Geothermal reservoir production and associated induced seismicity may experience pronounced attention in the near future, given the ambitious plans for reducing greenhouse gas emissions toward a carbon-neutral economy and society. At some geothermal sites, the occurrence of hazard- and risk-prone induced earthquakes caused by or associated with reservoir stimulation has resulted in project shutdown (e.g., Pohang, South Korea, and Basel Deep Heat Mining, Switzerland). At other geothermal sites, the maximum event magnitudes were successfully maintained below a threshold defined by local authorities (e.g., Helsinki St1 Deep Heat project in Helsinki, Finland). In this study, we review some of our results from seismological and geomechanical reservoir characterization at The Geysers geothermal reservoir in California, USA, the largest producing geothermal field worldwide. We relate our findings to other geothermal sites to better understand the variability of reservoir behavior. In particular, we obtain a constant and relatively low seismic injection efficiency at The Geysers, which is interpreted to be related to the large energy dissipation through thermal processes and additional dissipation through aseismic slip, the latter now being considered to play a fundamental role in earthquake nucleation. We discuss some characteristics of the seismicity from The Geysers that suggest stable reservoir seismic injection efficiency and possibly low potential to rupture into large induced earthquakes, reducing the associated seismic hazard.
Recently, Machine learning (ML) has been widely utilized for laboratory earthquake (labquake) prediction using various types of data. This study pioneers in time to failure (TTF) prediction based on ML using acoustic emission (AE) records from three laboratory stick-slip experiments performed on Westerly granite samples with naturally fractured rough faults, more similar to the heterogeneous fault structures in the nature. 47 catalog-driven seismo-mechanical and statistical features are extracted introducing some new features based on focal mechanism. A regression voting ensemble of Long-Short Term Memory (LSTM) networks predicts TTF with a coefficient of determination (R2) of 70% on the test dataset. Feature importance analysis revealed that AE rate, correlation integral, event proximity, and focal mechanism-based features are the most important features for TTF prediction. Results reveal that the network uses all information among the features for prediction, including general trends in high correlated features as well as fine details about local variations and fault evolution involved in low correlated features. Therefore, some highly correlated and physically meaningful features may be considered less important for TTF prediction due to their correlation with other important features. Our study provides a ground for applying catalog-driven to constrain TTF of complex heterogeneous rough faults, which is capable to be developed for real application.
Abstract The long‐term temporal and spatial changes in statistical, source, and stress characteristics of one cluster of induced seismicity recorded at The Geysers geothermal field (U.S.) are analyzed in relation to the field operations, fluid migration, and constraints on the maximum likely magnitude. Two injection wells, Prati‐9 and Prati‐29, located in the northwestern part of the field and their associated seismicity composed of 1776 events recorded throughout a 7 year period were analyzed. The seismicity catalog was relocated, and the source characteristics including focal mechanisms and static source parameters were refined using first‐motion polarity, spectral fitting, and mesh spectral ratio analysis techniques. The source characteristics together with statistical parameters ( b value) and cluster dynamics were used to investigate and understand the details of fluid migration scheme in the vicinity of injection wells. The observed temporal, spatial, and source characteristics were clearly attributed to fluid injection and fluid migration toward greater depths, involving increasing pore pressure in the reservoir. The seasonal changes of injection rates were found to directly impact the shape and spatial extent of the seismic cloud. A tendency of larger seismic events to occur closer to injection wells and a correlation between the spatial extent of the seismic cloud and source sizes of the largest events was observed suggesting geometrical constraints on the maximum likely magnitude and its correlation to the average injection rate and volume of fluids present in the reservoir.
<p>The occurrence of earthquake-repeaters, i.e. co-located seismic events of comparable magnitude with highly similar waveforms breaking the same fault patch with an almost identical mechanism, is generally regarded as an indication that the fault surrounding the earthquake asperity is (aseismically) creeping. Earthquake repeaters can either occur during transient loading, e.g. within the afterslip of large earthquakes, or during the constant tectonic loading of tectonic faults. In this study we consider the latter.</p><p>The Main Marmara Fault (MMF) belongs to the western part of the North Anatolian Fault Zone (NAFZ) between the Anatolian and Eurasian plates and runs close to the population centre of Istanbul below the Marmara Sea. While the main NAFZ branches to the east and west of the MMF ruptured in M>7 earthquakes in the last century, the MMF itself is regarded as a seismic gap with the potential to host an M>7 event in the near future. Knowledge about the amount of aseismic creep of the off-shore MMF strand is important for a better seismic hazard assessment for the city of Istanbul and is heavily debated.</p><p>Building on earlier studies that identified repeating earthquakes in the western part of the MMF, we investigate a newly compiled seismicity catalogue of the Sea of Marmara for repeating events along the complete MMF. The catalogue spans the time period 2006-2020, comprises almost 14,000 events in the magnitude range M0.3-M5.7 and was compiled from regional permanent stations operated by AFAD and KOERI. Phase onset times were automatically picked with a two-step procedure using higher-order statistics and an AIC-representation of the waveforms for crude and fine-tuned estimation of the P- and S-onsets. The resulting onset-times were used in the Oct-tree location algorithm of the probabilistic NLLoc software using a regional velocity model and station corrections to obtain the final hypocentres.</p><p>To search for earthquake repeaters, we divide the MMF into overlapping segments and perform a station-wise cross-correlation analysis for all available event waveforms in each segment. Correlated waveforms start 1 s before the P-wave arrival and include the complete waveform including the S-wave coda. Waveforms were bandpass filtered between 2 and 20Hz to retain a rather wide frequency spectrum. We apply strict selection criteria and identify repeating events only as those with a normalized cross-correlation coefficient larger than 0.9 at at least 3 stations and a temporal separation of more than 30 days to exclude bursts of highly similar events in aftershock sequences or earthquake swarms.</p><p>The highest density of repeating earthquakes is found below the western Marmara Sea (Central Basin and Western High) with a systematic decrease of repeaters towards the east (Kumburgaz Basin) and none at all in the presumably locked Princess Islands section of the MMF immediately south of Istanbul. These results for the first time provide a consistent image of the amount of creep along the entire overdue Marmara section of the NAFZ derived from permanent onshore stations refining earlier results obtained from individual spots using local seafloor deployments.</p>
The Geysers geothermal field located in California, USA, is the largest geothermal site in the world, operating since the 1960s. We here investigate and quantify the correlation between temporal seismicity evolution and variation of the injection data by examination of time-series through specified statistical tools (binomial test to investigate significant rate changes, cross correlation between seismic and injection data, b-value variation analysis). To do so, we utilize seismicity and operational data associated with two injection wells (Prati-9 and Prati-29) which cover a time period of approximately 7 yr (from November 2007 to August 2014). The seismicity is found to be significantly positively correlated with the injection rate. The maximum correlation occurs with a seismic response delay of ∼2 weeks, following injection operations. Those results are very stable even after considering hypocentral uncertainties, by applying a vertical shift of the events foci up to 300 m. Our analysis indicates also time variations of b-value, which exhibits significant positive correlation with injection rates.
Earthquake forecasting is a highly complex and challenging task in seismology ultimately aiming to save human lives and infrastructures. In recent years, Machine Learning (ML) methods have demonstrated progressive achievements in earthquake processing and even labquake forecasting. Developing a more general and accurate ML model for more complex and/or limited datasets is obtained by refining the ‘ML models’ and/or enriching the ‘input data’. In this study, we present an event-based approach to enrich the input data by extracting spatio-temporal seismo-mechanical features that are dependent on the origin time and location of each event. Accordingly, we define and analyze a variety of features such as: (a) immediate features, defined as the features which benefit from very short characteristics of the considered event in time and space, (b) time-space features, based on the subsets of acoustic emission (AE) catalog constrained by time and space distance from the considered event, and (c) family features, extracted from topological characteristics of the clustered (family) events extracted from clustering analysis in different time windows. We use AE catalogs recorded by tri-axial stick-slip experiments on rough fault samples to compute event-based features. Then, a random forest classifier is applied to forecast the occurrence of a large magnitude event (MAE>3.5) in the next time window. Results show that to obtain a more accurate forecasting model, one needs to separate background and clustered activities. Based on our results, the classification accuracy when the entire catalog data is used reaches 73.2%, however, it shows a remarkable improvement for separated background and clustered populations with an accuracy of 82.1% and 89.0%, respectively. Feature importance analysis reveals that not only AE-rate, seismic energy and b-value are important, but also family features developed from a topological tree decomposition play a crucial role for labquake forecasting.
ABSTRACT Earthquake source parameters provide key diagnostic observations to quantify the seismogenic environment and understand earthquake physics. Among them, earthquake stress drop plays an essential role in impacting the frequency content of ground motion. Accurate stress-drop estimation is conditioned on data quality, appropriate modeling of propagation effects, and selection of the source model and inversion techniques. One way to evaluate reliability of stress-drop assessments is to compare results combining different methodologies and assumptions. In this study, we calculate earthquake source parameters for micro- to moderate earthquakes in the Sea of Marmara region, northwestern Türkiye, where the Main Marmara fault encompasses a spectrum of slip behaviors from creeping to locked. We apply two approaches: (1) a spectral fitting approach to constrain the corner frequency, seismic moment, and quality factor, and (2) a nonparametric spectral decomposition approach to isolate source spectra from propagation and site effects. We then estimate the earthquake stress drop using a Brune source model. This leads to source parameter estimates for 1577 and 1549 earthquakes with ML (1.0–5.7) for the spectral fitting and spectral decomposition approaches, respectively. Despite the fundamental differences in methodologies, results from both methods are consistent, particularly in highlighting relative differences within the dataset. Small but statistically significant spatial stress-drop variations are observed along different fault segments of the Main Marmara fault. In particular, lower average stress drops are observed in fault segments partially releasing slip aseismically, with the lowest values observed surrounding earthquake repeaters, which may imply a weaker fault in the creeping region. The M ≥ 5 earthquakes along the Main Marmara fault within the last decade were not followed by significant changes in the stress drop, suggesting no significant reduction of fault stress level or fault strength due to their occurrence, supporting the presumably high stress level on this fault.
Abstract We investigate source processes of fluid‐induced seismicity from The Geysers geothermal reservoir in California to determine their relation with hydraulic operations and improve the corresponding seismic hazard estimates. Analysis of 869 well‐constrained full moment tensors ( M w 0.8–3.5) reveals significant non‐double‐couple components (>25%) for about 65% of the events. Volumetric deformation is governed by cumulative injection rates with larger non‐double‐couple components observed near the wells and during high injection periods. Source mechanisms are magnitude dependent and vary significantly between faulting regimes. Normal faulting events ( M w < 2) reveal substantial volumetric components indicating dilatancy in contrast to strike‐slip events that have a dominant double‐couple source. Volumetric components indicating closure of cracks in the source region are mostly found for reverse faulting events with M w > 2.5. Our results imply that source processes and magnitudes of fluid‐induced seismic events are strongly affected by the hydraulic operations, the reservoir stress state, and the faulting regime.
In this study we analyze the nano- and picoseismicity recorded during the Fatigue Hydraulic Fracturing (FHF) in situ experiment performed in Äspö Hard Rock Laboratory, Sweden. The fracturing experiment composed of six fractures driven by three different water injection schemes (continuous, progressive and pulse pressurization) was performed during the year 2015 inside a 28 m long, horizontal borehole located at 410 m depth. The fracturing process was monitored with two different seismic networks covering a wide frequency band between 0.01 Hz and 100000 Hz, including broadband seismometers, geophones, high frequency accelerometers and acoustic emission sensors. The combined seismic network allowed for detection and detailed analysis of nearly 200 seismic events with moment magnitudes MW < -4 that occurred solely during the hydraulic fracturing stages. We relocated the seismic catalog using double-difference technique and calculated the source parameters (seismic moment, source size, stress drop, focal mechanism and seismic moment tensor). The derived physical characteristics of induced seismicity are compared with the stimulation parameters as well as with the geomechanical parameters of the site.
Short term prediction of earthquake magnitude, time, and location is currently not possible. In some cases, however, documented observations have been retrospectively considered as precursory. Here we present seismicity transients starting approx. 8 months before the 2023 MW 7.8 Kahramanmaraş earthquake on the East Anatolian Fault Zone. Seismicity is composed of isolated spatio-temporal clusters within 65 km of future epicentre, displaying non-Poissonian inter-event time statistics, magnitude correlations and low Gutenberg-Richter b-values. Local comparable seismic transients have not been observed, at least since 2014. Close to epicentre and during the weeks prior to its rupture, only scarce seismic activity was observed. The trends of seismic preparatory attributes for this earthquake follow those previously documented in both laboratory stick-slip tests and numerical models of heterogeneous earthquake rupture affecting multiple fault segments. More comprehensive earthquake monitoring together with long-term seismic records may facilitate recognizing earthquake preparation processes from other regional deformation transients.
