Self‐noise estimates from spectral correlation analysis of ambient noise data recorded synchronously by three accurately coaligned broadband seismometers have become a de facto standard for comparison of instruments and assessing their performance. Nonetheless, many of the self‐noise spectra published suffer a spectral disturbance within a limited range of frequencies (0.1–1 Hz), corresponding to that of Earth’s secondary microseisms.
In this work, we present a method to improve self‐noise estimates by numerically rotating recorded traces three‐dimensionally prior to self‐noise computation. This increases signal coherency among the three sensors’ recorded traces, which, primarily due to errors in installation and manufacturing limitations in regard to orthogonality of their sensing axes, in most cases are imperfectly aligned during the experiment.
Rotating seismic traces by small angles has a strong effect on self‐noise spectra obtained by coherency analysis predominantly in the aforementioned frequency range. We use this effect to determine optimal angles of alignment between two sensors for each of their components by searching for minimum self‐noise during incremental trace rotation.
The technique is applied to data recorded by four STS‐2 sensors during a long‐term self‐noise monitoring experiment at the Conrad Observatory (Austria), and, for vertical components in the majority of cases, it results in complete removal of the self‐noise disturbances. For horizontal traces, a similar effect can be observed, but self‐noise spectra often show additional, long‐period disturbances, which are broader in frequency range and insensitive to rotation of seismic traces.
Our results further indicate that the method is able to reproduce inaccuracies in sensor alignment as small as 1/100 of a degree. For vertical components, angular resolutions on the order of two millidegrees have consistently been found, potentially rendering the method a means to verify alignment already during the preparation phase of self‐noise experiments.