ETD

• Einstein Telescope
• seismic noise
• Sicily
• wind turbines

The discovery of gravitational waves made through large-scale laser interferometers (e.g. LIGO and VIRGO) is having important implications in cosmology, astrophysics, and nuclear physics, opening a new era for the physical sciences. However, the observatories designed to detect gravitational waves are extremely sensitive to seismic noise originating from different sources such as earthquakes, ocean waves (microseisms), atmospheric disturbances (strong wind and storms), and anthropogenic activities such as industrial installations and wind farms. In particular, the installation of large-scale wind turbines introduces various environmental considerations, one of which is the generation of seismic noise that may have a pernicious effect on gravitational wave detectors. Understanding the implications of seismic noise generated by wind farms is crucial for mitigating the potential impact that this source of noise poses on their operational performance. Seismic noise level is generally dependent on the depth of the infrastructure and on the geology of the site. For this reason, site characterization both in terms of seismic noise and shallow subsurface is thus fundamental for the design of next-generation large-scale gravitational wave detectors and, eventually, to perform successful experiments. Within this context, the European scientific community has proposed the construction of a new gravitational wave detector with significantly higher sensitivity, the Einstein Telescope (ET). Among the different candidate sites to host ET, Italy has proposed the abandoned Sos Enattos mine, located in the municipality of Lula, in Sardinia. This site is among the 10% seismically quietest in the world, making it an ideal location to ensure the detection efficiency of ET. However, a recent project proposed by the company EDP Renewables plans to install a new wind park near this site, posing a risk for the Italian candidature given the significant increase in local seismic background noise caused by vibrations of the aerogenerators. This problem is being faced worldwide by the scientific community since wind exploitation is continuously growing due to the ongoing transition to a more sustainable energy supply for our society, and scientific instruments are susceptible to the disturbances generated by the operation of wind turbines.

To evaluate the effect that next-generation wind farms could have on the ET site, my thesis analysed the seismic noise generated by new types of wind turbines (taller and heavier than the old ones), which were operating at Fulgatore (TP) in northwest Sicily. During my internship, I installed nine broadband three-component seismometers for a one-month acquisition: four on the basement of four different turbines in Fulgatore and five in the archaeological park of Segesta, about 12–13 km away. Three stations in Segesta experienced technical problems, so the analysis was conducted using six stations in total. For correlations of seismic data with the wind farm activity I utilized turbine operational data with 10-minute resolution, kindly provided by EDP Renewables, and additional wind data from SIAS with 1-hour resolution. The typical turbine-induced noise is characterized within the frequency band 0.1–10 Hz. For each station, I calculated power spectral densities (PSD) on 10-minute waveforms and compared them with turbine operating regimes, classified based on rotor speed intervals. At the Fulgatore Array, the spectra showed narrow peaks in all channels (HHZ, HHE, HHN), some at constant and others at variable frequency. Spectrograms confirmed that below 2 Hz the dominant signal is represented by the Blade Passing Frequency (BPF), which ranges from 0.30 to 0.54 Hz depending on rotor speed, and its harmonics, while the 2-10 Hz is characterized by stable peaks at about 2.45 Hz and 5.25 Hz, plus a peak at 8.60 Hz only in the HHZ channel.

To study the near-field ground motion polarization, I filtered the waveforms around these frequencies and applied Principal Component Analysis using singular value decomposition of the data covariance matrices. Results show that BPF and harmonics generate linear oscillations with horizontal dip or inclinations lower than 30°, consistent with the higher amplitudes observed in the power spectral densities of HHN and HHE channels. The peak at 5.25 Hz shows linear polarization with an average incidence angle of approximately 60° and maximum amplitude in HHZ power spectral densities, while the 8.60 Hz peak is linearly polarized along the vertical direction with dips of about 85°: this is consistent with its presence only in the HHZ power spectral densities. The azimuth of the main polarization directions changes with the nacelle orientation, which rotates to face the wind and maximize its exploitation.

After characterizing the spectral features of the wind turbines, I extended the analysis to the Segesta site to identify correlations with the wind farm's operation. Results show that the wavefield radiated from the wind farm is attenuated and not detectable at 12–13 km under the local geological and noise conditions. During maximum turbine activity, PSD values at Segesta are approximately 40–60 dB lower than those at Fulgatore in the 2–10 Hz band. Instead, in the band below 2 Hz the difference is lower because all PSDs are affected by the secondary microseismic peak at 0.22 Hz.

Finally, I extracted a source time function representing the vertical motion of one turbine, made of four sinusoids at 0.54 Hz, 1.08 Hz, 2.45 Hz, and 5.25 Hz, each with a random phase and an amplitude retrieved from an RMS ground velocity statistical analysis during conditions of maximum operational regime of the wind farm. A cross-correlation analysis between station pairs at the Fulgatore array revealed that the phase relationships between different sources do not exhibit any temporal trend and can be assumed to be random. Additionally, after these findings, I carried out a pilot simulation of the wavefield generated by the wind farm using the spectral element method Salvus and the extracted source time function, adapting the approach proposed by Limberger et al. (2022) to the specific requirements of this study. This result could be employed for future studies at the Sardinian site to predict the characteristics of the wind farm-radiated wavefield under the knowledge of the local geological setting.