Reliable precipitation forcing is essential for engineering hydrology, yet in-situ gauge records are often spatially sparse in topographically complex regions. This study evaluates two global reanalysis products over a 38-year period (1985–2022): The European centre for medium-range weather forecasts (ECMWF) Fifth-Generation reanalysis (ERA5) and the National centers for environmental prediction (NCEP) Climate forecast system reanalysis (CFSR) and Climate forecast system version 2 (CFSv2), using observations from 46 rain gauges in the Upper Ping River Basin, northern Thailand. A multi-metric framework assessed rainfall performance at daily, monthly, and annual temporal scales. Raw reanalysis datasets exhibit a pronounced “drizzle effect,” with excessive rainfall frequency and underestimated extreme intensity. ERA5 shows better spatial consistency and monthly temporal correlation (r ≈ 0.80) than CFSR, particularly in mountainous areas. Empirical Quantile mapping (QM) substantially reduces volumetric bias, improves rainfall extremes and upper-tail rainfall characteristics, and enhances overall agreement with observations. Bias-corrected ERA5 provides the most consistent overall performance across spatial and temporal scales. By integrating diagnostics across temporal scales, together with spatial, continuous, categorical, and distributional diagnostics, this study systematically assesses raw and bias-corrected reanalysis rainfall in a tropical mountainous basin. The findings provide practical guidance for selecting and bias-correcting reanalysis rainfall for hydrological and water-resources applications where gauge observations are sparse or incomplete.