Using CORONA Imagery to Study Land Use and Land Cover Change—A Review of Applications

: CORONA spy satellites o ﬀ er high spatial resolution imagery acquired in the 1960s and early 1970s and declassi ﬁ ed in 1995, and they have been used in various scienti ﬁ c ﬁ elds, such as archaeology, geomorphology, geology, and land change research. The images are panchromatic but contain many details of objects on the land surface due to their high spatial resolution. This systematic review aims to stud the use of CORONA imagery in land use and land cover change (LULC) research. Based on a set of queries conducted on the SCOPUS database, we identi ﬁ ed and examined 54 research papers using such data in their study of LULC. Our analysis considered case-study area distributions, LULC classes and LULC changes, as well as the methods and types of geospatial data used alongside CORONA data. While the use of CORONA images has increased over time, their potential has not been fully explored due to di ﬃ culties in processing CORONA images. In most cases, study areas are small and below 5000 km 2 because of the reported drawbacks related to data acquisition frequency, data quality and analysis. While CORONA imagery allows analyzing built-up areas, infrastructure and individual buildings due to its high spatial resolution and initial mission design, in LULC studies, researchers use the data mostly to study forests. In most case studies, CORONA imagery was used to extend the study period into the 1960s, with only some examples of using CORONA alongside older historical data. Our analysis proves that in order to detect LULC changes, CORONA can be compared with various contemporary geospatial data, particularly high and very high-resolution satellite imagery, as well as aerial imagery.


Introduction
The government of the United States of America (U.S.) started the space-borne photography CORONA program in the 1950s, with the first successful image acquisition in August 1960 [1,2].Lasting until May 1972, CORONA was the first high-resolution satellite imagery [3] dedicated to military intelligence purposes.The mission was to monitor Soviet missile strength [4] and Chinese nuclear programs [5], but it had global spatial coverage, particularly in Eastern Europe and Asia [6].The CORONA program consisted of satellites launched by Thor-Agena rockets [5,7] that were equipped with various panoramic camera models (referred to as Keyhole, KH), providing a black and white 70 mm film [5], with KH-4B having the best spatial resolution [8,9] (Table 1).Cameras of the satellites of the CORONA program were calibrated by concrete targets on the ground in Arizona [10].After 12 years of service, the mission was replaced by HEXAGON, a Central Intelligence Agency (CIA) program [11].While the CORONA name typically refers to satellites using KH-1 to KH-4B cameras, two other camera systems, ARGON (KH-5) and The CORONA images were declassified in 1995 by Executive Order 12951 because the program was no longer considered to be crucial to national security [14].The primary archive of the imagery-analog photographic products containing film negatives and photo prints that include 800,000 photographs-is stored by the National Archives and Records Administration (NARA) at the National Archives in College Park, Maryland [14][15][16].Following the 1995 declassification of military imagery, the dataset was converted into digital form and made available by the United States Geological Survey (USGS), together with imagery of the follow-up missions in 1995 (Declass 1 dataset), 2002 (Declass 2 dataset) and 2011 (Declass 3 dataset) [14,17,18].
A single CORONA image is a narrow strip, with size depending on the camera type, and it is approximately 20 km wide and 200 km long [9,16,18,19].Unprocessed CORONA images have severe spatial distortions [20] and contain limited spectral information due to 8-bit panchromatic (single band) image depths [21].As a result of a dual-angle system of two panchromatic cameras at 15 degrees off-nadir in KH-4, KH-4A and KH-4B systems, it allows also producing a digital elevation model [16,22].
Due to the availability of CORONA imagery since 1995, many researchers used the data to study various environmental processes [23], for instance, past land cover in the Caucasus Mountains [24], forest habitat fragmentation in the Bucegi Mountains, Romania [25], and post-agricultural forest succession in the Carpathian mountains [26].In many studies, CORONA images were used jointly with other geospatial data to assess changes in land use and land cover (LULC) since the 1960s [27,28], with several studies reporting substantial difficulties when processing the data [6,29,30].Although CORONA images were widely used in many research fields and the number of studies is increasing, a comprehensive assessment of CORONA imagery applications in studying LULC changes is still missing.However, previous studies that reviewed CORONA imagery were mainly related to technical issues, such as camera construction, imaging and algorithms [16,31].In this review paper, we therefore aimed to find out how CORONA imagery was used so far in LULC change studies, particularly looking at how they compare to other historical or contemporary geospatial data and how they were analyzed to assess the LULC change.This is important as a broad spectrum of current spatial datasets differs substantially in nature from old spatial data that may offer high potential for LULC studies.

Methods
In our research, we followed a typical research design of the systematic review [32,33], starting first with querying a database of scientific papers, defining criteria of research paper selection, selecting research papers for analyses and examining the content of research papers according to a structured set of questions [34].
For our analysis, we used the SCOPUS database (www.scopus.com,accessed on 1 April 2023), a general database of research publications that has been used in various review papers [34,35].To select papers in which CORONA data could possibly be analyzed, we searched article titles, abstracts and keywords by using the following set of expressions connected by logical OR: To build the query, we used quotation marks in case Corona was used jointly with image, imagery or images, as our initial trials showed that a simple phrase such as corona image (or corona imagery or corona images) provided a high number of papers related, e.g., to astronomy or medical sciences (for instance, COVID-19-related papers).The search was carried out by the end of 2022, and we set no restrictions for the publication year.
The final set of selected publications was initially screened first to remove papers that did not use CORONA imagery and that were accidentally selected from the SCOPUS database.Next, we examined the remaining papers using CORONA data and identified papers with a focus on LULC studies.For the LULC studies, we used a set of research questions to analyze the content of publications in the context of LULC and LULC changes.These questions were grouped into spatial, temporal and thematic aspects of the selected studies: 1. Spatial aspects 1.1.In which regions and countries CORONA images were used?1.2.What was the size of the study area?1.3.How did CORONA images cover the study area (entire study area-wall-to-wall mapping or a part of the study area using pre-defined sampling)?For question 2.1, we used 7 basic LULC categories from global land cover mappings [36] (forest, grassland, cropland, built-up, bare land and sparse vegetation; wetland and water; and snow and ice).For question 2.2, referring to methods used for CORONA imagery analyses, we were interested primarily in the division of manual versus automated image interpretation, specifying classification approaches for the latter.For question 3.3, regarding geospatial data, we grouped the data used in various studies into 7 classes: topographic maps; low and medium spatial resolution satellite images (pixel size > 100 m; LR images); high-resolution satellite images (pixel size 10-100 m, HR images); very highresolution satellite and aerial images (pixel size < 10 m; VHR images); LULC thematic maps; field measurements; and others.

Database Query
The query expression returned 154 items in total: 147 were research articles, 3 were review papers and 4 were book chapters published since 1995.From the initial set of 154 research publications, we rejected 53 publications: 38 were not relevant to CORONA imagery (e.g., related to research domains such as nanotechnology or astronomy), 8 papers were in languages other than English (two Korean, five Chinese and one Italian), 4 were book chapters, and 3 were review papers.Book chapters and review papers presented overviews or technical aspects of the CORONA program mostly in the context of archaeological applications.In the remaining set of 101 research articles, we identified 54 papers dealing with LULC for further analyses (full list see Appendix A).The other 47 papers dealt with digital elevation modeling and geomorphology (22), archaeological applications (19) or technical aspects of the CORONA program (6) (Figure 1).Further on, we only analyze and discuss the set of 54 LULC papers.With respect to the papers examined in this study, 54 LULC-relevant papers were published in 37 journals, with only 2 journals publishing more than 2 papers: Remote Sensing (11 papers) and the International Journal of Remote Sensing (4).Publication years varied, starting in 2000 (Figure 2).Recent years (2020, 9 papers; 2021, 10 papers; 2022, 6 papers; Figure 2) show increasing interest in using CORONA imagery to study LULC compared to the first two decades after declassifying CORONA data in 1995.

Spatial Aspects
The case studies were located in 36 countries (Figure 3, Appendix A).Most studies were carried out in Russia, China, Central and Eastern Europe, the Middle East and Eastern Africa.However, there are substantial gaps with respect to country-level geographical coverage, for instance, in Western Europe.The size of the study area varied from a small area of around 2 km 2 [37] to a large area of approximately 484,000 km 2 [30], and in one case, the size was even more than 1 million km 2 [38].In 34 case studies, CORONA imagery covered the entire study area (wallto-wall mapping) while 19 studies used CORONA to investigate a part of the study area using a pre-defined sampling strategy.In one paper, a combined approach of a wall-towall timberline mapping and analysis of forest cover based on 43 sample small plots was used [39].For 22 papers, the area of case study sites was not explicitly given, and we estimated the areas based on maps published in the papers.
A wall-to-wall mapping of an entire study area using CORONA imagery was almost three times more common than carrying out sampling when the size of the case study area was less than 5000 km 2 .On the contrary, for study areas exceeding 5000 km 2 , a typical approach was to sample part of the study area for analysis (Figure 4).

Thematic Aspects
Most studies considered several LULC categories simultaneously, while 24 out of 54 papers focused solely on one single LULC category.Forests were the most frequently studied LULC category, followed by wetlands and water bodies, bare land and sparse vegetation land cover classes (Figure 5, Appendix A).
With respect to the studies reviewed here, 35 studies used manual interpretation (onscreen vectorization) and 13 studies used automated classification, while 6 studies combined manual interpretation and automated classification to extract LULC information from CORONA imagery (Appendix A).With automated approaches, forest and cropland were preferred over other LULC categories.We identified nine studies that used automated classification to map the entire study area (wall-to-wall mapping).The study area sizes in this subset of studies ranged from 8 km 2 [40] to 44,957 km 2 [41].Among automated approaches, object-based image analysis (OBIA) or various classification approaches using the textural features of CORONA imagery were applied (11 out of 13 studies).

Temporal Aspects
Five studies were confined to single-moment analysis of LULC at the moment of CO-RONA imagery acquisition (Appendix A).These studies were highly varied and included forest cover mappings carried out in the 1960s-1970s [42], the use of panchromatic images to generate enhanced multi-layer products [43], and testing land cover classification methods [42,44,45].
In total, 49 papers represented a multitemporal approach (Appendix A), with CO-RONA imagery being one of the datasets used in the time series spanning the entire study period.In 42 studies, the beginning of the study period took place in the 1960s or 1970s, with CORONA imagery being the oldest dataset used to assess LULC at the beginning of the study period.Seven studies analyzed LULC changes before the launch of the CO-RONA program using older data than CORONA, for instance, archival maps, historical or archival records and older aerial photographs, with CORONA imagery being an intermediate dataset of the entire time series.The study period extended in most cases into the 21st century (46 studies), with only three studies considering LULC changes solely in the 20th century (Figure 6).
Multitemporal studies used various geospatial data to detect LULC and its changes by comparing them with CORONA imagery (Figure 7).The dominant type of data used in multitemporal studies were very high-resolution images of various observation programs, platforms and data archives.However, among various types of imagery, Landsat was the most frequently used dataset that was analyzed jointly with CORONA (25 studies).
Multitemporal studies focused on forest cover changes (deforestation, forest expansion and degradation of dryland forests), urban expansion, changes in water bodies and glacier extents.To study and detect forest cover changes, high-resolution imagery, such as Sentinel-2, and Landsat as well as LULC thematic maps were mainly used [28,46].Only a small number of studies employed low-resolution satellite imagery; e.g., a moderateresolution imaging spectrometer (MODIS) was used to analyze changes in snow cover and ice extent [47].One interesting feature of using CORONA imagery is that the very high-resolution imagery of HEXAGON and GAMBIT (the follow-ups of the CORONA mission) was jointly used with CORONA to study changes in urban areas [5].

Discussion
Following the declassification of imagery acquired during the CORONA and subsequent missions in 1995, LULC studies using spy satellite image data started to increase in numbers, reaching a total of 54 papers by the end of 2022, with case studies located in different countries and covering all continents except for Australia and Antarctica; however, one of the early research papers published in 1998 used CORONA imagery in a glaciological study of the Ross Ice Shelf [48].Quite interestingly, some studies that had been included in the preliminary set relevant to the LULC analysis had to be removed after a detailed inspection of the papers, as it revealed that they were using imagery acquired in the CORONA programʹs follow-up missions (HEXAGON) and not the imagery of the CO-RONA program [49,50].
As the mission of CORONA was to provide images, especially of the Soviet Union and China [51], these two countries were featured in most LULC studies in which CO-RONA data were used.Our results showed that except for present-day Russia and China, CORONA imagery was also commonly used in India and Nepal; the Middle East (Iraq, Iran and Syria); Northern and Western Africa (Egypt, Mali and Senegal); and Central and Eastern Europe (Poland and Romania) [26,28,30,[42][43][44][52][53][54][55][56][57][58].Some of these countries (e.g., Poland and Romania) have good coverage of CORONA imagery due to the goals of the program and the importance of the countries during the Cold War era [59][60][61].One of the reasons that probably pushed researchers to use CORONA in several countries listed here is the poor availability of comparable aerial data that are not easily accessible via online archives, although they exist [5]: for instance, aerial data for Poland acquired in the 1950s or the 1960s.At the end of 2022, we could not find any LULC studies using CO-RONA imagery in various countries that potentially provide good coverage of these data (post-Soviet countries, such as, e.g., Ukraine or Belarus), suggesting that CORONA data have not been fully explored within the LULC context yet.This is changing, however, with new studies published for various post-Soviet areas, such as the Caucasus Mountains [24].
We noted two strategies for using CORONA imagery to study LULC and its changes in the selected case study areas.In approximately two-thirds of the studies, with a case study area that is typically less than 5000 km 2 , researchers attempted to find and later analyze full CORONA coverage for the entire study area.The other papers, typically with case study areas exceeding 5000 km 2 , used CORONA data only for selected fragments of their study area.The reason for the focus on small case study areas or fragments of large study areas is related to the size of a single CORONA strip (depending on the camera, approximately 4000 km 2 ) and its width (approximately 20 km), which pose various difficulties related to geo-rectifying the strips in larger study areas [22].Spatial distortions in CORONA images are difficult to reduce and may result in significant errors when overlaying and comparing them with contemporary very high-resolution satellite and aerial images [44,62,63] .Our findings confirm that spatial distortions are a major problem when using CORONA imagery, as the case studies mostly use the manual or semi-automated rectification and georeferencing of CORONA data.While no fully automated approach for this processing step was provided, Nita et al. [42] developed a CORONA imagery orthorectification method by relying on the structure from motion technology and were able to efficiently process more than 200 CORONA strips for a study area exceeding 200,000 km 2 .Mean absolute errors were less than 4 m in flat areas but increased to 10 m in hilly areas and over 20 m in mountainous areas [42].Song et al. [6] used the scale-invariant feature transform to match CORONA imagery to Landsat data and received errors close to one Landsat MSS pixel.In other studies, the reported root mean square errors were around 10 m [26,29,62].
Spatial distortions of CORONA images also contribute to difficulties in image mosaicking.In addition, receiving a seamless coverage of CORONA imagery for large areas is extremely demanding due to radiometric differences, acquisition frequency and cloudfree image availability, forcing researchers to analyze CORONA imagery rather than on a one-to-one basis [64].Leempoel et al. [65] and Gurjar and Tare [66] in their research mentioned that these specific problems in using and processing declassified CORONA images are quite challenging.A frequent approach, therefore, is to use CORONA imagery as a sample that covers only a part of the larger area, focusing, e.g., on specific objects or locations of interest.For instance, Marzolff et al. [67] used a sampling strategy to study changes in dryland woodlands in Morocco at the tree level, while Song et al. [30] employed sampling to assess forest cover change rates in Sichuan Province, China, over an area of 484,000 km 2 .
One possibility to increase the use of CORONA imagery in the research community is to provide access to geometrically corrected imagery.The CORONA Atlas & Referencing System, developed by the University of Arkansas' Center for Advanced Spatial Technologies (CAST) [68], carried out the orthorectification of a subset of CORONA imagery and provided free public access to the georeferenced dataset.The imagery covered mostly the Middle East, and while it was primarily dedicated to archaeological studies [63], it was also used in other domains of science.For instance, Saleem et al. [41] used the CORONA Atlas to map LULC in the 1960s in northern Iraq for the study area exceeding 40,000 km 2 , for which they produced a wall-to-wall LULC map from images available in the CORONA Atlas archive.
We found out that forests were the most frequently studied LULC category.Forests were studied in different climatic zones such as central Poland [28], southern Morocco [67], western Russia [52], Canada [69], southeastern China [30], Norway [39] or Brazil [6], which show that CORONA, due to its appropriate high resolution, was good enough to capture extent and properties of various forest types in diverse environmental settings.Another frequently studied category is built-up areas in urban and peri-urban settings.For example, Shalaby et al. [55] studied the directions of urban expansion and confirmed that the high spatial resolution of declassified images was sufficient to accurately map initial morphologic features.Pan et al. [70] monitored the urban expansion of Xining City in China and confirmed the capability of CORONA and its resolution to quantify urban expansion.
In the analyzed studies, manual interpretation (on-screen vectorization) was a dominant approach used to classify LULC based on CORONA images.Manual interpretation is preferred over automated approaches due to the risk of errors in the latter because of poor quality and the mismatching of trained samples [26].A minority of studies attempted to use automated classification methods.As already noted, OBIA or textural features of CORONA imagery were the most commonly used, with random forests (RF) and support vector machine (SVM) algorithms commonly implemented in image classification.For instance, Song et al. [30] applied various textural features of CORONA imagery and SVM to assess forest cover in the 1960s in a large-scale study of Sichuan Province, China.They confined the analysis using stratified random sampling based on forest cover and its changes occurring since the 1970s, which were assessed using Landsat data.Song et al. [6] used a similar approach to study forest cover changes in Eastern United States and Central Brazil.Chen et al. [71] proposed supervised LULC classification using a combination of the textural and spectral features of CORONA and SPOT images to study long-term changes in the wetlands of Taiwan.Deshpande et al. [44] employed convolutional neural networks and RF to receive a land cover map with five categories, and the latter approach also used the textural features of CORONA imagery.Among studies using automated approaches, most studies report land cover classification accuracies exceeding 90%, especially in the case of binary classifications, such as forest-non-forest, or classifications with a low number of land cover classes [6,26,30,52].In specific cases, accuracies may drop below 90% for some land cover classes.Chen et al., 2020 [71], observed low user accuracy with respect to pond delimitation (below 60%), while Agapiou in 2021 [45] found that the accuracy for some land cover categories, such as vegetated areas and salt lakes, was below 90%, although the overall accuracy of land cover classification reached 94%.
Quite interestingly, while OBIA seems to be well-suited for interpreting very highresolution CORONA data, it was used in less than 20% of all considered studies, prevailing, however, among studies preferring the automated approach.In addition, for approaches using a combination of automated and manual methods, image segmentation frequently preceded the visual interpretation of segments [72,73].One of the reasons for not using OBIA for processing CORONA data even more frequently is the need to maintain high geometric accuracy with respect to the segmentation outputs that typically require working with orthoimages, which are quite demanding in the case of CORONA.It was also relatively rare to use additional data, e.g., field measurements or aerial imagery, during that period to overcome the limitations of CORONA imagery interpretation [43,66,74].For instance, Andersen and Krzywinski [74] studied 19 sites, each containing a minimum of 30 individual trees that were mapped and measured.They interpreted features located in the CORONA image and compared the outputs to trees recorded in the field.
We found that most of the studies utilizing CORONA imagery are multitemporal, providing a comparison of various geospatial data to detect spatial changes in LULC in multitemporal analyses.The low cost and high spatial resolution of CORONA images offer a fairly effective and straightforward solution for studying long-term LULC evolution, especially in the case of slow and gradual forest cover expansion [64].Only seven studies assessed LULC and its changes before the operation of CORONA using historical records, archival maps and older aerial imagery.For example, Jabs-Sobocińska et al. [26] studied post-agricultural forest expansion in the Polish Eastern Carpathians in a multitemporal approach from 1944 to 2019 using war-time German aerial photography dated 1944, CO-RONA and Sentinel-2 imagery.Due to the high visibility of forest edges in images, CO-RONA was used specifically to map abandoned agricultural lands that transformed later into forests in the region of post-war resettlements and depopulation.The study of Chen et al. [71] on wetlands and water bodies (irrigation ponds) in Taoyuan City, Taiwan, included two periods: 1904-1960s and 1970-2000.For the first period, the authors used the Taiwan Ancient Land Map Digital Archive (TALM) created in 1904 and CORONA imagery acquired in 1969, while for the latter period, CORONA imagery and SPOT-5 were acquired in 2000.
For the majority of multitemporal case studies (49), CORONA imagery remained the earliest dataset used in the time series.In the analyzed studies, CORONA imagery was used alongside a variety of other geospatial data in various combinations.One of the CO-RONA features enabling the comparison of CORONA imagery to other geospatial data is its high spatial resolution (1.8-7.5 m), although the spatial resolution at the edges of an image could be much lower than in its central part [43].In various LULC studies carried out with CORONA imagery, the researchers preferred data with a spatial resolution similar to that of CORONA in order to obtain a comparable scale of measurement, the same sample size and identical geometry [75].Therefore, in LULC change studies, CORONA images were most frequently compared with very high-resolution satellites or aerial imagery.DeWitt et al. [76] used CORONA imagery jointly with very high-resolution remote sensing data, including IKONOS, WorldView-1 and GeoEye-1, to analyze local-scale LULC changes.Spiekermann et al. [77] used RapidEye imagery, offering images with spatial resolutions similar to CORONA (5 m) that are suitable for studying bare land and sparse vegetation and forests, while Racoviteanu et al. [47] used RapidEye and WorldView-2 to study glacier changes in Nepal.It was also common to compare CO-RONA data with high-resolution imagery, such as various Landsat satellites or Sentinel-2.For instance, Htwe et al. [46] used Landsat 5 and 7 data from 1989 to 2009 to detect LULC changes related to the transformation of farming systems in Myanmar.Saleem et al. [41] used Landsat Multispectral Scanner, Thematic Mapper, Enhanced Thematic Mapper Plus and Landsat Data Continuity Mission imagery in comparison with CORONA imagery to map and quantify long-term LULC changes in northern Iraq.
The variety of geospatial data used in various studies confirms that CORONA is potentially comparable to a range of recently developed datasets and may be a valuable dataset in LULC change detection studies.Various difficulties related to the processing chain, however, require that CORONA imagery is employed in a general framework of qualitative, post-classification comparison methods [78,79], with independent LULC thematic maps derived from various datasets and later compared to each other using simple thematic map overlays.
Several factors ought to be considered when different datasets are combined to study LULC and its changes, including the availability and quality of geospatial data, user experiences and proficiency of the procedures [80].Although CORONA imagery is widely used by researchers due to the various advantages it offers, there are also several drawbacks reported in the analyzed papers related to data acquisition and data quality that affect the choices of processing methods.The lack of digital metadata is another factor hampering the use of CORONA in effective large-area mapping [67].It increases the difficulties of georeferencing, frequently referred to as a time-consuming process that requires available ground control points [6,8,26,[28][29][30]44,64].The automated approach to classifying CORONA imagery likewise has many difficulties with respect to collecting training samples, characteristics of various phenomena, spectral limitations and radiometric distortions of CORONA [6,66].Overall, variations in the quality and temporal coverage of the CORONA imagery collection [64], geometric distortions [30,43,62,63,65] and resultant difficulties in mosaicking [62] have caused researchers to reduce the use of these images.

Conclusions
In this overview, publications related to CORONA imagery were systematically analyzed.Although our study provides evidence of the increasing usage of CORONA imagery, we note that several drawbacks prevent their wider application, with the most important barriers being a lack of georeferencing and highly variable radiometric quality.These shortcomings compel researchers to either limit the size of the case study area or use various sampling strategies to cover the study area with a sufficient amount of data, thus decreasing the effort dedicated to georeferencing.Moreover, researchers prefer manual interpretation methods, especially for relatively small areas, over automated classification approaches that may result in low accuracy and misclassifications.Nowadays, the potential of CORONA imagery is becoming increasingly known compared to the past.While our focus was on LULC, roughly 40% of studies that used CORONA imagery were dedicated to other fields such as terrain mapping and analysis, frequently focusing on glacier mapping and mass balance assessment, or archaeology, which was focused upon mostly in the Middle East.However, LULC-related studies prevail, and our analysis shows that the outputs of the CORONA program are suitable sources for analyzing LULC globally, encompassing various land cover categories (mostly forests, wetlands and water and built-up areas) and offering high spatial resolution data for the 1960s and the early 1970s.The declassified imagery has particular potential for detecting and analyzing historical forest changes, as forest cover change is a long-term process requiring historical depth offered by data such as CORONA [26,28,30].From the perspective of LULC change monitoring, the combination of various datasets with different spectral and spatial resolutions with declassified CORONA imagery was found to be useful, providing a reasonable accuracy with respect to LULC change detection.CORONA imagery was mostly compared to high or very high-resolution data (aerial photographs and satellite imagery) and was used primarily to extend LULC time series to the 1960s, especially in areas with poorly accessible aerial imagery with comparable quality.

Figure 1 .
Figure 1.SCOPUS query results and selection of the subset of LULC-related papers.

Figure 2 .
Figure 2. Number of LULC-relevant studies using CORONA imagery published per year.

Figure 3 .
Figure 3. LULC case studies with CORONA imagery: country distribution.

Figure 4 .
Figure 4. Size of the study area in relation to CORONA imagery coverage.

Figure 5 .
Figure 5. LULC categories studied in the analyzed papers.

Figure 6 .
Figure 6.Time periods in the analyzed case studies.Each row represents one case study; blue color indicates the beginning, and green indicates the end of the study period.

Figure 7 .
Figure 7. Various data types used with CORONA imagery to study LULC.HR: High-resolution images; VHR: very high-resolution images; LR: low-resolution images.
denotes the study areaʹs size approximated using maps and data published in the study. *