Abstract: Using hourly precipitation data with a spatial resolution of 0.1°×0.1° merged from automatic regional stations and CMORPH during 2008–2014, together with station observations from two extreme rainfall events in July 2021 and July 2023, we selected eight major heavy rainfall regions over North China and its surroundings. Based on the 1-arc-second ASTER GDEM Version 3 topographic data and the above precipitation data, the two-dimensional discrete cosine transform (2D-DCT) method was applied to investigate the spectral characteristics of topography and torrential rainfall. The results show that terrain spectra exhibit pronounced anisotropy across all regions. The dominant terrain wavelengths are mainly concentrated in the range of 60–200 km. At this scale, strong coherence is found between terrain and rainfall spectra, indicating that terrain features at these wavelengths play a key role in modulating heavy rainfall. In contrast, some regions display multi-scale characteristics, suggesting a combined influence of synoptic-scale systems and terrain forcing. Based on the fraction of terrain variance resolved by numerical models, a horizontal resolution of 1 km can capture more than 90% of terrain variance over most of North China. This provides a useful reference for selecting model resolution in rainfall simulation and forecasting. However, the optimal resolution still depends on regional terrain characteristic scales. Rainfall over the North China Plain is more sensitive to small-scale terrain (<100 km), and thus requires sub-kilometer resolution. In contrast, in the Taihang Mountain foothill regions, where dominant wavelengths are larger, coarser resolutions (e.g., 3–9 km) can still represent the major terrain effects. In addition, rainfall simulation is influenced not only by terrain forcing, but also by initial conditions and physical parameterization schemes.