Research on Wildﬁres and Remote Sensing in the Last Three Decades: A Bibliometric Analysis

: Evaluating the impact of wildland ﬁres on landscapes, a pursuit increasingly supported by remote sensing techniques, requires an understanding of wildﬁre dynamics. This research highlights the main insights from the literature related to “wildﬁres” and “remote sensing” published between 1991 and 2020. The Scopus database was used as a source of information regarding scientiﬁc production on these topics, after which bibliometric tools were employed as a means through which to reveal patterns in this network of journals, terms, countries, and authors. The results suggest that these subject areas have undergone signiﬁcant developments in the last three decades, having been the focus of growing interest among the scientiﬁc community. The most relevant contributions to the literature available have been made by researchers working in the areas of earth and environmental sciences (54% of the publications), primarily in the United States, China, Spain, and Canada. Research trends in this ﬁeld have undergone a signiﬁcant evolution in recent decades, explained by the strong relationship between the technological evolution of detection methods and remote sensing data acquisition. analysis of climate change. A second cluster involved developing systems to monitor, map and detect wildland ﬁres based on satellite imagery. The third cluster demonstrates the relationship between studies on the impact of emissions, especially in the various biomes. The fourth cluster demonstrates the relationship between studies that analyse burnt areas based on models, using remote sensing data as well as the inﬂuence of anthropogenic factors.


Introduction
Fire plays a key role in the ecology of many ecosystems [1,2], having been present throughout the history and the development of society. It can also be considered a determining factor in the distribution and dominance of savannas worldwide, even in places where the climate and soil could potentially sustain forests [3][4][5]. Thus, fire can be understood as a natural ecological factor that influences the structure and functioning of several ecosystems [6]. However, a distinction must be made between fires and wildfires, the latter being defined by their uncontrollable nature, causing significant losses [7].
A severe threat to many environments, wildfires are considered one of the most challenging phenomena in environmental sciences [8]. The increase in frequency and intensity of fires has been documented in several ecosystems around the world [9][10][11], producing a significant impact on global warming, as fires often result in a considerable loss of biomass and carbon, which can alter local climates [2,7,[12][13][14][15][16].
In this context, remote sensing techniques are a highly feasible and effective tool for describing patterns of the occurrence of fire in various ecosystems, as satellite imaging is an important means through which to delineate the perimeters of fire expansion and characterise the degree of intensity or severity reached by fires [12,13,[17][18][19][20][21].
The assessment of changes in vegetation patterns caused by fire over time can be conducted in various ways, from traditional field observations to monitoring carried out using Earth Observation Systems [22]. Earth observation is an important tool for monitoring land surface and vegetation dynamics at regional and global scales over significant periods of time [23,24]. Although their temporal, spatial, and spectral resolutions vary, the various Earth Observation Sensor systems available provide tools with which to assess vegetation

Materials and Methods
To answer the questions proposed, research was based on a bibliometric analysis approach. Thus, quantitative data from reference articles were used to produce graphs, tables, bibliographic data networks, and textual data networks. In terms of a methodological approach, the study was divided into two main stages. The first step included the choice of search database, identifying terms relevant to the subject and implementing filters to the search engine (Step 1). After this first step, a manual screening was performed, in which all the titles and abstracts of the articles returned in the search were read in order to identify and remove any unrelated articles. A standardisation of terms was then performed, in which any repeated words or terms were identified and excluded. The data network was then constructed of all the articles using VOSviewer software (Step 2). The methodological procedures employed to achieve the results required are presented in the following methodological flowchart ( Figure 1) and described below.

Bibliographic Basis
The database used in this study was extracted from Scopus Database [40]. Scopus is the largest multidisciplinary database of abstracts and references of scientific literature, accounting for more than 25,100 titles, with about 23,452 of which being taken from peerreviewed journals and 5000 from international publishers. Scopus offers the most comprehensive overview of world research production [27,40,41], incorporating efficient analysis tools to recover and aggregate information and export data in various formats, providing a comprehensive view of the total volume of global research produced in various areas of knowledge. All of these reasons contributed to Scopus being chosen as the bibliographic database used for this research [41].
This study was limited to an analysis of publications printed in journals, reducing the bias caused by duplicate publications and minimising false-positive results. Reviews, conference proceedings, book chapters, and books were not considered as they include works that may have been published more than once in different media sources [27].
To identify the articles published relating to the subject studied over the last 30 years, a search for related terms was performed using the Scopus [40] search tool. The possible textual variants that would allow for the research themes to be found in their totality were then determined, producing the following list: "Megafire" OR "Extreme fire" OR "Large forest fire" OR "Wildland fire" OR "Wildfire" OR "Forest fire" OR "Bushfires" AND "Remote Sensing". These terms were located in the titles, abstracts, and keywords of publications from between 1991 and 2020 on the 19 January 2021. Reviews and book chapters were removed from the search, as were articles unrelated to the subject.

Bibliometric Analysis
VOSviewer software [42,43] was used as a means through which to provide a visual representation of the data network. The tool is specifically designed for bibliometric analysis, and is used to view data returned by searches conducted in the Scopus Database (as well as other databases, such as Web of Science, Dimensions, PubMed) [42,43]. VOSviewer can be used to build networks of scientific publications, scientific journals, researchers,

Bibliographic Basis
The database used in this study was extracted from Scopus Database [40]. Scopus is the largest multidisciplinary database of abstracts and references of scientific literature, accounting for more than 25,100 titles, with about 23,452 of which being taken from peer-reviewed journals and 5000 from international publishers. Scopus offers the most comprehensive overview of world research production [27,40,41], incorporating efficient analysis tools to recover and aggregate information and export data in various formats, providing a comprehensive view of the total volume of global research produced in various areas of knowledge. All of these reasons contributed to Scopus being chosen as the bibliographic database used for this research [41].
This study was limited to an analysis of publications printed in journals, reducing the bias caused by duplicate publications and minimising false-positive results. Reviews, conference proceedings, book chapters, and books were not considered as they include works that may have been published more than once in different media sources [27].
To identify the articles published relating to the subject studied over the last 30 years, a search for related terms was performed using the Scopus [40] search tool. The possible textual variants that would allow for the research themes to be found in their totality were then determined, producing the following list: "Megafire" OR "Extreme fire" OR "Large forest fire" OR "Wildland fire" OR "Wildfire" OR "Forest fire" OR "Bushfires" AND "Remote Sensing". These terms were located in the titles, abstracts, and keywords of publications from between 1991 and 2020 on the 19 January 2021. Reviews and book chapters were removed from the search, as were articles unrelated to the subject.

Bibliometric Analysis
VOSviewer software [42,43] was used as a means through which to provide a visual representation of the data network. The tool is specifically designed for bibliometric analysis, and is used to view data returned by searches conducted in the Scopus Database (as well as other databases, such as Web of Science, Dimensions, PubMed) [42,43]. VOSviewer can be used to build networks of scientific publications, scientific journals, researchers, research organisations, countries, and keywords, for example. Items in these networks can be connected by co-authorship, co-occurrence, citation, bibliographic coupling, or co-citation links [27,28,30,43]. Creating a semantic network, the items are represented by nodes and edges. Nodes are objects such as co-authorship or co-occurrence of words and countries, for example. An edge can exist between any pair of nodes. An edge is a connection or relationship between two nodes. The distance between two nodes on the graph produced indicates the approximate relationship between search terms, and the relationship between the respective terms; smaller distances point to a greater number of co-occurrences. The size of a label on a node is determined by the weight of an item within a network (sizes have a direct correlation with frequency) [27][28][29][30]43].
All articles returned by the conducted search were analysed in terms of their textual and bibliographical data. Based on the textual data, co-occurrences were analysed between terms (in article titles and abstracts). Limiting the terms considered to the titles and abstracts of the articles reduced the risks of terms being repeated in different parts of the same documents being registered [28]. The research topics were categorised into four clusters based on when they were published: 1990s, 2000s, 2010s, and a comprehensive 30-year set.
The following analysis was performed using bibliographic data: bibliographic coupling (relationship between items based on the number of shared references); citation (relationship between items based on the number of times they cite each other); co-authorship (relationship of items based on the number of co-authored documents) and co-citation (relationship of items based on the number of times they are cited together) [28,43].
Each colour represents a cluster in the network visualisation maps obtained using VOSviewer software. These clusters were built by the software following the methodology described by Van Eck and Waltman [43] and Mourão [28]. VOSviewer constructs a map based on a co-occurrence matrix. The construction of a map is a process that consists of three steps. In the first step, a similarity matrix is calculated based on the co-occurrence matrix. In the second step, a map is constructed by applying the VOS mapping technique to the similarity matrix. Finally, in the third step, the map is translated, rotated, and reflected [43].
Other concepts presented in the figures and tables below are also explained in this publication. For example, a cluster is a group of items presented on a map; a link is a relationship between two items and average citations obtained by the documents within which an item appears.

Bibliometric Results and Discussion
The results found in the conducted bibliometric analysis shall be detailed in this section for sources, terms, countries, authors, citations, and co-citations. The number of documents obtained in the Scopus database (1722 documents, considering only articles and excluding books, conferences, and other documents) involving wildfires and remote sensing, published between 1991 and 2020.

General Information
An analysis of the annual distribution of the number of articles published on wildfires and remote sensing (see Figure 2) highlights the period between 2016 and 2020 as the most representative five-year period in the timeframe assessed, accounting for 40% of all publications. This period registers the highest growth rate in the number of publications. Due to the technological evolution that has occurred in recent years, with the development of new sensor systems, time series data, advances in image processing techniques, and the increased availability of free images have been established [26]. Of the 1722 publications, 951 (30.8%) were linked to the earth and planetary sciences subarea, 731(23.7%) with the environmental science subarea, 603 (19.6%) with the agricultural and biological sciences subarea, 193 (6.3%) with the social sciences subarea, 143 (4.6%) with the engineering subarea, 101 (3.3%) with the computer sciences subarea and the rest of the subareas present with low productivity on the subject. Figure 3 presents the relationship between journals based on the number of references they share (bibliographic coupling × journal sources). The Remote Sensing of Environment journal had an impact factor of 9.085 in 2019, and the International Journal of Remote Sensing registered an Impact Factor of 2.976 in 2019. These were the sources/journals with the most hits in the conducted search, accounting for half of the documents between them. These two publications were also the focus of most of the citations identified in the research. Strong relationships were registered between the following journals: Remote Sensing of Environment, Remote Sensing, International Journal of Wildland Fire and International Journal of Remote Sensing. The network was drawn up solely encompassing journals with at least eight publications, therefore containing only 43 journals and five clusters. Of the 1722 publications, 951 (30.8%) were linked to the earth and planetary sciences subarea, 731(23.7%) with the environmental science subarea, 603 (19.6%) with the agricultural and biological sciences subarea, 193 (6.3%) with the social sciences subarea, 143 (4.6%) with the engineering subarea, 101 (3.3%) with the computer sciences subarea and the rest of the subareas present with low productivity on the subject. Figure 3 presents the relationship between journals based on the number of references they share (bibliographic coupling × journal sources). The Remote Sensing of Environment journal had an impact factor of 9.085 in 2019, and the International Journal of Remote Sensing registered an Impact Factor of 2.976 in 2019. These were the sources/journals with the most hits in the conducted search, accounting for half of the documents between them. These two publications were also the focus of most of the citations identified in the research. Strong relationships were registered between the following journals: Remote Sensing of Environment, Remote Sensing, International Journal of Wildland Fire and International Journal of Remote Sensing. The network was drawn up solely encompassing journals with at least eight publications, therefore containing only 43 journals and five clusters.

Bibliographic Coupling and Journal Sources
The size of each node (dot) and link (line) is proportional to the intensity of the node or line link. Table 1 presents additional information, listing the journals in order, from those with the most citations to the least, and according to the number of articles published. Table 1 also shows other relevant values such as the cluster number of each source, the value of its links and the strength of links.  The size of each node (dot) and link (line) is proportional to the intensity of the node or line link. Table 1 presents additional information, listing the journals in order, from those with the most citations to the least, and according to the number of articles published. Table 1 also shows other relevant values such as the cluster number of each source, the value of its links and the strength of links.   VOSviewer generates clusters considering the individual publications in focus [43]. In this specific case, each source was analysed as an individual and was exclusively grouped into one community; each was built considering the relevance of the specific variable. Consequently, 43 items were displayed in five clusters. In total, 19 items were grouped into cluster 1, which is composed of the most cited sources. This phenomenon is not uncommon in bibliometric analyses [28,44] and reflects a certain publishing preference among journals, which publish articles that cite other works within certain groups of journals, especially those with similar impact factors [28,45].

Citation of Sources
In this case, connections between items were based on the number of times journals cited each other (   The Izvestiya-Atmospheric and Ocean Physics journal does not show a strong relationship to the others and is placed in an isolated position within the network. It is also important to note that Yaogan Xuebao/Journal of Remote Sensing did not link to any of the journals in the "citation x sources" analysis. However, this journal was still included in the journals with at least eight publications on the research subject.

Main Terms
Based on the textual data located in the titles, abstracts, and keywords of the articles, and using the frequency of co-occurrence to analyse the textual data, only the terms repeated at least 10 times were considered, repeated terms having been identified and adjusted in the pre-processing phase of the data. The research topics were categorised into four clusters (represented using different colours in Figure 5  The Izvestiya-Atmospheric and Ocean Physics journal does not show a strong relationship to the others and is placed in an isolated position within the network. It is also important to note that Yaogan Xuebao/Journal of Remote Sensing did not link to any of the journals in the "citation x sources" analysis. However, this journal was still included in the journals with at least eight publications on the research subject.

Main Terms
Based on the textual data located in the titles, abstracts, and keywords of the articles, and using the frequency of co-occurrence to analyse the textual data, only the terms repeated at least 10 times were considered, repeated terms having been identified and adjusted in the pre-processing phase of the data. The research topics were categorised into four clusters (represented using different colours in Figure 5 The network established for the 1990s (1991 to 2000, see Figure 5) allowed three clusters to be identified. In the first one, the most frequent terms were "data", "change", "wildfire", "analysis", and "effect". The second cluster contained the terms: "forest fire", "fire", "image", "burned area", "detection". The third contained the terms: "remote sensing", "use", "monitoring", "gis", "observation". These results reflect the major scientific interests present in the 1990s within the discussion on the techniques used for conducting image analyses on forest fire research, as well as on the development of remote sensing tools used to analyse changes in land use, monitoring and landscape dynamics related to forest fires [46,47].
If the development of the network as a whole, encompassing the entirety of the period studied, is considered (1991-2020, see Figure 5), the existence of a well-established network of terms is established, exposing a clear conceptual framework resulting from the interconnection of studies involving forest fires and remote sensing. In fact, in cluster one, the emergence of research topics related to the analysis of changes in forest areas following forest fires can be perceived based on the application of spectral indexes and concerning the impact of such events on climate change. Cluster two addresses topics associated with wildfire detection, monitoring and mapping based on satellite imagery. The third cluster demonstrates the relationship between terms connected to the impact of fire, fire emissions, pollution, distribution and value. The final cluster is centred on research topics that assess burnt areas by implementing models applied to remote sensing data. The influence of anthropic factors is also included in this cluster [7,26,46,49,[52][53][54][55][56].

Countries
The nationalities of co-authors were determined in order to pinpoint collaborations established between countries. The search conducted of the Scopus database returned publications from 106 countries and 26 publications with no country identified. Figure 6 demonstrates that the countries identified in the analysis were grouped into seven clusters (see Table 2). The criteria defined for the selection of countries were those that produced journals with a minimum of 10 published documents and a minimum of 50 citations. A total of 33 countries were selected as a result, with most co-authored documents being registered as from the USA (681 documents, 30% of the studies), China (192 documents, 8.4% of the studies), Spain (176 documents, 7.7% of the studies), and Canada (142 documents, 6.2% of the studies); these four countries make up around 52.3% of the studies among the 33 countries analysed. These analyses of the various co-authorships reveal the international cooperation between authors of different nationalities and affiliations in terms of the themes explored here and related to forest fires and remote sensing.   United Kingdom   1   91  4004  78  23  Russia  88  1661  31  16  Germany  77  3062  56  26  Greece  64  1385  33  16  France  51  1791  36  22  Finland  20  627  13  15  Switzerland  16  787  14  13  Norway  11  443  9  12  Austria  10  218  9  12  United States  681  27,655  212  31  China  191  2840  82  22  Canada  141  4496  67  25  Japan  40  783 26 17 Figure 6. Network visualisation map based on bibliographic data (co-authorship, countries). Table 2. Countries to which authors are affiliated.

Country Clusters Documents Citations Total Link-Strength Link
United Kingdom   1   91  4004  78  23  Russia  88  1661  31  16  Germany  77  3062  56  26  Greece  64  1385  33  16  France  51  1791  36  22  Finland  20  627  13  15  Switzerland  16  787  14  13  Norway  11  443  9 12 Austria 10 218 9 12 As expected, the network of collaborations between different nationalities had a positive correlation with the co-citation of works [28]. Table 2 presents additional information quantifying the data presented in Figure 6, displaying the way in which relationship clusters are organised according to the number of publications, citations, and strength of links. The table also shows the interrelations between countries, allowing the data to be considered in terms of trends through the network of publications among countries. It is also possible to observe that the countries with more significant link-strengths in the groups are: United Kingdom (cluster 1), United States (cluster 2), Spain (cluster 3), Australia (cluster 4), India (cluster 5), South Africa (cluster 6) and Turkey (cluster 7). Portugal and Brazil are part of cluster 3, collaborating more closely with Spain, Italy, Belgium, Chile, Mexico, and Argentina.
The USDA Forest Service ranks first among the institutes that have published the most (Table 3). Among the top 20 publishers, 15 are from the United States, two are Chinese, two Canadian, and one is a European centre based in Belgium.

Authors and Co-Citation
To understand whether the scientific topics addressed within this article tend to be more or less concentrated in terms of their authorship, this subsection shall analyse the most productive authors, co-authorships, and co-citations individually. The following Table 4 therefore presents the single authors with the highest number of publications and citations related to forest fires and remote sensing, as established by the conducted search.  When analysing the authors returned in the search from the angle of co-authorship, authors were required to have at least four documents. In order to create links, the relationship between items was based on the number of documents co-authored. Figure 7 presents the network visualisation map of these items and reveals that the various authors were grouped into 18 clusters. Some of the 245 items identified in the network did not connect to any others, with the largest set of interconnected items therefore consisting of 206 items.
In these groups, the authors with the most co-authored documents were Emilio Chuvieco (19 studies), Z. Li (19 studies), and A. T. Hudak (18 studies). Table 5 also details the number of citations of the ten most productive groups. All these authors have been cited a high number of times, demonstrating their significant contribution to the research topic of this paper.
authors were required to have at least four documents. In order to create links, the relationship between items was based on the number of documents co-authored. Figure 7 presents the network visualisation map of these items and reveals that the various authors were grouped into 18 clusters. Some of the 245 items identified in the network did not connect to any others, with the largest set of interconnected items therefore consisting of 206 items. In these groups, the authors with the most co-authored documents were Emilio Chuvieco (19 studies), Z. Li (19 studies), and A. T. Hudak (18 studies). Table 5 also details the number of citations of the ten most productive groups. All these authors have been cited a high number of times, demonstrating their significant contribution to the research topic of this paper.  For links focussing on co-citation, the relationship between the items was based on the number of times cited together. Only authors with a minimum of 50 citations were selected for the analysis. Figure 8 shows that L. Giglio  For links focussing on co-citation, the relationship between the items was based on the number of times cited together. Only authors with a minimum of 50 citations were selected for the analysis. Figure 8 shows that L. Giglio and E. Chuvieco are the most cocited authors (1273 and 1263 citations, respectively). Other authors with a significant number of citations and with a pronounced relatedness are C. O. Justice (995 citations), E. S. Kasischke (831 citations), and Y. J. Kaufman (708 citations).

Conclusions
The bibliographic revision performed within this work centred on publications on the subject of wildland fires and remote sensing in scientific journals that were published in the last three decades (1991-2020). It was possible to identify the leading journals, countries, authors, and institutions involved in the research conducted on the above-mentioned topics and map out the most relevant terms used over this period.
The results show that the number of publications on the subjects increased during the period analysed, demonstrating the growing interest of the scientific community. Of

Conclusions
The bibliographic revision performed within this work centred on publications on the subject of wildland fires and remote sensing in scientific journals that were published in the last three decades (1991-2020). It was possible to identify the leading journals, countries, authors, and institutions involved in the research conducted on the above-mentioned topics and map out the most relevant terms used over this period.
The results show that the number of publications on the subjects increased during the period analysed, demonstrating the growing interest of the scientific community. Of the total publications, 54% were linked to earth and environmental sciences, revealing the interest and connection between the topics analysed and the environmental dynamics. The research trends in this field include the significant developments in remote sensing techniques for studies on forest fires in recent decades. This evolution is explained by the fact that there is a strong correlation between the technological evolution of detection methods and remote sensing data acquisition.
It was also concluded that publications with the highest number of articles and citations were scientific journals, specifically Elsevier's Remote Sensing of Environment (IF 2019: 9.085) journal, Taylor and Francis' International Journal of Remote Sensing (IF 2019: 2976) and Csiro Publishing's International Journal of Wildland Fire (IF 2019: 2988).
In terms of researchers' countries of origin, the United States of America was the highest contributor, providing 681 co-authored documents, and was also the country with the highest number of international co-operations. China contributed 192 documents, Spain contributed 176 and Canada contributed 142. Consequently, the research institutions with higher contribution rates come from these countries: the USDA Forest Service coming in first place, followed by the Chinese Academy of Sciences and the University of Maryland.
Regarding the most frequent indicators in publications, evolution of the theoretical and methodological field was noted over the three decades analysed. A well-established network was also found to exist between the terms, creating four major relationship clusters. The first cluster involved analysing changes in forests caused by wildland fires over the years and the use of spectral indices in the analysis of climate change. A second cluster involved developing systems to monitor, map and detect wildland fires based on satellite imagery. The third cluster demonstrates the relationship between studies on the impact of emissions, especially in the various biomes. The fourth cluster demonstrates the relationship between studies that analyse burnt areas based on models, using remote sensing data as well as the influence of anthropogenic factors.
Considering the authors with the highest levels of published scientific work identified in this paper, it can be concluded that Emilio Chuvieco has produced the most on the topic studied, both individually and collaboratively. The most cited authors are Y. J. Kaufman This work presents certain limitations, some of which will serve as a basis for future research. First, the bibliometric analysis could also be developed using other quantitative or qualitative tools (for example, Web of Science or Google Scholar), which may present some differences, especially concerning citations. As databases are not updated immediately once an article is published, slight variations may exist in the number of articles present in the WoS and Scopus databases.