ratio rose with increasing distance from the tree trunk, with the maximum variability observed near the trunk and the minimum variability in the middle of the canopy. 3) Throughfall showed moderate to strong spatial autocorrelation, with a nugget-to-sill ratio ranging from 0.035 to 0.304. The autocorrelation range varied significantly with the phenological period: the range was the largest (3.87 m) at the peak growing season and the smallest (1.48 m) at the late growing season. 4) At the 95% confidence level and within a 5% relative error, a minimum of 8 and 13 rainfall collectors with a cross-sectional area of 510.7 cm² were required for the peak and late growing seasons, respectively. Conclusions For the entire growing season, it is recommended to adopt a combined radial and grid layout with a minimum of 8 to 13 collectors. For individual Robinia pseudoacacia trees, collectors should be arranged on a concentric ring at 1/2 of the canopy radius. The results of this study can provide a scientific basis for optimizing the throughfall observation scheme of Robinia pseudoacacia forests, improving observation efficiency, and parameterizing relevant hydrological models.