Attempt at creating a high-noise and low-noise model for both the North and South Island of NZ
To characterise the background ambient seismic noise at a station over a range of frequencies the construction of a noise model reflected from the stations power spectral density (PSD) is often the standard. Different approaches have been taken over the years to extract a low-noise and high-noise baseline/threshold but by far the most commonly used has been that of Peterson (Peterson 1993) in which he computed a New Low Noise Model (NLNM) and New High Noise Model (NHNM) by estimating the PSD's of the background noise from 75 stations in the Global Seismograph network (GSN). McNamara and Buland (McNamara and Buland 2004) further extended Petersons noise estimation method by examining the distribution of PSD's using a probability density function (PDF) which is often referred to as a probabilistic power spectral density (PPSD). Due to the age of Petersons NLNM/NHNM (20+ years) and the fact that extremely quiet GSN stations from large contiental interiors were used (ANMO/QSPA) it is often seen that such low noise levels of the NLNM are unattainable for other national seismic networks. With population growth and increased urbanisation near stations these noise levels from 20/30 years ago will progressively become an unrealistic threshold thus McNamara and Buland suggested that for routine monitoring purposes a low noise model produced from a stations PPSD statistical mode (highest probability) should be used to create a Mode Low Noise Model (MLNM). The MLNM is created from the minimum of all station PPSD mode values (per octave) of the vertical component to form a combined low noise threshold for the particular network under examination and may change with time as population increases and technology evolves.
For McNamara and Buland their particular focus for creating a MLNM was the Advanced National Seismic System (ANSS) in the United States. S. J. Rastin (S. J. Rastin 2012) further investigated the concept of a MLNM presented above for the North Island of New Zealand by analysing 5 years of recordings (2005-2009) from the New Zealand National Seismograph Network (NZNSN). A North Island MLNM was produced from processing done in PQLX and a comprehensive discussion had on the comparisons of each station to this low-noise mode baseline.
My desire here is to do two things really:
-
"Pythonise" the creation of the MLNM so it can be done for any seismic network and output the model into a csv and numpy binary format so it can be read in with any seismic PSD plot produced in python/obspy for a realistic low noise baseline comparison for that particular station(s) and related seismic network.
-
Create a MLNM and HNM (high-noise model) for both the North and South Islands of New Zealand from NZNSN recordings in 2020 and save each in the formats above.
First you need to define a list of stations to create the model. I have created two .txt files; NI_NN_list.txt defines the North Island stations of the NZNSN and SI_NN_list.txt defines the South Island NZNSN stations.
Run create_folders.sh to create the station folders from the .txt station lists. The newly created individual station folders are used to store the daily ppsd files. Some changes to the path may be needed depending on how you name your folder (i.e. different seismic network).
The script used to produce the daily station ppsd's is create_ppsd_npz_v2.py. It is designed to run with parrallel processing and therefore you can change the number of cpu cores to use. As I was processing seismic data for the entire year of 2020 (for all NI/SI stations) I ran this externally within a more powerful compute space of 80 cores of which I used 30. For example processing all the North Island station daily ppsd's for 2020 takes 5040 seconds (84 minutes) with 30 cores used. You can also edit the station list, network, channel/component, FDSN client, path, and timeframe.
Finally, to create the MLNM, plot, and save it in csv/npy format run nz_noise_mod.py. Again editing the timeframe, station list, and file paths will be needed. The repo in it's present state has all of the North Island station ppsd's for 2020 already stored in their folders so you can run the script as is to create the North Island MLNM for 2020.
Between S. J. Rastin's NI MLNM published in the 2012 BSSA article and my NI MLNM from 2020 there are a few similarities/differences.
Firstly, two new sites have been added to the NZNSN; GRZ - Sanderson Farm on Great Barrier Island and RTZ - Ruatahuna in the Te Urewera's. As in Rastin's model sites KNZ and PXZ have the largest primary microseism (10-16s) levels again due to their proximity to the coast in the eastern Hawkes Bay region that is likely dominated by stronger swells year round. KNZ had quite high noise levels within the longer periods (30s+) of the earlier model which in 2020 has now diminshed, a sensor swap in 2009 may explain this whereby the previous 30s corner period sensor was swapped for a broadband sensor extending the corner period to 60s. Rastin took this into account and only used 7 months of data from the 60s broadband seismometer (Guralp CMG-3ESPC) but potentially recording for the 5 extra months (Jan-May) in 2020 has the effect of reducing the mode of the long period noise for KNZ.
The greatest similarity between both models 10+ years apart is that of the performance of sites WAZ - Tokomaru West Road in Wanganui and WEL - Dominion Observatory in Wellington. Within the secondary microseism passband (5-10s) the background ambient noise at WAZ is greatly amplified due to the geological setting exceeding the Peterson NHNM at periods slightly lower than 5s (1-3s). Sitting upon weak siltstone and kilometeres of sediment in the Wanganui Basin this strong source of ambient noise is amplified due to low impedance from the unconsolidated nature of the seidmentary geology and the trapping of seismic energy from total internal reflection. KNZ and PXZ again have have the next largest secondary microsiesm peaks due to their coastal proximity. Within the shorter period passband where cutural and higher frequency environmental noise tends to dominate WEL (and WAZ) are the worst performers in both models throughout time. This would not be unexpected for WEL as the site is located rather close to Wellington City subjecting it to strong cultural noise sources.
GeoNet sites SYZ - Waikawa Valley and TUZ - Tuapeka near Invercargill/Dunedin show very high noise levels in the longer periods beginning to break away from the trend of other South Island sites at about 9s and exceeding the Peterson NHNM from 19/20s and above. The cause is likely sensor degradation rather than an envrionemntal source as both sensors are rather old (2006 and 2003 installs respectively) and this may be the dominating contributor to poor long period performance. As of October 2021 a new Nanometrics broadband sensor (TH120-1) has been installed at TUZ and SYZ is scheduled for replacement soon. Other interesting observations are DCZ - Deep Cove's sharp elevated noise at the primary microseism peak and OPZ - Highcliff Hill alongside PYZ - Puysegur Point showing elevated noise over a wide higher frquency passband from 2-0.1s (0.5-10hz).
S. J. Rastin, C. P. Unsworth, K. R. Gledhill, D. E. McNamara; A Detailed Noise Characterization and Sensor Evaluation of the North Island of New Zealand Using the PQLX Data Quality Control System. Bulletin of the Seismological Society of America 2012;; 102 (1): 98–113. doi: https://doi.org/10.1785/0120110064
Daniel E. McNamara, Raymond P. Buland; Ambient Noise Levels in the Continental United States. Bulletin of the Seismological Society of America 2004;; 94 (4): 1517–1527. doi: https://doi.org/10.1785/012003001