Machine learning prediction of groundwater heights from passive seismic wavefield

SUMMARY Most of water reservoirs are underground and therefore challenging to monitor. This is particularly the case karst aquifers which knowledge mostly based on sparse spatial temporal observations. In this study, we propose a new approach, supervised machine learning algorithm, Random Forests, continuous seismic noise records, that allows prediction river height. The study site aquifer in Jura Mountains (France). An accessible through an artificial shaft instrumented by hydrological probe. generated recorded two broadband seismometers, located (20 m depth) at surface. algorithm succeeds predicting height thanks signal energy features. Even weak river-induced such as surface can be detected used algorithm. Its efficiency, expressed Nash–Sutcliffe criterion, above 95 per cent 53 for data from stations, respectively.