From Sparse Stations to AI and Fiber Optics: 35 Years of Earthquake Monitoring Mapped

Earthquake monitoring has undergone a quiet but profound revolution over the past three and a half decades, and a new bibliometric analysis has now charted that transformation in unprecedented detail. By combing through 2,854 scientific publications indexed in the Web of Science Core Collection between 1990 and 2024, researchers Shuhuai Liu, Lei Wu, and Cheng Liao traced how the field evolved from sparse, station-based observation into a data-rich, multi-sensor, and computationally intensive enterprise. Their study, published in Discover Geoscience, used the network-mapping tools VOSviewer and CiteSpace to dissect publication trends, collaboration networks, co-citation structures, and keyword evolution, revealing three distinct technological eras and a striking recent surge of interest in artificial intelligence, fiber-optic sensing, and satellite-geodetic integration.

The first era, spanning 1990 to 1999, was a period of modest output and constrained ambition. Annual publication counts remained low, reflecting the limitations of sparse seismic networks, limited computational resources, and predominantly analog or early digital instrumentation. Monitoring during this decade was largely bound by data availability and processing capacity, so research clustered around waveform interpretation, source characterization, and station-based physical analysis within the classical seismological framework. It was, in essence, an instrument-centered discipline in which a handful of carefully placed stations and expert analysts defined the frontier of what could be observed about the Earth’s shaking.

The second era, from 2000 to 2018, brought sustained growth driven by an observational expansion that fundamentally changed what seismologists could see. Broadband seismic networks proliferated, dramatically improving the fidelity and coverage of ground-motion recordings. At the same time, space-based geodesy matured: continuous Global Navigation Satellite System (GNSS) observations enabled near-real-time tracking of crustal deformation, while Interferometric Synthetic Aperture Radar (InSAR) offered systematic mapping of how the ground shifted across vast regions after earthquakes. These techniques pulled geodesy and seismology into closer collaboration, broadening the research agenda from single-event analysis toward regional-scale monitoring, deformation tracking, and multi-parameter interpretation of seismic processes.

The third and most recent era, beginning in 2019, is defined by an unmistakable acceleration. Publication output rose sharply, and keyword and co-citation patterns became dominated by three emerging fronts: deep-learning-based earthquake detection and phase picking, Distributed Acoustic Sensing (DAS), and the joint use of GNSS and InSAR data. Benchmark studies such as PhaseNet and Earthquake Transformer, which apply neural networks to the automated identification of seismic phases in continuous waveform streams, have become among the most heavily co-cited works in the field. DAS, meanwhile, converts ordinary fiber-optic cables into dense seismic arrays capable of sampling ground motion every few meters over tens of kilometers, on land and across the seafloor. The authors are careful, however, to frame this surge as evidence of rising research attention rather than proof of mature, field-wide operational adoption, since the phase covers only six years and recent publications have had little time to accumulate citations.

Beyond technology, the analysis exposes a deeply uneven global landscape. Although researchers from 105 countries and regions contributed to the corpus, the United States and China together account for roughly half of all publications. The American lead reflects decades of investment in seismic and geodetic infrastructure, early adoption of open-data policies, and sustained methodological innovation. China’s rapid ascent, by contrast, corresponds to large-scale modernization of national monitoring networks, integrated GNSS-InSAR observation systems, and heavy investment in AI-enabled monitoring and earthquake early warning. A second tier of productivity is occupied by Italy, France, and Germany, and several European countries, notably France, Germany, and England, achieve higher citation rates per publication, suggesting an emphasis on methodologically integrative and internationally collaborative work.

The collaboration network analysis places the United States at the center of the field’s international architecture, with high betweenness centrality indicating a bridging role that connects otherwise separate regional research communities. At the institutional level, 2,372 organizations contributed to the corpus, led by national and transnational bodies that operate large observational systems, including the United States Department of Energy, the French National Centre for Scientific Research (CNRS), the China Earthquake Administration, and the Russian Academy of Sciences. Organizations such as INGV in Italy, GFZ in Germany, and the Helmholtz Association exemplify the fusion of operational monitoring mandates with strong research missions, occupying the interface between fundamental seismology and applied hazard mitigation. Notably, institutional productivity and average citation impact do not follow the same rankings, hinting at a structural distinction between data-producing operational agencies and conceptually influential research institutions.

Perhaps the most intellectually interesting finding concerns continuity. Despite three decades of technological upheaval, the co-cited author network remains anchored by foundational contributors to seismic source theory, wave propagation, event location, and earthquake mechanics. Modern monitoring workflows, in other words, still rest on classical seismological principles; the new tools have expanded the scale and speed at which physically grounded analyses can be performed rather than displacing the underlying physics. At the same time, a growing cluster of machine-learning researchers has become prominent in the co-citation structure, indicating a rebalancing of methodological priorities toward automated detection, phase picking, denoising, and large-scale catalog construction. The authors also caution that the growth of AI-related literature reflects not only genuine algorithmic innovation but also the diffusion of established machine-learning methods across new regional datasets and case studies.

The co-cited reference analysis identifies three benchmark directions that now structure the field’s intellectual agenda. The first is automated detection and phase-picking, exemplified by deep-learning models capable of processing vast continuous waveform archives. The second is dense and distributed sensing, where DAS benchmark studies demonstrate that existing fiber-optic infrastructure, including submarine telecommunications cables, can be repurposed as large-scale seismic observation arrays. The third is integrated and shareable data resources: curated benchmark datasets such as STanford EArthquake Dataset (STEAD) have enabled systematic model development, cross-dataset comparison, and reproducible evaluation, while GNSS-InSAR integration increasingly links seismic signals with deformation observations. Together, these clusters suggest that earthquake monitoring is converging on automation, dense observational systems, and multi-sensor data infrastructure as interconnected pillars.

The study’s authors are candid about the limitations of bibliometric evidence. Relying exclusively on the Web of Science may underrepresent regional and non-English journals, and the deliberately broad search query captured some methodologically adjacent industrial and reservoir monitoring work; a conservative sensitivity analysis flagged 376 such records, or 13.17 percent of the corpus, and confirmed that the main growth trajectory and core themes remained stable in the trimmed subset. Citation bias, time-lag effects, and database-driven visibility all complicate interpretation, and the authors stress that co-occurrence and co-citation indicators measure visibility and intellectual influence, not operational performance. No bibliometric signal can establish whether AI genuinely improves detection across all networks, whether DAS outperforms conventional arrays in practice, or whether GNSS-InSAR fusion improves warning performance.

What the analysis does offer is a roadmap. The convergence of seismic, geodetic, and fiber-optic themes points toward testable opportunities in heterogeneous monitoring: combining waveform detections with GNSS displacement for rapid magnitude characterization, using InSAR to map coseismic deformation when cloud-free acquisitions permit, and deploying DAS to densify observations along terrestrial and submarine fiber routes. The authors call for rigorous validation of AI models on geographically and temporally external networks, transparent comparison against classical baselines, error stratification by magnitude and noise regime, and reproducible benchmarks. They also highlight a representation gap: bibliometric visibility concentrates in well-funded systems, while many high-risk developing regions remain underrepresented, underscoring the need for interoperable regional catalogs, shared archives, and benchmark datasets suited to sparse networks and complex terrain. As earthquake monitoring moves toward hybrid paradigms that fuse computational intelligence with physical understanding, this study provides the clearest quantitative picture yet of where the field has been, and where its attention is now heading.

Subject of Research: Bibliometric analysis of technological evolution and research trends in earthquake monitoring from 1990 to 2024

Article Title: Technological evolution and emerging research hotspots in earthquake monitoring from 1990 to 2024 revealed by bibliometric evidence

Article References: Liu, S., Wu, L., & Liao, C. (2026). Technological evolution and emerging research hotspots in earthquake monitoring from 1990 to 2024 revealed by bibliometric evidence. Discover Geoscience, 4(1), Article 387. https://doi.org/10.1007/s44288-026-00744-7

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00744-7

Keywords: earthquake monitoring, bibliometrics, artificial intelligence, Distributed Acoustic Sensing, GNSS, InSAR, deep learning, seismic networks, earthquake early warning, crustal deformation, Web of Science, research trends