Abstract: Dense-array observation has become a key method for obtaining high-resolution subsurface structures. However, the spatiotemporal heterogeneity of the noise source distribution significantly reduces the signal-to-noise ratio (SNR) of noise correlation functions (NCF), thereby affecting the extraction of dispersion curves. This study analyzed the noise level characteristics in low-frequency (0.1-1 Hz) and high-frequency (1-10 Hz) bands across dense arrays in typical tectonic regions of eastern and western China, including the Ganyanchi pull-apart basin in Ningxia on the northeastern Tibetan Plateau, the Sanshui Basin in Guangdong, and the Beiliu seismic source area in Guangxi. The Sanshui Basin and Beiliu seismic source area are located on the northern continental margin of the South China Sea and exhibit relatively uniform noise levels in the low-frequency band. High-SNR Rayleigh wave signals were successfully extracted in the 0.1-1 Hz range, enabling effective derivation of dispersion curves using the Frequency-Bessel transform method. Conversely, the Ganyanchi pull-apart basin in the northeastern Tibetan Plateau shows low noise intensity in the low-frequency band, where the basin-mountain coupled topography induces precursor wave interference with near-zero time delays in the NCF waveforms. At high frequencies, localized nonstationary noise wavefields result in low-SNR Rayleigh wave signals within a 4 km range, accompanied by chaotic, non-convergent signals that hinder dispersion curve extraction. Thus, prior to ambient noise data acquisition, the distribution characteristics of ambient noise sources and regional tectonic features must be analyzed to determine the observation scheme.