Korean Journal of
Veterinary Service

Outbreaks of diseases that jump from animals to humans have increased fears of another pandemic. While such zoonoses have been around for millennia, they have become more common in recent decades due to damage to the environment, such as deforestation, mass livestock cultivation, climate change, and other human-induced upheavals of the animal world. Given this situation, it is necessary to clarify the direction of research in this area around the world (Oh et al, 2021; Zhou et al, 2022). Advancements in the field of artificial intelligence (AI) are expected to provide optimal solutions to various social problems (Cath et al, 2018; Tomašev et al, 2020). The growth of AI had been stagnant since the 1950s; however, a significant leap forward occurred when machine learning addressed the technical limitations of the existing expert systems and neural networks in the mid-2000s, ushering in a new golden era (Haenlein and Kaplan, 2019).

The utilization of AI in the field of environmental damage and animal diseases continues to increase, thereby expanding its application area and scope to related domains such as water waste, air pollution, solid waste, and animal diseases (Fan et al, 2018; Mo et al, 2019; Abdallah et al, 2020; Ye et al, 2020). AI technologies are widely used in environmental damage monitoring and prediction, damage source simulation and analysis, and intelligent damage control (Ye et al, 2020). Although this technology has emerged as an effective countermeasure against environmental damage, systematic analysis of the research flows and themes, major research countries, and research collaboration in this field remain poorly understood. Identifying the intellectual structure and network of AI research in environmental damage related to animal epidemics will not only help us to understand the knowledge base of related studies, but will also provide insights for future research.

This research was based on statistical analysis of bibliographic data for publications, including articles. Bibliometric analysis is useful for identifying a research overview, characteristics, evaluation, and collaboration (Ellegaard and Wallin, 2015; Mao et al, 2018; Oh et al, 2021). Combined with network analysis, it is widely used to analyze and visualize the concentration, distribution, and impact relationships among research themes or researchers (Hou et al, 2008; Oh et al, 2021). A bibliometric and network analysis helps us to understand the flow, growth, and decline of research on sub-topics, the collaboration and influence of the actors conducting the research, the research hotspots, and the directions of future research (Oh et al, 2021).

In the field of environmental impacts on veterinary research, bibliometric and network analysis has been performed on a few subjects, including the use of biosorption technology in water treatment (Ho, 2008), lead in drinking water (Hu et al, 2010), solid waste (Fu et al, 2010; Yang et al, 2013), estuary damage (Sun et al, 2012), soil contamination (Guo et al, 2014), industrial wastewater (Zheng et al, 2015), non-point source damage (Yang et al, 2017), sources of atmospheric damage (Li et al, 2017), heavy metal damage (Ouyang et al, 2018), nitrate leaching (Padilla et al, 2018), agricultural waste management (He et al, 2019), electronic waste management (Andrade et al, 2019), and air pollution and human health (Dhital and Rupakheti, 2019). However, there have been few related prior studies applying AI in the domain of environmental damage causing animal diseases (Zhao et al, 2020).

Therefore, to address the aforementioned research gap, the present study used bibliometric and network analysis to examine the intellectual structure and collaboration network of AI application studies conducted over the past two decades (2000∼2019) in the field of environmental damage. To this end, bibliographic information from articles on AI applications in the field of environmental damage and animal diseases was collected from the Web of Science (WoS), a scholarly database. After pre-processing the data, we identified the distributions and influences by year, journal, subject category, and country/territory of the article publication and citation. Next, we visualized the linkages between the main keywords and the collaborations between the countries/territories using VOSviewer, a network analysis software. Finally, we present suggestions for conducting AI-applied study for further research on animal pandemics related to environmental damage.

In this study, bibliographic data was collected from the Web of Science (WoS) Core Collection, which is a scholarly database run by Clarivate Analytic. On April 25, 2020, the search was performed using the following search formula:

(TS=(“artificial intelligence*” OR “deep learning*” OR “machine learning*” OR “neural network*” OR “expert system*”) AND TS=(“damage*” or “pollution*” or “pollutant*”)) AND WC=(environment*)

When deriving the formula, we first selected the terms “artificial intelligence”, “deep learning”, “machine learning”, “neural network”, and “expert system”. Next, we selected “pollution” and “pollutant”, as these terms were related to pollution. Using the formula thus obtained, i.e., topic search (TS), we retrieved the bibliographic data of publications containing the aforementioned terms in their abstracts, author keywords, and titles. The subject scope was limited to environmental science, under the WoS category (WC). To select the appropriate keywords and category, we consulted three experts with at least five years of education in the field of AI and environmental science. We also referred to the studies conducted by Gao et al. (2019), Abdallah et al. (2020), and Zhao et al. (2020).

To analyze the journal articles, only the documents included in the Science Citation Index Expanded and Social Sciences Citation Index of the WoS index were selected, and the document type was limited to only articles. The publication timespan was set to twenty years, from 2000 to 2019, and documents specified as early access, review, correction, or editorial material were removed. Finally, the bibliographic data of 1,191 articles were collected for the analysis.

We performed pre-processing on the author keywords. Author keywords are words or terms selected by the authors for the critical topics covered in their articles and are often used to identify hotspots in research in bibliometric and network analysis (Hu et al, 2010; Andrade et al, 2019; Dhital and Rupakheti, 2019; Oh et al, 2021). The typographical errors in these keywords were corrected, plural expressions were converted into singular ones, and abbreviations were spelled out. Using Microsoft Excel and KnowledgeMatrix Plus (http://mirian.kisti.re.kr/km/) developed by the Korea Institute of Science and Technology Information (KISTI), we performed our statistical analysis for the bibliometric analysis. We used the VOSviewer program (https://www.vosviewer.com/) for visualization of the process of the network analysis and to understand the research landscape.

Research landscape: annual publications and citations

The year-wise trends in the publication of AI-based articles in the field of environmental damage related to animal diseases from 2000 to 2019 were illustrated (Fig. 1). The average annual number of publications was 59.6 articles (s.d.: 55.2). An increasing trend was observed and expressed as an approximately exponential function equation: y=11.193e^0.13x, with R²=0.9212. Of the total of 1,191 articles, 273 articles (22.9%) and 918 articles (77.1%) were published in the first phase (2000∼2009) and second phase (2010∼2019), respectively. There were 3.4 times more publications in the 2010s than in the 2000s. Article publications in the last five years (2015∼2019) accounted for 55.3% of the total, and a peak was observed in the year 2019, with a total of 239 articles (20.1% of the total). These findings revealed that research on the application of AI in the field of environmental pollution has increased significantly since 2015, on an annual basis.

Figure 1. Article publication and citation trend of AI applications in environmental damage.

Additionally, Fig. 1 illustrates the citation trend for AI-based articles in the field of environmental pollution, from 2000 to 2019. A total of 1,191 articles were cited a total 19,383 times. The average number of citations per article was 16.3 citations (s.d.: 27.9). Of these, 1,022 articles were cited more than once, accounting for 85.8% of all the articles. An increasing trend in the number of citations was observed and expressed as the following power trendline equation: y=2.016x^2.4489, with R²=0.9821. In the first phase (2000∼2009), 2,215 citations were made, which accounted for 11.4% of the total citations. In the second phase (2010∼2019), 17,168 were made, which accounted for 88.6% of the total. Furthermore, the number of citations in the second phase was 7.6 times more than that in the first phase. The number of citations made since 2015 was 12,649, which accounted for 65.3% of the total citations made, and this number has continued to increase significantly every year. The most cited papers had a total of 312 citations (1.6% of the total citations). A total of 21 articles were cited more than 100 times. These articles were cited 3,516 times and accounted for 18.1% of the total citations. Of the total 1,191 published articles, the number of citations for 119 articles corresponding to more than 10% was 9,503, which accounted for 49.0% of the total number of citations. These findings showed that the citation of AI-based articles in the field has increased significantly every year, and 10% of all these articles have an extremely high share in the total number of citations.

Leading journals in publications and citations

A total of 155 journals have published AI application articles in the field of environmental pollution related to animal diseases. These journals have produced an average of 7.7 articles each (s.d.: 13.2). Fig. 2 shows the top 15 journals publishing topical articles over the last twenty years. Atmospheric Environment, Science of the Total Environment, Environmental Science and Pollution Research, Environmental Modelling Software, Environmental Monitoring and Assessment, Environmental Pollution, Atmospheric Pollution Research, Chemosphere, Fresenius Environmental Bulletin, International Journal of Environmental Research and Public Health, Journal of Environmental Management, Journal of Cleaner Production, Journal of the Air Waste Management Association, Sustainability, and Air Quality Atmosphere and Health are the top 15 journals that have produced the most articles over the last 20 years. These top journals have published 579 articles (mean per journal: 38.6, s.d.: 22.6), accounting for 48.6% of all the articles. The top two journals, Atmospheric Environment (104 articles, 8.7% of the total) and Science of the Total Environment (81 articles, 6.8%), have a large share of the research productivity. The findings revealed that articles on related topics have been published in various journals, and the top 15 journals carry nearly half of the publications.

Figure 2. Top 15 journals in article publications and citations.

The 155 journals that published articles were cited at an average of 125.1 times (s.d.: 388.3) per journal, with 114 journals cited more than once, and these citations accounted for 92.9% of the total number of citations made. Fig. 3 also shows the top 15 most cited journals and the number of citations per article in these journals over the last twenty years. Atmospheric Environment, Environmental Modeling & Software, Science of the Total Environment, Chemosphere, Environmental Science & Technology, Environmental Pollution, Journal of Environmental Management, Water Research, Atmospheric Pollution Research, Environmental Science and Pollution Research, Environmental Monitoring and Assessment, Journal of the Air & Waste Management Association, Remote Sensing of Environment, Water Air and Soil Pollution, and Journal of Hazardous Materials were the top 15 ranked journals in citations over the last 20 years. These journals were cited 13,443 times (mean per journal: 896.2, s.d.: 933.7), which accounted for 69.4% of all the citations. In particular, Atmospheric Environment (3,850 citations, 19.9% of the total citations), Environmental Modelling & Software (1,982 citations, 10.2%), and Science of the Total Environment (1,736 citations, 9.0%), which have more than 1,000 citations each, are highly influential journals considering their impact factors. The average number of citations per article in the top 15 journals was 27.6 citations (s.d.: 18.9), which is 2.8 times more than the average 9.0 citations per article of other journals (s.d.: 10.5). Notably, Remote Sensing of Environment (84.3 citations per article), Environmental Modelling & Software (50.8 citations), and Environmental Science & Technology (50.8 citations) had the highest citations per article. These findings revealed that the number of citations in the top journals was also remarkably high.

Figure 3. Top 15 countries/territories in article publications and citations.

Leading research countries/territories in publications, citations, and collaborations

A total of 86 countries/territories have published articles on AI applications in environmental pollution. These countries/territories produced an average of 19.1 articles each (s.d.: 16.5). Fig. 3 shows the top 15 countries/territories in article publications over the last twenty years. People’s Republic of China (PR China), the United States (USA), Iran, Spain, Italy, India, England, Turkey, Greece, Germany, Canada, Poland, South Korea (S. Korea), Australia, and France were the top 15 countries/territories that have produced the most articles over the last 20 years. These countries published an average of 82.8 articles each (s.d.: 72.2), with an average share of 7.0% per country of the total articles published. The top two countries, PR China (316 articles, 26.5% of the total) and the United States (196 articles, 16.5%), held a very high share of the research productivity. These findings revealed that AI applications in the field of environmental pollution are significant in several countries/territories, with the top 15 countries/territories having a remarkably high share (especially PR China and the United States).

Fig. 3 also shows the number of citations in the top 15 most cited countries/territories and the average number of citations per article in these countries/territories. The average number of citations in these countries was 1,146.4 (s.d.: 1,059.9), which accounted for 5.9% (s.d.: 5.5%) of the total number of citations. The number of citations was the highest in PR China, followed by USA, India, Germany, Iran, France, Finland, Turkey, Canada, Chile, Oman, Portugal, S. Korea, and Poland. Among them, PR China (4,130 citations, 21.3% of total citations) and the USA (3,206 citations, 16.5%) were the top two countries, with a total of more than 3,000 citations. Finland (70.7 citations per article) and Oman (69.0 citations) had the highest number of citations per article. The findings revealed that the share of citations is very high in PR China and the USA, and some countries/territories, such as Finland and Oman, have strong influences per article.

The top 15 countries/territories that conducted research by cooperating with other countries were in the following order of the total link strengths (TLS) as calculated by VOSviewer: USA (108 TLS), PR China (99 TLS), England (49 TLS), Italy (35 TLS), Iran (34 TLS), Spain (32 TLS), Germany (30 TLS), Canada (28 TLS), Australia (26 TLS), France (22 TLS), Netherlands (18 TLS), S. Korea (18 TLS), Greece (15 TLS), India (14 TLS), and Malaysia (12 TLS). Fig. 4 presents a visualization of the research collaboration across the countries. The USA and PR China, Iran, Spain, the United Kingdom, India, France, Switzerland, Belgium, Canada, Australia, and Denmark actively collaborate with other relatively more diverse countries to conduct research.

Figure 4. Research collaboration among countries/territories.

Author keywords and co-occurrence networks

Fig. 5 show the most frequent occurrences of author keywords focusing on environmental pollution and AI/methodology. The keyword of air pollution appeared in 114 articles, followed by PM 2.5, PM, air quality, ozone, PM 10, water quality, nitrogen dioxide, pollution, remote sensing, heavy metal, geographic information system, nitrogen monoxide, pollutant, and nitrate. Among the top 15 keywords, air-related pollution appeared eight or more times. However, pollution associated with water or land was relatively infrequent. Artificial neural networks (266 articles) appeared most frequently as author keywords. Neural network, machine learning, multiple perception, random forest, self-organizing map, support vector machine, and deep learning were among the top 15 professional techniques used, and they are all related to artificial intelligence.

Figure 5. Top 15 author keywords in environment damage and AI/methodology category.

Fig. 6 presents a visualization of the network relationship among the author keywords. Water pollution and heavy metals were observed to be relatively well researched, with genetic algorithms and self-organizing maps being the most widely used AI techniques. Ozone, pm, nitrogen monoxide, and nitrogen dioxide were the most actively investigated using neural networks or multilayer perceptrons. Air quality has been mostly investigated using deep learning. They also have a high degree of connection within specific fields, such as animal diseases.

Figure 6. Author keyword network map.

This study analyzed the knowledge structure and collaborative network of AI application research conducted over the last twenty years in the field of environmental pollution related to animal diseases. The discussions and implications based on the results obtained are as follows.

First, the number of studies applying AI in the field has grown rapidly over the last five years (since 2015). This may be due to the golden age of AI, which began in the 2010s. Furthermore, the technological evolution of AI has recently accelerated due to the advancement in the technologies of machine learning and deep learning; these developments are expected to expand the scope of AI in various other domains of environmental science (Ye et al, 2020). Considering the fast and efficient data processing, reasoning, and prediction associated with AI, it is highly likely that AI will have several applications in pollution source monitoring, environmental impact assessment, and future environmental pollution simulation. In this regard, related research is also expected to rapidly expand in a quantitative and exponential manner, and it is also expected to undergo qualitative expansion as various researchers collaborate to address the challenges associated with environmental impacts on contagious diseases.

Second, over 150 journals have published studies combining environmental impacts and AI; however, the top 15 journals’ share of articles is close to half of this total number of publications. A similar trend was also observed in the citation rate. Although high productivity and influence of the top journals are common in most academic fields, it is important to ensure academic diversity to promote related research. It is also necessary to create a self-sustaining cycle of knowledge production and diffusion, in which articles are published and cited in a greater number of journals. To this end, the topic of AI application must be more widely discussed by journals, although the features and scopes of different journals may vary.

Third, over 80 different countries/territories have published related articles, but the top 15 have a relatively high share of the publications and citations. In particular, the publication productivity and influence in PR China and the United States are extremely high. Interestingly, emerging powers in the field of AI (Iran, India, South Korea) were also ranked in the list of the top 15. In addition, the influence of articles was high in countries/territories comprising various regions (Chile, Oman, Portugal). Although environmental pollution is a universal topic at the pandemic era, it can also be regional or country-specific. If AI-based studies to address topics related to environmental issues increase, the productivity and influence of the top countries will be gradually mitigated.

Fourth, PR China and the United States formed the core of research cooperation; more active collaborative network needs to be formed amongst various countries/territories. Furthermore, to strengthen the diversity of research, collaborative research among various countries/territories will be crucial. Efforts towards academic collaboration must be made to solve various topics related to environmental issues at the global level.

Finally, in related studies, AI was used not only for general techniques such as artificial neural networks, machine learning, and deep learning, but also for various other learning and neural network techniques, such as self-organizing maps and support vector machines. However, in the field of environmental pollution, the application of AI has focused mainly on air pollution, rather than on soil or water pollution, and therefore, the scope must be expanded to include both soil and water pollution. This will also lead to an increase in the thematic diversity.

The last step of such environmental pollution is climate change, and the factor that has the greatest influence on the climate change is air pollution (Seinfeld et al, 1998; D’Amato et al, 2014). As can be seen from our result, air pollution occupies the highest proportion of the keywords in environmental pollution-related papers. Environmental pollution or climate change has caused a dramatic shift in the ecosystem. It is apparent that this environmental shift has prompted the migrations of reservoirs or vectors which can be important factors for animal pathogen transmission, and made each animals vulnerable to pathogen exposure (Patz et al, 2003; Altizer et al, 2011; Losacco et al, 2018). This pattern is the most common phenomenon of emerging infectious diseases from animals that have recently appeared, and it has been discussed that this trend will gradually accelerate.

Therefore, follow-up studies should consider performing bibliometric and network analysis of the application of AI in various field of environmental science; in particular, environmental pollution. The analysis results could then be compared, based on which insights and new research directions can be developed to enhance the use of AI across various fields. Furthermore, comparative studies between the major countries/territories are also needed. In particular, comparing the research topics of major countries/territories is also expected to be highly useful for promoting research cooperation at the global level.

This work was supported by the Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry (IPET), through the Animal Disease Management Technology Advancement Support Program, funded by the Ministry of Agriculture, Food and Rural Affairs (MAFRA) (Project No. 122013-2).

No potential conflict of interest relevant to this article was reported.