Tomographic imaging based on long-term ambient seismic noise measurements, mainly the phase information from surface waves, has been shown to be a powerful tool for geothermal reservoir imaging and monitoring. In this study, we utilize seismic noise data from a dense nodal array (192 3C nodes within 20km2) over a ultra-short observation period (4.7 days) to reconstruct surface waves and determine the high-resolution (0.2km) three-dimensional (3-D) S wave velocity structure beneath a rural town in Zhejiang, China. We report the advantage of cross-coherence over cross-correlation in suppressing pseudo-arrivals caused by persistent sources. We use ambient noise interferometry to retrieve high quality Rayleigh waves and Love waves. Body waves are also observed on the R-R component interferograms. We apply phase velocity dispersion measurements on both Rayleigh waves and Love waves and automatically pick more than 23,000 dispersion curves by using a Machine Learning technique. 3-D surface wave tomographic results after depth inversion indicate low-velocity anomalies (between -1% and -4%) from the surface to 2km depth in the central area. Combined with the conductive characteristics observed on resistivity profile, the low-velocity anomalies are inferred to be a fluid saturated zone of highly fractured rock. Joint interpretation based on HVSR measurements, and existing temperature and fluid resistivity records observed in a nearby well, suggests the existence of the high-temperature geothermal field through the fracture channel. Strong correlation between HVSR measurements and S wave velocity model sheds light on the potential of extraction of both amplitude and phase information from ambient noise.