Spatial data quality and uncertainty publication patterns and trends by bibliometric analysis

1.1 Spatial data quality and uncertainty – overview of research development

The heterogeneity of the real word, technologies for data acquisition and processing, database management tools and platforms leads to a large amount of duplicated, inconsistent, ambiguous, and incomplete spatial data. Therefore spatial data quality and uncertainty (SDQ&U) is an increasingly important issue in geographical information science with thousands of publications in countless journals, conferences, and books. At the early stage of GIS development (from the 1960s until the mid-1980s), the imperfection of spatial data was mainly expressed as errors in geographical position and topology [1, 23]. In the following years, the development of GIS technology and its use in the decision-making process have fostered increasing interest in the quality and uncertainty research [4, 5]. Issues of quality and uncertainty of spatial data have become even more important because data imperfections propagated through spatial analysis affect the decision-making process. Recently, researchers have focused their scientific attention on uncertainty modelling and the final impact of data imperfection on the spatiotemporal analysis [6]. Moreover, since 2004 data delivered by volunteers (crowdsourcing data) have become the subject of extensive research. These studies have a broader context, analysing not only different types of errors

but also the behaviour of volunteers expressed in form of their mapping activity in a given area e.g: [7, 8, 9, 10, 11, 12].

Although the concepts of uncertainty in spatial data and spatial data quality are similar, the standards have evolved only for the latter. The quality of data is understood as ‘degree to which a set of inherent characteristics fulfils requirements’ [13] with respect to the immanent attributes related to the geospatial nature of the data, like positional accuracy or spatial resolution. This definition was further adapted in ISO 19100 series for geographical information, e.g ISO 19101-:2002 Geographic information – Reference model [14151617]. However, the issue how to evaluate and describe data uncertainty still remains open.

Many international organisations include theory of quality and uncertainty of spatial data in their research programmes. The International Cartographic Association (ICA) research agenda focuses on visualization of data quality in general, and spatially varying quality in particular [20]. The International Federation of Surveyors (FIG) deals with the quality of data as part of the work of the Commission III Spatial Information Management in the field of establishing data-quality-standards relevant to spatial information management in cooperation with international spatial data standard committees [21]. Great interest in data quality and uncertainty is also visible in the International Society of Photogrammetry and Remote Sensing (ISPRS) program, where several Commissions deal with this issue, including problems related to quality and uncertainty modelling as well as the semantic modelling and linking of ontologies. ISPRS-WII/4 working group has recently introduced the new term ‘trust in spatial data’ [6], which refers to the level of matching real-world phenomena with their abstraction in a database. Data quality and data usability are a priority in the Association of Geographic Information Laboratories in Europe (AGILE) scientific programme [22] with the number of papers on spatial data quality growing consistently [23]. Regardless of the conferences organized by the aforementioned associations (e.g International Symposium on Spatial Data Quality (ISSDQ) or International Workshop on Spatial Data Quality), several national seminars and conferences on data quality issues take place every year. Furthermore, SDQ sessions are arranged within the framework of most conferences related to geographic information, spatial data infrastructure, data analysis, and management.

The steady rise of SDQ&U research results in a large and growing number of publications. Analysis of these publications is possible thanks to bibliometrics, i.e a statistical analysis of publications.

1.2 Bibliometrics

The term ‘bibliométrie’ was coined by the Belgian businessman Paul Otlet in 1934 and defined as ‘the measurement of all aspects related to the publication and reading of books and documents’ [24]. Thirty-five years later, the term was popularised by Pritchard in the publication entitled “Statistical Bibliography or Bibliometrics” [25].

Although there are numerous bibliometric databases, the most commonly used ones are: Web of Science (WoS), Scopus and Google scholar [29], until 2004, the Science Citation Index Expanded (SCIE), the Social Science Citation Index (SSCI) and the Arts and Humanities Index databases, available together by using the WoS, were the only exhaustive sources of data citation. An alternative to WoS is Scopus developed by Elsevier. It indexes more than 15,000 refereed journals, but its reference lists have been indexed consistently only since 1996. Both databases belong to commercial providers and require an access fee. They vary in terms of the covered period: WoS has been archiving publications since 1900, while Scopus – since 1966. WoS and Sopus have been dominant among the academic society, mainly through the annual release of the journal impact factor, an essential tool that evaluates the significance and scientific position of a given publication. On the other hand, Google Scholar, developed by Google Inc., is freely accessible. This database indexes most of scientific literature available on the Web, e.g: books, preprints, journals, reports, books of abstracts, as well as files from digital archives. There is a debate in academic community on whether Google Scholar might become an alternative to the commercial citation databases as a source for evaluating studies, primarily due to the unknown quality of the resources indexed and overall policy [26, 27, 28, 29, 30].

Bibliometrics is perceived as an effective tool for analysing and monitoring scientific achievements and research trends in numerous disciplines of science, technology, and humanities. An essential component of bibliometrics is citation analysis, as it is used to indicate the impact of publications and expresses the significance of the obtained results for other, later studies. Garfield [32] enumerated fifteen reasons why scientists refer to existing publications. Among them giving credit to pioneers and peers, providing background for methodology, as well as correcting own work and the work of others are of utmost importance.However, many researchers criticised citation analysis and demonstrated its limitations. The main drawbacks, as found by Smith [3334]. It comprises a set of methods to create an overview and to depict the development of a given research field.

While many bibliometric studies have been carried out in disciplines dealing with spatial data including remote sensing [37] and GISciences in general [38] the issues of spatial data quality and uncertainty have been broadly summarised only during a panel discussion that took place at the Sixth International Symposium on Spatial Data Quality (ISSDQ), held in Canada, in July 2009 [18].

This paper aims to provide a statistical overview of spatial data quality studies by bibliometric analysis of publications indexed in WoS from 1990 until the 23^rd of May 2018. WoS was selected after an in depth literature review. This decision was supported by the following reasons: multidisciplinary i.e wide and exhaustive indexing of the journals, books, and proceedings, the time span (since 1990), as well as a warranty of high quality of indexed publications [26, 29, 31, 39, 40].

We reveal the patterns and trends of scientific publications, geographical distribution, as well as the most productive journals and authors along with the most cited papers. The research output and impact in specific fields of research was also determined. The analysis provides several insights which may aid researchers, data providers, and public administration in understanding the development of the field. The next section (section 2) provides a description of methods and data used. Then the results are presented and discussed in sections 3 and 4. The paper ends with a brief concluding section.

3.1 Publication outputs

The total number of publications (TP) on spatial data quality and uncertainty from 1990 till May 23 2018 equals 2,090, till the end of 2017 – 2,069. Only one paper related to SDQ&U was published in 1990, a year later – as many as 6 papers. During the last decade of the 20^th century the total number of publication grew to 155, at the end of the first decade of the 21^st century it rose 6-fold, and in the beginning of 2018 (23 May) 2.25-fold, and reached the value of 2,140. The annual increase in the global number of publications on SDQ&U could be described by the second degree polynomial (R square = 0.9485), with three obvious peaks in 2003, 2009-2010 and 2016, marked in dark blue in Figure 1. Research articles accounted for 59.4%, followed by

Figure 1

The increase in the annual number of publications since 1990 till May 2018

proceedings papers (34.1%), book chapters (3.2%), and reviews (2.1%). The remaining 1.2% were editorial materials; book reviews, books (as many as 5), and meeting abstracts.

The average article length varied slightly with the coefficient of dispersion equal to cv = 0.172. A typical publication had approximately 12 pages (RMS=0.38) and was written by 3 authors affiliated to 2 or 3 countries.

The publications on SDQ&U were cited 22,737 times by 20,352 articles. The rate of self-citations was just 4.4%. The average number of references per article (excluding self-citations)was 10.60, and it fluctuated moderately with the coefficient of dispersion cv = 0.53. As many as 781 publications (35.5%) were not cited, while 223 (which is 10%) were referred to only once. The percent of publications that were quoted at least 10 times slightly exceeds 25. Only 37 scientific papers were cited at least 100 times. The distinct growth in the number of publications and the citations shows the steady increase and communication in the SDQ&U research during the past three decades (Table 1). The correlation between the number of authors who conducted research on SDQ&U and the number of journals is significant, Pearson’s r=0.989.

Moreover, a strong positive correlation was found between the annual total number of publications and the cumulative references count. On the other hand, cited references per publication are characterized by a moderate, negative correlation with the number of publications and the total count of references, and a strong negative correlation

with the time that passed between the publication year and the year of our analysis (Table 2).

The first paper indexed in WoS, entitled “Generation of Project Base Maps Using US geodata for the National Water-Quality Assessment Program”, was written by Mladinich [45], and published in Technical Papers of 1990 ACSM-ASPRS Annual Convention, vol 2: Cartography. This article has not been cited yet. The first cited articles were published a year later, in 1991. Since May 2018 they were cited 313 times without self-citation, with an average number of 62.6 citations per item.

3.2 Subject categories, journals and conferences

Global research on spatial data quality and uncertainty spanned over 145 WoS research categories, which accounts for 41.4% of all research areas. For 27 categories, the number of publications was one, and for the next 14 only two publications. The median of the number of publications in the WoS categories equals 7, while the mean value - 32.15. 90% of publications on SDQ&U fall into WoS categories. This diversification is best reflected by Lorentz curve and Gini coefficient (Figure 2). The blue line shows the distribution of publication output among the WoS categories from 1990 till the 23^rd of May 2018, while the thin, grey line at the 45^∘ angle presents perfectly equal distribution of publications. The further away from the diagonal (e.g: Geosciences Multidisciplinary, Geography, Computer Science Interdisciplinary Applications, and Imaging Science Photographic Technology), the more unequal is the distribution.

Figure 2

Statistical dispersion of SDQ&U outputs across subject categories, journals and proceedings

Figure 3

Number of publications by authors

Remote Sensing accounted for 18.8% of all publications (402 papers), with an exponential increase in the number of publications in 5-year windows. The next most productive WoS category - Computer Science Information Systems was attributed to 17.5% papers, and until the year 2010 the number of these publications has grown steadily. In the second decade (2011-2018), the number of publications in this category fell by almost 50%. Geography physical and Geography comprised 10.9% and 17.0% output, respectively.

In both categories, a gradual, linear increase in the number of publications was observed. On the other hand, the outputs denoted to Environmental Sciences and Geosciences Multidisciplinary categories grew gradually and covered respectively 16.5 and 13.5% of publications on SDQ&U.

SDQ&U research works were published in 832 SCI indexed journals in the years 1990-2018 (May). There is a significant concentration of articles in 17 journals, which take 98% of all publications on SDQ&U (Table 3). This large dispersion is visible on the Lorenz curve (Figure 2b), and also confirmed by a high value of the coefficient of dispersion cv = 1.82.

The International Journal of Geographical Information Science published the most, i.e as many as 75 articles on SDQ&U, which constitutes 3.6% of the total output. It is followed by ISPRS International Journal of Geo-Information (24 papers, 1.5%), Environmental Modeling

Software (20, 0.96%), and Computers Geosciences (20, 0.96%). However, SDQ&U related articles are a small fraction of all papers, assuming the highest percentage (4.08%) for International Journal of Geographical Information Science and 3.07% for Transaction in GIS respectively (Table 3. column 3). On average, the percentage of papers on SDQ&U in the aforementioned journals slightly exceeds 1%. The most influential journals that published research papers on data quality were ISPRS Journal of Photogrammetry and Remote Sensing as well as Science of the Total Environment, with the 5-year impact factor of 6.457, and 5.102 accordingly. At the same time, Photogrammetric Engineering and Remote Sensing and International Journal of Digital Earth provide the highly cited articles with CPP equals 35.44 and 29.54 respectively. It is worth noting that publications related to crowdsourcing data, mainly OpenStreetMap, are most frequently cited.

Articles published in 17 top journals were cited 18.70 times on the average, achieving a considerable influence on further research. The most frequently cited paper with an average annual number of citations amounting to 32.89 is entitled “Crowdsourcing geographic information for disaster response: a research frontier”, and it was published in 2010 in International Journal of Digital Earth. The article was written by Goodchild, M. F. and Glennon, J. A. from University of California Santa Barbara, USA, and since 21 May 2018 it has been cited 289 times.

Research works on SDQ&U were presented at 448 international conferences, with the average number of papers equal to 2.56. The dispersion of the number of papers in conference proceedings measured by Gini coefficient equals 0.64 (Figure 2c) and is lower than WoS categories and journals (Gini – 0.56 and 0.41 respectively). The five most productive conference proceedings are presented in Table 4.

The highly cited conference paper “Resolving conflict in eutherian mammal phylogeny using phylogenomics and the multispecies coalescent mode” written by Song Sen, Liu Liang; Edwards Scott V. and Wu Shaoyuan was published in 2012 in Proceedings of the National Academy of Sciences of the United States of America. The paper was cited 179 times, while the average number of citations per year equals 26.42. It is a result of co-operation between American and Chinese universities.

3.3 The most influential authors

Eight out of 6,053 authors were the most productive and they have made outstanding achievement on SDQ&U research. Among these authors Shi W.Z. from Hong Kong Polytechnic University and Goodchild M.F. from University of California published the largest number (25 and 22) articles, while Gelfand A.E. received the highest citation rate (28.88). Over 87.5% researchers have published just one article, which indicates that quality issues were treated as part of larger projects. The most productive authors in the field have conducted their research at universities in the USA (4 people), Europe (3 people: Netherlands, Spain, Germany), Canada (1), China (1), and Tunisia (1)) (see Table 5).

Papers related to spatial data quality had the highest percentage share in the publishing output of Beaubouef T. (from Se Louisiana Univ, United States). 10 out of 23 publications indexed in ISI (43%) addressed the SDQ&U issue. For comparison, with Goodchild M.F., who published the most on spatial data quality, this percentage was just over 10% (see Table 5). The most productive authors conducted their research in Remote Sensing, Geography physical, and Computer Science.

Only 40 scholars grouped in 7 clusters have worked on spatial data quality in broader national or international networks, which is reflected in the number of common publications equal or greater than 5. The collaborations network of these authors is presented in Figure 4. The circle size shows the authors’ co-operations measured by the number of links between other scholars, while the thickness of lines demonstrates the co-operation strength expressed by the strength coefficient s_ij. The distance between the authors indicates the relatedness of collaboration.

Figure 4

Network of scholars’ co-authorship in the field of spatial data quality

The most collaborative authors are: Wang J. (s_ij = 13), Li L. (s_ij = 13), Liu H. (s_ij = 10), Shi W. (s_ij = 9), Chen Y. (s_ij = 7), and Stein A. (s_ij = 7). However, only the authors grouped in the red cluster represented extensive international co-operation (China, the Netherlands, Canada, USA, and the United Kingdom). The quantity of multi-author papers roughly corresponds to the content, accounting for only 14.6% of all publications in the field.

3.4 Institutions and countries productiveness and geographic distribution

There were 1,928 research institutions that participated in spatial data quality research, out of which 234 conducted neither national nor international cooperation in the field of spatial data quality. Scholars from the Chinese Academy of Agricultural Sciences and Wuhan University (China) published the most papers related to research on spatial data quality. They are followed by two Australian universities: of Melbourne and Queensland. The Hong Kong Polytech Univ. was ranked in the fifth position. The 50 top scientific institutions conducting research in the field were dominated by 16 universities from the United States. The top ones included five universities from China (including Singapore and Hong Kong) and 16 from Europe, e.g: universities from the UK, Germany, Italy, Netherland, and France as well as international organisations like the International Union for Conservation of Nature (IUCN), located in Switzerland. The co-operation network is presented in Figure 5. The size of the circle and the label are determined by the potency of an institution. The distance between two institutions indicates their relationship in terms of co-citation links: the closer two institutions are the stronger the co-operations. Collaboration in the SDQ&U research is frequent, and is dominated by national co-operations. The most internationally collaborative countries are: USA-Canada-Australia-UK-Netherlands. China, Germany, France, Italy, and India have started broad international co-operation 2010 onwards. Five years later, Eastern European countries such as the Czech Rep., Poland, and Romania also joined international cooperation in the field of spatial data quality research.

Figure 5

Network of institutional co-operation

Research on spatial data quality and uncertainty has been conducted in 95 countries from all continents (Figure

Figure 6

Geographical distribution of countries involved in SDQ&U research

6). The first rank belongs to the USA (557 publications), followed by: China, Germany, the UK, Australia, Canada, Italy, the Netherlands, France, and India (Table 6). Poland, a Central European developing country, was ranked as the twelfth country (88.2 %), after Spain (89.3%). Norway was the most influential country, if average citation per publication is considered (CPP=45.14). The CPP was 21.26 and 18.31 for the United States and the United Kingdom respectively. Chinese publications are relatively the least cited (CPP=4.73), which shows that they have had low impact on world science so far (Figure 6).

On the other hand, scholars from 19 countries have published just one publication, 13 of which have not been referenced till May 2018. Nine countries with two publicationswere ranked in the 67^th position (21.2 percentile of the output).

In general, a linear growth in the number of publications was observed in the 5-year window starting in 1990 (Figure 7). The exceptions are China, Canada and the Netherlands, where after 2009 a slight decrease in the number of publications was visible.

Figure 7

Temporal trend in the output of the field, by the selected country

The analysis of previous bibliometric studies concerning publications in GIScience journals (after [3835, 3646]) indicates a very strong correlation between these research areas and spatial data quality (R²=0.98 and R² adjusted amounts 0.96). Moreover, the number of publications is strongly positively correlated with country investment in research and development: the Pearson’ r coefficient equals 0.92. This dependence is also underlined by journals and conference proceedings in which scientists mainly publish (see Figure 3, Tables 3 and 4).

3.5 Co-words analysis

The word that dominated among 8,327 key words used by authors and key words plus in the research output related to spatial data quality and uncertainty was GIS (Table 7). It was used 312 times and strongly co-coupled (s_ij = 218) with the other frequently used keywords (Figure 8). In relation to papers on SDQ&U, GIS was mainly associated with computing environment as well as data and databases management, much less with analyses and applications. Uncertainty was the second most frequently used word. It appeared 167 times, out of which 120 times in the publication

Figure 8

Key-words co-network in 1990-2018

titles (GIS only 68). Uncertainty was mainly collocated with spatial data, information, modelling, assessment, management, and visualization. The term has been used more frequently since 2005, when spatial data users became more aware of the high influence of data uncertainty on the results of spatial analysis. Quality as a key word occurred 116 times (but only 75 times in the titles). Its usage is strongly connected with growing availability of volunteered geographic information, mainly OpenStreetMap (Figure 8 red cluster).

Key words analysis revealed that they could be clustered in 4 groups. The first group contains such frequently used concepts related to data imperfection as: spatial data quality, completeness, geometry, uncertainty assessment, trust, topological relations. It includes mainly papers on the basic problem of imperfection of spatial data in relation to established standards (national or international) or reference, more reliable data. The second group is focused on uncertainty related to any type of data processing. The concepts underline both raster and vector data analysing, as well as methods of simulation, interpolation,

kriging, modeling, data mining, fuzzy set theory, spatial statistics, and interpolation. The third key words group comprises uncertainty co-located with data storage, management, and delivery; while the fourth - applications for which imperfection is pertinent, like: agriculture, forest, land cover/land use, and geology. The key words groups ordered by total link strength (s_ij) coefficient are shown in Table 8.

The co-occurrence of relationships among key words used at least 10 times is shown in Figure 8. The colour indicates the period when the words were mainly used, while the size is proportional to the occurrence frequency. The lines depict the relationships between the words: the more lines the stronger the connection between the words.