Fast and Slow Groundwater Reservoir Dynamics Revealed by Seismic Velocity Changes and Bayesian ICA in a Taiwan Mountain Ridge

Mountain block recharge (MBR) is an important but difficult-to-observe component of mountain water budgets, especially in steep terrain where groundwater measurements are sparse. In such settings, relative seismic velocity changes (dv/v) derived from ambient-noise interferometry provide a potential proxy for hydrological storage variations, but interpretation is complicated because each frequency band may reflect a mixture of processes acting at different depths and spatial scales. Here, we investigate whether multi-frequency dv/v can be used to constrain groundwater storage dynamics at the mountain front. We analyse continuous seismic data from a ridge-scale array in Wanrong, Taiwan, together with precipitation, discharge, and borehole groundwater records, for the period March 2015 to June 2016. Daily dv/v time series are estimated in four frequency bands. All bands show pronounced seasonal and event-scale variations, including a strong velocity decrease during the typhoon season, but they differ markedly in amplitude and temporal evolution. Lower-frequency bands correlate more strongly with groundwater levels in the Hualien plain, whereas higher-frequency bands correlate more strongly with groundwater levels near the mountain front, indicating that the observed dv/v signals reflect a superposition of hydrological responses with different characteristic depths and spatial supports. To separate these mixed signals, we apply a Bayesian Independent Component Analysis approach in which the multi-band dv/v observations are represented as a linear mixture of latent reservoir responses. The inversion resolves two distinct characteristic timescales, with posterior median values of 69 and 160 days. Higher-frequency bands are dominated by the fast component, while lower-frequency bands show a larger contribution from the slow component. Event-wise partitioning of precipitation further suggests that a larger fraction of summer rainfall is routed to the slow reservoir during the typhoon season. These results show that multi-frequency seismic velocity monitoring, combined with Bayesian source separation, can resolve fast and slow groundwater storage dynamics in complex mountainous terrain. The approach provides a new observational constraint on MBR-related processes where direct hydrogeological observations are limited.