Radon Anomalies and Earthquake Prediction: Trends and Research Hotspots in the Scientific Literature

Abstract

Radon anomalies have long been explored as potential geochemical precursors to seismic activity due to their responsiveness to subsurface stress variations. However, before this study, the scientific progression of this research domain had not been systematically examined through a quantitative lens. This study presents a comprehensive bibliometric analysis of 379 articles published between 1977 and 2025 and indexed in Scopus and Web of Science. Utilizing the Bibliometrix R-package and its Biblioshiny interface, the analysis investigates temporal publication trends, leading countries, institutions, international collaboration networks, and thematic evolution. The results reveal a marked increase in research output since 2010, with China, India, and Italy emerging as the most prolific contributors. Thematic mapping indicates a shift from conventional geochemical monitoring toward the integration of artificial intelligence techniques, such as decision trees and neural networks, for anomaly detection and predictive modeling. Notwithstanding this methodological evolution, core research themes remain centered on radon concentration monitoring and the analysis of environmental parameters. Overall, the findings highlight the coexistence of traditional and emerging approaches, emphasizing the importance of standardized methodologies and interdisciplinary collaboration. This bibliometric synthesis provides strategic insights to inform future research and strengthen the role of radon monitoring in seismic early warning systems.

1. Introduction

Earthquakes constitute one of the most unpredictable and potentially devastating natural hazards, exerting profound impacts on infrastructure, economic systems, and human life. Despite considerable advances in geophysical research and seismic monitoring technologies, the accurate and timely prediction of seismic events remains a formidable scientific challenge [1,2,3,4,5,6,7].

In this context, the radioactive gases radon (²²²Rn) and thoron (²²⁰Rn)—natural decay products of uranium and thorium, respectively—have been proposed as potential seismic precursors. Both gases can be emitted through geological fractures, and changes in the tectonic stress regime may influence their release rates. Radon, with a half-life of 3.8 days, enables the detection of broader temporal patterns. In contrast, thoron, with its significantly shorter half-life of just 55.6 s, is more suitable for capturing highly localized and recent variations. These distinct properties have led to their widespread use in monitoring studies involving soil gas, groundwater, and atmospheric samples to identify anomalies that may precede seismic events [8,9,10,11,12,13,14,15]. Furthermore, recent theoretical and experimental advances have linked radon-induced air ionization to variations in atmospheric electric fields, thermal anomalies, and ionospheric irregularities, supporting its role in the lithosphere–atmosphere–ionosphere coupling model (LAIC) for short-term earthquake forecasting [16].

Over the past decades, numerous empirical studies have investigated the geochemical behavior of soil gases in relation to seismic activity, consistently reporting anomalous variations, particularly in radon and thoron concentrations, preceding moderate-to-large earthquakes across diverse tectonic settings [9,15,17,18,19]. Research conducted in regions such as the Himalayas, the Apennines, and the Eastern Mediterranean has employed both continuous and passive monitoring strategies in soil gas, groundwater, and cave environments, frequently utilizing solid-state nuclear track detectors and scintillation-based systems [20,21,22]. Moreover, these studies have identified statistically significant anomalies in radon emissions that occur days to weeks prior to seismic events and, in several cases, have proposed empirical relationships linking anomaly amplitude to epicentral distance and earthquake magnitude. To further enhance signal reliability, multivariate statistical techniques and meteorological corrections have been widely applied to distinguish long-term tectonic signal (associated with plate movement) from short-term seismic signals linked to individual earthquakes or swarms and to separate both from environmental noise [23,24].

Despite significant advances in understanding the behavior of radon in seismic contexts, previous reviews have predominantly adopted qualitative or technical-experimental perspectives, focusing on the geological, physicochemical, and environmental factors that influence radon concentrations in soil and groundwater [25,26]. For instance, the study [27] provides a critical overview of ground-based observations of earthquake precursors, including radon emissions, thermal anomalies, and electromagnetic disturbances, highlighting the heterogeneity of methodologies and the lack of statistical consistency across case studies. Several studies have critically examined the limitations arising from methodological heterogeneity, the lack of long-term time series, and the sparse distribution of monitoring networks. These works underscore the need for more robust statistical methodologies and the deployment of broader sensor networks to improve the reliability of radon-related observations [28]. Moreover, other investigations—such as [29,30]—have concentrated on transport mechanisms, the classification of radon anomalies, and the influence of hydrometeorological conditions on radon variability. These contributions have advanced the understanding of radon behavior under varying environmental and tectonic scenarios. However, applying a structured and quantitative perspective is crucial to deepen our understanding of the scientific development of this field, map international collaboration patterns, and explore the evolution of key research themes.

In order to address this gap, the present study includes a comprehensive bibliometric analysis of 379 articles indexed in Scopus and Web of Science from 1977 to 2025. Using the Bibliometrix R package version 4.3.3 and its Biblioshiny interface [31], this review examines publication trends, leading contributors, international collaboration, and thematic evolution, providing a data-driven synthesis and strategic outlook to inform future research and policy on seismic early warning systems.

2. Methodology

2.1. Study Design

This study corresponds to a descriptive and exploratory bibliometric review performed to identify the main scientific trends, thematic lines, leading authors, and international collaboration networks related to analyzing radon anomalies as potential precursors of seismic events. In order to ensure transparency, comprehensiveness, and reproducibility in the process of study identification and selection, the review followed the guidelines established by the PRISMA 2020 statement (Preferred Reporting Items for Systematic Reviews and Meta-Analyses), which provides a standardized methodological framework for conducting systematic reviews [32].

2.2. Database Search Strategy

The systematic search for documents was conducted using the Scopus and Web of Science (WoS) databases, both widely recognized for their extensive coverage in physical sciences, geosciences, and multidisciplinary research. The search was carried out up to April 2025, with no initial restrictions on the year of publication, to capture the historical and temporal evolution of the field.

In Scopus, the query was applied to the Title, Abstract, and Keywords fields, while, in WoS, it was applied using the thematic field TS (Topic Search). The search equation was structured into three thematic groups, which were combined using Boolean operators:

• Group 1: Radon
  (“radon anomaly” OR “radon anomalies” OR “radon concentration” OR “radon gas” OR “radon emission” OR “radon exhalation” OR “soil gas radon” OR “Rn-222”);
• Group 2: Seismicity
  (“earthquake” OR “earthquakes” OR “seismic activity” OR “seismic event” OR “seismic events” OR “seismic precursor” OR “precursory signal”);
• Group 3: Analytical focus
  (“correlation” OR “correlations” OR “predictive value” OR “prediction” OR “forecasting” OR “forecast” OR “statistical analysis” OR “data analysis”).

In addition, to avoid the inclusion of studies unrelated to the geophysical scope of radon, a Boolean AND NOT operator was used to exclude publications focused on mathematical approaches, particularly those addressing the Radon transform. The following terms were excluded:

• Exclusion group: Mathematical focus
  (“Radon transform” OR “Radon integral” OR “Radon mathematical”).

Therefore, the final search equation applied in both databases was (Group 1) AND (Group 2) AND (Group 3) AND NOT (Exclusion group).

2.3. Eligibility Criteria and Screening Procedure

Specific eligibility criteria were applied to define the final dataset for analysis and ensure the thematic relevance and quality of the selected documents.

• Inclusion criteria:

  –

  Articles published in peer-reviewed journals.

  –

  Documents written exclusively in English.

  –

  Availability of complete metadata in the Scopus or Web of Science databases.

• Exclusion criteria:

  –

  Duplicate records across both databases.

  –

  Documents written in languages other than English.

  –

  Documents such as review articles, conference proceedings, book chapters, etc.

These criteria were applied sequentially during the screening and eligibility assessment phases to ensure the thematic consistency, academic relevance, and comparability of the records included in the analysis.

As outlined in Section 2.1, the data collection and screening were conducted according to the PRISMA 2020 guidelines [32]. Figure 1 displays the corresponding PRISMA flow diagram, which details each process stage.

• In the identification phase, 726 records were retrieved, 391 of which came from Scopus and 335 from the Web of Science. Subsequently, 229 duplicate records were removed, resulting in 497 unique records.
• During the screening phase, the titles and abstracts of these records were examined. At this stage, 31 documents were excluded due to language criteria, as they were written in languages other than English: 20 in Chinese, 3 in Russian, 2 in Japanese, and 6 in other languages.
• It is important to note that no documents were classified as “not retrieved” (reports not retrieved = 0). The bibliometric approach adopted in this study based the analysis exclusively on the metadata exported from the selected databases. Therefore, it was not necessary to access the full texts of the articles, as no detailed content assessment was performed. Consequently, 466 records were retained for further evaluation.
• In the eligibility assessment phase, 87 documents were excluded for not meeting the predefined inclusion criteria. These documents were categorized as follows: 61 conference proceedings, 18 review articles, 6 book chapters, 2 documents classified as “other”.

As a result, 379 scientific articles met all inclusion criteria and were incorporated into the bibliometric analysis. This selection process ensured the quality, thematic relevance, and consistency of the final dataset, confirming that each included study explicitly addresses the relationship between radon anomalies and seismic events. The complete list of these documents can be found in Supplementary File S1: Bibliographic dataset, provided as a CSV file.

2.4. Semantic Cleaning of the Dataset

Once the final corpus—comprising 379 articles—was established, the set of extracted keywords underwent a semantic cleaning process. This step aimed to reduce informational noise and enhance the analytical precision of the co-occurrence studies and thematic network construction. In order to achieve these goals, terms considered transversal, redundant, or overly frequent were removed, as their presence lacked the discriminative power to identify specific subtopics within the field. These terms were grouped into four categories:

• Generic methodological or structural terms: article, parameters, field, samples, time, exposure, analysis, identification, monitoring, forecasting, prediction, model, mechanism, system, systems, modelling, carrier, level, phenomena;
• Generic geographical, topographical, or spatial terms: area, region, zone, location, valley, surface, fault, faults, crust, rock, soil, spring, waters, thermal springs, volcano, eruption;
• Overly broad or transversal physical or environmental variables: temperature, pressure, flow, transport, water, gas, air, groundwater, stress, deformation, evolution, permeability, magnitude, hazard, atmosphere, abnormal;
• Duplicate concepts or high-frequency terms lacking discriminative value: radon, earthquake, earth quake, earthquakes, precursor, precursors, variation, variations.

The term “earth quake” appeared likely due to a typographical or parsing error during keyword extraction by the software and was treated as equivalent to “earthquake” for semantic cleaning purposes. Therefore, these terms were excluded due to their limited analytical value, as their inclusion impeded the identification of meaningful thematic patterns. Specifically, they hindered semantic segmentation and generated uninformative nodes in the co-occurrence maps, thereby diminishing the interpretive quality of the results.

2.5. Bibliometric Analysis

The bibliometric analysis was conducted using the Bibliometrix package and its graphical interface Biblioshiny, both developed within the R statistical environment [31]. This tool enables the assessment of scientific output, the exploration of collaboration networks, and the examination of the thematic evolution of a research field. The main analyses performed covered the following aspects: scientific production over time, scientific output by country, an author keyword word cloud, co-occurrence of author keywords, collaboration networks among authors and institutions, collaboration networks among countries, a general thematic map and a thematic map of the last decade, corresponding authors’ countries, most relevant institutional affiliations, most cited authors, sources, and articles.

Finally, these analyses enabled us to identify the patterns of scientific production, the key contributors to the field, the structure of academic collaborations, and the existing research gaps in the study of radon anomalies as potential seismic precursors.

3. Results

This section is organized into four subsections: Section 3.1 systematically examines the temporal evolution of scientific output, Section 3.2 the geographical distribution and international collaboration networks, Section 3.3 the identification of the most influential sources and articles, and Section 3.4 the principal research topics along with their thematic evolution over time. The findings offer a historical overview of the development of this research field and facilitate the identification of collaboration patterns, key contributors, and emerging thematic areas. Hence, this analysis provides a comprehensive and nuanced understanding of the current state of the field and the prevailing research trends.

3.1. Scientific Production and Temporal Evolution

The evolution of scientific output concerning radon anomalies and their potential association with earthquake prediction shows a consistently upward trend over time, as shown in Figure 2. The earliest publications date back to 1977, and there was low and sporadic production during the initial two decades, with annual outputs not exceeding eight articles. However, from the year 2000 onwards, a more sustained growth in research activity is evident. This upward momentum became particularly pronounced starting in 2009, when—for the first time—the number of annual publications reached double digits (10 articles). This output level continued throughout the subsequent decade, albeit with moderate fluctuations. Between 2010 and 2023, annual scientific production stabilized at a relatively high level, culminating in a peak of 23 publications in 2022. As of April 2025, eight articles have already been published, indicating a likely continuation of this positive trend, provided the current publication rate is maintained. Moreover, the cumulative production curve further corroborates this ascending trajectory, surpassing 370 articles by the end of the study period. These findings underscore a growing scientific interest in the topic, likely driven by technological advancements in radon detection, improvements in predictive modeling, and a heightened global awareness of seismic risk management.

3.2. Geographical Distribution and International Collaboration

3.2.1. Production by Countries

The geographical distribution of scientific output on radon anomalies associated with earthquakes shows a notable concentration in Asian and European countries, shown in Figure 3. A quantitative analysis revealed that China leads the field significantly, with 216 publications, followed by India with 194 and Italy with 130. Collectively, these three nations account for a substantial portion of the total scientific production, indicating a strong consolidation of research capacities within these regions.

At the second level, countries with intermediate output levels include Pakistan and Turkey, each contributing 77 documents, and Japan with 43. These figures underscore the active involvement of nations in seismically active zones, consistent with their strategic interest in seismic monitoring and prediction. Moreover, several European countries, such as Greece (38), Spain (26), and Slovenia (24), although producing fewer publications, continue to make sustained contributions to the advancement of knowledge in this area.

Countries like the United States (33) and Iran (29) also emerge as significant contributors. In contrast, Latin American countries—such as Mexico (11) and Peru (4)—alongside various African and Oceanian nations exhibit lower levels of scientific output. This pattern largely reflects differences in seismic risk exposure, the availability of scientific infrastructure, and institutional investment in early warning systems. The corresponding heatmap visually represents this distribution, with more intense shades observable for Asia and Southern Europe—particularly China, India, Italy, Turkey, and Pakistan—and lighter tones across the Americas, Africa, and Oceania. However, it is important to note that countries with lower publication rates may be underrepresented due to limited access to indexed databases or the predominance of publications in local languages.

In this section, publications are assigned to countries based on the affiliations of all contributing authors. The latter allows for capturing the full extent of international collaboration. A separate analysis based on the affiliation of the corresponding author is presented in Section 3.2.2.

Hence, this geographical analysis is further enriched by examining the countries of the corresponding authors, as discussed in the subsequent subsection. This additional perspective is crucial for identifying leading academic institutions and prevailing authorship dynamics.

3.2.2. Corresponding Author Countries

The analysis of the countries of corresponding authors offers valuable insights into the nations at the forefront of research initiatives on radon anomalies in earthquake prediction (see

Table 1. Corresponding author countries.

Country	Publications	Publications (%)	SCP	SCP (%)	MCP	MCP (%)
India	67	17.7 %	63	94.0 %	4	6.0 %
China	49	12.9 %	37	75.5 %	12	24.5 %
Italy	39	10.3 %	34	87.2 %	5	12.8 %
Turkey	24	6.3 %	20	83.3 %	4	16.7 %
Pakistan	18	4.7 %	8	44.4 %	10	55.6 %
Japan	10	2.6 %	9	90.0 %	1	10.0 %
Greece	9	2.4 %	5	55.6 %	4	44.4 %
Slovenia	9	2.4 %	7	77.8 %	2	22.2 %
Iran	8	2.1 %	8	100.0 %	0	0 %
Russia	8	2.1 %	6	75.0 %	2	25.0 %
Korea	7	1.8 %	6	85.7 %	1	14.3 %
Romania	7	1.8 %	7	100.0 %	0	0 %
Spain	7	1.8 %	7	100.0 %	0	0 %
Indonesia	5	1.3 %	5	100.0 %	0	0 %
United States	5	1.3 %	2	40.0 %	3	60.0 %
Germany	4	1.1 %	3	75.0 %	1	25.0 %
Israel	4	1.1 %	3	75.0 %	1	25.0 %
Poland	4	1.1 %	3	75.0 %	1	25.0 %
Croatia	3	0.8 %	3	100.0 %	0	0 %
France	3	0.8 %	2	66.7 %	1	33.3 %
Mexico	3	0.8 %	3	100.0 %	0	0 %

). India is the leading country, with 67 publications, accounting for 17.7% of the total analyzed. Notably, 94% of these publications involve exclusive national collaboration (SCP, Single-Country Publications), while only 6% reflect international co-authorship (MCP, Multiple-Country Publications), indicating a robust internal research capacity.

China, which previously led in total publication volume, ranks second in corresponding authorship, with 49 articles (12.9%). In contrast to India, China demonstrates a higher degree of international engagement, with 24.5% of its publications produced in collaboration with foreign institutions. Italy follows in third place, with 39 publications (10.3%), of which 87.2% are nationally produced and 12.8% are from international partnerships.

Several other countries also exhibit notable levels of output, including Turkey (24 publications), Pakistan (18), Japan (10), Greece (9), Slovenia (9), Iran (8), and Russia (8). Among them, Pakistan stands out for its high proportion of internationally co-authored publications (55.6%), reflecting a research strategy emphasizing global collaboration. Conversely, countries such as Iran, Romania, Spain, Indonesia, Mexico, and Croatia reported 100% national output, which may be attributed to institutionally centered research models or constraints in establishing international networks.

Moreover, countries like the United States, although contributing a smaller number of publications as corresponding authors (5), show significant international collaboration, with 60% of their output involving co-authorship with other nations. Similar trends are observed for Greece, France, and Russia, where international partnerships play an important role in the research output.

This characterization of corresponding authors complements the broader analysis of geographical distribution presented in the previous subsection. It enables a more apparent distinction between research volume and leadership in authorship. Furthermore, it provides the basis for examining the structure and dynamics of collaborative networks, a topic that will be addressed in the following subsection.

3.2.3. Country Collaboration Network

Figure 4 depicts the international collaboration network among countries engaged in research on radon anomalies and their association with earthquake prediction. Each node represents a country, with its size proportional to the frequency of its participation in international collaborations. The links between nodes indicate co-authorship relationships, while the colors group countries into clusters characterized by stronger mutual interactions.

Countries exhibiting the highest centrality and volume of connections include India, China, the United States, and Italy. This pattern is consistent with their leadership in scientific output and the number of corresponding authors, as discussed in the preceding subsections. Notably, India and China constitute the core of a dense collaboration network, particularly collaborating with nations such as Pakistan, the United States, Iran, Vietnam, Germany, Ireland, and Lebanon. These countries form a highly integrated cluster, represented in green.

Italy, meanwhile, leads another prominent cluster (depicted in brown) characterized by strong ties with Slovenia and the Czech Republic. The latter suggests the presence of well-established regional research networks in Central and Eastern Europe. Similarly, Turkey, Greece, Russia, Georgia, Japan, and the United Kingdom constitute a third cluster (orange), reflecting a pattern of intercontinental and transregional collaboration.

In other areas of the network, less densely connected groupings are observed. For instance, the cluster led by France acts as a bridge between various European countries (e.g., Portugal, Israel, Spain, Poland) and Latin American nations such as Peru and Colombia, with additional connections to Canada. This structure indicates a more dispersed, but nonetheless meaningful, collaborative dynamic particularly relevant to transatlantic research initiatives. At the periphery of the network, several countries—including Egypt, Iraq, South Korea, Australia, and the United Arab Emirates—display a lower density of connections. This suggests a more limited or narrowly focused engagement in bilateral collaborations.

Overall, the network reinforces earlier findings: countries with leading roles in scientific production and authorship also dominate the landscape of international collaboration. This interconnected structure facilitates the dissemination of knowledge and supports the advancement of comparative research across regions with varying seismic and geological conditions.

Since these collaborations are often facilitated by academic institutions and highly productive research centers, examining the most relevant affiliations within the field is essential. The following subsection, therefore, focuses on identifying these key institutions, offering a complementary perspective on the organizational actors that play a central role in driving knowledge production in this scientific domain.

3.2.4. Most Relevant Affiliation

Table 2. Most relevant affiliations.

Affiliations (Country)	Articles/Affiliation
University of Azad Jammu and Kashmir (Pakistan)	32
Firat University (Turkey)	24
Guru Nanak Dev University (India), Mizoram University (India)	23
National Cheng Kung University (China—Taiwan), National Center for Research on Earthquake Engineering (China—Taiwan)	21
China University of Geosciences (China)	19
Jadavpur University (India), National Taiwan University (China—Taiwan)	14
National Centre for Physics (Pakistan), Universitas Gadjah Mada (Indonesia)	13
Jožef Stefan Institute (Slovenia)	12
University of West Attica (Greece)	11
University of Catania (Italy), Sapienza University of Rome (Italy)	10
National Central University (China—Taiwan), Tohoku University (Japan), University of Turin (Italy), Wadia Institute of Himalayan Geology (India)	9
Institute of Seismological Research (India), Seoul National University (South Korea), University of Bari (Italy)	8
Afyon Kocatepe University (Turkey), Ankara Yıldırım Beyazıt University (Turkey), Ege University (Turkey), Institute of Earth Sciences (China—Taiwan), National Institute for Earth Physics (Romania), Sakarya University (Turkey), Universidad de Alicante (Spain), Universidad de La Laguna (Spain)	7
Aristotle University of Thessaloniki (Greece), Bhabha Atomic Research Centre (India), Kerman Graduate University of Technology (Iran), Kobe Pharmaceutical University (Japan), Lanzhou Institute of Seismology (China), Ocean University of China (China), Quaid-i-Azam University (Pakistan), University of Michigan (US)	6

shows that the University of Azad Jammu and Kashmir leads the ranking with 32 publications, establishing it as a key reference institution in South Asia—an outcome that aligns with the high research productivity previously reported for Pakistan. It is followed by Firat University in Turkey, with 24 publications, reflecting the country’s significant role in scientific output and participation in international collaboration networks. Guru Nanak Dev University and Mizoram University, based in India, are in third place, with 23 publications. This further underscores the central role of India as a leading contributor to global scientific production and a prominent institutional actor.

Moreover, several of the most relevant affiliations are located in countries that serve as major hubs of international collaboration. For example, the National Cheng Kung University and the National Center for Research on Earthquake Engineering (both in Taiwan) collectively account for 21 publications. At the same time, the China University of Geosciences contributes an additional 19. This trend reinforces earlier observations: institutional prominence is strongly associated with geographic influence and international engagement.

Notable institutions also emerged in Europe, such as the Jožef Stefan Institute in Slovenia (12 publications), the University of West Attica in Greece (11 publications), and the University of Catania and Sapienza University of Rome in Italy (together accounting for 10 publications). These results reflect the consolidation of the Italy–Slovenia–Greece axis within the broader European research landscape.

Furthermore, the range of institutions with seven or more publications highlights the growing diversity of institutional participation. This group includes affiliations from Turkey, India, Taiwan, Romania, Spain, and Indonesia. Among them, the University of Alicante and the University of La Laguna (Spain) stand out, demonstrating the active involvement of Iberian institutions in this scientific field.

Hence, the analysis reveals that many of the most influential institutions are closely aligned with countries, leading to scientific production, corresponding authorship, and international cooperation. This alignment reaffirms the existence of well-established regional research hubs that play a pivotal role in shaping the field.

Having identified the most prominent institutions, the next logical step is to analyze the most cited authors, sources, and articles. The latter will provide a deeper understanding of the organizational structure, intellectual foundations, and leading contributors in the field.

3.3. Most Influential Sources and Articles

3.3.1. Most Relevant Sources

The analysis of publication sources, shown in

Table 3. Scientific journals with the highest number of publications.

Source(s)	Articles/Journal
Natural Hazards (Q1), Applied Radiation and Isotopes (Q2)	20
Journal of Radioanalytical and Nuclear Chemistry (Q2)	19
Radiation Measurements (Q2)	17
Journal of Environmental Radioactivity (Q2)	15
Annals of Geophysics (Q2), Pure and Applied Geophysics (Q2)	11
Geophysical Research Letters (Q1), Scientific Reports (Q1)	9
Nuclear Geophysics (discontinued)	8
Tectonophysics (Q1), Applied Geochemistry (Q1), Arabian Journal of Geosciences (not currently indexed in WoS or Scopus)	6
Applied Geochemistry (Q1), Environmental Earth Sciences (Q1), Natural Hazards and Earth System Sciences (Q1), Current Science (Q2), Journal of Atmospheric and Solar-Terrestrial Physics (Q2), Geochemical Journal (Q2), Geofluids (Q3), Nuclear Tracks and Radiation Measurements (discontinued)	5
Water (Q1), Atmosphere (Q2), Acta Geophysica (Q2), Journal of Physics of the Earth (discontinued; merged into Earth, Planets and Space in 1998)	4

, enabled the identification of scientific journals that have served as the primary channels for disseminating research on radon anomalies associated with earthquakes. The two most prolific journals, each publishing 20 articles, are Natural Hazards (classified in Q1) and Applied Radiation and Isotopes (Q2). Both reflect the interdisciplinary nature of the field, integrating aspects of geophysics, environmental radiation, and natural hazard assessment. These are followed closely by the Journal of Radioanalytical and Nuclear Chemistry (Q2), which published 19 articles, highlighting the strong presence of research employing nuclear analytical techniques applied to geological phenomena. It is important to note that the quartile rankings were verified using the Scimago Journal & Country Rank (SJR) portal.

Radiation Measurements (Q2) and the Journal of Environmental Radioactivity (Q2) also demonstrated significant contributions, with 17 and 15 articles, respectively. Other relevant journals include Annals of Geophysics and Pure and Applied Geophysics (both Q2), each with 11 articles, as well as Geophysical Research Letters and Scientific Reports (both Q1), each with 9 articles—indicative of the publication of high-impact findings in widely recognized and multidisciplinary outlets.

However, it is also important to consider the presence of journals that are now discontinued or delisted. For example, Nuclear Geophysics accounts for eight articles despite ceasing publication in the 1990s. Similarly, Nuclear Tracks and Radiation Measurements produced five articles, although it was succeeded by Radiation Measurements, which remains active. The Journal of Physics of the Earth, with four articles, was absorbed in 1998 by Earth, Planets and Space. Another notable case is the Arabian Journal of Geosciences, which contributed six articles to the dataset. Although it was previously indexed in Scopus and Web of Science, it was removed from both databases in 2021 due to concerns regarding editorial integrity and currently lacks an active quartile classification.

In summary, the results indicate that scientific production on radon and earthquake-related research is predominantly published in Q1 and Q2 journals, with a strong thematic focus on geosciences, environmental radiochemistry, and risk analysis. Moreover, the presence of discontinued journals reflects the historical evolution of the field and the transition of its academic output to more contemporary publication platforms.

3.3.2. Most Relevant Articles

Identifying the most cited articles provides valuable insight into the studies that have attained the highest levels of visibility and impact within the scientific community.

Table 4. Most cited documents.

Article	Year	Citations	Citations per Year
[33]	2011	398	26.53
[34]	1980	158	3.43
[35]	2002	136	5.67
[36]	2007	111	5.84
[37]	1988	108	2.84
[38]	1994	105	3.28
[39]	1987	104	2.67
[21]	2004	102	4.64
[22]	2005	100	4.76
[8]	2015	81	7.36
[24]	2003	81	3.52

presents the documents with the highest citation counts in the analyzed dataset, along with their average number of citations per year, to offer a time-adjusted perspective. This combined approach highlights both the long-standing influence and the recent academic relevance, underscoring the pivotal role of these works in understanding radon anomalies and their potential connection to earthquake prediction. Hence, a qualitative assessment of the most influential studies is presented below, focusing on their main contributions, methodological frameworks, and application contexts.

The article by Jordan et al. [33] is the most cited document within the analyzed corpus, with 398 citations. Commissioned after the 2009 L’Aquila earthquake, the paper outlines the conceptual and regulatory foundations of operational earthquake forecasting (OEF) developed by the International Commission on Earthquake Forecasting for Civil Protection (ICEF). Unlike deterministic prediction, OEF promotes real-time, probabilistic seismic forecasting based on multidisciplinary data, emphasizing transparent and evidence-based communication for civil protection agencies. The study synthesizes scientific, institutional, and societal aspects of earthquake forecasting, offering recommendations on model validation, decision-making under uncertainty, and public communication strategies. Although centered in the Italian context, it has gained international relevance and remains a foundational reference in the evolution of seismic risk management frameworks. Its influence extends to the broader discourse on early warning systems, including those involving geochemical precursors such as radon.

The second most cited article within the corpus, with a total of 158 citations, is the work by Chi-Yu King [34]. This study stands as one of the earliest and most systematic attempts to empirically establish a correlation between anomalous radon concentrations in soil gas (particularly along active fault zones) and the occurrence of earthquakes. The study implemented a monitoring network spanning 380 km of the San Andreas Fault system in California, progressively installing over 60 stations beginning in 1975. The track etch technique was used to measure alpha radiation density emitted by radon accumulated in cups buried 70 cm. This method offers a key advantage: it allows for the integration of measurements over one-to-two-week intervals, thereby reducing the influence of short-term meteorological variability.

The principal finding was the detection of spatiotemporal episodes of increased radon concentration which correlated reasonably well with local earthquakes of magnitudes greater than 4.0. These anomalies were observed across fault segments extending up to 100 km, suggesting large-scale deformation processes that enhance gas exhalation rates from the Earth’s crust. Notably, the study convincingly rules out meteorological fluctuations and local uranium content as primary causes. Instead, it proposes a mechanism based on episodic increases in crustal compression possibly linked to deep aseismic slip events.

This work was pioneering not only for its empirical results but also for its experimental design. It introduces strategies to isolate tectonic signals from environmental noise and establishes robust quality control protocols for radon measurement. Moreover, it advances the novel concept that soil gas radioactivity can indirectly indicate fault mechanics. This idea has since been widely adopted and further developed in the field. Hence, the article provided preliminary but compelling evidence of a correlation between radon anomalies and seismic activity and laid critical methodological foundations for future investigations. Its systematic design, extended monitoring period, and stringent control of environmental variables position it as a foundational reference in the study of geochemical earthquake precursors.

The third most cited article, with 136 citations, is the study by Baubron et al. [35]. This work introduces the use of soil gas profiles—specifically, ²²²Rn, helium (⁴He), and carbon dioxide (${\mathrm{CO}}_2$)—as a tool for characterizing active tectonic structures. The research was conducted in the French Pyrenees Jaut Pass area, a region known for moderate seismic activity and a well-defined fracture network.

The authors systematically sampled four transects, integrating geological observations with in situ geochemical analyses. Significant anomalies in helium and radon concentrations were detected along fractures oriented ${140}^{\circ }$ N–${150}^{\circ }$ E and ${30}^{\circ }$ N–${50}^{\circ }$ E, which were interpreted as deep-seated structures connected to the basement. In contrast, fractures oriented $0^{\circ }$ N–${20}^{\circ }$ E exhibited only shallow concentrations of ${\mathrm{CO}}_2$ and radon, suggesting limited structural connectivity.

Based on the spatial distribution of these anomalies, the authors propose a pull-apart tectonic model in which soil gas degassing patterns correspond to zones of active shear and slow deformation. Moreover, by integrating geochemical data with morphotectonic analysis, the methodology allows for a clear distinction between active fractures and those lacking deep connectivity.

Hence, this study contributes a robust approach to identifying and interpreting subsurface tectonic activity, offering valuable insights into fault behavior through near-surface gas measurements.

The fourth most cited article, the study by Ciotoli et al. [36], cited 111 times, focuses on the geostatistical analysis of soil gases in the Fucino intermontane basin, a seismically active region in central Italy. Utilizing over 800 sampling points distributed across an area of 220 ${\mathrm{km}}^2$, the research assessed concentrations of ²²²Rn, ⁴He, ${\mathrm{CO}}_2$, and methane (${\mathrm{CH}}_4$) to identify both surface and subsurface active tectonic structures through the application of statistical and geospatial methodologies.

The study integrated regional-scale surveys with high-resolution transects, employing advanced geostatistical techniques such as variogram analysis and kriging to model the spatial distribution of soil gases. The results revealed pronounced anisotropy in radon and ${\mathrm{CO}}_2$ concentrations, aligning with active faults oriented 305–315° N, through the structural configuration of the Apennine system. Notably, in the eastern sector of the basin—characterized by deep, well-defined faults—the anomalies exhibited a clear directional pattern. Conversely, in the western sector, where listric faults and thinner sedimentary cover predominate, the anomalies appeared more diffuse across multiple spatial scales.

In addition, the study examined gas transport mechanisms and identified a dominant advective pattern driven by carrier gases such as ${\mathrm{CO}}_2$ and ${\mathrm{CH}}_4$, which act as vectors for trace elements like radon and helium. This transport dynamic facilitates the migration of deep geochemical signals to the surface in spatially localized zones, in line with the conceptual model proposed in [40]. Furthermore, the morphology and distribution of the anomalies were found to depend not only on the structural geometry but also on the current tectonic activity of the faults. Overall, the study constitutes a rigorous and large-scale application of soil gas analysis as a tracer of brittle deformation. It underscores the value of integrating geochemical data with spatial modeling techniques to enhance the structural characterization of tectonically active regions.

With 108 citations, the study by Wakita et al. [37] ranks as the fifth most cited article within the analyzed corpus. This research explores short- and medium-term geochemical anomalies as potential seismic precursors, with particular emphasis on fluctuations in ²²²Rn concentrations in groundwater. Drawing primarily on empirical data from Japan and complemented by observations from China, the study addresses the methodological and conceptual challenges inherent in detecting reliable precursor signals—particularly given the high natural variability and environmental noise that characterize geochemical datasets.

One of the most illustrative cases examined is the Izu-Oshima-Kinkai earthquake (M 7.0, 1978), during which a pronounced radon anomaly was recorded. Radon concentrations began to decline several weeks before the event, reached a minimum at 6 days before the earthquake, and then exhibited a sudden increase immediately preceding the seismic shock. This pattern was consistent with variations observed in other geophysical parameters—such as groundwater temperature, water level, and crustal deformation measured by extensometers—suggesting the presence of a shared tectonic process operating in the lead-up to the earthquake. The study also revisited several historical events, including the Shizuoka earthquake (1965) and the Great Kanto earthquake (1923), where unusual hydrological phenomena—such as spontaneous eruptions of water from wells and geysers—were reported weeks or even months in advance. Although these cases did not always include radon measurements, the authors contend that such manifestations may be linked to a common tectonic origin associated with accumulating and releasing deep-seated stress.

From a methodological standpoint, the study emphasizes the necessity of long-term monitoring, multivariate analytical frameworks, and the development of physical models capable of elucidating the mechanisms driving observed geochemical variations. Moreover, it highlights the irregularity and non-reproducibility of precursor signals: in some instances, even within the same geographic region or monitoring station, anomalies fail to recur in subsequent seismic events, posing significant limitations for their operational use in earthquake forecasting. Taken as a whole, the article provides a critical and thorough assessment of the current state of knowledge regarding geochemical precursors. It offers practical guidelines for their interpretation and strongly advocates for integrating geophysical and geochemical observations within comprehensive seismic monitoring strategies.

Ranking sixth in the analyzed corpus, the article by Virk et al. [38], with 105 citations, provides one of the earliest systematic investigations into the potential correlation between radon anomalies and the occurrence of the Uttarkashi earthquake (mb = 6.5; Ms = 7.0), which struck the Garhwal Himalaya region of India on 20 October 1991. The study was conducted across five monitoring stations located in Amritsar and the Kangra Valley, approximately 330 to 450 km from the epicenter, and employed two complementary detection methods: the plastic film (track etch) technique for weekly integrated measurements and the emanometry technique for daily recordings.

The findings revealed significant increases in radon concentrations in soil gas and groundwater, observed between one and two weeks before the earthquake. Most notably, at the Palampur station, radon levels in soil gas increased by 154% above the historical mean. In comparison, groundwater radon levels rose by 68%. Both were detected 5 days before the seismic event. These anomalies were consistently recorded across different stations and using both detection techniques, reinforcing the hypothesis of a shared tectonic response. An empirical threshold was defined to classify an anomaly as any value exceeding the historical average by more than two standard deviations.

Moreover, the study carefully evaluated the potential influence of meteorological variables, ultimately dismissing their relevance due to the climatic uniformity of the region and the lack of correlation with other environmental factors. The authors further support their interpretation by referencing the Fleischer dislocation model, which posits that tectonically induced changes in crustal porosity and radon migration—either upward or downward—can explain such anomalies, even at significant distances from the epicenter. Hence, through continuous, multisite, and multimethod monitoring, the study demonstrated that radon anomalies can be detected consistently and in advance of moderate-to-strong seismic events. It offers compelling empirical support for using radon as a potential seismic precursor in tectonically active regions.

With 104 citations, the study by Rikitake [39] ranks as the seventh most cited article within the analyzed corpus. The author analyzed 418 records of potential seismic precursors collected in Japan, encompassing geodetic, geochemical, seismic, and electromagnetic phenomena, with the aim of identifying spatial and temporal patterns in their behavior relative to the magnitude of the mainshock. In particular, variations in radon concentration were examined within the category of “precursors of the quasi-1st kind,” which tend to appear closer in time to the earthquake event than geodetic or magnetic precursors. An empirical relationship was established between the precursor occurrence time with respect to the earthquake (T) and the earthquake magnitude (M), modeled in logarithmic form as $logT=a+bM$ for the case of radon, with estimated parameters $a=-0.47\pm 0.13$ and $b=0.28\pm 0.12$. The relatively modest slope suggests a limited window of anticipation for higher-magnitude events.

Moreover, the study highlights that the maximum detection range of these precursors varies depending on the type of signal. In the case of radon, the detection distance is notably shorter than that of deformation- or magnetic field-related precursors. The latter is likely due to radon’s sensitivity to localized stress fields and its short half-life, which constrain its spatial detectability. Furthermore, the study proposes a sequential model for precursor emergence based on underlying physical processes. According to this model, signals associated with deformation and piezomagnetism tend to appear first, followed by seismic anomalies, and, ultimately, phenomena linked to fluid migration, such as radon emissions. This hierarchical progression is grounded in the dilatancy model [41] and offers a valuable framework for interpreting the appearance of different types of precursors in the evolving seismic process. Hence, the study provides not only a systematic classification of precursor types but also a theoretical basis for understanding their temporal and spatial behavior, contributing significantly to the field of earthquake prediction research.

Ranked eighth, with 102 citations, the study by Planinić et al. [21] presents the findings of a four-year continuous soil radon monitoring campaign (1998–2002) conducted in Croatia to assess the potential of radon as a seismic precursor. Monitoring was performed using LR-115 nuclear track detectors (produced by Kodak-Pathe, as reported in the original study) installed at depths of 0.5 and 1 m across three observation stations. These measurements were complemented by barometric pressure, temperature, and precipitation records to evaluate the influence of meteorological variables on radon concentration levels.

The authors identified radon anomalies that preceded several seismic events and applied statistical criteria—such as the two standard deviation threshold—to differentiate natural fluctuations from potential precursor signals. Notably, such anomalies were detected prior to all earthquakes with magnitudes $M\ge 3$ occurring within a 200 km radius of the monitoring site. To correct for meteorological effects, the authors developed a multiple linear regression model which can be expressed as follows: $c=-0.156p+0.481h-0.306t+178.17$, where c represents radon concentration, and p, h, and t correspond to atmospheric pressure, precipitation, and temperature, respectively. This model enables the removal of atmospheric influences and the isolation of radon anomalies of tectonic origin. Moreover, the study examined the empirical relationships between earthquake magnitude (M), epicentral distance (D), and lead time (T). The following equation was fitted: $log\left(DT\right)=0.63M+b$, with an estimated average value of $b=1.68$. Additionally, the area under the radon anomaly peak (S) was used to estimate earthquake magnitude through the relation $M=k\sqrt{S}$. An expression was also derived to estimate the anticipated epicentral distance as a function of lead time and anomaly characteristics: $D={\displaystyle \frac{1}{T}}·{10}^{b+0.63}·k\sqrt{S}$, where k is a correction factor determined experimentally.

In summary, the study offers a comprehensive and methodologically robust approach that integrates geochemical observation, meteorological adjustment, and empirical modeling. It provides compelling evidence for the utility of radon as a potential early indicator of moderate seismic activity in tectonically active regions.

Ranked ninth, with 100 citations, the study by Walia et al. [22] presents a comprehensive analysis of radon concentrations in soil and groundwater as potential seismic precursors in the northwestern Himalayan region of India, spanning the period from June 1996 to September 1999. Monitoring was conducted at two stations—Palampur and Dalhousie—located near the Main Boundary Thrust (MBT) using a scintillation-based emanometry technique. Daily time series of radon concentrations were generated and compared with local seismic activity while accounting for meteorological variables such as temperature, precipitation, relative humidity, and wind speed.

The analysis revealed 126 radon anomalies, of which 74 were correlated with seismic events ranging in magnitude from 2.1 to 6.8 and located within 200 km of the monitoring sites. A statistical threshold of $\pm 2$ standard deviations from the seasonal average was applied to define an anomaly. Notably, the majority of anomalies were classified as pre-seismic (76%), followed by co-seismic (14%) and post-seismic (10%). However, radon peaks did not consistently align between the two stations or the two media (soil and water), a discrepancy attributed to differences in local geology, rock type, saturation levels, and gas transport mechanisms.

Regarding exogenous influences, radon concentrations in soil exhibited positive correlations with temperature, relative humidity, and precipitation and negative correlations with wind speed. These results indicate that, while meteorological variables exert a significant influence, their effects can be effectively mitigated through seasonal normalization and the application of robust statistical criteria. Moreover, the study incorporated empirical scaling relationships to examine the interplay between seismic magnitude (M), epicentral distance (D), and anomaly amplitude (A). The authors proposed the following expressions: for $2.0\le M\le 3.5$: $log\left(AD\right)=0.30M+3.00$ and for $M>3.5$: $log\left(AD\right)=0.22M+3.13$. These can be generalized as $log\left(AD\right)=aM+b$, where the coefficients a and b vary according to local geological and structural conditions.

In conclusion, the study offers an integrated framework combining empirical data, statistical correlations, and algorithmic modeling to assess the potential of radon as a seismic precursor. Although current models present certain limitations, the authors advocate for enhancements through the concurrent monitoring of additional carrier gases (e.g., ${\mathrm{CO}}_2$ and ${\mathrm{CH}}_4$) and the incorporation of continuous, multivariable datasets that could improve the discrimination between tectonic signals and environmental noise.

Ranked tenth in total citation count, with 81 citations, the study by Oh and Kim [8] stands out with a citations-per-year rate of 7.36, indicating a strong and sustained academic impact since its publication in 2015. The article introduces an innovative approach to earthquake prediction through the simultaneous monitoring of a radon isotopic pair: ²²²Rn and ²²⁰Rn. The monitoring campaign was conducted over a full year (2010–2011) in Seongryu Cave, South Korea—an underground environment characterized by relatively stable conditions, effectively minimizing the influence of external meteorological factors. Notably, while ²²²Rn has a half-life of 3.82 days, thoron (²²⁰Rn) exhibits an extremely short half-life of 55.6 s, making it a highly localized and responsive tracer for short-term tectonic activity.

During the observation period, anomalous peaks in both isotopes were recorded exclusively in February 2011, preceding the Tohoku-Oki earthquake (M 9.0), which occurred in Japan in the March of the same year. While ²²²Rn peaks were also observed during the summer—likely due to well-documented seasonal influences such as ventilation and humidity—²²⁰Rn peaks occurred only in winter and showed no correlation with meteorological conditions. This distinction reinforces their interpretation as potential tectonic precursor signals. Moreover, a positive correlation between ²²²Rn and ²²⁰Rn was observed during the anomalous period, likely due to the concurrent release of carrier gases such as ${\mathrm{CO}}_2$. Under normal conditions, however, the correlation between the isotopes was negative. Additionally, the behavior of ²²⁰Rn was found to be strongly dependent on the sampling point’s proximity to the ground surface, a consequence of its short half-life. This further underscores its utility as a tracer for recent and highly localized geodynamic phenomena.

The study concludes that the joint analysis of ²²²Rn and ²²⁰Rn allows a more accurate differentiation between genuine tectonic signals and environmental noise. This dual-isotope approach represents a methodological advancement by introducing isotopic pairing to enhance the specificity of seismic precursor detection. Furthermore, the findings support the feasibility of developing underground monitoring networks based on this strategy, which may contribute to more reliable early warning systems in tectonically active regions.

Finally, the study ranked tenth is the work by Zmazek et al. [24], which explores the application of decision trees as a predictive tool for analyzing radon concentrations in soil gas to assess their potential as seismic precursors. Monitoring was carried out at six stations in the Krško Basin, Slovenia using Barasol-type sensors installed in boreholes at depths ranging from 60 cm to 5 m. Hourly radon measurements were recorded from April 1999 to February 2002 alongside key environmental parameters such as barometric pressure, air and soil temperature, and precipitation.

The predictive model was constructed using data from periods without seismic activity (non-SA), employing regression trees implemented on the WEKA platform—open-source software developed by the University of Waikato (New Zealand) for data mining and machine learning, widely used for classification and regression tasks. Several approaches were evaluated, including linear regression, instance-based regression, and traditional regression trees. Among these, model trees (MT) achieved the best predictive performance, yielding correlation coefficients above 0.80 and considerably lower root mean square errors (RMSE). However, during seismic activity periods (SA)—defined as the 7 days before and after an earthquake—the model’s predictive accuracy declined significantly. This decline in the correlation between predicted and observed radon values and the marked increase in RMSE were interpreted as indicative of system perturbations caused by tectonic stress. For example, at Station 6, the correlation coefficient decreased from 0.80 in non-SA periods to 0.22 during SA periods, accompanied by a 72% increase in RMSE. Systematic discrepancies between measured and predicted radon concentrations were also detected before local seismic events (with magnitudes between 0.8 and 3.3), sometimes up to 7 days before the event. Importantly, the decision tree-based modeling approach not only facilitated the detection of potential precursor anomalies but also enabled the quantification of the relative influence of each environmental parameter on radon concentration. One of the key strengths of this methodology is that it does not require explicit prior knowledge of functional relationships among variables, as such dependencies are automatically inferred from the data.

In addition to the documents listed in Table 4, two recent studies stand out due to their high annual citation rates, highlighting their growing influence in the field. The work of Singh et al. [42], with 64 citations since its publication in 2017, achieves a citation rate of 7.11 per year. It applies artificial neural networks and multiple linear regression to analyze soil radon data in Northeast India, showing promise in isolating geophysical signals from meteorological noise. Likewise, the study by Jena et al. [43], published in 2021 with a rate of 7.00 citations per year, demonstrates the potential of convolutional neural networks (CNN) for assessing earthquake risk by integrating deep learning and geospatial analysis. Although these articles do not appear among the top ten in terms of total citations, their substantial annual impact and methodological innovations make them notable contributions to the evolving research landscape on radon anomalies and seismic forecasting.

In summary, the study highlights the potential of machine learning techniques, particularly decision trees, for identifying anomalous radon behavior linked to seismic processes. The latter is especially relevant for detecting low-magnitude events, underscoring the value of data-driven approaches in developing seismic precursor identification frameworks.

3.4. Research Topics and Thematic Evolution

3.4.1. Word Cloud of Author Keywords

The analysis of keywords used by authors facilitated the identification of recurring concepts and the main thematic axes that characterize the corpus of documents analyzed. Figure 5 presents a graphical representation in the form of a word cloud, in which the size of each term is proportionally scaled to reflect its appearance frequency. The most frequently occurring terms are earthquake precursor and earthquake prediction, underscoring the field’s central objective of identifying signals that may alert us to seismic events in advance. These core concepts are closely associated with terms such as soil gas, radon anomaly, soil radon, and radon concentration, which highlight the relevance of radon and its measurement in gaseous media as a valuable geophysical tool. Moreover, the substantial presence of terms related to environmental conditions—such as environmental parameters and meteorological parameters—as well as anomalies and seismicity, reveals the growing emphasis on differentiating tectonic influences from environmental noise. The latter underscores the importance of accurately accounting for external variables in interpreting radon data.

Equally noteworthy is the emergence of methodological and technological terms, including machine learning, neural networks, deep learning, linear regression, and decision trees. These keywords reflect the increasing integration of advanced computational techniques for time series analysis, pattern recognition, and earthquake forecasting. This trend is further corroborated by the presence of supporting terms such as time series analysis, correlation coefficient, and anomaly detection. In addition, keyword analysis reveals references to specific geographic regions (e.g., Taiwan, NW Himalayas, Northern Pakistan) and detection technologies (e.g., SSNTD, LR-115 films, CR-39), suggesting both the spatial distribution of case studies and the instrumental methodologies employed in the field.

Therefore, this constellation of terms delineates a thematic landscape where geophysical radon analysis, environmental monitoring, regional seismicity, and artificial intelligence converge. The next subsection, dedicated to the analysis of keyword co-occurrence patterns, further explores this terminological and conceptual diversity.

3.4.2. Co-Occurrence of Author Keywords

Figure 6 presents the co-occurrence network of author keywords, where nodes represent individual terms and links indicate their co-appearance. The size of each node reflects its frequency of use, while the thickness of the connecting lines denotes the strength of the association between terms. Furthermore, colors group nodes into semantic clusters according to their thematic affinity. The most prominent cluster, shown in green, is centered on the term earthquake precursor, which is closely associated with concepts such as soil radon, anomaly, time series analysis, radon in soil, Taiwan, TEC, and NW Himalayas. This grouping suggests a research orientation toward using radon as a geochemical indicator in tectonically active regions, particularly in temporal monitoring and detecting anomalous patterns.

The second major cluster, represented in blue, revolves around the term earthquake prediction, which is primarily linked to environmental parameters, neural networks, radon in soil gas, and machine learning. This cluster reflects a methodological trend emphasizing the integration of computational and artificial intelligence techniques in predicting earthquakes using geo-environmental data.

Another relevant cluster, shown in red, is organized around the keyword radon anomaly, which appears in conjunction with earthquake magnitude, Afyonkarahisar, and well water. This pattern suggests studies focused on specific geographic regions where radon in groundwater is employed as a geodynamic tracer. A separate cluster in brown centers on the term soil gas and its connection with anomalies. Additional smaller clusters include a violet group composed of multidisciplinary monitoring and radon anomalies and an orange cluster linking LR-115 films and meteorological parameters, all of which illustrate the diversity of experimental settings and instrumental techniques explored in the literature.

Overall, the structure of the network reveals a clear thematic segmentation, where instrumental, methodological, and applied approaches converge around a common interest in radon as a potential seismic precursor. This segmentation will be further examined in the following subsection, focusing on the general thematic map and exploring the dimensions of centrality and density to characterize the core and emerging thematic lines within the field.

3.4.3. General Thematic Map

The general thematic map shown in Figure 7 categorizes the key terms in the literature based on two dimensions: centrality, which reflects the degree of connection with other topics, and density, which indicates the level of internal development. This classification enables the differentiation between motor themes, basic themes, niche themes, and emerging or declining topics, thereby providing a strategic overview of the conceptual structure of the field.

Among the niche themes located in the upper-left quadrant (high density and low centrality), prominent terms include artificial neural networks, machine learning, and deep learning. These concepts point to an emerging methodological strand marked by a high degree of technical specialization. Although these themes are not yet central to the field, their high density suggests the formation of consolidated subdomains focused on time series processing, automated anomaly detection, and nonlinear seismic forecasting. This pattern aligns with previous insights from the keyword and co-occurrence analyses, revealing the increasing integration of advanced computational techniques.

Additionally, this quadrant includes terms such as Hurst exponent and uranium, which reflect the application of sophisticated mathematical models for characterizing the persistence of radon time series, along with studies emphasizing the geochemical origin of radon. Their position suggests that, while technically robust, these topics have not yet attained central visibility within the broader research agenda. By contrast, the quadrants associated with higher centrality are dominated by classical operational concepts such as earthquake prediction and soil gas, which have been thoroughly addressed in earlier sections. These themes serve as structural pillars of the field, yet they no longer represent the primary focus of recent methodological innovation.

Hence, the general thematic map not only confirms the continuity of traditional research lines but also highlights the emergence of computationally advanced approaches as specialized areas of development. Although currently peripheral, these niche themes may evolve toward greater centrality as they are integrated into applied and interdisciplinary research.

The evolution, consolidation, or possible decline of these themes will be further explored in the next subsection, which presents the Last Decade Thematic Map and examines thematic trajectories over the past ten years.

3.4.4. Last Decade Thematic Map (2014–2025)

The thematic analysis for 2014–2025 (April) revealed a sustained evolution in the principal conceptual axes and the consolidation of new methodological frameworks of the field, as shown in Figure 8. Within the motor themes quadrant, core concepts such as earthquake precursor, soil gas, and environmental parameters remain prominent, indicating that these elements continue to form the structural backbone of research. Their enduring presence suggests that, despite the emergence of new technologies, the field remains fundamentally anchored in geochemical observation and environmental monitoring.

In the basic themes category, terms such as radon concentration, anomalies, soil radon, and meteorological parameters stand out for their consistent usage. However, these terms display a lower degree of conceptual development, likely reflecting their role as recurring variables or cross-cutting elements across a wide range of studies rather than as centers of innovation in their own right.

Compared to the general thematic map, a notable transformation is the growing prominence of artificial intelligence-based approaches. Terms such as artificial neural networks, machine learning, and deep learning have emerged as well-developed niche themes. Their positioning points to the increasing integration of sophisticated computational models into the analysis of radon data—particularly for time series interpretation, anomaly detection, and seismic event forecasting. This methodological shift is consistent with the patterns identified in the keyword co-occurrence analysis and underscores the transition toward more automated and data-driven monitoring systems.

Conversely, terms such as CR-39, SSNTD, and uranium appear among the emerging or declining themes. Their placement may indicate a decline in relevance, likely due to the progressive replacement of traditional experimental techniques by digital sensors and multivariate analytical tools.

In summary, the thematic map of the past decade depicts a research domain that, while maintaining its classical conceptual foundations, is undergoing a methodological transformation toward predictive automation. This shift reflects the increasing adoption of data science tools poised to reshape research practices in the years ahead.

4. Discussion

4.1. Main Findings

The analysis reveals a sustained and growing scientific interest in the relationship between radon anomalies and earthquake prediction, with a marked intensification beginning in 2010. As depicted in Figure 1, this trend coincides with the increasing availability of higher-resolution sensors and the incorporation of multivariable processing techniques in data analysis. Notably, 2022 recorded the highest level of scientific output, suggesting a moment of thematic consolidation in the field.

From a geographical perspective, China, India, and Italy account for the largest share of publications, reflecting their significant exposure to seismic hazards and their well-established scientific infrastructures in this domain. The predominance of corresponding author publications from India and China is consistent with overall production patterns. However, it is important to note the lower level of international collaboration in the case of India—94% of its publications are SCP—in contrast to countries like Pakistan, where 55.6% of studies are co-authored with international partners (MCP). This discrepancy may reflect a tension between strong domestic research capabilities and the openness of scientific networks. The international collaboration network, illustrated in Figure 4, supports this interpretation by revealing clusters led by Asian countries with active connections to institutions in Europe and North America. It is important to note that several researchers have made notable contributions to the field of radon anomalies in the context of earthquake prediction, including authors such as R. Tiwari, V. Walia, H. Virk, and S. Singh. While bibliometric software helps identify recurring names, challenges in disambiguating common surnames and abbreviated author entries may affect the accuracy of author-level metrics. For this reason, detailed author rankings and co-authorship visualizations were excluded to avoid misinterpretation. Instead, author contributions are discussed descriptively to acknowledge their role in shaping the research landscape.

Despite the high seismic hazard in Latin American countries such as Mexico, Peru, and Chile, their scientific output on this topic remains minimal or non-existent. In this review, only one article was identified for Mexico and one for Peru, while no indexed contributions were found for Chile. Other seismically active countries in the region, such as Ecuador and Colombia, also have no presence in the literature. Institutional factors, such as limited public investment in research and seismic monitoring infrastructure, may partially explain this regional underrepresentation. Additionally, the tectonic configuration of countries such as Mexico, Peru, and Chile is characterized by active subduction zones, where an oceanic plate descends beneath a continental plate [44,45,46]. In these settings, earthquakes often originate offshore or at intermediate-to-great depths, which reduces the likelihood of shallow crustal fracturing and hinders the migration of radon to the surface. In contrast, regions like the Himalayas—characterized by continental collision and shallow crustal faulting—offer more favorable conditions for detecting radon anomalies prior to seismic events. This geotectonic distinction may help explain the limited number of radon-based studies in high-risk subduction zones despite their intense seismic activity. Moreover, conducting radon monitoring in such settings often requires more sensitive instrumentation, extended measurement periods, and integration with additional geophysical or geochemical indicators—such as ${\mathrm{CO}}_2$ or helium emissions—to enhance the reliability of detection. These scientific and technical demands may further limit the development of long-term monitoring programs in subduction zones, especially in regions with restricted research funding.

At the institutional level, the most productive affiliations are concentrated in South and East Asia, led by the University of Azad Jammu and Kashmir (32 publications), Firat University (24), and Guru Nanak Dev University (23). This pattern underscores the regional significance of radon-related research in tectonically active zones. Furthermore, the involvement of European institutions—such as the Jožef Stefan Institute in Slovenia and several Italian universities—highlights the enduring role of Europe in the instrumental and methodological development of radon monitoring.

Regarding publication venues, most research has been disseminated through high-impact journals classified as Q1 or Q2, including Natural Hazards, Applied Radiation and Isotopes, and Radiation Measurements. Nonetheless, a substantial proportion of studies also appeared in journals that have since been delisted or discontinued, such as Nuclear Geophysics and the Arabian Journal of Geosciences. This observation raises important questions about editorial standards and the evolving criteria for indexation in a field still undergoing consolidation.

The most frequently cited articles reflect foundational contributions and methodological innovation. For example, the seminal work by Jordan et al. (2011) [33], the most cited, does not directly address radon, yet it provides a crucial normative framework for operational earthquake forecasting. In contrast, empirical studies such as those by King (1980) [34], Baubron et al. (2002) [35], and Ciotoli et al. (2007) [36] offer robust evidence supporting the use of radon as an indicator for active fault characterization and earthquake anticipation. More recent studies, including those by Oh and Kim (2015) [8] and Zmazek et al. (2003) [24], incorporate advanced techniques such as isotopic pair analysis and decision trees, indicating a gradual transition toward more computational and automated research approaches.

In terms of thematic orientation, the keyword cloud (Figure 5) and the co-occurrence network (Figure 6) reveal a convergence between traditional geophysical concepts—such as earthquake precursor, soil gas, and radon anomaly—and emerging analytical techniques, including machine learning, deep learning, and neural networks. This integration is further illustrated in the general thematic map (Figure 7), where these newer methodologies appear as niche themes that are technically well-developed but still limited in thematic centrality. However, the thematic map for the last decade (Figure 8) shows a notable shift of these terms toward higher density, suggesting that they are increasingly consolidating as methodological pillars within the field.

The findings indicate that research on radon as a potential seismic precursor has evolved from largely descriptive and exploratory efforts to more integrated, computationally advanced approaches characterized by enhanced control over external variables. Nevertheless, a classical thematic core—centered on geochemical observation and environmental monitoring—remains essential to the empirical grounding of the field. The coexistence of these traditional and emerging approaches suggests that the field is undergoing a methodological transition. The principal challenge will be the cross-validation of predictive models through robust, replicable datasets across diverse geological contexts.

4.2. Comparison with Previous Reviews

In contrast to previous studies that adopted predominantly qualitative approaches or focused on case-based analyses, the present bibliometric review provides a quantitative, systematic, and global perspective on the scientific production related to radon anomalies as potential earthquake precursors. Past reviews [25,26,30] share a descriptive orientation, emphasizing the characterization of local events and radon emission mechanisms. However, they do not employ bibliometric techniques or examine patterns of scientific collaboration or thematic evolution. Similarly, although [28,29] offer a critical perspective on instrumental advances and highlight the relevance of multivariable approaches, they do not engage in a quantitative analysis of methodological trends within the field.

By contrast, the present bibliometric study is distinguished by its analysis of 379 articles published between 1977 and 2025 indexed in Scopus and/or Web of Science using tools such as Bibliometrix and Biblioshiny. These tools enable the identification of scientific productivity patterns, collaboration networks, and emerging research trends. Moreover, this study evidences a paradigm shift in the field—from conventional geophysical monitoring toward the progressive adoption of artificial intelligence techniques, including decision trees, neural networks, and deep learning, for the automated detection of anomalies. This transition is reflected in the thematic clusters and semantic evolution observed over the last decade.

Despite these methodological differences, there are substantive points of convergence between this review and the aforementioned studies. All recognize the potential of radon as a geological precursor within multivariable monitoring frameworks and agree that anomalies tend to cluster in tectonically active regions such as Japan, India, China, and Italy. They also share concerns regarding the difficulty of establishing robust causal relationships between radon emissions and seismic events, primarily due to the scarcity of long-term time series and the geological heterogeneity of monitoring sites. Furthermore, there is consensus on the need to filter environmental influences adequately and to standardize criteria for anomaly detection—an ongoing challenge widely acknowledged in the specialized literature.

Complementarily, the work [47] offers an extensive thematic review on electromagnetic and radioactive earthquake precursors grounded in a rigorous analysis of methods such as DFA, power spectra, entropy, and multifractals. Nonetheless, their approach remains narrative and does not include a bibliometric assessment of the field’s evolution or its institutional actors. Although the potential of computational tools is acknowledged, their growing adoption is not explored in depth, unlike in the present review, which further contributes by providing strategic cartography of key research lines, institutions, and emerging challenges.

In summary, although this review adopts a more quantitative and structural approach, it converges with previous studies in highlighting the value of radon as a seismic indicator, the importance of rigorous analytical methodologies, and the need to strengthen monitoring infrastructure through expanded networks and interdisciplinary approaches.

5. Limitations

While this bibliometric review offers a robust overview of the research landscape on radon anomalies and seismic prediction, certain limitations must be acknowledged. These include the exclusion of grey literature and non-indexed studies, potential biases due to reliance on author-assigned metadata, and the inherent limitations of citation-based indicators. Furthermore, the study identifies—but does not critically evaluate—the shift toward AI-based approaches. These considerations highlight the need for complementary research that integrates qualitative assessments and methodological comparisons. Moreover, the exclusion of full-text content analysis limited the depth of thematic exploration, which could be addressed in future mixed-method reviews.

6. Future Perspectives

Building on the insights derived from this bibliometric analysis, several promising directions for future research emerge that may substantially enhance the predictive utility of radon anomalies in seismic monitoring. First and foremost, standardizing data collection protocols and anomaly classification criteria is imperative. At present, methodological heterogeneity significantly limits the comparability and reproducibility of findings across studies. Hence, the development of unified frameworks for long-term radon monitoring—encompassing both soil and groundwater measurements—would strengthen the robustness and consistency of precursor signal detection across diverse tectonic settings.

Second, future investigations should emphasize the cross-validation of empirical models using high-resolution, multisite datasets. Such an approach would enable researchers to test the generalizability of observed correlations between radon anomalies and seismic events while also refining scaling relationships that involve earthquake magnitude, epicentral distance, and lead time.

Third, the growing availability of continuous time series data offers fertile ground for the application of artificial intelligence and machine learning techniques. The expanded use of decision trees, neural networks, and ensemble methods—especially those trained on labeled seismic and non-seismic periods—holds considerable potential for enhancing anomaly detection capabilities and minimizing false positives. Moreover, incorporating real-time meteorological corrections could further improve model accuracy and reliability.

Additionally, integrating radon monitoring with complementary geophysical and geochemical indicators—such as ${\mathrm{CO}}_2$ flux, helium concentrations, crustal deformation, and groundwater dynamics—may contribute to the development of more robust and multidimensional early warning systems.

Ultimately, it is crucial to consider the socio-institutional aspects of implementing seismic early warning systems based on radon anomalies. The latter includes developing effective public communication strategies, evaluating ethical considerations, and conducting cost–benefit analyses for deploying monitoring networks in high-risk regions. Interdisciplinary collaboration among geoscientists, data scientists, engineers, and policy-makers will be crucial for translating scientific advancements into operational tools for seismic risk mitigation.

7. Conclusions

This review provides a comprehensive and systematically structured synthesis of the scientific literature on radon anomalies as potential precursors to seismic events, encompassing nearly five decades of academic output indexed in Scopus and Web of Science. The findings indicate a sustained increase in scholarly interest since 2010, a trend that aligns with the advent of enhanced radon detection technologies and the implementation of advanced data analysis methodologies. While this growth appears driven by developments specific to the field, it may also reflect, in part, the broader expansion in global scientific output observed during the same period. China, India, and Italy emerged as the most prolific countries in terms of research output, albeit with differing degrees of international collaboration. Notably, institutional and authorship leadership is concentrated in South and East Asia, with European research networks also playing a significant and complementary role.

Moreover, the thematic analysis reveals the coexistence of well-established research lines—primarily focused on geochemical observations and soil gas monitoring—with an emerging interest in computational approaches, including machine learning, neural networks, and decision tree algorithms. Although these data-driven methodologies remain relatively peripheral, they demonstrate strong internal momentum and are progressively consolidating as innovative analytical paradigms in the domain of radon-based earthquake forecasting.

In conclusion, the present study enhances the current understanding of the scientific landscape related to radon anomalies and seismic prediction. It delineates both consolidated areas of research and emerging avenues for future exploration. Notably, the integration of traditional observational techniques with modern computational models represents a critical opportunity to improve the reliability and practical application of radon signals in the context of seismic risk assessment and early warning systems.