[Objective] River basins with complex terrain and sparse ground-based rainfall observations often fail to fully capture the spatial heterogeneity of rainfall. This introduces substantial uncertainties into hydrological runoff modeling, flood disaster risk management, and related applications. The Nujiang River Basin upstream of the Gongshan station in southwestern China exemplifies this challenge: while it covers a vast area with highly rugged topography, it is plagued by a scarcity of rainfall gauges and their uneven distribution. This study aims to address the challenge of design storm estimation in such data-sparse river basins by exploring the use of high-resolution reanalysis rainfall data. [Methods] A systematic performance assessment was first conducted for three precipitation datasets: the China Meteorological Forcing Dataset (CMFD), the Multi-Source Weighted-Ensemble Precipitation product (MSWEP), and ERA5. Daily rainfall values from each dataset were compared with available gauge observations using four statistical indicators: Pearson correlation coefficient (R), relative bias (BIAS), root mean square error (RMSE), and mean absolute error (MAE) to identify the dataset with the best overall performance. Subsequently, for the assessed optimal dataset, the L-moment method was applied in a basin-wide, grid-scale batch processing approach to compute design storms. The specific steps included: (1) determining the most appropriate probability distribution type for each grid point through goodness-of-fit tests; (2) validating the reliability of the method using leave-one-out cross-validation based on observed data. In contrast to conventional analyses typically conducted separately at individual stations, this grid-scale batch processing approach enabled consistent treatment across the entire spatial domain of the river basin, thereby providing a more intuitive reflection of the spatial distribution characteristics of rainfall. Then, annual maximum rainfall series for 1-day, 3-day, and 7-day durations were extracted for each grid cell to compute grid-based design storm values for specified return periods according to the locally optimal probability distribution type and parameters. Finally, spatial analysis was conducted on the design storm values to reveal their spatial distribution patterns. [Results] The evaluation of daily areal average rainfall revealed substantial performance differences among the three reanalysis datasets. ERA5 tended to significantly overestimate rainfall in the Nujiang River Basin, with a BIAS of up to 56%, indicating substantial deviation from ground observations. MSWEP performed better than ERA5. CMFD showed the best performance, displaying a strong correlation with observed station data (R=0.82), an exceptionally low BIAS (1%), and the smallest RMSE (1.38 mm) and MAE (0.78 mm) among the three datasets. The accuracy assessment results at each station further confirmed the superiority of CMFD. Nevertheless, all three datasets exhibited their largest significant errors at Gongshan station—an outcome consistent with previous studies indicating that both reanalysis and merged products struggled to maintain accuracy in areas of steep relief and high spatial rainfall variability. The derivation of design storms using the grid-scale L-moment method driven by CMFD exhibited significant reliability, specifically manifested in the following aspects.At the station scale,the design storm values under the 5% design frequency,estimated by the L-moment method based on CMFD grid data,were highly consistent with results obtained by the Pearson Type Ⅲ distribution’s visual curve-fitting method. At the areal scale, for the 1-day, 3-day, and 7-day annual maximum design storms in the river basin, under the four return periods of 0.1%, 1%, 2%, and 5%, the relative errors between the calculation results of the CMFD-based grid method and the areal design storm values of the river basin derived by the traditional visual curve-fitting method were within 10%. Spatial analysis of the CMFD-derived design storm maps revealed three distinct high-value zones of design storms across the river basin. The high-value zone surrounding Gongshan station was corroborated by observed data, confirming its reliability. The other two potential high-intensity zones were located in areas lacking adequate ground observations and thus required further verification through targeted field campaigns. [Conclusion] The grid-based high-resolution design storm estimation framework in this study overcomes the limitations of traditional station-based methods in data-scarce basins. By integrating optimally evaluated reanalysis precipitation data with the L-moment method applied in a spatially explicit manner, the approach yields detailed precipitation extreme maps that preserve local variation. Compared with the conventional approach that relies solely on limited station data to produce basin-average or single-station design storm estimates, the generated design storm maps deliver substantially enhanced spatial detail and accuracy.