The degradation of the railway subgrade is a common cause for train accidents in Russia because of the extensive railway system and large areas of weak soils. We present a seismic monitoring technique that utilizes moving trains as signal sources to probe the properties of the subgrade. The broadband (periods up to 100 s) seismic sensors recorded signals over several weeks during a non-stop monitoring experiment. The large statistic allows us to identify the signals properties and thus these signals are processed by an automated system. We defined the parameters of the low-frequency signal generated by a passing train that are sensitive to changes in the subgrade state. These are the ratio of amplitudes of horizontal components, and the time interval between the end of the train passage and the maximum amplitude surge in the component transversal to rails. We propose an analytical model to describe the interaction between a moving train and the subgrade that takes into consideration the viscosity of a substrate layer. The application of this model produces a consistent explanation of processes in the media and enables an “in situ” estimation of soil elasticity and viscosity, caused by seasonal thawing.