When meshed-meteorological datasets are used as reference values for climate scenarios, their climatological accuracy and physical consistency are often limited in regions with sparse observational station networks. In particular, estimated precipitation in mountainous areas are subject to large uncertainties, which may affect the formulation of future strategies for “River Basin Disaster Resilience and Sustainability by All”. One promising approach to addressing this issue is the use of high-resolution, long-term regional reanalysis data, which can help compensate for the spatial inhomogeneity of observational gauge networks and support the development of more robust climate scenario.In this study, we utilize a high-resolution, long-term regional reanalysis dataset (RRJ-Conv) as reference information to improve the accuracy of gridded precipitation in mountainous regions. For comparison, two precipitation reference datasets were constructed: (1) a conventional reference generated using the inverse distance weighting (IDW) method, and (2) an alternative reference generated using a selective inverse distance weighting method that accounts for the spatial correlation structure of daily precipitation represented in RRJ-Conv. Using each reference dataset, river discharge were simulated over Japan at a spatial resolution of 1.8 km for the period 2002–2019, and the simulated monthly mean discharge was compared with observations at each dam sites.The results show that the latter approach especially improved reproduction of peak river-discharge during the summer flood season, particularly at upstream dams located in high mountainous areas of the Tone River basin in the Kanto region. These findings suggest that incorporating spatial precipitation structure information from regional reanalysis data into the generation of reference precipitation fields may mitigate limitations imposed by sparse observational networks and improve the quality of reference precipitation dataset used in developing climate scenarios.