Earthquake Source Characterization in Aotearoa, New Zealand using Joint Analysis of Seismological and Geodetic Data

Data from modern seismological and geodetic networks enable the source properties of earthquakes to be studied in detail but, typically, these different data are analysed independently. In Aotearoa, New Zealand, little systematic analysis of earthquake sources is performed other than the routine seismological analysis of moderate-magnitude earthquakes and the more comprehensive but largely separate seismological and geodetic analyses of large earthquakes.

As such, the opportunity exists to explore routine workflows for finding source parameters that (a) enable seismological and geodetic data to be used independently and jointly and (b) generate comprehensive estimates of uncertainties. In this thesis, we develop a workflow that generates centroid moment tensor (CMT) and rectangular finite source models for moderate-to-large earthquakes in Aotearoa that incorporate both seismological and geodetic observations. Models are computed in a Bayesian bootstrapping framework that generates meaningful uncertainties for model parameters. Using this workflow we analyze (1) the advantages and drawbacks of single- and multiple-dataset source inversions, (2) the impact of the velocity model chosen on resultant source models, and (3) the suitability of our workflow for wider use throughout Aotearoa.

We applied this workflow to the 2014 Eketāhuna, 2013 Lake Grassmere, 2007 George Sound, and 2022 Lake Taupō earthquakes which all encompass distinct tectonic regimes. Joint-dataset models incorporating seismological and geodetic observations were consistently more robust than single-dataset models. The inclusion of seismic data improved rupture duration constraints by ∼87% and better-constrained fault geometry for CMT models. Geodetic observations produced more accurate centroid locations, with the inclusion of GNSS data decreasing depth uncertainty estimates by ∼7% in CMT models of the Lake Taupō earthquake. Geodetic data led to improved depths and faulting geometries in finite source models, while including seismic data decreased uncertainty in fault dimensions by ∼1.5 km. The inclusion of HRGNSS receivers into sparse seismic networks produced centroid depths and dip angles more accurate to published Lake Grassmere source models by ∼10 km and ∼20◦.

The choice of velocity model had little effect on inversions incorporating static GNSS offsets only as source-receiver distances were small. It becomes more important when including stations at greater distances from the source. Velocity models encompassing a small region can artificially increase the prominence of localized geological features, while those spanning large areas have an insufficient resolution at large source-receiver distances. The best-performing velocity models covered the extent of the North or South Island of Aotearoa. Source models computed largely agreed with previous publications, except for that of the Lake Taupō earthquake. Disagreement is suspected to be due to the difficulties in modelling shallow volcanic sources, rather than insufficiencies in the workflow. Uncertainties for the Lake Taupō models were approximately 66% smaller than those for the George Sound models, reflecting the increased availability of geodetic data. All models disagreed with the depths of previous publications. Model depths for the Eketāhuna and George Sound earthquakes were on average ∼8 km deeper than published models, while those for Lake Grassmere and Lake Taupo were shallower by ∼7 km and ∼1 km respectively. We suspect that the inclusion of geodetic stations has led to improved depth constraints.