Median Filtering Methods for Non-volcanic Tremor Detection

Various properties of median filtering over time and space are used to address challenges posed by the Non-volcanic tremor detection problem. As part of a "Big-Data" effort to characterize the spatial and temporal distribution of ambient tremor throughout the Northern San Andreas Fault system, continuous seismic data from multiple seismic networks with contrasting operational characteristics and distributed over a variety of regions are being used. Automated median filtering methods that are flexible enough to work consistently with these data are required. Tremor is characterized by a low-amplitude, long-duration signal-train whose shape is coherent at multiple stations distributed over a large area. There are no consistent phase arrivals or mechanisms in a given tremor's signal and even the durations and shapes among different tremors vary considerably. A myriad of masquerading noise, anthropogenic and natural-event signals must also be discriminated in order to obtain accurate tremor detections. We present here results of the median methods applied to data from four regions of the San Andreas Fault system in northern California (Geysers Geothermal Field, Napa, Bitterwater and Parkfield) to illustrate the ability of the methods to detect tremor under diverse conditions.