Analyzing the violation of drone regulations in three VGI drone portals across the US, the UK, and France

ABSTRACT

Drone technology opens the door to major changes and opportunities in our society. But this technology, like many others, needs to be administered and regulated to prevent potential harm to the public. Therefore, national and local governments around the world established regulations for operating drones, which bans drone use from specific locations or limits their operation to qualified drone pilots only. This study reviews the types of restrictions on drone use that are specified in federal drone regulations for the US, the UK, and France, and in state regulations for the US. The study also maps restricted areas and assesses compliance with these regulations by analyzing the spatial contribution patterns to three crowd-sourced drone portals, namely SkyPixel, Flickr, and DroneSpot, relative to restricted areas. The analysis is performed both at the national level and at the state/regional level within each of the three countries, where statistical tests are conducted to compare compliance rates between the three drone portals. This study provides new insight into drone users’ awareness of and compliance with drone regulations. This can help governments to tailor information campaigns for increased awareness of drone regulations among drone users and to determine where increased control and enforcement of drone regulations is necessary.

KEYWORDS:

• User-generated content
• drone
• regulation
• SkyPixel
• DroneSpot
• flickr
• social media
• point pattern

1. Introduction

Drones are one of many emerging technologies that can legitimately be both celebrated and feared (Snead and Seibler 2016). Despite the development of drone technology and its ability to assist with challenging tasks, such as medical services, package delivery, Search and rescue operations, precision agriculture, or industrial inspection (Radoglou-Grammatikis et al. 2020; Tezza and Andujar 2019; Kim and Moon 2019; McRae et al. 2019; Angurala et al. 2020), it is subject to various laws and regulations worldwide that limit its usage in order to reduce risks and preserve privacy. That is, governments around the world introduced drone regulations and banned drone flights over selected facilities and locations in support of the national security and people’s safety (Clarke 2014; Farber 2014). Also, since drone technology is developing rapidly, there is the need to update drone regulations accordingly. For example, the US congress legislated the recreational use of drones in 2012. In 2015, the Federal Aviation Administration (FAA) mandated to register drones, including those for recreational use, due to the increasing number of drones and the inherent threat of incidents and the lack of collision insurance (Snead and Seibler 2016). Both a professional segment that leverages new Unmanned Aerial Vehicle (UAV) technology and a UAV hobbyist market have evolved over the years where consumers, also without prior knowledge of aviation, have been able to acquire powerful UAVs that can fly several thousand feet high (Pyrgies 2019). An online survey that aimed to identify factors that concern the public regarding the increasing number of drone users in the US revealed equal concern between privacy and safety (Zwickle, Farber, and Hamm 2019). Among various proposed regulations for drones mentioned in that survey, most respondents supported regulations that limit the operation of drones in some way, e.g. prohibit flying drones over private properties. However, the respondents were less inclined toward more severe policies, which ban drone use either temporarily, e.g. just at night, or completely.

The main objectives of our study are (1) to map no-fly zones for drones and (2) to assess compliance of contribution to three drone image/video sharing portals (SkyPixel, DroneSpot, and Flickr) with nationally and regionally issued restrictions. Using drone positions shared on crowd-sourced photo portals gives access to a new source of spatio-temporal data, which can be used to analyze violations of drone regulations even before any incidents occur. Crowd-sourcing portals are only one source of geotagged data that is actively shared by the Web community, which is often referred to as Volunteered Geographic Information (VGI) (Goodchild 2007). Whereas numerous studies analyze the distribution of crowd-sourced photos, which were taken at the ground level (e.g. by smartphones), this study analyzes patterns of VGI-based pictures taken from above-ground through drones.

Three countries (US, UK, and France) were chosen for this analysis given their high volumes of image contributions to the three selected drone portals, their relatively transparent drone regulation systems, and the availability of mapped restriction areas at the federal level. The unification of regulations in the EU in 2020 will in the future also allow to compare compliance more easily within the framework across a larger number of EU countries. All drone contributions located inside restricted areas during the time period when regulations were active are considered to be in violation of drone regulations. This study provides insight into the user behavior of drone pilots that contribute to any of the three portals. This knowledge gained can be used to develop targeted information campaigns about drone regulations through federal or local administrations, especially in regions of high violation rates.

2. Related work

This section summarizes related studies that review drone-related regulatory frameworks (Herrmann 2017; Stöcker et al. 2017), safety concerns (Ghosh et al. 2021; Huttunen 2019; Solodov et al. 2018; Zhi et al. 2020), and incident analyses (Pyrgies 2019; Truong and Choi 2020). It also reviews previous studies that analyze contribution patterns to crowd-sourced drone photo portals (Mandourah and Hochmair 2021; Hochmair and Zielstra 2015).

Drones, despite their wide range of potential applications in everyday life, can also pose a significant threat to citizens and infrastructure. Harmful behavior may be conducted deliberately, e.g. through acts of sabotage, carrying explosives, spreading chemicals, smuggling of narcotics, and delivering mobile phones into prisons (Solodov et al. 2018). Harm can also be caused unintentionally, such as through improper use (e.g. flying over crowds) or technical malfunctions, which can lead to accidents and crashes (Zhi et al. 2020). Tethered UAVs are provided with a stable power supply through a connection to the ground. This type of UAV can therefore overcome some technical limitations of battery-driven UAVs, such as limited power resources, attenuated signals, and path loss (Saif et al. 2021). In smart city environments, UAV can collect data from Internet of Things (IoT) devices on the ground, and various intelligent techniques, such as Artificial Neural Network (ANN) based solutions have been employed to accurately monitor and suitably adjust signal strength to maintain the network connectivity, to provide the desired quality of service, to identify the drone coverage area and to optimize the location a drone in order to gather more data from IoT framework applications (Alsamhi et al. 2021). Drones themselves or their payload can also be subject to unlawful interference, leading to possible security threats (Huttunen 2019). Another common concern is that drones can invade the privacy of the public since they possess a wide range of sensing techniques, which can be used for surveillance tasks with better agility and accuracy than mounting static cameras (Ghosh et al. 2021).

A study of 139 drone incidents on or near airports in the US, the UK, and other countries analyzed risk factors associated with these incidents and proposes mitigation measures (e.g. prevention, deterrence, denial, detection, and neutralization) to bring risks to an acceptable level (Pyrgies 2019). The study showed that UAV incidents occurred more often at higher altitudes and farther from airports than expected and that they happened primarily in the landing phase, rather than the departure phase, of an airplane. Another study used seven machine learning algorithms to develop predictive models for UAV violation incidents in the National Airspace System (NAS), based on FAA’s unmanned aircraft systems (UAS) sighting reports between October 2016 and September 2017 (Truong and Choi 2020). Analyzing three UAV violation types (flying above 400 feet, flying within 5 miles from an airport, and flying in a restricted airspace), it was found that location, distance to the airport, state, UAV altitude, airport type, and aircraft type are the most influential factors to predict the UAV violation type. To address the increasing danger of incidents involving UAVs near airports and other critical infrastructure, another study reviews counter-drone technologies to prevent such incidents, including RF jamming, GPS jamming, UAV Net Capturing, high power microwave, or laser guns (Lykou, Moustakas, and Gritzalis 2020).

To safely incorporate UAVs into the national airspace, many countries and regions around the world task agencies to issue aviation laws. Examples include the FAA for the US or the European Union Aviation Safety Agency (EASA) for the European Union (EU). A national review and comparison of UAV regulations between the US and Europe and their impact on the construction industry is provided elsewhere (Herrmann 2017). The review finds that, although UAV regulations are categorized differently in both frameworks, e.g. based on weight, maximum flight height, or operator experience and certification, both frameworks limit the weight of the UAV to approximately 55 pounds and prohibit the use of the UAV around people who are not involved in its operation. Another study reviews 19 national drone regulations, including those of the US, the UK, China, Germany, and Austria, from the perspectives of past, present, and future along criteria such as applicability (scope), technical prerequisites (e.g. mandatory instruments), operational limitations (restrictions), application procedure, qualification of pilot, and ethical constraints (e.g. privacy and data protection regulations) (Stöcker et al. 2017). Almost all countries incorporate a maximum take-off mass, whereas only two countries introduce a minimum threshold in their regulations. UAVs that are heavier than 150 kg are usually not regulated by national aviation authorities but by international bodies like the EASA in Europe and tend to be regulated similarly as manned aircrafts. All analyzed countries except for China and Nigeria allow only low-altitude flights in the range between 90 m (Canada) and 152 m (Colombia) above ground level. An overview of first releases of UAV regulation shows, for example, that the UK and Australia were the first nations that introduced regulations in 2002. Despite commonalities, such as mandatory portal registration, obligatory insurance coverage and pilot licensing procedures, a distinct heterogeneity of national regulations exists, e.g. in terms of minimum distance requirements around people or airports, privacy policies, and insurance coverage.

Governments aim to educate drone users and make drone regulations including no-fly zones accessible to the public in their administrative jurisdictions. Austria, for example, provides a web map application, which allows for safe drone mission planning. For this purpose, legal definitions of four types of areas of operation (i.e. undeveloped, uninhabited, populated, and densely populated) which are provided in narrative form by authorities were translated into spatially delineated operational zones (Paulus et al. 2017). A probabilistic approach was used to generate a risk map of accidents with unmanned aircraft. It considers several layers including population density, sheltering factor, no-fly zones, and obstacles (Primatesta, Rizzo, and Cour-Harbo 2020). The paper mentions different types of no-fly zones, including those determined by national regulation agencies (e.g. airports, military areas), nature-sensitive areas (e.g. national parks), or security zones (e.g. areas with public events).

The proliferation of drones among recreational and hobby users led to a rapid increase in drone photo and video sharing portals over the past few years. Analyzing the spatial distribution of user-generated content of this kind allows to identify hot spots and factors associated with an increase or decrease of user contribution activities, including proximity to airports or coastlines, population density, or presence of water bodies (Hochmair and Zielstra 2015; Mandourah and Hochmair 2021). These studies demonstrate that the number of drone contributions on social media portals is rapidly increasing, which allows analysts to better understand user behavior of drone pilots and contribution patterns to the cloud.

3. Materials and methods

3.1. Drone contributions

This section describes how point data from crowd-sourced photo and video contributions were extracted and processed for each of the three analyzed portals. No scripts were necessary for data access. However, Python scripts were used to handle downloaded data, e.g. for conversion between JSON file types and GeoJSON format.

3.1.1. SkyPixel

SkyPixel is owned by DJI, which was founded in Shenzhen, China, in 2006. DJI founded SkyPixel in November 2014 as a global community application for sharing aerial photos and videos. SkyPixel contributions were collected for the three analyzed countries, resulting in photos uploaded between 10 October 2014 and 28 September 2020. Upon upload, users can specify the content type (photo, video, or 360-degree photos). SkyPixel detects all geo-place names (such as country, city, landmark) that match the positional information stored in the media EXIF file, which are then suggested as tags to be chosen by the user. The locations will not be specified if users leave the location window blank. These geo-place tags can be used to extract desired media data when included in a SkyPixel API query. Results are limited to 20 pages with a maximum of 256 drone contributions per page, resulting in a maximum of 5120 hits per tag in total. To obtain all contributions from an area (e.g. country) which includes more than 5120 photos, the request needs to be broken down into smaller geographic sub-regions, such as states or cities, and search results need then be subsequently combined. The SkyPixel website provides a search window that uses the user input to find matched contribution keywords, including geographic terms. The API query to the SkyPixel server allows to specify various search parameters, including geo-place name, content type, number of contributions, and the page number (offset). For example, the image information for the latest 20 contributions in Miami can be accessed through the following query URL: https://www.skypixel.com/api/v2/searches/photos?lang=en-US&platform=web&device=desktop&keyword=miami&limit=20&offset=0. The browser receives a server response in JSON format, which contains all information displayed on the media web page including content type, title, description, date, drone model, and contribution coordinates. The above-described procedures resulted in 17,382 contributions for the US, 3238 for the UK, and 2621 for France.

3.1.2. Flickr

Flickr is a photo and video sharing portal for contributions from both ground and above-ground levels, which does not focus specifically on drone media. However, it gives users the ability to submit their photos and videos to Flickr groups that host contributions related to a specific topic. Based on the description (needs to be drone-related) and size (at least 1500 contributions) of Flickr groups, we identified 21 drone groups, which contained a total of over 165,000 worldwide drone photos and videos on the day of data download (3 September 2019). Flickr contributions to groups contain Photo-ID, title, owner, Flickr-group name, date taken, URL, latitude, and longitude. The data were downloaded worldwide through the Flickr API (Juhász, Rousell, and Arsanjani 2016). Since numerous images and videos were assigned to multiple groups, duplicates were removed, resulting in about 26,000 unique, geotagged contributions. Out of these, 6273 contributions came from the US, 3162 from the UK, and 602 from France. Among all Flickr contributions used in this study, drone videos constitute a mere 1.3% share, and photos a 98.7% share. Flickr uses positional information stored in media EXIF files for geocoding, but it also allows to pinpoint the geographic position of uploaded media on a world map (Zielstra and Hochmair 2013). That study showed that Panoramio photos had a better positional accuracy than Flickr photos for ground-level images. We expect therefore also differences between contribution patterns, e.g. with regards to violation rates, from images uploaded to different drone portals.

3.1.3. DroneSpot

This portal was founded in France in December 2018 as a video-sharing application, focusing on collecting drone contributions from several European countries and other parts of the globe. DroneSpot informs drone pilots in selected countries about the geographic locations of drone regulations in an interactive map. This map displays restricted areas and highlights recreational sites for drone pilots, which are clear of restrictions. DroneSpot contributions were extracted on 29 June 2020, which resulted in 3623 videos, out of which 2658 were shared in France. To contribute a video, it must be first uploaded to a video-sharing portal, e.g. YouTube. Then, in DroneSpot, the geographic location of the contribution can be specified by pinpointing the video position on the world map. Contributions located in flight restriction areas will not be published. Data access can be achieved through reading an xhr file with coordinates and attributes which is returned as part of the server response when the world map is requested through a browser. Media links and additional user-provided information (e.g. video titles, geographic coordinates) will be included in the response to a php request where the video ID needs to be included as a query parameter. For example, the details for video with ID 297349, such as upload date or geographic coordinates, can be accessed through https://www.drone-spot.tech/fiche-spot.aspx?id=297349. Although DroneSpot intends to publish only photos located outside restricted areas, its data were included in our study to test the effectiveness of this restrictive data handling and to be able to compare the resulting numbers to those of the other two portals.

3.2. Restricted areas

Restricted areas for civil drone users, i.e. no-fly zones, including prohibited facilities, were compiled as GIS layers from different governmental websites at the federal and state level, as described further below. The mentioned resources will allow the research community to access data, which allows it to construct GIS flight restriction datasets on their own for further analysis as needed.

3.2.1. United States

The FAA is the official authority for civil aviation legislation in the US and regulates all aspects of civil aviation in the country. In 2012, the US President signed the FAA Modernization and Reform Act of 2012 which mandated FAA to develop a comprehensive plan to integrate UAS into the national airspace (Elias 2016). FAA was also tasked to issue regulations pertaining to small commercial drones and to develop standards for the operation and certification of unmanned aircraft operated by federal, state, or local governments. The drones were classified into three types based on weight, which are micro-UAS (4.4 pounds or less), small UAS (up to 55 pounds), and large UAS (more than 55 pounds). The rule for operating drones under 55 pounds in the NAS is regulated in FAA’s 14 CFR (Code of Federal Regulations) Part 107 (FAA 2020). However, FAA’s regulatory framework does not apply to the operation of small UAS if they are carried out strictly for recreation since there is a limited statutory exception that provides a basic set of requirements. That is, recreational flights are limited to flying within the visual line of sight, need to give way to and do not interfere with manned aircraft, need to fly at or below 400 feet in uncontrolled airspace, need to obtain prior authorization to fly in controlled airspace, need to comply with airspace restrictions and prohibitions, and need to follow safety guidelines of an FAA-recognized Community-Based Organization (FAA 2021f), which includes items such as not flying over unprotected people, moving vehicles, and occupied structures.

FAA restricted areas for drone flights are mapped and displayed on a UAS Facility Map (FAA 2021d), which maps temporary and permanent restrictions. This study focuses on permanent restrictions on facilities, which include airport airspaces and national security zones within the FAA regulations. Flying over national security zones for recreational purposes is not allowed at any flight height. As opposed to this, the allowable flight height in airport airspaces depends on the distance from the airport and ranges between 0 ft (no flights at all on or close to airports) to 400 ft (at least a few miles away from the airport to the limits of the controlled area). Since the crowd-sourced drone data lack height information, our analysis considers only the strictest airport airspace regulated zone, which forbids flights at any flight height. The two polygon datasets (national security zones and airport airspaces), which give a combined area of 73,000 km², were downloaded from FAA’s UAS Open data website (FAA 2019).

A subset of 21 US states have introduced additional drone regulations over the years around certain facilities, most of which were integrated into our analysis. Some regulations were excluded in our analysis due to the unclear semantics of certain facility types, such as freight transportation facilities. Seven general facility categories were considered in our study, namely correctional, gas, water, telecommunication, electrical, petroleum, and chemical (911 Security 2019). Since the facilities from the last two categories share many commonalities (e.g. restrictions issued in the same year) and locations, we grouped these facilities into one category. Table 1 lists the subtypes (e.g. refineries and terminals) for each of those categories that were used to map restriction locations in the different US states. Most GIS data representing these subtypes were obtained from the Homeland Infrastructure Foundation-Level Data Website (HIFLD 2021). Dam locations were the only data set downloaded from the Socioeconomic Data and Applications Center (SEDAC 2021). These features come either with point, polyline, or polygon geometries (Table 1). The area covered by a facility varies by category and subtype. For example, a petroleum refinery is typically larger than a natural gas storage, which itself covers a larger area than a telecommunication tower, and so on. Furthermore, the facility size may even vary between individual facilities that fall into the same subtype. To obtain realistic sizes of restricted objects, some edits to feature geometries were performed, such as changing the data type from point to polygon through digitizing the feature outline from ESRI’s aerial imagery layer or through buffering (Table 1).

Table 1. Categories of restricted facilities considered in state regulations and their subtypes.

Facility categories (in bold) with their subtypes	Geometry	Feature converted to
Petroleum & Chemical (AZ, FL, KY, NV, OK, OR, TN, TX)
Refineries	Point	Polygon by digitizing
Terminals	Point	Polygon by buffering 500 ft
Plants	Point	Polygon by digitizing
Transloading facilities	Point	Polygon by digitizing
Pipelines	Line
Telecommunication (AZ, AR, FL, OK, OR, TN, TX)
Land mobile commercial trans towers	Point
Land mobile broadcast towers	Point
FM trans towers	Point
BRS EBS transmitters	Point
Cellular towers	Point
TV digital station transmitters	Point
Natural Gas (AZ, FL, KY, OK, OR, TX)
Terminals^1	Point	Polygon by digitizing
Underground storage^2	Point	Polygon by buffering 300 ft
Processing plants^2	Point	Polygon by buffering 500 ft
Pipelines	Line
Electrical (AZ, FL, KY, NV, OK, OR, TN, TX)
Substations	Point
Transmission lines	Line
Correctional (AZ, CA, DE, GA, KY, LA, MI, NV, NJ, NC, OR, SC, SD, TN, TX, UT, VT, WI)
Prison boundaries	Polygon
Water (AZ, KY, NV, OK, OR, TN, TX)
Dams	Point
Wastewater treatment plants	Point

Furthermore, all features (original point/line as well as converted features) need to be buffered to produce final restricted areas around them. The choice of the buffer radius was based on drone regulations issued by states, which mention a default radius of flight restriction zones around facilities, such as 250 feet in Tennessee, 400 feet in Oklahoma, or 500 feet in Nevada, North Carolina, and South Carolina (911 Security 2019). Since most states use a 500-foot radius, we adopted this buffer radius also for those states, which did not specify a buffer radius in their regulations.

Figure 1 illustrates for each state the number of facilities falling into the six facility categories mentioned in state regulations (Table 1), and into the two layers based on FAA restrictions (airport airspaces and national security zones). States without state regulations feature generally a low number of restricted facilities. Only three states (Arizona, Oregon, and Texas) have restrictions in all eight restricted facility types. For Texas, this fact, combined with the state’s large area, explains the highest facility count for this state.

Figure 1. Number of restricted facilities for US states.

3.2.2. United Kingdom

Drone regulations in the UK are issued by the Civil Aviation Authority (CAA). Since 31 December 2020, the rules for flying drones in the UK and all European member states are the same. The new rules divide drone flights into three categories, which are Open (low-risk flights), Specific (flights in high-risk areas, e.g. urban area), and Certified (for large drones) (UKCAA 2020). Most recreational flights with a drone under 25 kg will fall under the Open category. Its key rules limit the flight height above ground to 400 ft (120 m), require a line of sight, and mandate keeping clear of airspace restrictions (including around airfields) unless permission has been obtained. The Open category has three subcategories (A1, A2, and A3) based on the weight of the drone. The subcategories come with certain privileges (e.g. flying over people with a drone if it is less than 250 g) or requirements (such as drone registrations and passing some pilot tests).

Since this study is concerned with drone contributions before 2021, past regulations at that time will be considered for the analysis instead, which include among others a 5 km circular restriction around airports and airfields and restrictions around a series of 5 km rectangular zones from the end of each runway (Dronesafe 2021). The Dronesafe website provides an interactive map displaying the locations of UAS airspace restrictions. KML data files of these restriction zones can be downloaded from the NATS Website (NATS 2021). Additional sources were used to complement restrictions in the UK, including DJI’s website, which allows site visitors to select a particular region, based upon which flight restrictions for that area will be displayed in a map (DJI 2021). The interactive map focuses on certain facilities such as airport airspaces, military bases, nuclear power plants, prisons, and stadiums. The UK no-fly zones on the DJI interactive map were scraped from xhr files, which were returned from the server in response to a browser request. The returned data contains facility geometries with attributes such as facility id, name, type, latitude and longitude of the polygon centroid, and coordinates of the polygon points. Stadiums, which are displayed in the DJI map as a 500-m radius circle from the center and downloadable as center points, were converted to areas using a 500-m buffer radius. Another regulatory source was NetworkRail, which states that drone pilots cannot fly within 50 mof the railroad network and within 150 m of railroad built-up facilities, respectively (Network Rail 2021). Since it is unclear which buildings or structures railway built-up facilities entail, only railways were considered as restricted areas for that facility group. To map this restriction, OSM railroad geometries for the UK were obtained from Geofabrik (https://download.geofabrik.de/). A 50-m buffer was applied around the OSM railroad geometries. Figure 2 visualizes the number of flight-restricting facilities for the 12 UK regions (nine official governmental regions in England besides Scotland, Wales, and North Ireland). The railroad facility count in each region represents the number of different track names.

Figure 2. Number of restricted facilities for UK regions.

3.2.3. France

The Directorate General for Civil Aviation (DGAC) operates, organizes, and secures aviation activities in the French airspace. The DGAC established flight restrictions for different altitudes, which are displayed as an interactive map on a governmental website (Géoportail 2021). The website does, however, not offer an option for data download. To do so, no-fly zones were scraped from uMap (uMap 2021), which is based on OSM data. Figure 3 shows the combined zero altitude flight restriction zones obtained from uMap. Analyzes related to drone restriction violations will be conducted for France as a whole and its 22 administrative regions.

Figure 3. No-fly areas in France obtained from uMap.

3.3. Statistical tests

The violations of drone regulations were analyzed for each drone portal and each of the three countries by determining the number of contributions located inside and outside restricted areas, based on center coordinates of shared images. As a result, a contingency table was created for each country, which contains for each portal the number of drone contributions inside and outside restricted areas. To analyze if the proportion of images posted in restricted areas differed significantly between the compared portals in a country, a chi-square test of independence was conducted for each country. The Null hypothesis is that the portal has no effect on this proportion. In general, a Chi-square test is used to analyze the differences between groups without requiring equality of variances (McHugh 2013). Using a binomial test, a subsequent, refined analysis determined for each sub-region of a country, i.e. the 50 US states, 12 UK regions, and 22 administrative regions of France, if the proportion of drone contributions falling inside restricted areas is higher than what can be expected based on the proportion of the restricted area within the sub-region of interest alone. The binomial test is an exact test that is applied when an experiment has two possible outcomes (such as an image falling inside or outside a restriction region). It evaluates the statistical significance of the difference between observed test results and expected results.

4. Results

A drone contribution is considered to be in violation of drone regulations if it falls inside a restricted area polygon after a regulation has been established. For example, Florida began to prohibit drone flights over natural gas facilities in 2017. There is a clear relationship between the proportion of restricted area and drone violation rate in a country, as can be seen when moving downwards through the rows in Table 2. This means that jurisdictions who declare more no-fly zones will also need more code compliance officers or information campaigns to ensure compliance with these restrictions. Whereas the three center columns list violation rates for the individual drone portals, the right-most column computes the violation rate based on contributions to all three portals. SkyPixel has higher violation rates than Flickr in all countries. DroneSpot has the lowest violation rate in France since this portal does not allow to publish drone images contributed to restriction zones. The few DroneSpot images located in restricted zones must, for some reason, have bypassed this intended filter.

Table 2. Comparison of violation rates (in %) between portals and countries.

	Restricted area (%)	SkyPixel	Flickr	DroneSpot	All portals
US	1.7	9.2	8.0	-	7.0
UK	4.4	11.1	7.6	-	9.3
France	16.9	55.6	24.6	0.4	27.5

4.1. Violations by state and region within analyzed countries

This section charts and maps violation rates for the different administrative regions within the US, the UK, and France. The brightness value of the gray tone in the background choropleth map in Figure 4 indicates the proportion of restricted areas (based on state and federal restrictions) in each US state. The four states with the highest proportion of areas with restrictions are Florida (8.2%), Oklahoma (7.5%), Texas (7.2%), and Kentucky (7.0%). Dots in yellow (SkyPixel) and orange (Flickr) indicate the locations of contributions from these sources that are in violation of restrictions. Violation clusters can be observed around urban areas (Dallas, Houston) in Texas, along the coasts in Southern California and Florida, and in some northeastern states, especially Connecticut. The size of a pie chart corresponds to the total number of drone media contributed to SkyPixel and Flickr in each state, which is highest for California (3152), followed by Texas (2074) and Florida (1874). These numbers follow the ranking of states by population (CA: 39.5 mio., TX: 29.0 mio., FL: 21.5 mio.). A red wedge expresses the proportion of contributions that are in violation of drone regulations, which is highest in Arizona (25.0%), followed by Connecticut (21.9%) and Texas (21.3%). For some of the 21 states that issue state regulations in addition to federal FAA rules (Figure 5), this leads to high overall violation rates. For example, state regulations alone cause a 21.6% violation rate in Arizona (out of a total of 25.0%), a 20.2% violation rate in Texas (total: 21.3%), and a 11.0% violation rate in Oregon (total: 12.8%).

Figure 4. Violation rates in the US.

Figure 5. Violation rates for 21 US states based on different types of flight restrictions.

In the UK map (Figure 6), most regions in England reveal a higher proportion of restricted areas than the three regions outside of England. The London region shows most violations, both in absolute counts (142) and relative counts (41.8%) because it is the smallest region and has many restricted facilities, leading to 23.3% of London’s area being restricted. Scotland is the largest region in the UK and has the smallest restriction rate (1.9%) and the highest number of contributions (1455). Among all regions, Northern Ireland has the lowest number of total contributions (98), and Wales has the lowest violation rate of 3.6%.

Figure 6. Violations in the UK.

In France, Île-de-France, which contains Paris, is the administrative region with the highest proportion (40.7%) of restricted areas (Figure 7). Despite this, its violation rate of 50% is lower than that for the Limousin region, located in south-central France, which has France’s highest violation rate with 67%. Rhone Alpes has most violation counts in absolute numbers (258), but a lower violation rate (28%) due to having most contributions (924). Auvergne has the lowest violation rate of 4.4% with 7 out of 158 contributions falling inside restricted areas. Figure 7 is the only violation map among the three analyzed countries, which shows DroneSpot contributions (green dots).

Figure 7. Violations in France.

4.2. Violations by facility type in the US and the UK

Figure 8 shows the number of violations in each US state, separated by facility type. Regarding FAA regulations, there is generally a higher proportion of airport airspace violations than national security zone violations. At the state level, violations of electrical, telecommunication, and water facility restrictions are most common. Six states have over 100 violations (Arizona, California, Florida, Nevada, Tennessee, and Texas), all of which also have state regulations (Table 1).

Figure 8. Number of violations by facility type in US states.

For the UK, violations of flight restrictions occur most often over railroads and airport facilities (Figure 9). No violations were observed for prison and nuclear power plant restricted zones. Due to the abundance of restricted facilities, London experiences the highest number of violations.

Figure 9. Number of violations by facility type in UK regions.

4.3. Comparison of violation rates among portals

Table 3 shows the contingency tables for all three countries, where the upper three are for US alone. Contribution counts to each portal are split into two categories: The upper category (“In”) expresses the number of drone contributions falling inside restricted zones (denoting a violation), whereas the lower category (“Out”) counts the number of drone contributions outside restricted zones. The bold numbers in each set denote the portal with the highest proportion of violations. Except for state-wide regulations in the US, SkyPixel had the highest proportion of violations in all regulation scenarios. The right-most column provides the chi-square test statistic for each contingency table and its associated level of significance. It shows that for each of the five analyzed situations, the association between violation behavior and portal is significant at a 5% level of significance or less and that violation rates in SkyPixel are higher for most of the regulatory frameworks.

Table 3. Cross-tables and chi-square statistics for violations of drone regulations in different countries.

Regulations	Location	SkyPixel	Flickr	DroneSpot	Chi-square
US (State-wide)	In	603	291	-	5.4*
	Out	8497	3454	-
US (FAA)	In	1030	218	-	55.4***
	Out	16,352	6055	-
US (All)	In	1592	499	-	8.3**
	Out	15,790	5774	-
UK	In	358	240		22.7***
	Out	2880	2922
France	In	1458	148	10	2017.4***
	Out	1163	454	2648

Significance codes: *** 0.001, ** 0.01, * 0.05

4.4. Comparison of violations among states and regions within each country

Table 4 summarizes for each drone portal and regulatory framework the number of states or regions without any violations observed. It also shows the number of states (regions) with a significantly higher proportion of violations than the proportion of restricted areas within a state (region), as determined by the binomial test. The table also reports the unweighted mean, median, and maximum violation rates across the states (regions) in a country.

Table 4. Statistical summary for binomial test results.

	SkyPixel	Flickr	DroneSpot
	violations	violations	violations
State-wide regulations in the US (21 states)
No violation regions	10	10	-
Significant regions	6	5	-
Mean violations across 21 states	5.1%	3.1%	-
Median of violations across 21 states	0.2%	0.2%	-
Maximum violation of a state	26.8%	18.3%	-
FAA regulations in the US (50 states)
No violation regions	8	22	-
Significant regions	27	11	-
Mean violations across 50 states	5.2%	2.6%	-
Median of violations across 50 states	3.1%	0.7%	-
Maximum violation of a state	23.0%	15.9%	-
All regulations in the US (50 states)
No violation regions	7	18	-
Significant regions	32	14	-
Mean violations across 50 states	7.3%	3.8%	-
Median of violations across 50 states	4.7%	1.7%	-
Maximum of violations across 50 states	28.8%	21.7%	-
Regulations in the UK (12 regions)
No violation regions	0	0	-
Significant regions	6	6	-
Mean violations across 12 regions	10.6%	9.0%	-
Median of violations across 12 regions	6.5%	8.0%	-
Maximum of violations across 12 regions	45.5%	17.5%	-
Regulations in France (22 regions)
No violation regions	0	2	13
Significant regions	20	6	0
Mean violations across 22 regions	51.9%	31.4%	0.6%
Median of violations across 22 regions	49.1%	23.1%	0.0%
Maximum of violations across 22 regions	92.9%	88.9%	3.6%

4.4.1. United States

For the US, three regulation frameworks were considered where state-wide regulations were applicable to 21 US states. Only five states had significantly higher violations than expected in Flickr compared to SkyPixel with six, and the unweighted average violation rate across these states was smaller for Flickr as well. This can be primarily attributed to a high violation rate of 26.8% (148 out of 552 images) for SkyPixel in Arizona, mostly due to water-related violations (water treatment plants and dams). However, country-wide, these state regulations were more frequently violated in Flickr than in SkyPixel (compare first row in Table 3). For the national FAA regulations, the results of the binomial tests for the 50 states are in-line with country-wide results, i.e. a higher violation rate for SkyPixel (see second set of records in Table 3). Compare, for example, 27 states, which exceed the expected rate of violations based on the area proportion in SkyPixel with the 11 states in Flickr. Along the same line, states without violations are more abundant in Flickr, and all other descriptive state metrics (mean, median, and maximum) are higher for SkyPixel as well. When both federal and state regulations are combined, metrics also point toward SkyPixel to be more frequently in violation of drone regulations than Flickr.

Figure 10 shows the density plots of violation rates for the 50 US states for SkyPixel (a) and Flickr (b) under consideration of combined FAA and state regulations. Distribution densities are plotted separately for states that revealed a significant excess of violations compared to the restricted area proportion, as determined by Binomial test results (p < 0.05), and those that did not (p > 0.05). As expected, the peak and mean of violation proportions are higher for the group of significant states (pink density curve). As Figure 10 demonstrates, even some of the states that are marked as significant show very low violation rates (e.g. 1.8% for SkyPixel in North Dakota), which is, however, because of an even smaller proportion of land being marked as restricted for these states (e.g. 0.4% in North Dakota). In concordance with the numerical results presented in Table 4 (third record group), the density distribution plot also reflects that with SkyPixel some states reach violation rates close to 30% in SkyPixel and more than 20% in Flickr.

Figure 10. Density plots of US drone violations rates for (a) SkyPixel and (b) Flickr.

4.4.2. United Kingdom

In the UK, six regions were found to have significantly more violation rates than expected from area proportions for SkyPixel and Flickr (Table 4). Only two regions were simultaneously statistically significant in both SkyPixel and Flickr, namely Scotland and Northwest in England. Other descriptive values (mean, median, and maximum) are higher for SkyPixel, which leads to SkyPixel having a higher proportion of violations across the UK than Flickr (compare Table 3 for the UK). The density plots in Figure 11 show a clear distinction between regions with (p < 0.05) and without (p > 0.05) significantly higher violations than expected. In SkyPixel, one region (London) stands out with a high violation rate of 45.5% (peak in pink area in left figure). In Flickr, London also stands out with a high violation rate of 16.3%. However, this is still less than the expected violation rate based on area (23.3%), which makes this value to be shown under the green density plot (p > 0.05) on the upper tail.

Figure 11. Density plots of UK drone violations for (a) SkyPixel and (b) Flickr.

4.4.3. France

France is the only country where contributions from three drone portals were compared due to the provision of contributions to DroneSpot. All metrics in Table 4 points toward fewest violations on DroneSpot, followed by Flickr and SkyPixel. DroneSpot is the only portal that has a smaller proportion of violations than expected (based on area proportions) for each administrative region and has therefore also no significant region. The latter finding renders the density plot for DroneSpot (Figure 12(c)) in France different from other charts in the sense that it contains only one area (p > 0.05). Also, its range of violations, with up to only 3%, is much smaller compared to other drone portals. As far as SkyPixel (Figure 12(a)) and Flickr (Figure 12(b)) goes, their density plots show a clear distinction between significant and non-significant administrative regions. The Flickr density plot shows several peaks, with the highest violation rate at 88.9% (8 out of 9 observations in restricted areas) for Île-de-France.

Figure 12. Density plots of France drone violations for (a) SkyPixel, (b) Flickr, and (c) DroneSpot.

5. Discussion

This study compares drone violations between three countries that feature different types of restriction areas. Although the land size of France is only 6% of that of the US, the acreage of drone-restricted areas in France reaches about two-thirds of that in the US. This results in the highest density of restricted areas (16.9%) for France among the three compared countries (US: 1.7%; UK: 4.4%). The UK features the highest number of shared drone contributions per million residents between 2015 and 2019 (86.3), followed by the US (65.0) and France (37.6).

5.1. Summary of violation rates

Drone contribution counts as well as the violation rates differ over the years and between the analyzed countries. For Flickr and SkyPixel combined, France reveals the highest violation rate (43.6%), followed by the UK (8.8%) and the US (8.7%) between 2015 and 2019. This high violation rate for France can likely be attributed to the high proportion of areas designated as no-fly zones in that country compared to the US and the UK. Figure 13 illustrates annual image contribution counts per million residents (bars) and violation rates (lines) between 2015 and 2019 for Flickr and SkyPixel. In all three countries, a steep decline in SkyPixel contributions can be observed in 2017 together with a decrease in the violation rate for that portal. Overall, SkyPixel contributions increase for France (Figure 13(c)), whereas they remain relatively even for the US (Figure 13(a)) and the UK (Figure 13(b)). Flickr contributions per million residents show a general increase for the US and the UK, but not so for France, where SkyPixel instead seems to gain in popularity over the years. Violation rates tend to generally increase over the years in all countries and portals, and they are generally higher for SkyPixel than for Flickr (compare also Table 4). DroneSpot for France (not charted in Figure 13) demonstrates almost perfect compliance with regulations (compare Table 4) since this portal bans drone images that are in violation with restricted zones. Such stringent control of compliance rules before publishing a contributed image may encourage drone users to adhere to federally mandated drone flight restrictions.

Figure 13. Contributions per million residents and violation rates for analyzed portals and countries between 2015 and 2019.

For the US, some states with additional state regulations showed high violation rates, such as Arizona or Texas. A possible explanation is that many recreational pilots are not aware of both federal and regional flight restrictions. In conclusion, decision and policy makers can use regional violation maps (Figures 4, 6, and 7) and charts (Figure 8) to identify locations with higher violation rates (especially high-population density areas) and the type of facilities often associated with violations (e.g. electrical utilities) to apply targeted information campaigns (e.g. through additional signage, online portals, or pamphlets) to better to educate drone users in the affected areas.

The fact that violation rates are higher for SkyPixel than for Flickr may point toward different types of users. Whereas both portals appear to attract both hobbyists and professional photographers, the user communities may differ in experience, socio-economic background, or other motivations to contribute. Such differences have been found for other VGI portals, such as OpenStreetMap (Heipke 2010) or between different social media apps (Juhász and Hochmair 2019). Since violations of flight restrictions have been observed both in SkyPixel and Flickr for all three analyzed countries, mentioning the existence of no-fly zones and provision of links to further regulatory guidelines posted on the VGI drone portals could increase compliance rates.

5.2. Factors associated with violations

To more closely analyze which socioeconomic and infrastructure-based factors contribute to the violation of drone restrictions, negative binomial regression models for violations in SkyPixel and Flickr were developed at the zip code level for Florida. Candidate independent variables include UAS registry enrollment counts per zip code; the percentage of restricted area by FAA and by state (US only); number of UAV images posted per zip code; population per zip code; shortest distance (in km) to attractions or infrastructure, including airports, inland water bodies (lakes and rivers), and state parks; presence of populated place (city centers); and presence of coastline. The registry enrollment counts denote the number of users (hobbyists and non-hobbyists) of registered drones from 2015 to 2018 per zip code (FAA 2021b). The locations of airports, water bodies, city centers, and coastlines were downloaded from the Natural Earth Website (Natural Earth 2021). Zip code population data and state park locations were obtained from other Florida data sources (FDEP 2019; FGDL 2021). Since in negative binomial models the offset is the natural log of the exposure, the natural log of the zip code area (in km²) was used as the offset. Regression models were built in a stepwise manner with the goal to maximize model fit while avoiding collinearity among predictor variables. The results are reported in Table 5. Although the number of statistically significant variables differs between the SkyPixel and Flickr models, coefficients points in the same direction were significant. That is, the number of violations increases with the number of image contributions in a zip code area, the percentage of a zip code area that falls under some drone restriction, and the presence of a city center. Violations decrease with distance between zip code and nearest airport (further away from airports means fewer opportunities to break restriction rules) and increase with distance from water bodies (implying that there are generally few restrictions around water bodies). Registry enrollment count was not a significant predictor.

Table 5. Negative binomial regression models predicting violation numbers.

	Flickr		SkyPixel
	Estimate	Z-score	Estimate	Z-score
(Intercept)	−11.87 ***	−14.55	−9.96 ***	−24.11
Number of contributions	0.78 ***	6.59	0.36 ***	17.37
Restricted area by state (%)	0.18 ***	6.08	0.13 ***	7.39
Restricted area by FAA (%)	n.s.	n.s.	0.04 ***	8.12
Distance to airport	n.s.	n.s.	−0.01 *	−2.13
Distance to water bodies	n.s.	n.s.	0.01 ***	4.61
Presence of city center	2.47 *	3.25	1.22 *	2.16
McFadden Pseudo R^2	0.971		0.836
N	959		959

Significance codes: *** 0.001, ** 0.01, * 0.05

5.3. Challenges associated with mapping of restricted areas

Quantifying violations of drone regulations in the three analyzed countries required a spatial representation of restricted area. The level of detail used in federal and state regulations to define these no-fly zones varies between countries. To obtain a more comprehensive picture of no-fly zones, it was also necessary to consult and compare multiple sources for each country. While the UK and France define regulations at the federal level only, 21 US states provide additional regulatory frameworks with additional restrictions around specific infrastructure and facilities (e.g. natural gas terminals and telecommunication towers). Since no facility restriction maps are provided for the US states, it takes a combination of different GIS data sets to map restricted areas around these facilities. As opposed to this, various mobile apps and websites, both provided through governments and industry, were established to map restricted areas at the federal level for selected countries around the world. These maps can be used as a starting point for secure drone flight planning. Still, statewide and other local restrictions are missing from such maps. Therefore, we recommend cross-checking different online maps for a region of interest to get a complete as possible picture of no-fly zones. Another challenge in the analysis of violations is that not all online Web mapping applications facilitate downloading restriction zones in a GIS data format. A workaround for some Websites is to run a Web parser to gradually build a dataset of no-fly areas from multiple server requests. Also, the presentation of flight regulations could be improved by providing a one-stop access point to obtain all relevant regulatory documents (e.g. federal and local) for drone flights in a given area.

5.4. Spatial uncertainty

Measuring violations of drone regulations in this study relies on point coordinates obtained from drone imagery shared on online portals. The reliability of these coordinates can be affected in mainly two ways, namely by readout precision (number of decimal digits) of obtained coordinates and by actual positional offsets between the provided coordinates and the location at which an aerial image was taken, i.e. point accuracy. Readout precision differs between drone portals. That is, Flickr and DroneSpot display geographic coordinates with six decimal degrees, which corresponds to a precision of about 10 cm. SkyPixel coordinates are limited to the nearest thousandth of a decimal degree, which corresponds to a ground distance of about 100 m. Whereas SkyPixel and Flickr contributions are coded in an EXIF contribution files DroneSpot does not use EXIF files but instead, drone users are asked to specify the image location by pinpointing it on the world map upon image upload. To determine if these uncertainties in image coordinates could affect our judgment of whether an image falls inside or outside a restricted zone, the positional offset of shared drone contributions from their true position was manually assessed using a random sample of 20 photos from each of the three portals. More specifically, through comparison of the image content with an aerial background map, the distance offset for each analyzed image could be determined. Heavily zoomed-in or underexposed photos as well as photos taken above large uniform landscapes without distinct landmarks were excluded from this validation process.

The analysis resulted in a mean positional error of 76 m for SkyPixel, 4 m for Flickr, and 48 m for DroneSpot (“Position offset” columns in Table 6). Out of the 20 photos, 18 Flickr, 5 SkyPixel, and 8 DroneSpot contributions were found to be mapped at the correct location within about 5 m. To assess if these positional errors of drone imagery would affect estimated violation rates for the different portals, the mean nearest neighbor distance from each drone image to the closest restriction area boundary was determined (“Area boundary” columns in Table 6). Mean nearest neighbor distances are at least a magnitude larger than position offsets. Therefore, image offset error will rarely affect estimated violation rates for each portal and thus violation rates can be considered reliable.

Table 6. Positional errors and distance to nearest restriction area for a sample of drone images.

	SkyPixel		Flickr		DroneSpot
	Area boundary	Position offset	Area boundary	Position offset	Area boundary	Position offset
Distance (mean) [km]	10,709.6	76	8295.1	4	725.6	48
Distance (Std.-dev.) [km]	18,069.5	107.7	14,203.2	14.0	869.2	65.2
Points at correct location	5		18		8

5.5. Limitations of the study

The analyses in this study combine the locations of drone contributions and mapped aviation regulations. Errors in any of these sources can affect results and conclusions drawn. Regarding the drone contributions, the downloaded data covers different time periods, due to the different inception years for the different portals. Since user contribution patterns may change over time, analyzing violation rates across different time spans may lead to some small inconsistencies. However, using as many data as possible outweighs this potential drawback. SkyPixel limits the number of contributions that can be extracted per geographic place tag to 5120. This limit was reached in only one situation, namely when extracting contributions that use the “United States” tag. Due to SkyPixel having three contribution types (photo, video, and 360 photo), the “United States” tag was used three times to extract different data types separately. As a result, all SkyPixel videos and 360 photos were extracted, but not all standard photos could be extracted, with the oldest downloaded photo dating back to January 2016. However, this limitation did not seem to have a noticeable effect on overall download numbers, since the contribution count for SkyPixel peaks in 2015 (Figure 13(a)). Another limiting factor on the temporal aspect is that only upload dates but not image capture dates are shared with drone contributions. This can be relevant if no-fly zones change over time, e.g. for US state regulations.

Certain aviation regulations specify location, altitude, and time. Due to the lack of flight height in the shared drone data, this study considered for airports only those no-fly areas, which prohibit flights at all altitude, which may lead to an underestimation of the number of violations around airports. It should also be noted that the same drone contribution could be shared on more than one drone portal by a user, which means that contributions to the portals are not necessarily independent. Since each portal is analyzed separately in terms of violation rates and contribution patterns, this issue does not affect the obtained compliance rates. It might affect comparison of compliance rates between portals within a country through chi-square statistics (Table 3), but even with potential double counts all identified differences were found to be significant.

Regarding the second influencing factor, i.e. aviation regulations, knowledge of regulation updates is important for accurate compliance analysis. Aviation regulations are updated periodically and rapidly, especially for US states (911 Security 2019). Regarding national regulation updates, we were not able to obtain the geographical extent of earlier restricted areas in all three countries. The availability of such information may improve the accuracy of assessed violation rates. Spatial uncertainty in regulations is another aspect that may affect the study results. For example, regarding US state regulations, only 5 out of 21 states mentioned the radius of buffer zones around restricted facilities (911 Security 2019). As a workaround, the most frequent buffer distance from among the five states was applied to create no-fly zones for the other 16 states as well.

6. Conclusions

Drone pilots use these vehicles in a variety of innovative and critical operations, such as emergency response, humanitarian efforts, parcel delivery, policing and border patrol, or personal transportation (Applin 2016; Engberts and Gillissen 2016; Marin 2016; Martini et al. 2016; Michaelides-Mateou 2016). Involving drones in new operation types cannot be done without regulating these activities. One of the proposed recent drone regulations is FAA’s Notice of Proposed Rulemaking on Remote Identification (remote ID) of Unmanned Aircraft Systems, which was published on 31 December 2019 (FAA 2021e; Wu et al. 2021). The remote ID is a novel approach that facilitates drone integration into the US airspace under consideration of concerns related to safety, national security, and law enforcement by sharing drone information, such as drone identification, geographic coordinates, and altitudes for both the vehicle and its control station. A Standard Remote ID drone is produced with built-in remote ID broadcast capability in accordance with the remote ID rule’s requirements. The final rule became effective in the US on 16 March 2021, and in the EU on 1 July 2020 (EASA 2019a, 2019b; FAA 2021c).

Governments already provide online information and maps about restrictions and regulations. For example, the FAA created an official airspace application called B4UFLY (FAA 2021a) which includes an online map of restrictions, which allows users to specify the flying location and check restrictions and airspace status. It is widely used among drone enthusiasts with around 1600 reviews on Google Play and 220 reviews in the Apple Store. One drawback of the application is, however, the limitation to federal FAA regulations, which neglect regulations posed by individual states.

For future work, the study can be extended to other countries. Since the EU adopted a unified integrated drone system guided by EASA at the beginning of July 2020 (EASA 2019a, 2019b), comparison of regulation violations will become easier between EU countries. Also, other drone portals, such as Dronestagram and Travelwithdrone, can be considered in similar follow-up studies.

Data availability

The point data of images and videos on the VGI drone portals used for the analysis are available as csv files in figshare at https://figshare.com/s/98bbc2f980346be076ff, DOI 10.6084/m9.figshare.13634549.

Disclosure statement

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

Additional information

Notes on contributors

Ammar Mandourah

Ammar Mandourah received his master’s degree in Geographic Information Systems from the University of Redlands in 2016 and is currently a Ph.D. candidate in the Geomatics program at the University of Florida. His research evolves around the geospatial analysis on user contributions in different social media portals.

Hartwig Hochmair received his PhD from the Technical University of Vienna, Austria, and is a Geomatics Faculty member 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.