Seismic Network Noise Correlation Workflow

The seismic instrumentation in Australia has increased gradually over the years from tens of stations in the early 1990s to ~300 active stations in recent years (supplementary Fig. S1a). Among these stations, several long operating networks (e.g., AU and S1) form the backbone array, which are augmented by temporary deployments with an operation period of 1–2 years in target regions across the continent (Figure S1b; also see Fig. 1b). The C² workflow is well suited for networks in Australia where permanent stations are distributed near the coastal areas, surrounding the temporary deployments further inland (see Fig. 1b). We briefly summarize the key processing steps employed to extract the noise correlation functions (NCFs) between synchronous and asynchronous stations. As a first step, conventional ambient noise correlation (i.e., C¹) is conducted between synchronous station pairs. The continuous seismic recordings are cut into 1-hour segments with a 30-min overlap between consecutive time windows. After removing the mean and linear trends, we apply a low-pass filter with a corner frequency of 1.25 Hz and downsample the data to 2.5 Hz. The amplitude spectra of traces are normalized (i.e., spectral whitening) to broaden the frequency content. A daily NCF is obtained by cross-correlating the preprocessed segments and stacking the resulting cross-correlation functions from all (48) time windows. Similarly, daily NCFs are stacked to form a monthly stack. With an ensemble of monthly stacks, we conduct quality control by examining the consistency of NCFs, whereby correlation coefficients between all NCF pairs are computed and those with a below-average value are discarded. The accepted NCFs are stacked to obtain the final NCFs (i.e., C¹ functions), which form the basis for bridging asynchronous stations using the C² approach.
The C² workflow invokes source-receiver interferometry (SRI) to project the energy from one receiver via the surrounding backbone arrays to the other receiver^{41 (link)}. The application of C² does not directly cross-correlate the noise recordings between the two target receivers, hence simultaneous operations of the two stations are not required. This idea can be applied to effectively tie asynchronous arrays (supplementary Figure S2). For two temporary arrays deployed at different time periods, we are able to retrieve the inter-array NCFs functions with the aid of the surrounding long-term stations via a three-step process. First, the C¹ is computed between temporary array A and the surrounding stations (supplementary Fig. S2a). Second, temporary array B, which is deployed after the extraction of temporary array A, is cross-correlated with the same set of stations. These two steps effectively turn the surrounding long-term stations into common virtual sources that illuminate both temporary arrays (supplementary Fig. S2b). Finally, for a target station pair, the two C¹ functions from the same virtual source are cross-correlated again to form a C² function, and all C² functions, each corresponding to a different virtual source, are stacked to obtain a final C² estimate. We perform a weighted stacking scheme based on the Voronoi cell tessellation and implement radial and azimuthal tapering as proposed in ref. ^{42 (link)} to improve the stacking. We refer readers to ref. ²² for implementation details. The C² workflow thus provides an indirect approach to retrieve the NCFs between asynchronous stations (or arrays), a situation that cannot be achieved with the conventional C¹ approach. The additional ray paths from C² connect asynchronous stations and provide complementary information to the C¹ dataset (supplementary Fig. S3a; also see Fig. 2). The waveforms of C¹ and C² both show clear surface wave energy with a similar move-out over a large (0–3500 km) distance range (supplementary Fig. S3b, c). We obtain a total of 230,788 and 696,046 NCFs from C¹ and C², respectively.