In recent years, the use of unmanned aerial vehicles (UAVs) steadily grew in many countries, including Poland, primarily as a result of rapid developments in the field of unmanned flight and drone-borne sensors for geospatial data collection. One impediment to the wider spread of UAVs is airspace regulation [1
]. Currently, legislative and regulatory authorities worldwide are faced with the challenge of striking a balance between different UAV-related economic, information, and safety needs [3
]. If regulatory constraints are adjusted to respond better to actual social needs, UAVs may find use in a greater number of industries [5
Studies were already published regarding applications of UAV-collected data in areas such as forestry and agriculture. Guerra-Hernández et al. [6
] tested the applicability of this technology to the stock-taking of forests. In Japan, UAV-based remote sensing was successfully used to monitor real-time wheat growth status, and to map within-field spatial variations of wheat yield for smallholder wheat growers [7
]. In Brazil, UAVs were deployed over sugarcane fields. Wachholz de Souza et al. [8
] described an object-based image analysis (OBIA) procedure for UAV images, designed to map and extract information about skips in sugarcane planting rows. German researchers examined the prospects of monitoring biophysical parameters and nitrogen content in wheat crops using images taken from UAVs [9
], while their Swedish colleagues researched ways to identify aquatic plants which serve important environmental functions, and thus, should be monitored to detect changes in ecosystems [10
]. In Reference [11
], high-resolution thermal imagery acquired by an unmanned aerial vehicle was used to map plant water stress and its spatial variability. This technology can also be applied widely in engineering [12
] and environmental protection. An interesting application is presented in Reference [14
], which describes the integration of an off-the-shelf laser-based methane detector into a multi-rotor UAV, and demonstrates its efficacy in generating an aerial methane concentration map of a landfill. In order to perform plant protection operations, an automatic spraying system based on unmanned aerial vehicles (UAVs) was designed in China [15
Nevertheless, the authors of this article were mostly interested in landslides. Most publications focus on applying UAVs to monitoring and assessing landslide dynamics [16
]. The UAV demonstrated its capability for producing valuable landslide data, but improvements are required to reduce data processing time for the efficient generation of ortho-mosaics based on photogrammetric digital terrain models (DTMs), in order to minimize geo-referencing errors.
This study, however, addresses a different issue, namely the collection of spatial data for creating and updating cadastral databases with respect to landslide sites. The potential applicability of UAVs to the acquisition of such data was already indicated in the literature [22
]. According to Reference [26
], with the exception of Reference [27
], cadastral mapping is not mentioned in review papers on application fields of UAVs [3
]. As is suggested in Reference [26
], this might be due to the small number of case studies within this field, the often highly prescribed legal regulations relating to cadastral surveys, and the general novelty of UAV use in mapping. Nevertheless, all existing case studies underline the high potential of UAVs for cadastral mapping, in both urban and rural contexts, for developing and developed countries.
This study was designed to answer the question of whether the statutorily required accuracy is attainable for difficult-to-access landslide sites where considerable differences in terrain elevation may hinder the use of photogrammetric data for updating cadastral databases. Analyses were conducted to compare the accuracy of different UAV data processing methods in determining the coordinates of points which form the framework of a cadastral database. Relevant tests were carried out with a varying number of ground control points (GCP) used for developing an orthophoto map and a digital surface model (DSM), as well as with varying accuracy of determining the coordinates of such points (static or kinematic Global Navigation Satellite Systems (GNSS) survey of control points).
3. Results and Discussion
The respective mean errors of determining the coordinates of the control points and the check points were as follows:
These values were determined in relation to the base control network, which includes ASG-EUPOS reference stations.
A comparison of coordinates determined using two independent methods (Table 4
) shows that the obtained values were free of gross errors, and that the maximum difference in the coordinates did not exceed 0.050 m.
The UAV-collected data were processed using the Agisoft PhotoScan Professional software [69
] in four variants (Table 5
), differing from each other in terms of the number of control points used (Figure 5
) and the method of their measurement.
The first step of image processing was to align the images. At this stage, the images were uploaded to the software and were given the initial orientation by adding approximate coordinates of image projection centers. The alignment was competed using an accuracy parameter set to “high”. It ensured the use of the original image resolution. Additionally, control points were indicated on the images. This was preceded by uploading coordinates of the terrestrial photogrammetric control to the software. Each marker was indicated on all photos where it was visible. The block of photographs for each variant was finally aligned, and, at the same time, the camera calibration parameters were determined. In this process, the following values were determined: the principal distance (c
), the location of the principal point (cx
), and the distortion parameters (k1
, and p1
). As a result of the alignment, the root-mean-squared errors of the control points were obtained (Table 6
The assumed control point identification error at the data processing stage (interpretation error) was 1 pix (20 mm). Accordingly, the ultimate marker (control point) mean spatial error was 60 mm for RTK GNSS, and 25 mm for static GNSS.
It must be noted that both the number of photogrammetric control points used, as well as the survey method applied to measure them, affects the obtained error values. The control point mean-squared error for Variant 1, which was considered to be the least accurate (eight control points surveyed using RTK GNSS), was more than twice the analogous parameter determined for Variant 4 (15 control points surveyed using static GNSS). However, none of the errors was greater than 0.100 m. The obtained values of spatial errors were considerably affected by altitude errors, which were negligible for the purposes of updating a cadastral database. The mean-squared errors of the planar coordinates showed a similar tendency and range from 0.026 m (Variant 4) to 0.048 m (Variant 1).
The final step of image processing in the Agisoft PhotoScan software was to create dense point clouds with the method of dense matching. Then, DSMs were created. These were used as a basis to conduct orthorectification of images and to create orthophoto maps. As a result of the data processing conducted, four orthophoto maps with a GSD of 19.7 mm and four DSMs with a resolution of 39.5 mm were generated in the PL-2000/7 rectangular planar coordinate system.
In order to determine the accuracy of the obtained photogrammetric products and their suitability for updating a cadastral database, the positions of 18 check points on the orthophoto map were measured, with their altitude coordinates read from the DSM. The obtained coordinates were compared with the data received by means of the direct static GNSS and RTK GNSS surveys. The comparison of the coordinates was carried out for the datasets marked by (+) in Table 7
. The computed accuracy assessment parameters for check points surveyed using static GNSS, compared to an aligned block of photographs, are presented in Table 8
. Similar calculations for check points surveyed using RTK GNSS are presented in Table 9
It must be noted that, regardless of the check point survey method, the standard deviation of the differences in spatial coordinates did not exceed 0.100 m for any of the calculation variants, or 0.050 m for the planar coordinates. Such values met the accuracy requirements for determining the coordinates of boundary points laid down in applicable Polish laws and regulations [45
]. The mean difference in the coordinates is also worth noting, as it may indicate the occurrence of a systematic error. In Variants 2, 3, and 4, this parameter was near 0 for each coordinate, while, in Variant 1, it was −0.050 m for the altitude coordinate, which is a significant value. Analysis of the maximum and minimum differences in the coordinates shows that they ranged from −0.100 m to 0.100 m except for Variant 1, for which the obtained results were the most unsatisfactory (as much as −0.200 m for ΔH
When studying the obtained results, one must consider the locations of the check points in relation to the control points (Figure 5
). Some of the check points were located on or even outside the external boundary of the survey area as defined by the extreme control points (points 606, 609, 613, 513, 507, 305, 306, 311, and 312). Such locations were chosen intentionally, in order to see how the corresponding deviations would trend. It was at precisely those points that the maximum/minimum differences in coordinates were recorded for Figure 6
For planar coordinates, the greatest east–west deviation was recorded for point 606. Depending on the calculation variant, it ranged from −0.087 m to −0.052 m. The greatest north–south deviations occurred for points 507, 311, and 312, from 0.062 m to 0.049 m for point 507, from 0.024 m to 0.071 m for point 311, and from 0.034 m to 0.071 m for point 312. For point 305, similar results (0.067 m and 0.060 m) were obtained only in Variants 3 and 4.
When ready, the orthophoto map was compared with the cadastral database. In order to carry out relevant spatial analysis, the boundary of the landslide was outlined on the orthophoto map. It was established in line with the landslide visible in the photographs. If the boundary of the landslide ran close to cadastral boundaries, it was drawn over the boundaries of relevant plots in order to prevent the creation of new objects with a small surface area in the cadastre.
It should be emphasized here that any interference with the cadastral database must be preceded by appropriate settlements that are carried out in the field with the owners of the subject properties. Entities of the real estate cadastre express their consent for entering new objects into the database or updating the existing ones, by signing relevant documents that are necessary to carry out surveying and legal procedures.
Given the fact that the survey was conducted following the demolition of all the designated buildings, this study focused solely on plots of land. Once the vectorized outline of the landslide was imposed on the cadastral map (Figure 7
), it was found that the landslide covered 25 plots in whole and 34 plots in part. These analyses were carried out manually. There are studies aimed at exploring options for automatically delineating boundaries for UAV-based cadastral mapping [26
]. However, in the opinion of the authors, the algorithms described in Reference [70
] to extract boundaries automatically would be ineffective in the case of areas subject to landslides. Cadastral boundaries in areas affected by landslides are assumed to be not visible, as they may not coincide with natural or manmade object contours.
For the plots which were entirely covered by the landslide, data should be changed with respect to land use classes and soil quality classes, i.e., such plots should be marked as wasteland, excluded from soil classification. With regards to the other plots, their land use classes and soil quality classes should be properly adjusted. Under applicable Polish laws and regulations [45
], this means that the boundary of the landslide running over relevant plots should be surveyed with an accuracy of at least 0.50 m in relation to the nearest points of the geodetic control network and the survey network, which can be attained with a large margin of error on the basis of UAV-acquired data.
Unfortunately, this process cannot be automated due to the fact that surveys using UAV technology are only a part of the surveying and legal procedures that lead to the introduction of changes in the cadastral database.
If the landslide is declared undevelopable land area, it will be possible to register future changes with respect to all or part of the properties covered by the landslide. In such a case, the outline of the landslide, which would become a new boundary, would have to be surveyed with an accuracy of 0.10 m in relation to the nearest points of the geodetic control network and the survey network [45
], which is also attainable, as proven above.
Given the accuracy of the generated orthophoto map, it can be concluded that UAV-collected data may be sufficiently accurate to be used in surveying and legal procedures aimed at updating the cadastral database.