The earthquake dynamics is determined by the preset conditions of fault zone. However, the fault properties at seismogenic depth are hard to constrain, especially for the preseismic phase, during which the fault motion is relatively slow. To better characterize the fault behavior at different stages, we utilize high-resolution seismic catalog to image fault geometry, make statistics on seismicity, and analyze the source spectra. Through case studies spanning continental faults, foreshock sequences, and large earthquakes, this dissertation underscores the pivotal role of microseismicity in illuminating the connection of fault behavior before, during, and after a large earthquake.In the first part of the dissertation, I introduce two earthquake detection workflows: (1) PALM (Zhou et al., 2021b), which integrates phase Picking, Association, Location, and the Matched filter technique to achieve high-completeness detection for intense sequences; (2) LoSAR (Zhou et al., 2024), which Localized a Self-Attention RNN picker to build long-term catalogs with detectability comparable to that of matched filter. These tools offer a generalized approach for earthquake detection, making them suitable for seismic networks commonly deployed worldwide. Systematic tests demonstrate that our methods achieve higher detection completeness and accuracy compared to contemporary algorithms. Additionally, their high temporal stability and computational efficiency make them particularly helpful for studying fault zones and earthquake physics. In the second part of the dissertation, I conducted case studies on fault zones and seismic sequences across diverse tectonic settings, including the East Anatolian Fault Zone in SE Turkey, Xiaojiang fault zone and Yangbi foreshock sequence in SE Tibet, and Ridgecrest-Coso region in California. Our findings revealed key seismicity characteristics related to large earthquakes: (1) Major fault zones often feature subsidiary structures that produce increased microseismicity and elevated b-values, potentially skewing overall b-value estimates; (2) Quiescence along major faults prior to large earthquakes is common, typically accompanied by low b-values, serving as indicators of high seismic hazard; (3) Seismicity depth distribution remains stable before and after large events, useful for inferring fault locking depth; and (4) Foreshock sequences can be more than a triggered cascade, with both inter-event stress transfer and aseismic process play important roles.