The application of civic technologies in a field survey of landslides

Slope failures are financially devastating natural hazards that contribute to land degradation in many areas throughout the world. The adaptation of civic technologies (Google Tango) in a field survey of landslides was examined. Data acquired from different resources and processed using different technologies were merged into a single model to concurrently demonstrate the interoperability and scalability of these data and the model quality. Reference control points were established using a survey‐grade Topcon Hiper SK global navigation satellite system receiver and a Topcon GPT 9003 M total station. An aerial survey was performed in an area of approximately 30,000 m2 using airborne laser scanning (9 points/m2) and aerial photogrammetry using a remotely piloted aircraft system (500 points/m2). The models suffered from data gaps in less visible areas, and micro‐scale landforms reflecting landslide activity were poorly visible. The missing details were supplied using data obtained from close‐range photogrammetry (9,132 m2; 92,300 points/m2) and a Lenovo Phab 2 Pro running Google Tango, which acquired detailed point clouds in near real‐time conditions (1,847 m2; 109,000 points/m2). Scans using the phablet provided point clouds with homogeneously dispersed data gaps, but the spatial accuracy was lower. However, the ergonomics of its field use and its low cost made it competitive with other technologies. The results confirmed that models based on point clouds acquired using different technologies allow the identification and measurement of micro‐scale landforms that may indicate landslide activity.

amount of time (Marvin et al., 2016). From a different point of view, public interest in open government data contrasts with vague ideas about how data relate to the lives of common people (Schrock & Shaffer, 2017).
Google's Tango project was developed by Google's Inc. Advanced Technology and Projects team in 2014. It was primarily designed for augmented reality (AR) gaming (Google, 2017). The adoption of Google's Tango technology in a field survey of landslides might provide a new perspective on the application of cutting-edge *civic technologies in the investigation of land degradation processes. We are aware that the effects of fine-scale land degradation would not be buried in the statistics for larger areas (Warren, 2002). From another perspective, the mapping and measuring of micro-scale landforms using common geodetic techniques often provide sparse or insufficient data to identify subtle landforms, particularly when a site is located in forested or remote terrain (Ortu o et al., 2017). Digital models developed from aerial survey data obtained using image-based technologies often suffer from data gaps in less visible areas (rocky and steep slopes), and micro-scale landforms indicating landslide activity cannot be properly recognized (Kasai, Ikeda, Asahina, & Fujisawa, 2009;Olyazadeh et al., 2017).
In this context, a survey performing the supplementary mapping of missing data in digital models of slope failures was initiated. For this purpose, we employed the Lenovo Phab 2 Pro using Google Tango technology. The first tests of this technology in forests (Hyyppä et al., 2017;Tomaštík, Saloň, Tunák, Chudý, & Kardoš, 2017) suggested that the accuracy of its acquired 3D data might be sufficient; however, its application for mapping and measuring slope failures had not been broadly tested for forested land and rugged terrains. The Tango technology provides an integrated package that offers processing capabilities (using a smartphone's own computational resources), connections (wireless local area network, global system for mobile communications), and other sensors (such as global navigation satellite systems; Hyyppä et al., 2017).
These attributes make Tango efficient and user-friendly for data acquisition by the general public and not only by experts in geodesy.
We emphasize several characteristics that make Google'sTango technology an appropriate tool for mapping and measuring the micro-scale landforms of landslides. The Google Tango project applies a methodical 'Markerless AR' approach, which means that this technology has no prior references about the captured environment. Therefore, it offers a complete map of the 'as-built' scene in near real time, which is very important for AR applications (Kopsida & Brilakis, 2016). The use of AR enables the visualization of construction progress in an image by superimposing a 3D model on the actual construction scene (Jadidi, Ravanshadnia, & Alipour, 2014). During all phases of scene building, the 3D geometric scene of the site is reconstructed, and progress photographs are georegistered in a virtual environment (Golparvar-Fard, Pe a-Mora, & Savarese, 2009).
The process of as-built scene building means that a scene is replenished by the data generated during the capturing phases. In the case of the Google Tango Project, both the pose data and 3D reconstructed point clouds can be used along with iterative optimization methods to refine the alignment between the as-planned and as-built data. This technology uses motion sensors such as inertial measurement units, which can significantly improve the tracking performance when combined with an embedded depth sensor and a point cloud (Kopsida & Brilakis, 2016).

| THE AIM
The investigated landslide is located in a deep valley with rugged terrain. The site is covered by a 90-year-old forest stand that mainly comprises beech and hornbeam. Present gaps in the forest canopy are mostly caused by the activity of the studied landslide. Landslide processes pose a unique risk for existing infrastructure-the road from the district town of Zvolen to the village of Železná Breznica.
Because it is difficult to access the inventory site, it is impossible to investigate the entire area of interest on the same scale using only one type of an image-based contactless technology, and using data generated by different aircrafts and using different techniques is recommended for achieving a precise digital terrain model (Hsieh, Chan, & Hu, 2016;Le et al., 2016). Two research objectives are addressed in this article based on the precise results from previous studies conducted using smartphone global navigation satellite system (GNSS) positioning in forests (e.g., Tomaštík, Tomaštík, Saloň, & Piroh, 2017) and the potential applications of civic technologies, specifically the Lenovo Phab 2, for landslide mapping.

| Examining the Lenovo Phab 2 data quality and interoperability across different scales and amongst digital models
The comparison of different techniques/technologies for landslide mapping has been broadly discussed (Raouf et al., 2017). We assume that the data collected by the Lenovo Phab 2 may supply missing details from less or nonvisible areas. To confirm the scalability of digital models, the dataset accuracy was examined. Data gaps, which are caused by terrain characteristics (e.g., vertical surfaces), interfering objects (e.g., trees), and other factors, are often present in the digital terrain models (DTMs) and digital surface models (DSMs) that are developed from the data obtained by conventional techniques, including airborne laser scanning (ALS) light detection and ranging technology (LiDAR), imagery from remotely piloted aircraft systems (RPAS), and close-range photogrammetry (CRP; up to approximately 300 m). All these technologies were applied in the field survey. The data gaps present in different models of CRP and Tango were visually compared, and their occurrence was explained.
Generally, accuracy assessment of geometric analysis is difficult to perform. A completely perfect landslide inventory map against which to compare the results does not exist, as landslide inventory maps created from LiDAR analysis by different experts result in inventory maps with considerable differences (Eeckhaut et al., 2007). Nevertheless, data need to be sufficiently accurate and precise. For the purposes of geographic information system (GIS) environmental applications, the residual errors and root mean square errors of control points are measured to assess the geometric transformation accuracy. Residual error is a measure of the fit between the true locations and the transformed locations of the control points (Zhu, 2016).
The method selected for examining the residual errors of transformations did not yield an actual error assessment in this article. However, the method yielded valuable results regarding the models' accuracy in the short term without requiring time-consuming and expensive ground control point (GCP) measurements and position accuracy testing. This method is important because it may be easily implemented in field surveys for the purposes of territorial planning. *Cutting-edge technology for the general public.

| Assessing the applicability of landslide digital models in participatory GIS
The surfaces of active landslides are characterized by scarps, areas of temporary or permanent water ponding, and ridges, which are generally short-lived landforms that can quickly be destroyed (Parise, 2003). Therefore, the adoption of an easy-to-use device for mapping and measuring micro-scale landforms may accelerate the data collection of these ephemeral but very important landforms. The efficiency of the Lenovo Phab 2 Pro for performing landslide inventory was examined (time savings, low cost, and ease-of-use aspects), acknowledging that participatory planning approaches permit updates of national databases on natural hazards by the general public ( Figure 1).
Gathering spatial data across different spatial and temporal scales through a user-friendly interface would help planners address the complexity of planning procedures. The current techniques for automating the progress of data collection promise to eliminate the labour-intensive tasks associated with manual data collection (Golparvar-Fard et al., 2009;Matta & Serra, 2016).

| Geoinformation system on slope failures in Slovakia
A systematic study of slope failures in Slovakia (over a nearly 50year-long period) was coordinated by the state administration body of the Division of Geology and Natural Resources of the Ministry of Environment of the Slovak Republic and resulted in the publication of The Atlas of Slope Stability Maps (1:50,000; Šimeková et al., 2006). The State Geological Institute of Dionýz Štúr (ŠGÚDŠ is the Slovak abbreviation used in Figure 2) provides a geospatial geological database that is available as part of the 'Geological Information System' project, which was launched at the end of 2005 (Liščák & Káčer, 2013). Altogether, 21,190 slope failures that cover an area of 2,575.912 km 2 are registered in The Atlas of Slope Stability Maps (1:50,000; Šimeková et al., 2006) and pose either damage or threat to 5.25% of the area of the Slovak Republic, and landslide areas constitute 77.68% of slope failures (Bednárik & Liščák, 2010). The landslide databases of France, Italy, and Slovakia are considered the most complete landslide databases in Europe (Eeckhaut & Hervás, 2012).
Despite this fact, many landslides are not included in these official databases. Here, one of these landslides is characterized in a case study from Central Slovakia.

| The landslide localization and its natural settings
The investigated landslide is located on the borders of three cadastral districts (Železná Breznica, Budička, and Tŕnie; 3020.73 ha), and it is located in the Kremnické vrchy Mts. These mountains belong to the Neogene volcanic region in Central Slovakia. According to Bakon et al. (2015), landslides may threaten up to 60% of the peripheral areas of neovolcanic regions in Slovakia, where their slope deposits comprise clay, clay-sandy, clay-stony, sandstone-rocky to rock-bearing slope sediments, and debris (Maglay & Pristaš, 2002). Slope deposits The investigated landslide (which was identified during a field survey in 2013) is located above the road extending from the district town of Zvolen to the village of Železná Breznica. It represents an active complex with a gully-related landslide system undercutting the landslide slope in its northern part ( Figure 2). Field analysis focused on its short-lived micro-scale landforms, which comprise a gully system, terrain depressions (sometimes with wet areas), minor scarps, and undulating terrain, thus indicating landslide activity and a head scarp. The terminology of the partial landslide landforms is adopted from Cruden and Varnes (1996).

| Applied technologies in data acquisition
The selection of a mapping method depends on the availability of datasets and the local context (Ahmed & Dewan, 2017). Natural conditions influence the data acquisition process (Hsieh et al., 2016;Pirasteh & Li, 2016). Forests constitute a specific environment for surveying landforms using image-based remote sensing technologies.
Although trees indeed represent an obstacle to the visibility of the

| Aerial photogrammetry performed using a RPAS (Spring, 2017)
Imagery was acquired using the Phantom 3 Professional RPAS with an average flight height of 43 m above ground level. A total of 852 images were obtained with a spatial resolution (ground sampling interval) of 14 mm. The camera had the following parameters: sensor: 1/ 2.3" CMOS; effective resolution: 12.4 M (total pixels: 12.76 M); lens: field of view 94°20 mm (35 mm format equivalent) f/2.8 focus at ∞.
The imaging material was processed using the Agisoft PhotoScan Professional 1.2.6 software 144 (as described by, e.g., Turner, Lucieer, & Wallace, 2014). The investigated area was 29,617 m 2 . Eleven GCPs were used for georeferencing, which formed approximately three slopewise lines, namely, left (4 points), middle (3 points), and right (4 points). However, the actual positions of the GCPs were dependent on the occurrence of gaps in the forest canopy.

| CRP (Spring, 2017)
A calibrated SLR EOS 5D Mark II digital camera with EF 16-35 mm f/ 2.8 L II USM from Canon was employed for CRP. A total of 1,253 images were obtained with geometric resolutions ranging from 0.65 to 1.54 mm. The camera offered a full-frame CMOS sensor (36 mm × 24 mm) with a resolution of 21.1 megapixels. The focal length was set as 35 mm, and the aperture, sensitivity, and shutter speed were adjusted depending on ambient conditions in the forest stand. We used the Gig tube Wireless II viewfinder to optimize the image axis when the camera was placed on a pole. According to the capturing method, the number of frames in the block ranged from 80 to 400. These frames were subsequently connected into a single model. The scanned area was 9,132 m 2 .
The applied photogrammetric survey was based on the principles of the 'Structure-from-Motion' method. This method operates under the same basic tenets as stereoscopic photogrammetry, namely, that 3D structures can be resolved from a series of overlapping, offset images. This approach fundamentally differs from conventional photogrammetry, in which image acquisition involves capturing overlapping photographs of multiple locations (Westoby, Brasington, Glassera, Hambreya, & Reynolds, 2012). Using a preplanned route ensures the best photo collection and avoids data losses that are common in forest environments (Chudy et al., 2014). Using this method, a complex series of detailed photos was collected from deep cracks, scarps, and gullies. Continuous imaging running under a remote control in the camera menu was set-up. Thus, the motion and rotation paths around the axis of the camera were secured. The elevated position of the projection centre was stabilized using a telescopic pole, thus ensuring a minimal distance of 3 m to the surveyed objects. The adjustment of the photographing parameters ensured that there was a sufficient depth of sharpness to create high-quality images in the sloping and rugged terrain under vegetation exhibiting irregular shading.

| Reality capturing by the Lenovo Phab 2 Pro (Spring, 2017)
For detailed surveying performed on micro-scale landforms in an area of 1,847 m 2 , the Lenovo Phab 2 Pro was employed. This device uses Google Tango technology for reality capturing. An RGB-D camera with a 'Time-of-Flight' infrared sensor, which uses a given principle to determine the distances to objects (currently, it mainly uses infrared radiation), ensures the function of 'depth perception'; the function of 'motion tracking' is enabled by the embedded sensors in the device that allow position and motion tracking; and the function of 'area learning' means that the device looks for the same objects within already existing 3D models and real space (Google, 2017). This device was handheld during the field survey, and no stabilization was used.

| The point field and GCPs (Spring, 2017)
The point field was established using coded targets as reference control points to assess the accuracy of DTMs and DSMs. Using identical check points in digital models, point clouds were transformed into the  Figure S2.
In practice, analytically driven methods typically rely on numerical methods for a complete evaluation (Heuvelink, 1998). The results of previously conducted actual error assessments of Tango technology  Table 1) were considered to be indicative of the estimated values of acceptable accuracy of the digital models in this article. The characteristics of the forest environment were very similar, and the devices employed by researchers were identical.
The root mean square coordinate errors in Table 1 were calculated as follows: where Δx i , Δy i , and Δz i are the differences between the reference coordinates and the coordinates determined from the point cloud, and n is the number of points in the set. The number of control points was greater than 30 in all the referenced studies. The RMSE x and RMSE y errors were used for the calculation of the root mean square horizontal error RMSE xy as follows:  The overall RMSEs were calculated from the mean errors acquired for individual coordinates in this article; for example, the root mean square spatial error RMSE XYH was calculated as follows: The positional (horizontal; RMSE XY ), vertical (RMSE H ), and spatial (RMSE XYH ) accuracies were calculated for individual models. Additionally, the errors associated with the methods applied to acquire the coordinates of GCPs (i.e., GNSS and total station) were also considered here.
Data gaps were visually identified and compared in DSMs obtained from CRP and the Lenovo Phab 2 Pro. The reasons for their occurrence and their characteristics are explained.

| The assessment of digital model applicability in participatory GIS
First, the data interoperability between different digital models on several scales was examined.
The DTMs and DSMs processed using different technologies were merged into a single model, thus concurrently demonstrating the interoperability and scalability of the data. Differences in the point cloud density of the scaled models enable visualizations of the study site, from a general overview to specific micro-scale landforms indicating activity (e.g., minor scarps), which may be important for landslide activity monitoring. The GNU-licenced CloudCompare software was used for this purpose. This step was possible only after all partial models were transformed into a uniform coordinate system: in this study, the S-JTSK (Datum of Uniform Trigonometric Cadastral Network) coordinate system, which is the obligatory system for surveying in Slovakia. The process resulted in some overlapping (duplicate) parts from differing methods. These parts were not edited in this study; however, future research could refine this methodology to prioritize the most precise and most complete data during this process. The least accurate dataset must be considered when determining the accuracy of the entire model.
Second, the effectiveness of data acquisition during the mapping process was compared amongst different technologies and evaluated considering the following criteria: a. costs, acquisition time, and processing time; b. data quality; and c. relevance of specific environmental conditions (e.g., forest stands in a phase of vegetation growth, rugged terrain, method interoperability, and the mapping of micro-scale landforms of active landslides).

| Comparisons of data accuracy and visibility of landslide details in digital models
Additional adjustment of the scanned parameters was impossible because LiDAR data were obtained from a contracted mapping service (specified in detail in Table 2). The average point cloud density was approximately 9 points/m 2 ; the value of RMSE XYH declared by the provider was 0.047 m. The distance of the scanning laser device (700 m) from the mapped object allowed the borders and major landforms of the landslide to be identified (i.e., a head scarp, ridges, major scarps, undulating terrain, and a gully-related landslide system; Figure 4a); however, it was impossible to identify the particular micro-scale landslide landforms being reflected.
The aerial photogrammetry performed using RPAS produced a much denser point cloud (500 points/m 2 ), while the area (29,617 m 2 ) was comparable to that covered by LiDAR; the value of RMSE XY was 0.05 m and that of RMSE H was 0.03 m. In both cases, data gaps occurred in deeply rugged terrain that is indicative of active landslides (major scarps and steep slopes of ridges; Figure 4b). CRP

| The visual comparison of data gaps in digital models
Differences in data gaps between different models were evident. The point clouds obtained from CRP showed the accumulation of data gaps in heavy rugged terrain (Figure 5a), whereas the point clouds acquired by the phablet showed data gaps that were more or less homogeneously dispersed in DSMs (Figure 5b).

| The applicability of the involved technologies to landslide mapping and measuring
The different point cloud densities in the DTMs and DSMs allow view scaling from the whole model to its details, as is illustrated by the interactive models in Figure 4. The DTMs and DSMs developed by different procedural steps from point clouds acquired by different technologies may be merged into a single model after undergoing a reliable transformation into a uniform coordinate system. Such an 'integrated' model can be further processed in the GIS environment by using its countless applications.
An analysis of the effectiveness of landslide mapping and measurement (Table 2) indicates that ALS is the most suitable technology for the mapping of 'previously known' landslides (based on maps or previous studies), primarily due to its optimal relationship between its appropriate data accuracy and costs. The Google Tango technology is suitable for mapping smaller areas and individual objects; these data can be merged with the DTMs and DSMs created from the point clouds acquired using different techniques ( Figure S3). Data obtained from CRP are appropriate for rapidly mapping and measuring short-lived micro-scale landforms representative of landslide activity; minor scarps are displayed in 3D interactive models (Figure 6a,c); a slope exhibiting undulating terrain with tension cracks under an older, stabilized landslide is presented in a 3D interactive model ( Figure 6b); a gully is displayed in Figure 6 d. CRP technology is more time consuming and expensive than the Google Tango technology. The Lenovo Phab 2 Pro is suitable for the fast and low-cost mapping of small-scale landforms in great detail (see 3D interactive model in Figure 6e, documenting a minor scarp).
Furthermore, it is suitable for mapping larger but not very extensive landforms; one example is a gully with an uncovered soil substrate in its upper part (Figure 6f). 6 | DISCUSSION

| Advantages and constraints of technologies involved in the landslide survey
Landslide deformation measurements are very important; however, these point-based measurements can be time consuming if the required data density is high (Prokešová, Kardoš, & Medveďová, 2010 (Tomaštík, Mokroš, et al., 2017). A study on Google Tango accuracy    (Hyyppä et al., 2017).
Using various integrated technologies in a field survey allows research project scalability (Marvin et al., 2016). Digital models purchased from mapping services (LiDAR scanning or RPAS survey) are usually inexpensive only for large localities and represent an easy-touse alternative to field surveys. vakia. Slope failures (see Figure 2) limit the future urbanization of the territory of the concerned municipalities. Although the studied landslide was documented in field photos in 2013, it has not been added to any database until now, and it has never been officially monitored.
Currently, it does not affect the road below its body. Nevertheless, slope movement in its upper parts (a head scarp) was documented in photos taken during the field surveys in 2013 and 2017 ( Figure S4).
Involving local authorities is particularly important during the process of anticipating landslides (Spizzichino, Margottini, Trigila, & Iadanza, 2013). Currently, most people wear tracking devices, and available tracking data are increasing (Drummond, Joao, & Billen, 2006). Measurements made using Kinect-or Tango-type systems could also be applied in a crowdsourcing context (Hyyppä et al., 2017), allowing non-experts to undertake specialized tasks for certain purposes more quickly and at lower costs (Capineri, 2016). Future landslide hazard studies require the use of multiscale and multitemporal spatially referenced data from a wide variety of sources that are shared through web-based platforms (Hou, Lu, Wu, Xue, & Li, 2017). Computer-aided territorial planning and accessibility to geographic data might support the development of breakthrough ideas in spatial planning and related decision-making processes (Matta & Serra, 2016).
In particular, when there is no evidence of the extent of a landslide observed in the LiDAR data or any photographic record of its existence, the identification of slope failure indicators (scarps, bent trees, and cracks) is highly recommended (Pirasteh & Li, 2016). The measured parameters of these landslide formations (e.g., the height of the tear-off, the depths of cracks or the depth, and width of the erosive gully through which water flows from the landslide), which are hard to capture using common geodetic methods, may be rapidly evaluated using near real-time Tango technology to determine whether the landslide is active or even dangerous. It is only a matter of time before landslide details (with data) are recorded in cloud-computing services that can be downloaded by experts or by the public.
Eventually, these data could be used for landslide monitoring or generating a 3D model from associated point clouds.

| CONCLUSIONS
In conclusion, Google Tango is one of a number of emerging low-cost 3D scanning technologies that could become a key element in allowing citizens and the community to become engaged in decisionmaking processes concerning environmental issues (Counsell & Nagy, 2017). Territorial planning is a unique tool for creating well-maintained and well-functioning landscapes. Local people are aware of the driving forces behind land degradation, and the use of GIS proves its added value in the participatory process of integrated land use planning (Hessel et al., 2009). Experts as well as the general public often go into the field and discover uncharted landslides that can be recorded using