WIZ/WSRZ cross-correlation quality analysis, issue in 2016

[elidana]

@KTJacobs
has found some issues with seismic data from White Island (WIZ and WSRZ), potentially caused by an equipment change at the stations. Reporting below her email

Yates et al 2018 (https://doi.org/10.1029/2018GL080580) showed that there was a change in 2016, potentially attributed to a service at the seismic sites. I don’t know if he got the information from GeoNet field technicians. In the paper, Alec had stacked all of the component pairs, but when I looked at them individually it is clear to see that the change in 2016 is related to the East component at WIZ (see below). Obviously this is similar timing to the Kaikoura eq, but eq signals are typically seen across all component combinations. This helps identify the time of the change, but also the mechanism, since it doesn’t affect the North component I don’t think it can be a rotation for instance. It doesn’t diagnose the problem but can identify a rough time period and maybe range of problems?

A few plots showing a similarity analysis for ambient noise auto-correlations at WIZ and WSRZ.
The autocorrelations have been done as single station analysis (auto) between cross component pairs so there are three for each stations (EZ,NZ,EN). This requires the definition of a “reference” stack and then small time-window stacks through the whole dataset. Below a reference stack of the first 6 months of 2018 is used and the similarity is calculated by crossing the big reference stack against each 10-days stack with time. Something stable with time and close to a value of 1 (roughly similar) is what we’d expect to see. This is mostly observed except for short time periods. However there is a marked change in late 2016 in the East component of WIZ.

WIZ N-Z comps

WIZ E-Z comps

WSRZ N-Z comps

WSRZ E-Z comps

A zoomed view of the 2019 correlation values at each of the separate component pairs. The red blue plots are just colored wiggle plots, so red is an “up” on the trace and blue is “down”. These should be similar through time but a marked change exists in November 2019. The stacks are backward looking (value at each day is that day and the last x-days before that date, 10 days for most plots, 5 days for the red/blue interferogram plots).