Establishing high-frequency noise baselines to 100 Hz based on millions of power spectra from IRIS MUSTANG

Advances in seismic instrumentation have enabled data to be recorded at increasing sample rates. This has in turn created a need to establish higher-frequency baselines for assessing data quality, as the widely-used New High (NHNM) and Low Noise Models (NLNM) of Peterson (1993) do not extend to frequencies above 10 Hz. To provide a baseline for higher frequencies (10-100 Hz), we examine power spectral density probability density functions (PSDPDFs) for high-sample-rate stations available from the Incorporated Research Institutions for Seismology Data Services (IRIS DS) MUSTANG quality control system. We compute high-frequency high and low noise baselines by matching the appropriate composite PSDPDF percentile points to NHNM and NLNM power levels at overlapping frequencies (1-10 Hz) and then extending to higher frequencies (10-100 Hz) with piecewise linear fits to the matching PSDPDF percentile.

We find that the Peterson NLNM remains an accurate representation of the lower bound of global ambient Earth noise since it is matched by only 0.1% of Global Seismographic Network (GSN) PSDs. We present high-frequency high and low noise baselines intended primarily for use by temporary networks targeting high-frequency signals (e.g. monitoring of aftershocks or induced seismicity) based on statistics of PSDPDFs from all publicly available high-sample-rate data.

Most publicly-available high-sample-rate data is recorded by temporary deployments, and the experiment design and scientific targets of these deployments strongly influence the observed statistical distribution of high-frequency noise. We anticipate that the noise baselines presented here will be useful in automated quality control of high-sample-rate seismic data. However, we note that establishing a low noise model that accurately represents the lowest possible ambient Earth noise at frequencies up to 100 Hz will require additional continuous high-sample-rate data from high-quality permanent stations in low-noise environments.