Analysing the use of OpenAerialMap images for OpenStreetMap edits

ABSTRACT

OpenAerialMap (OAM) is a crowdsourcing platform for uploading, hosting, sharing, displaying, searching, and downloading openly licensed Earth image data from around the world. OAM was launched with the primary goal to facilitate rapid disaster response mapping after natural events, such as floods or hurricanes. Images contributed to OAM can be used as a background layer in OpenStreetMap (OSM) editors to edit map features, such as buildings or facilities, which may have been affected by such events. This study analyzes how the provision of OAM images is associated with changes in underlying editing patterns of OSM features by comparing the number of edited OSM objects between 2 weeks before and 2 weeks after an image has been shared through OAM. The comparison also involves other aspects of OSM editing patterns, including geometry types and OSM primary feature types of edited objects, type of feature editing operations, edits per user, and contribution distribution across continents. Results show that the number of point features added to OSM more than quadrupled within the first 2 weeks after OAM image upload compared to that of 2 weeks before, and that the number of ways added almost doubled. This suggests that the OSM community utilizes provided OAM images for OSM map updates in the respective geographic areas. The study provides a showcase which demonstrates how information from one crowdsourcing platform (OAM) can be used to enhance the data quality of another (OSM) with regards to completeness and timeliness.

KEYWORDS:

• Crowdsourcing
• aerial images
• open source
• data quality
• cross-linkage

1. Introduction

Numerous terms have been introduced to describe the different facets of sharing crowdsourced geospatial information, such as collaborative mapping, neogeography, participatory sensing, citizen science, or volunteered geographic information (VGI) (Bertolotto, McArdle, and Schoen-Phelan 2020; Goodchild 2007; Karim et al. 2020; Lin 2020; Lotfian, Ingensand, and Brovelli 2021; Saralioglu and Gungor 2022; See et al. 2016). All these concepts share one paradigm, which is that anyone even without cartographic skills and experience can collect and disseminate geographic information about a location or carry out other activities related to user-generated spatial content. OpenStreetMap (OSM) is the most popular and widely used open-source mapping platform worldwide with currently over 11 million registered users. It can be used as a tool for crisis response and planning activities (Bertolotto, McArdle, and Schoen-Phelan 2020), for example, in connection with earthquakes (Poiani et al. 2016; Soden and Palen 2014) or forest fires (Meier 2012). Initially, OSM contributions relied on the contributors’ local knowledge of the area and required field mapping techniques, which included uploading and integrating GPX track records from handheld GPS units. However, the provision of image service APIs and the development of custom Web mapping services allowed OSM contributors to trace roads and other features remotely in OSM editors. For example, Yahoo granted OSM the right to use its aerial image in late 2006 (Coast 2010; Haklay and Weber 2008), and Microsoft granted royalty-free use of the Bing Maps Imagery Editor API in November 2010 (Coast 2010). Over the years, additional providers of aerial and satellite images, including Esri, DigitalGlobe, and Mapbox made their data available to the OSM community for object tracing in OSM (Anderson, Sarkar, and Palen 2019; OpenStreetMap Wiki 2022). OAM, after ideas dating back to 2007, was relaunched in 2015 by the Humanitarian OpenStreetMap Team (HOT) with a grant from the Humanitarian Innovation Fund. It is operated on HOT’s computing infrastructure and provides a platform which facilitates the upload of current aerial images, in particular, to support humanitarian mapping with OSM (Humanitarian OpenStreetMap Team 2017). OAM hosts aerial images from the Open Imagery Network (OIN), which is a federated network of public storage endpoints on cloud services. It provides API and browser access to the entire image collection through a common OIN index file. Users can navigate the OAM map and click on any cell to display footprints, overviews, and details for each image. Once an image has been selected, it is embedded in a map layer which can be opened in an OSM editor, such as iD or JOSM, and be used as a background image to trace geographic features. OAM hosts currently around 15,000 images shared by over 1300 users worldwide. Images posted on the platform were acquired through satellite, drone, aircraft, balloon, or kite.

The goal of this study is to assess how OSM editing patterns change in the short-term and long-term in areas where OAM images are shared. The three following related tasks will be addressed:

1. Describe user contribution patterns to OAM through modeling the number of OAM contributions per user and the time span between a user’s first and most recent image upload to OAM.

2. Compare OSM editing patterns between 2 weeks before and 2 weeks after an OAM image has been uploaded for OSM features falling inside the area of the uploaded OAM image.

3. Quantify OSM editing patterns during the first 26 weeks after an OAM image upload through counting the weekly number of mappers editing OSM features in the area covered by the uploaded OAM image.

This study sheds light on the interaction between two widely used VGI platforms and, more specifically, the benefits of openly licensed Earth image for OSM.

The remainder of the paper is structured as follows. Section 2 reviews previous work on the use of airborne images and other sources for OSM updates. Section 3 describes data preparation and analysis methods used in this study, which is followed by the presentation of results in Section 4. Section 5 discusses the results and limitations of the study. This is followed by conclusions and directions for future work in Section 6.

2. Related work

Timely, high-resolution aerial imagery is critical for recording and monitoring environmental changes. The relative affordability of unmanned aerial vehicles (UAVs) and Unmanned Aerial Systems (UAS) has lowered barriers to obtain customized aerial images, resulting in a global supply of aerial images (Johnson, Ricker, and Harrison 2017). This led to the establishment of various image hosting platforms which provide a repository of UAV images to be shared with others. Data quality assurance, licensing, broad engagement of contributors, and provision of processing tools are considered ongoing challenges with such repositories. OAM is an open service for the sharing and use of images. The original copyright remains with the holder of the image but also grants OAM to license the image as CC-BY 4.0, with attribution as contributors of the Open Imagery Network (OpenAerialMap 2022). OAM has been assessed and used in a variety of research studies for different purposes. An online survey among 45 participants with the aim to assess the advantage and disadvantage of open-source images in archeology sites taken by drones found that the majority of respondents regarded volunteered images to be of better quality than traditional means of data collection, e.g. satellite images (Jorz 2019). However, only one out of three in-person interviewees in the study who had experience with OAM found the images of that platform to be of adequate quality. The author concludes that having multiple images available for the same area would allow to more reliably assess the data quality of OAM images, similar to how multiple contributors working in the same area on different VGI platforms can improve data quality, according to Linus’ Law (Haklay et al. 2010). Another study selected and analyzed five OAM aerial images covering different environments to detect vegetated areas by examining several RGB indices (Agapiou 2020). Since these images were downloaded as is, it was necessary to apply a radiometric calibration before further processing. Linear regression models showed that the green leaf index (GLI) was the best index, providing best feature detection with limited metadata. The authors conclude that ready-to-fly drones with RGB cameras will continue to play an important role for small survey campaigns and applications, such as vegetation extraction. Another study applied deep learning semantic segmentation methods for delineating building footprints in refugee camps from OAM drone imagery, demonstrating the applicability of this method in complex humanitarian applications (Chan et al. 2022).

Like in the case of OAM and OSM, the joint use of multiple VGI platforms can lead to improved data quality in one platform or both platforms. For example, Mapillary road-level image is included as a layer option in two common OSM editors (iD and JOSM), which makes it convenient for OSM mappers to use street-level photos for data editing in OSM. One study analyzed to which extent OSM feature edits use Mapillary street view-level data, based on tag information of edited OSM features and changesets (Juhász and Hochmair 2016a). This process of referencing the source dataset in tags or descriptions is referred to as cross-linkage. The study identified cross-linkage for 21 out of the 26 primary feature categories in the examined OSM dataset and found that Mapillary was most frequently associated with changesets rather than with individually edited features.

The OSM community uses various data collection techniques, such as field surveys, GPS trackers, and on-screen digitizing from aerial images to create verifiable information on the ground (Haklay 2013). Especially tracing from aerial images allows OSM contributors to effectively update features in a study area of interest even without prior mapping experience (Juhász and Hochmair 2018b). Desktop (or armchair) mapping from aerial image sources is commonly used to update OSM maps after natural disasters in remote areas, with only a small proportion of mappers actually traveling to the affected regions (Ahmouda, Hochmair, and Cvetojevic 2018). With desktop mapping, contributors do not need to be physically present at the location to be mapped, so that such mapping efforts are collaborative and distributed to volunteers around the world (Herfort et al. 2021). Mapping efforts are typically organized through web tools, such as the HOT Tasking Manager.

Part of OAM contributions consist of shared drone images. Drone contributions in different image sharing platforms (Dronestagram, Travelwithdrone and Flickr) are more likely to be found in richer countries, near water bodies (coastal regions, lakes) and airports, and at lower elevations (Hochmair and Zielstra 2015; Mandourah and Hochmair 2021). The distribution of users’ average daily contributions of drone images can be approximated through a power law, where most users contribute less than one dataset per day on average, and only a few users contribute more than 10 drone images per day on average. This pattern of participation inequality is known from other VGI platforms as well (Juhász and Hochmair 2016b; Neis and Zielstra 2014).

3. Materials and methods

3.1. Data collection and processing steps

3.1.1. OpenAerialmap

The OAM website provides a world map divided into tiles, and each tile shows the number of contributions. Upon sending the browser request to open the world map, the browser receives a fetch file in JSON format as part of the server response. The file contains image bounding box coordinates, all the meta-information the contributor provides, and an id assigned to the image contribution. During the upload process, users can provide information related to used equipment (such as the platform, sensor, and camera), the image (such as title, acquisition start and end time, and tags), and the contributor (such as name and contact). After uploading an aerial image, OAM assigns a unique URL to the image, which is provided in the extracted fetch file and can subsequently be used for image download. We extracted the metadata for all 10,340 OAM aerial images available on 24 May 2021. The metadata showed that around 40% of the images were taken by drones, 37% by aircraft, and 23% by satellite. Only four images were taken by balloon, and no image was taken by kite because this method was added to the platform options after 24 May 2021. Some users did not specify the correct image acquisition platform because drone model numbers were occasionally found in the sensor field of images contributed by aircraft. This indicates that these images were actually acquired by drone and not aircraft. Metadata showed drone model numbers in 84 out of 3,848 images which were stated to be taken by aircraft.

As examples for OAM datasets, Figure 1 shows three images taken by drone (a), aircraft (b), and satellite (c). The image characteristics follow OAM requirements and therefore provide a vertical view. The drone image has the best spatial resolution (3 cm), the smallest coverage area (0.92 km²), and features the largest map scale which shows most details. The aircraft image covers a larger area (2.6 km²) at a smaller spatial resolution (0.5 m). Lastly, the satellite image covers the largest area (73.9 km²) also at a 0.5 m spatial resolution. OAM recommends obtaining aerial photos from drone cameras linked to onboard GPS to record location and orientation information. This is because non-linked drone cameras sync GPS data in EXIF tags, which may lead to a positional delay of up to 2 seconds. An OAM article specifies recommendations for drone equipment for this purpose (World Bank and Humanitarian OpenStreetMap Team 2019).

Figure 1. Examples of aerial images from different acquisition platforms uploaded to OAM, taken by drone (a), aircraft (b), and satellite (c).

As a first processing step, 60 images with a low resolution, i.e. a pixel size of over 100 meters, were removed manually from analysis. OAM images that are overlapping with each other were also removed since in this case edited OSM features cannot be unambiguously allocated to one specific image, especially if the images were taken within a short period of time. In addition, all OAM contributions that lacked upload date or bounding box coordinates were excluded. These filtering steps are summarized in the left sequence in Figure 2, which results in a set of OAM images used for Task 1.

Figure 2. Data processing workflow.

These filtering steps reduced the workable data set to 5,067 OAM images. The extent of these filtered images is used to filter OSM changesets whose nodes and ways will be analyzed in tasks 2 and 3. Each OAM image contains three timestamps, which are acquisition start, acquisition end, and image upload. This study uses the third timestamp for all three tasks since this is when an OAM image becomes available within an OSM editor.

3.1.2. OSM

The full OSM history dump file was downloaded in PBF format. The file includes the editing history compartmentalized into changesets up to 1 November 2021. The file size was over 103 GB and included 113,184,258 changesets. For further analysis, only OSM changesets with a coverage area smaller than 225 km² were retained, which, due to its compactness, resulted in more focused edited features that tended to be associated with OAM images. ChangesetMD, which is a python-based XML parser, was used to import the OSM changeset data into a PostgreSQL database. Each changeset comes with information including id, timestamp, bounding box, number of changes, and tags, as well as contributor information consisting of id and username. The different versions of a geographic feature (node, way) or relation can then be extracted by specifying the changeset id in an API call, such as for changeset ID 17871446.¹ In this study, only changes in nodes and ways, but not relations, were considered. A node carries at least nine values (id, visible, version, changeset id, timestamp, user, user id, latitude, and longitude). Ways have similar attributes, except that they do not carry latitude and longitude but a list of node ids instead. The “visible” field indicates if a node or way exists or has been removed, and “version” is an incremental counter that starts with 1 in case of a newly created feature.

Each feature edit in the changeset can be classified into one of three activities based on their “version” and “visible” attribute values, namely create, modify, and delete. That is, newly created features carry version 1, deleted features have a false visible value, and modified features have a version larger than one and a true visible value. As opposed to feature creation and deletion, feature modification can be further sub-divided into different operations (Ahmouda, Hochmair, and Cvetojevic 2018; Rehrl et al. 2013) for nodes and ways (Table 1). A geometry change of a node means a shift of its location, whereas for a way it means a change in the list of nodes, or the coordinates of a node associated with a way. Tags can be added, modified, or deleted both for nodes and ways. Determining the nature of feature modification requires extracting two consecutive versions of a feature, followed by the comparison of some of their key-value pairs. The history of a feature can be obtained through the API based on a URL which requires provision of a node or way ID, such as for way 45209144.² The API returns feature details of each version in XML format. For task 2, the creation, modification, and deletion of OSM nodes and ways was analyzed for features that fell within the bounding box of an OAM image and for which changes took place between 2 weeks prior and 2 weeks after OAM image upload. For task 3, the weekly number of OSM mappers who edited features in the first 26 weeks since an OAM image upload were obtained for any OSM changeset that carried an “OpenAerialMap” source tag (for JOSM users) or an “OpenAerialMap” imagery_used tag (for iD users).

Table 1. Classification of modification operations on OSM nodes and ways considered for the analysis of contribution patterns.

Feature	Field	Operation
Node	Geometry	Change coordinates
	Tag	Add tag
		Modify tag
		Delete tag
Way	Geometry	Add node to way
		Modify node on way
		Delete node on way
	Tag	Add tag
		Modify tag
		Delete tag

3.2. Data preparation

To identify OSM feature edits for task 2 and 3, a spatial overlay function was applied which returns changesets intersecting with the set of selected OAM images. Among these intersecting changesets those which were submitted within up to 2 weeks before and after an OAM image upload were retained for further analysis. For task 3, changesets with “OpenAerialMap” source or image tags were extracted, independent of their provision date. These changeset filters related to task 2 and 3 resulted in 29,418 changesets for which edited POIs and ways were to be analyzed. Node extraction from these changesets underwent two filtering stages. First, nodes had to fall within bounding boxes of selected OAM images. Second, only nodes with tags were included in the statistical analysis of nodes, since they represent real-world features, such as traffic signs, fountains, or mailboxes, i.e. a point of interest (POI). As opposed to this, untagged nodes are typically part of ways and were therefore only counted in connection with way statistics. This approach resulted in 1,150,585 nodes, which include 18,940 POIs. The remaining nodes are part of 147,603 ways. In some cases, ways intersect with more than one OAM image. In this case, images were only considered if their upload time stamp differed by less than 24 hours so that the effect of image provision on OSM edits could be accurately determined. This filter reduced the number of ways considered in the analysis to 147,552.

Figure 2 illustrates the workflow of processing POIs and ways for tasks 2 and 3. The spatial filter applied to changeset selection (maximum area, required to overlap with an OAM image) is the same for both tasks. However, for task 2, an addition temporal filter, and for task 3, and additional tag filter is applied. After this, for both tasks nodes inside OAM images are extracted to identify POIs (i.e. nodes with attributes) and ways (having at least one node inside an OAM image extent) for further analysis.

Figure 3 provides a map with filtered OAM image locations and the corresponding intersecting changesets used for task 2 of the analysis.

Figure 3. World map of selected OAM image bounding boxes based on a download from May 24, 2021 (red) and bounding boxes of OSM changesets in the ± two-week period around OAM image downloads overlapping with OAM image extents (green).

3.3. Analysis methods

3.3.1. OpenAerialMap contribution patterns

A power law approximation can be used to model the distribution of OAM image uploads per user and the number of days between a user’s first and last day of OAM image upload. The power-law function can be formulated as

(1) $\mathit{P}\left(\mathit{\delta }\right)\mathit{\hspace{0.17em}}=\mathit{\delta }\exp \left(-\mathit{\beta }\right)$(1)

where δ is the contribution variable of interest and β is the exponent. For the analysis of the time span of a user’s OAM contributions, only users whose first contribution to OAM dates back at least a year before the OSM data extraction date will be considered.

3.3.2. OSM editing patterns

Most OSM-related analyzes relate to task 2 which compares contribution patterns before and after OAM image upload. Various aspects of edits will be discussed in the comparison. This includes an analysis of how the nature of edits on POIs and ways changes with respect to edit types (create, modify, delete), location (continents), OSM primary feature types, and number of edited features per user.

In the downloaded OAM image set, upload dates range between 7 October 2015, and 21 May 2021. Therefore, edited OSM features considered for task 2 were extracted between 23 September 2015, and 4 June 2021. The two-week pre-post analysis of OAM-related OSM editing patterns includes a comparison of the number of edited POIs and ways between these different periods, where a Mann-Whitney U test will be used to determine if a change in node and way contributions is statistically significant. Edited features within these two-week periods were also assigned to the closest continent.

Based on these data, a chi-square test of independence will be conducted to determine if there is a statistical association between the proportion of POIs and ways that are contributed to a continent, and the two-week period of analysis (i.e. before and after OAM image upload). OSM currently offers 29 primary feature types that newly added features can be assigned to by tagging the feature with a corresponding key-value combination. The proportion of created nodes and ways falling into the different primary feature types will be compared between the 2 weeks before and after an OAM image upload.

For task 3, the weekly number of OSM mappers who edited data are extracted for the 26 weeks following an image upload. For this task, only edited OSM features associated with OAM images that were created at least 26 weeks before the OSM changeset download were considered.

4. Results

4.1. OpenAerialMap contributions

Figure 4 shows the annual number of uploaded OAM images between 2016 and 2020, separated by image acquisition platform. Contributions by drone show an upwards trend over the years. Satellite contributions are less than 400 annually except for 2017. DigitalGlobe plays a major role in satellite-based image contributions, with a 27.8% share in 2016, a 92.7% share in 2017, a 72.1% share in 2018, and an 86.7% share in 2019. Analysis of the peak in contributions in 2017 shows that 1215 DigitalGlobe images were uploaded to OAM within 3 days, i.e. September 1, 11, and 12, for areas in the Caribbean mostly affected by Hurricane Irma. The hurricane made its first landfall on Barbuda on September 6, followed by landfall on the British Virgin Islands, Little Inagua Island on the Bahamas, the archipelago of Turks and Caicos, and on Cayo Romano island off of the northern coast of Cuba (BBC 2017). This data upload has been part of the maxar open data program which releases open data for sudden onsets on major crises events (maxar 2022). The peak for aircraft contributions in 2018 is a result of a concentrated upload of 1215 images covering Denver through the Denver Regional Council of Governments (DRCOG) in the week between October 18 and 25 of that year.

Figure 4. Annual number of uploaded OAM images between 2016 and 2020, separated by image acquisition platform.

Figure 5(a) plots the total annual number of OAM images uploaded between 2015 and 2020, separated into analyzed images and excluded (due to overlap, low resolution) images. The peak in 2018 can be ascribed to the bulk image upload through the DRCOG. Figure 5(b) provides the annual number of OAM contributors associated with these images, which shows a steady growth of the OAM community over the years.

Figure 5. Number of annual OAM contributions (a) and OAM contributors (b).

Corresponding to the analysis for task 1, Figure 6(a) illustrates the power law approximation of the total number of OAM contributions per user in a log-log plot. The majority of OAM image contributors (49.7%) uploaded one image only, whereas the highest number of contributions made by one user (the DRCOG) was 3172. Figure 6(b) shows the power law approximation of the difference in days (in log of 100’s) between the first and last contribution date for users who contributed at least two images and who have been registered to OAM for at least 1 year. Results show that among this group, 32.7% of users uploaded OAM images only on a single day, whereas 18.2% of users contributed for over a year after their first contribution.

Figure 6. Power law approximation of OAM contributions (a) and time span of contributions for OAM contributors (b).

4.2. OSM edits

4.2.1. Comparison between pre and post OAM image upload

Related to the analysis for task 2, the total number of features that were edited (create, modify, delete) within the 2 weeks before and after OAM image upload inside OAM image bounding boxes was 17,939 for POIs and 103,167 for ways (Table 2). The majority of edited features can be found within the 2 weeks after an OAM image upload both for POIs (73.6%) and ways (66.4%). This corresponds to a 279% increase in edits between before and after image upload (within ± 2 weeks) for POIs, and an 198% increase for ways. When looking at newly created features only, the number of POIs added to OSM increased by 305%, and the number of ways added increased by 96% within the same time period. Table 2 further reveals that the feature number associated with each editing activity (create, modify delete) is higher in the 2 weeks after OAM image upload than before. The proportion of edited features in the post-period ranges between 58.5% (delete POI) and 80.2% (create POI) with proportions for ways in-between. These numbers suggest that provision of OAM images triggers OSM map updates in the underlying areas. The bottom row in the upper and lower table halves show the number of edited POIs and ways beyond 2 weeks that were recorded in OAM-tagged changesets, which gives insight into the long-term effects of OAM image provision. Numbers in parentheses express the average number of POIs and ways edited (i.e. added, modified, deleted) by an OSM mapper in the 2 weeks before and after OAM image upload. The increase in this number between two time periods is expressed in % in brackets below. The high increase in number of modified POIs (143.6%) and ways (110.1%) compared to POIs and ways that were added (91.3% or less) or deleted (26.6% or less) suggests that OAM facilitates data quality improvement specifically for existing, i.e. already mapped, OSM data. Percentage numbers in the “Total” columns demonstrate heightened activity levels of individual mappers following the uploading of OAM images by handling on average 93.9% more POIs and 54.2% of ways after than before OAM upload.

Table 2. Number of new, modified, and deleted OSM POIs and ways in the 2 weeks before and after OAM image upload, and beyond these two periods. Numbers in parentheses show the average number of features edited per user, and numbers in brackets show its increase in the 2 weeks after OAM image provision compared to before.

OAM feature	Period	New	Modify	Delete	Total
POI	2 weeks before OAM	942 (21.9)	1666 (22.5)	2124 (17.6)	4732 (27.8)
	2 weeks after OAM	3812 (41.9) [+91.3%]	6407 (54.8) [+143.6%]	2988 (18.7) [+2.3%]	13207 (53.9) [+93.9%]
	Additional with OAM tag	637	272	92	1001
Way	2 weeks before OAM	26582 (129.7)	4417 (11.9)	3658 (15.8)	34657 (74.2)
	2 weeks after OAM	52224 (131.2) [+1.2%]	10927 (25.0) [+110.1%]	5359 (20.0) [+26.6%]	68510 (114.4) [+54.2%]
	Additional with OAM tag	36047	3778	4560	44385

Table 3 provides a count of tag and geometry-related modification operations on OSM POIs and ways 2 weeks before and after OAM image upload. Whereas Table 3 counts each individual operation, Table 2 counts the number of edited POIs and ways, some of which may have undergone multiple edits. Results in Table 3 show that in each operation category the number of edits is larger after OAM image provision, again suggesting that OAM images are used for OSM updates. The last row in Table 3 reports the percentage of modification operations related to tags (add, delete, modify) among all modification operations. The annotation of a geographical feature in OSM requires the digitization of the object and the attribution of the tag (Vargas-Munoz et al. 2020). Whereas the outline of an object to be digitized is generally visible on an aerial image, semantic object information, such as name or building type, is much less accessible. This means that the tag-related object modification operations on OSM edits that are based on aerial image sources alone are limited. Since Table 3 reveals only minor changes in the percentage of tag-related operations on OSM POI and way features after OAM image upload compared to before (10.6% vs. 10.1% for POIs; 17.7%. vs. 19.0% for ways), this suggests that Earth image data were also used as a primary data source for OSM object modification operations before OAM image upload, and that therefore digitization from OAM images would not lead to a decline in the completeness of semantic tag information.

Table 3. Number of modification operations on OSM POIs and ways in the 2 weeks before and after OAM image upload.

Detected operation	POI		Way
	Before	After	Before	After
Add tag	87	384	505	1425
Delete tag	49	179	445	1163
Modify tag	39	41	309	1254
Add node to way	–	–	1712	5771
Delete node from way	–	–	1735	5490
Change coordinates	1566	5111	1930	6605
% tag operations	10.1	10.6	19.0	17.7

Figure 7 shows the daily number of edited (create, modify, and delete) POIs (Figure 7(a)) and ways (Figure 7(b)) for the 2 weeks before and after an OAM image upload, thus providing a refined visualization of edit statistics from Table 2. The charts reveal peaks for the number of new features created within the first 24 hours after OAM image upload, demonstrating that the provision of aerial images is primarily used for rapid mapping of objects previously missing from OSM. Overall, the daily number of edited features is higher in the 2 weeks after image upload than before.

Figure 7. Daily number of edited POIs (a) and ways (b) within 2 weeks before and after OAM image upload.

4.2.2. Long-term effects of OAM image upload

Related to task 3, Figure 8 charts the weekly number of OSM mappers who contributed an OAM-tagged changeset within 26 weeks after OAM image upload. Edited POIs and ways in these changesets were found up to 164 weeks after an OAM image upload. Both charts show that the majority of mappers is active in the first week after an image upload. The charts also show that the majority of OSM mappers contribute in only one specific week after OAM image provision. The small spike in way edits for week 7 (Figure 8(b)) seems to be a random coincidence, since the edits relate to different OAM images uploaded for different locations around the world.

Figure 8. Weekly number of OSM mappers contributing at least one POI (a) or way (b) in a changeset with an OAM tag within 26 weeks after an OAM image upload.

4.2.3. Spatial patterns

OAM images were contributed to seven continents; Antarctica revealed only two images and no feature edits. The spider chart in Figure 9 shows for the remaining continents the proportion of the analyzed OAM images shared worldwide and the proportion of worldwide OSM edited POIs and ways in the 2 weeks before and after an OAM image upload. North America registers most shared OAM images (76.2%), followed by Europe (8.5%) and South America (6.9%). The proportions of OSM edited POIs (51.8%) and ways (51.0%) in North America are smaller than that of its OAM contribution share. In all other continents, the proportion of edited POIs or ways is higher than their global share of OAM contributions (except for the proportion of edited POIs in Asia and Oceania).

Figure 9. Share of OAM image contributions and edited OSM POIs and ways across continents (Antarctica not shown).

Figure 10 shows for the same six continents the proportion of edited OSM POIs (Figure 10(a)) and ways (Figure 10(b)) in the 2 weeks before and after an OAM image upload, together with the absolute number of edited features for each continent for these 4 weeks. All continents except for Oceania (which includes Australasia, Melanesia, Micronesia, and Polynesia) show a higher proportion of edited objects after the OAM image upload. This is supported through a chi-square test of independence which showed that the proportion of OSM features edited on continents is significantly associated with the time period of analysis (i.e. 2 weeks before and after OAM image upload) for POIs, X² (5, N = 6) = 1889.47, p < .0001, and ways, X² (5, N = 6) = 19,574.80, p < .0001.

Figure 10. Number of edited POIs (a) and ways (b) across continents (Antarctica not shown).

4.2.4. OSM primary features

Figure 11 plots the proportion of primary feature types for newly created POIs and ways for the 2 weeks before and after OAM image upload, where numeric labels show the total number of features added to each category. Four primary feature types (Geological, Healthcare, Route, and Telecom) were not mentioned in the newly created nodes and ways. Moreover, three more types (besides the previously mentioned four) were not used for newly created nodes, which are Boundary, Landuse, and Water. Most primary feature types experience a higher contribution proportion after OAM image upload. Exceptions are primarily those types with a small sample size, e.g. Craft, Emergency, Public Transport, and Sport nodes or Craft for ways.

Figure 11. Proportion of new POIs (a) and ways (b) tagged with different primary feature keys within 2 weeks before (blue) and after (orange) of OAM image upload.

4.2.5. Individual editing behavior

The box plots in Figure 12 show the distribution of edited features per OSM mapper for POIs (Figure 12(a)) and ways (Figure 12(b)) in the 2 weeks before and after OAM image upload. The number of edited POIs per user in the 2 weeks following OAM image upload (Mdn = 1) was higher than in the 2 weeks before OAM image upload (Mdn = 0). A Mann-Whitney test indicated that this difference was statistically significant, W(N_after = 355, N_before = 355) = 77850, p < 1.542e-08. Similarly, the number of edited ways per user in the 2 weeks following OAM image upload (Mdn = 2) was higher than in the 2 weeks before OAM image upload (Mdn = 0). A Mann-Whitney test indicated that this difference was statistically significant, W(N_after = 939, N_before = 939) = 519891, p < 2.206e-12.

Figure 12. Edited POIs (a) and ways (b) per mapper before and after OAM image upload.

Log-log plots in Figure 13 show that the distribution of edited POIs and ways per user before and after OAM upload can be approximated by a power law function. This pattern reveals participation inequality, which is known from other crowdsourcing platforms (Tang and Liu 2016). Smaller absolute β exponent values in Figure 13(b,d) compared to Figure 13(a,c), respectively, indicate longer tailed distributions and hence a larger number of users editing more features after OAM upload than before. For example, 34.7% of the users edited POIs only once before OAM image upload, which dropped to 30.6% after OAM image upload. Corresponding numbers for ways are 23.3% and 15.2%, respectively.

Figure 13. Power law approximations of edited POIs per mapper before (a) and after (b) OAM image upload; and of edited ways per mapper before (c) and after (d) OAM image upload.

5. Discussion

This study analyzed various aspects of OSM mapping behavior and showed that OAM Earth images provide a well-utilized shared data source for OSM map updates. Previous work has already demonstrated successful integration of airborne imagery (e.g. Landsat images) and OSM data, for various purposes, such as land use/land cover (LULC) mapping (B. Johnson and Iizuka 2016). Satellite images have also been used for the validation of OSM data quality (Xie et al. 2019). A previous case study, which was conducted for three Caribbean islands, used several remote sensing measurements with emanated classification procedures to assess the completeness of OSM spatial data as a preparation for future natural hazards (Goldblatt, Jones, and Mannix 2020). Despite these efforts, the role OAM-based images for OSM feature mapping has not been previously discussed, which is therefore a novel aspect of the presented research. One distinct aspect revealed in this study is the discrepancy between OAM and OSM editing activity between continents, where North America had by far the largest share of OAM images, but only about half of OAM-related POIs and ways edited. As opposed to this, edited POIs in Europe had a higher proportion (24.2%) than OAM images (8.5%) globally. This may be ascribed to the long-standing tradition of OSM mapping especially in Europe (Neis and Zielstra 2014) where the project was founded.

Previous studies examined OSM newcomer retention rates after organized mapping events or mapping tasks (Dittus, Quattrone, and Capra 2017; Juhász and Hochmair 2018b), showing that mapathons work more effectively than data import tasks to retain newcomers. Whereas our study has not directly examined retention of newcomers, it shows that OAM image addition comes with long-term edits of OSM features and is partially contributed by recurring users. OAM is listed as an important image source in the temporal plot of the evolution of humanitarian mapping in OSM besides other imagery sources, such as Mapillary street-level imagery and DigitalGlobe (now Maxar Technologies) imagery (Herfort et al. 2021). A recent study evaluated various forecasting models for OSM data contribution activities, including time-series decomposition, temporal linear correlation, vector autoregression, and random forest methods (Novack, Vorbeck, and Zipf 2024). The authors note that events with long-term influence on mapping activities (which includes the free provision of online imagery data for feature tracing, such as OAM) do generally not hamper the applicability of such models. Time-series analysis was also used to identify variations in seasonal, random, and trend components of OSM registration rates, finding largest bursts of registration to be mostly correlated with external factors, such as media broadcasts, mapping events, and online communities changing their background maps from Google to OSM, or technological updates such as the release of a new OSM editor (Bégin, Devillers, and Roche 2017a). Survival analysis and time-series analysis showed that withdrawals from OSM were based on both external events, such as Ordnance Survey beginning to release data for free use, and internal processes, such as license changes (Bégin, Devillers, and Roche 2017b).

Despite OAM’s usability as a data source for a more complete OSM dataset, it should be noted that tracing of features from Earth image data comes with its own sets of challenges and thus may lead to limited OSM data quality of digitized objects both in terms of geometric and attribute information (Anderson, Sarkar, and Palen 2019). For example, the digitizing of buildings may result in incorrectly representing a group of adjacent buildings as one building in OSM (Moradi, Roche, and Mostafavi 2023), or small buildings may be occluded by trees which creates difficulty in digitizing them (Fan et al. 2014). In terms of semantic attribute information object characteristics, such as building type or road surface/name may not be readily discernable from aerial images, but may instead need to be obtained through field surveys or the use of street-level photographs (Kang et al. 2018). The same is true for OSM relations, such as turn restrictions.

One limitation of the study is that the OAM image ID is not specified in the changeset tag, which led us to remove intersecting images to avoid confusion as to which OAM image would trigger editing of OSM features. Furthermore, it can be assumed that not all changesets that use OAM images are OAM-tagged, which might lead to an underestimation of edited features in the long-term analysis (task 3). Also, whereas OAM image provision was clearly associated with increased OSM editing activities in this study, it is unclear if users who frequently edited OSM features based on OAM images were also involved in OAM image upload processes, or more specially, if an image uploaded to OAM and the digitization of OSM objects within the OAM image area are performed by the same user. This is because users use different methods to identify themselves in the respective user profiles of the two platforms. That is, in OAM users provide their (or their organization’s) real name, whereas in OSM users choose an arbitrary username, which makes matching these two identities difficult. A possible workaround could be to compare the similarity (e.g. using overlap indices) of activity spaces based on user contributions between users from both platforms to identify matching users (Juhász and Hochmair 2018a). Even if such an approach led to a sufficient number of matches, information about users contributing to both OAM and OSM would not necessarily affect findings of this study, but provide some insight as to how well the OAM platform is already known among OSM users, and if there is need to further promote this platform in the OSM user community.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data of aerial images on the OAM platform and OSM features used for the analysis are available as csv files in figshare at: https://doi.org/10.6084/m9.figshare.20486289.v3, DOI 10.6084/m9.figshare.20486289

Additional information

Funding

This research received no external funding.

Notes on contributors

Ammar Mandourah

Ammar Mandourah received his Ph.D. in Geomatics from the University of Florida and is a Professor of Geomatics at King Abdulaziz University. His research evolves around the geospatial analysis on user contributions in different social media platforms.

Hartwig H. Hochmair

Hartwig Hochmair received his Ph.D. from the Technical University of Vienna, Austria, and is a Professor of Geomatics at the University of Florida. His research focuses on the spatial analysis of geographic information to address key issues of sustainable transportation, the usability of crowd-sourced geodata, and the spread of invasive species in South Florida.

Notes

1. https://www.openstreetmap.org/api/0.6/changeset/17871446/download.

2. https://www.openstreetmap.org/api/0.6/way/45209144/history.