Advanced Search

Indexed by SCI、CA、РЖ、PA、CSA、ZR、etc .

Volume 31 Issue 1
Jan.  2020
Article Contents
Turn off MathJax

Citation:

New Framework Based on Fusion Information from Multiple Landslide Data Sources and 3D Visualization

  • Recent monitoring techniques employ multiple sources of information for the characterization of the phenomenon to be studied, being the coupling and adjustment of multi-source data one of the first challenges to consider and solve. The authors propose a new framework of the multi-source and mul-ti-temporal data-oriented fusion for the characterization of landslide events. The main objective is to generate 3D virtual models (in the form of dense point clouds) and feed them back with the characteristic of soil and subsoil information. The scheme consists of three main steps. The first one is on-site data collection (geological characterization, geophysical measurements, GPS measurements, and UAV/drone mapping). The second step is generation of a high-resolution 3D virtual model (~1-inch spatial resolution) from the frames acquired through the UAV using the structure of motion (SfM) processing; the developed virtual model is optimized with GPS measurements to minimize geolocation error and eliminate distortions. The last step is assembling of the acquired data in the field and densified point cloud considering the different nature of the data, re-escalating procedure and the information stacking layer.
  • 加载中
  • Abidin, H., Heri, A., Mai, G., et al., 2004. On the Use of GPS Methods for Studying Land Displacements on the Landslide Prone Area. FIG Working Week 2004, May 22-27, 2004, Athens, Greece. https://www.fig.net/resources/proceedings/fig_proceedings/athens/papers/ts16/TS16_6_Abidin_et_al.pdf
    Al-Rawabdeh, A., He, F. N., Moussa, A., et al., 2016. Using an Unmanned Aerial Vehicle-Based Digital Imaging System to Derive a 3D Point Cloud for Landslide Scarp Recognition. Remote Sensing, 8(2): 95. https://doi.org/10.3390/rs8020095
    Alsadik, A., 2014. Guided Close Range Photogrammetry for 3D Modelling of Cultural Heritage Sites: [Dissertation]. University of Twente, Enschede
    Arosio, D., Longoni, L., Papini, M., et al., 2014. Analysis of Microseismic Activity within Unstable Rock Slopes. In: Scaioni, M., ed., Modern Technologies for Landslide Investigation and Prediction. Springer, Berlin, Heidelberg. 141-154
    Auge, M., 2008. Métodos Geoeléctricos para la Prospección de Agua Subterránea: [Dissertation]. Universidad de Buenos Aires, Buenos Aires
    Bogoslovsky, V. A., Ogilvy, A. A., 1977. Geophysical Methods for the Investigation of Landslides. Geophysics, 42(3): 562-571. https://doi.org/10.1190/1.1440727
    Carrera, H. J. J., Levresse, G., Lacan, P., et al., 2016. A Low Cost Technique for Development of Ultra-High Resolution Topography: Application to a Dry Maar*s Bottom. Revista Mexicana de Ciencias Geológicas, 33(1): 122-133
    Dong, S. C., Samsonov, S., Yin, H. W., et al., 2018. Two-Dimensional Ground Deformation Monitoring in Shanghai Based on SBAS and MSBAS InSAR Methods. Journal of Earth Science, 29(4): 960-968. https://doi.org/10.1007/s12583-017-0955-x
    Ganz, J., 1914. Die Gipfelbewegung der Rosablanche. Swiss Journal of Surveying and Rural Engineering, 21(10): 233. https://doi.org/10.5169/seals-188068
    González, N., G. A., Molina Garza, R. S., Aranda Gómez, J. J., et al., 2012. Paleomagnetismo y edad de la Ignimbrita Panalillo Superior, Campo Volcánico de San Luis Potosí, México. Boletín de la Sociedad Geológica Mexicana, 64(3): 387-409. https://doi.org/10.18268/bsgm2012v64n3a9
    Grayson, B., Penna, N. T., Mills, J. P., et al., 2018. GPS Precise Point Positioning for UAV Photogrammetry. The Photogrammetric Record, 33(164): 427-447. https://doi.org/10.1111/phor.12259
    Labarthe, H. G., Jiménez López, L. S., Aranda, J. J., 1995. Reinterpretación de la Geología del Centro Volcanico de la Sierra de Ahualulco, S. L. P
    Niethammer, U., James, M. R., Rothmund, S., et al., 2012. UAV-Based Remote Sensing of the Super-Sauze Landslide: Evaluation and Results. Engineering Geology, 128: 2-11. https://doi.org/10.1016/j.enggeo.2011.03.012
    Othaman, Z., Wan, A. W., Anuar, A., 2011. Evaluating the Performance of GPS Survey Methods for Landslide Monitoring at Hillside Residential Area: Static vs Rapid Static. IEEE 7th International Colloquium on Signal Processing and Its Applications, CSPA 2011. March 4-6, 2011, Penang. 453-459
    Pirotti, F., Guarnieri, A., Masiero, A., et al., 2014. Micro-Scale Landslide Displacements Detection Using Bayesian Methods Applied to GNSS Data. In: Scaioni, M., ed., Modern Technologies for Landslide Investigation and Prediction. Springer, Berlin, Heidelberg. 123-138
    Reshetyuk, Y., Mårtensson, S. G., 2016. Generation of Highly Accurate Digital Elevation Models with Unmanned Aerial Vehicles. The Photogrammetric Record, 31(154): 143-165. https://doi.org/10.1111/phor.12143
    Rodríguez, D. F., 2015. Estudio de Técnicas Electromagnéticas de Prospección de Subsuelo.[2019-7-27]. https://upcommons.upc.edu/bitstream/handle/2117/78151/memoria.pdf?sequence=1&isAllowed=y
    Sato, M., 2015. Near Range Radar and Its Application to near Surface Geophysics and Disaster Mitigation. Journal of Earth Science, 26(6): 858-863. https://doi.org/10.1007/s12583-015-0595-y
    Scaioni, M., 2015. Modern Technologies for Landslide Monitoring and Prediction. Springer. http://doi.org/10.1007/978-3-662-45931-7
    Stumpf, A., Malet, J. P., Allemand, P., et al., 2015. Ground-Based Multi-View Photogrammetry for the Monitoring of Landslide Deformation and Erosion. Geomorphology, 231: 130-145. https://doi.org/10.1016/j.geomorph.2014.10.039
    Telford, W. M., Geldart, L. P., Sheriff, R. E., 1990. Applied Geophysics (Vol. 1). Cambridge University Press, Cambridge
    Teixidó, T., Quintana, Á. R., 2013. Aplicación de la Tomografía Eléctrica en la Caracterización del Deslizamiento de Doña Mencía: [Dissertation]. Instituto Andaluz de Geofísica, Granada, Spain. 56
    Tian, Y. Y., Xu, C., Ma, S. Y., et al., 2019. Inventory and Spatial Distribution of Landslides Triggered by the 8th August 2017 MW 6.5 Jiuzhaigou Earthquake, China. Journal of Earth Science, 30(1): 206-217. https://doi.org/10.1007/s12583-018-0869-2
    Turner, D., Lucieer, A., Wallace, L., 2014. Direct Georeferencing of Ultrahigh- Resolution UAV Imagery. IEEE Transactions on Geoscience and Remote Sensing, 52(5): 2738-2745. https://doi.org/10.1109/tgrs.2013.2265295
    Zhong, C., Li, H., Xiang, W., et al., 2012. Comprehensive Study of Landslides through the Integration of Multi Remote Sensing Techniques: Framework and Latest Advances. Journal of Earth Science, 23(2): 243-252. https://doi.org/10.1007/s12583-012-0245-6
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(9)

Article Metrics

Article views(49) PDF downloads(7) Cited by()

Related
Proportional views

New Framework Based on Fusion Information from Multiple Landslide Data Sources and 3D Visualization

    Corresponding author: José Tuxpan, ose.tuxpan@ipicyt.edu.mx
  • 1. Division of Applied Geosciences, Scientific and Technological Research Institute of San Luis Potosi, San Luis Potosí 78216, Mexico
  • 2. Cátedras-CONACyT, National Council for Science and Technology, CDMX 03940, Mexico
  • 3. Centre for Scientific and Technological Studies, National Polytechnic Institute, Guanajuato 37358, Mexico

Abstract: Recent monitoring techniques employ multiple sources of information for the characterization of the phenomenon to be studied, being the coupling and adjustment of multi-source data one of the first challenges to consider and solve. The authors propose a new framework of the multi-source and mul-ti-temporal data-oriented fusion for the characterization of landslide events. The main objective is to generate 3D virtual models (in the form of dense point clouds) and feed them back with the characteristic of soil and subsoil information. The scheme consists of three main steps. The first one is on-site data collection (geological characterization, geophysical measurements, GPS measurements, and UAV/drone mapping). The second step is generation of a high-resolution 3D virtual model (~1-inch spatial resolution) from the frames acquired through the UAV using the structure of motion (SfM) processing; the developed virtual model is optimized with GPS measurements to minimize geolocation error and eliminate distortions. The last step is assembling of the acquired data in the field and densified point cloud considering the different nature of the data, re-escalating procedure and the information stacking layer.

0.   INTRODUCTION
  • There are historical records on the incidence of landslides; however, the first historical antecedent that considers a systematic methodology applying basic remote sensing through photographs dates to the early 1900s by Ganz (1914).

    Nowadays photogrammetric techniques allow generating a 3D reconstruction of the surface of the terrain at sub-metric scale, ideal for discriminating temporal changes on the observed surface and for having detailed information on its morphology (Reshetyuk and Mårtensson, 2016; Stumpf et al., 2015). The model is obtained from correctly oriented two- dimensional images with adequate light, position and distance conditions (Alsadik, 2014). Short-distance photogrammetry is used for the measurement of ground deformations in landslides. This approach has significant advantages such as the coverage of large areas of land using ABDM (area-based deformation measurements) methods, a high degree of automation, rapid data acquisition and generally digital photogrammetry is cheaper than 3D scanning technology (Arosio et al., 2014). The geodesic measurement techniques, for example, Global Positioning System (GPS), allow the acquisition of accurate information on the geometry and the current state of the slip position. The sampling method can be static or kinematic depending on the pre-survey valuation, obtaining a set of position data in the XYZ planes (Othaman et al., 2011). The GPS survey networks are particularly important when the magnitude of landslide is very small (Pirotti et al., 2014; Abidin et al., 2004). As a rule a landslide shows a change in the morphology of the internal structure of the affected land mass, so it is feasible to use passive seismic methods as a technique to follow the propagation of rift within a mass of rock as a result of a loading stress or water freezing (Arosio et al., 2014).

    Also, electromagnetic methods have been applied in different investigations of landslides (Teixidó and Quintana, 2013; Auge, 2008; Telford et al., 1990; Bogoslovsky and Ogilvy, 1977), mainly to determine the geometry of the phenomenon and to obtain the depths of the rock basement. The number of studies published from Ground Penetrating Radar data (GPR) has increased significantly in recent years (Scaioni, 2015; Arosio et al., 2014). This success, due to its range of operating frequencies (typically between 100 MHz and 1 GHz), allows information collection in metric/centimetric resolutions, which is useful in near-surface studies of disaster monitoring, profiling and detecting (Sato, 2015).

    Currently, unmanned aerial vehicles (UAV) offer an efficient platform for the research of landslide phenomena and constitute a novel platform for remote information acquisition. The versatility and low cost of UAV, compared to other platforms such as satellites and airplanes, make these vehicles one of the main options to be considered in research tasks. For example, images taken by a small radio-controlled multi-rotor UAV in Super-Sauze France (Niethammer et al., 2012) have been used to produce a high- resolution orthomosaic and digital terrain models (DTM) of several regions affected by landslides. Also, the ability of UAV to detect surface cracks and slip surface displacement was evaluated, as well as image processing approaches to the adequate geo- rectify process of data. This makes it possible to perform a very accurate and very-high-resolution topographic surveys, for example a digital model of the Maar crater bottom located in Parangeo, Mexico (Carrera et al., 2016), in which a methodology was used to develop a Digital Surface Model (DSM) of ultra- high resolution for observing structures caused by deformation such as slopes, fractures, and domes. In another case Al- Rawabdeh et al. analyzed the problem of the characterization of landslides with emphasis on the identification of fractures and escarpments (Al-Rawabdeh et al., 2016). InSAR techniques (Dong et al., 2018) were able to measure the deformation of the terrain and the subsidence with precision whenever the images with the appropriate resolution were available. A recent study complements the use of remote images with seismic information and work in the field to create a detailed inventory of landslides after the 2017 Jiuzhaigou Earthquake occurred on August 8, 2017 (Tian et al., 2019).

    In this work, an integrative methodology of multi-source data fusion is proposed for later applications of characterization, monitoring, and simulation of landslide movements. A dense point cloud model was combined with the implicit physical attributes of the area of interest to robustness the digital scheme. Resulting in a geo-located three-dimensional virtual scenario supported with properties of depth, resistivity, thickness and density of materials in the area, in a standard.xyz format for processing and recreation of simulations and situations that identify and prevent the occurrence of landslides.

1.   MULTI-SOURCE FUSION STRATEGY APPROACH
  • The proposal pursues the incorporation of a new processing geophysical scheme and geospatial information of sensitive areas to landslides. As shown in Fig. 1, it is based on the collection of data and geological characterization, GPS technology, subsoil measurements and the aggregation of high-resolution geospatial information acquired by UAV/drones, also supported by digital photogrammetry techniques for the typical treatment of data.

    Figure 1.  Conceptual map of the methods and tools used in the proposed methodology.

    An essential aspect of the proposed methodology is that it looks for the formalization of a virtual laboratory for simulation of landslide events. The first step reported in this article is to establish the foundations and necessary components for the adequate characterization of the phenomena. It highlights the use of multiple techniques to acquire the highest amount of available information and obtain a virtual model of the scene improved with the a priori information recollected from the measurements. It is the final objective of the second stage and of this methodology, to execute simulations of behavior-controlled landslide to identify the necessary conditions that trigger it. The methods and tools used for the development of the first stage are briefly described below.

  • The test area (Fig. 2) is located near the city of San Luis Potosí, Mexico, at 21 km on Highway 63. The region is a mountain slope prone to deformations due to moderate slopes and steep slopes, which are easily visible from the road. The landslide has significantly modified the topography and environment of the place as well as causing damage to the road that crosses the test area producing monetary losses for the state and government. The landslides present in this area have a complex behavior due to the intersections of multiple interrelated factors highlighting geological and geomorphological characteristics, soil properties, surface cover characteristics and hydrological processes (Labarthe et al., 1995).

    Figure 2.  Sketch maps showing (a), (b) the location of the study area and (c) the topographic map of the test zone with isolines every 20 m; (d) a panoramic photograph of the landslide front, the capture location is marked on the map (camera icon); and (e) two images showing road damage and fractures in the area.

2.   MATERIALS AND METHODS
  • The information supplied to the model and the interaction of the data is illustrated in stage 1 of Fig. 3. It begins with the acquisition of geophysical parameters measured through geophysical and geodesic techniques. The geological structures are characterized by geological observation, sampling and laboratory analysis. Finally, a topographic survey is carried out using a UAV for the generation of a realistic three-dimensional virtual model of very high spatial resolution (approximately 1 inch per pixel). Next, we describe the field work performed in the test area and integration of the parameters in the model.

    Figure 3.  Flowchart of the information collected and interoperability.

  • Geophysical methods are an essential tool for knowing the characteristics of the subsoil quickly and with acceptable precision (Rodríguez, 2015; Bogoslovsky and Ogilvy, 1977). The main advantages are: (ⅰ) their flexibility, speed, and robustness, (ⅱ) they are non-invasive methods and can provide information on the internal structure of the soil or rock mass, and (ⅲ) they have a functional coverage area. For the development of the methodology, electrical methods were considered, in particular, vertical electrical soundings (VES) and electrical resistivity tomography (ERT).

    Due to the complex topography of the study area, suitable sites for VES and ERT were selected carefully. Two VESs and three ERTs were carried out. The location of each survey is shown in Fig. 4 next to the corresponding map. In the case of the VESs, the AB/2 opening was 100 m, and a depth of 35 m was reached. The distance between VES 1 and VES 2 is 400 m reaching approximately 50 m in depth. The layer models basedon the resistivity of the materials are also shown in Fig. 4.

    Figure 4.  Location of the electrical soundings VES and ERT methods with their respective models.

    In VES 1, six layers were identified. Layer 1 of them has a thickness of 3 m and a resistivity of 100 Ω·m, and it is related to dry soil. Layer 2 is 1 m thick and a resistivity of 300 Ω·m, and it is associated with vitrophyre. Layer 3 has a thickness of 3.5 m and resistivity of 15 Ω·m, and it is interpreted as a volcanic gap horizon. Layer 4, which is 17 m thick with a resistivity of 300 Ω·m, is considered to belong to a vitrophyre horizon. Layer 5, which is 30 m thick and 10 Ω·m, is related to a volcanic gap. Finally, layer 6, being unable to penetrate and has a resistance of 100 Ω·m, is considered an andesitic rock. In VES 2, three layers were identified; the first, has a thickness of 0.2 m and a resistivity of 30 Ω·m that is related to dry soil; layer 2 is 28.8 m thick and a range of resistivity between 11 to 15 Ω·m associated with volcanic gap; layer 3, has a resistivity of 100 Ω·m, that is interpreted as a volcanic gap horizon.

    The prospecting sites were determined considering the significant presence of fractures and cavities in the terrain (Fig. 2), which represents an indication of the possible occurrence of landslides that may affect the lower elevation territory and the road is depicted in the map as a yellow line).

    The results of the tomography scans (ERT) of sites A, B and C are shown in the second part of Fig. 4. To characterize and obtain parameters that will feed the numeric model, three categories were established: (ⅰ) sliding material (values lower than 10 Ω·m), (ⅱ) drag material (100-500 Ω·m), and (ⅲ) fractures and cavities (resistivity higher than 2 500 Ω·m). According to this classification, ERT A shown in Fig. 4 indicates the presence of cavities and the existence of material blocks and boulders behind (rhyolite) shown in green and yellow polygons. In site B, there is a more significant presence of blocks and boulders of material and stable material; however, very high resistivities are observed in site C, indicating the presence of deep cracks in addition to the presence of material prone to move.

  • Another vital component of the methodology corresponds to the identification and characterization of the geological structures and stratigraphic units of the site to be monitored. Through geological observation and laboratory analysis, lithological materials were determined, and physical properties were obtained as the density of the rocks identified. In this stage also, geological maps were generated as preliminary elements of the identification of sites susceptible to landslides. The process used for the formulation of landslide risk mapping consisted of: (ⅰ) regional survey of the area through the synthesis of available information and the identification of general problem areas, (ⅱ) generation of cartography at local scale detailing at surface and sub-surface level the regions identified as susceptible, and (ⅲ) integration of the information collected.

    The geological study in this first stage consisted of an exhaustive exploration of the study area to find and identify the different rock units located in the area. Four different rock types were determined (Fig. 5a): conglomerate, Panalillo ignimbrite, Portezuelo latite, and Casita Blanca andesite. Samples of each type of rock were collected for laboratory characterization. Also, the fractures were located and an inventory was generated. The data was taken with the help of a topographic map, a GPS navigator and a geological compass, as shown in Fig. 5a. Figure 5b shows a geological section with the geological units of the subsoil. The oldest geological unit corresponds to the shale and limestones of the Caracol Formation. The Casa Blanca andesite is in this one. Altered materials of the Portezuelo latite are deposited on the andesitic formation.

    Figure 5.  (a) Geological map from field observations with fracture inventory; (b) geological section of the study area, the figure shows the stratigraphy and the structures that give rise to the landslides; (c) to (h) visualization of the different materials under microscope, (c) andesite, (d) altered andesite, (e) Potezuelo latite, (f) vitreous material, (g) Panalillo ignimbrite, and (h) lime-clay material. Objective: 1X, scale: 1 000 µm.

    Figure 5c shows the andesite of the Casita Blanca, considered the oldest rock in the volcanic sequence of the Ahualulco Mountain Range. The Casita Blanca andesite is formed by spills of lavas that rest on the Caracol Formation. It*s reported thickness is approximately 50 m, towards the top, the andesitic outcrops contain abundant vesicles filled mainly with calcite in the Casita Blanca, and the andesitic ones are Portezuelo latite (Labarthe et al., 1995). At the top of the stratum of the andesite outcrops, there is evidence of hydrothermal alteration as shown in Fig. 5d. Figure 5e shows the Portezuelo latite which is composed of intermediate reddish lava, with a holocrystalline porphyritic texture containing 5% to 10% of hematocrit dark-red subhedral phenocrysts, as well as 30% of subhedral crystals of milky white color (feldspar-sanidine). The matrix is hypocrystalline (devitrified quartz) with 1% to 3% dark reddish pyroxene. Also, chlorite alteration (Cl) is observed. Outcrops of vitreous material (Fig. 5f) were found within the Portezuelo Formation, this dark-greenish rock present mainly contains quartz components in the form of spherulites, in addition to feldspar and hematite oxide, supported in a vitreous matrix. According to the literatures (González et al., 2012; Labarthe et al., 1995), it is interpreted as vestiges of the vitreous shell of these volcanic structures and that in certain places can reach up to 20 m thick. This material is abundant mainly between the contacts of Portezuelo latite with other strata and has a silty-sandy structure. It is observable along the slopes and fractures of the higher parts. In the lower parts, they exhibit an extremely plastic behavior similar to plasticine, as the matrix is very tinny, and the collected samples showed high levels of humidity, revealing the great ease of the material to absorb water (Fig. 5h). This material would be behaving as a lubricant between the layers of andesite and latite, causing instability in the area. The apparent density of the rocks found in the area and shown in Figs. 5c-5h are: (c) unaltered andesite 2.68 g/cm3, (d) altered andesite 2.40 g/cm3, (e) Portezuelo latite 2.38 g/cm3, (f) Panalillo ignimbrite 2.41 g/cm3 and clay 1.46 g/cm3.

  • The proposed methodology uses GPS measurements with a dual purpose. The first is to establish a network that functions as a monitoring system that defines reference values (geolocation) and to direct the movement (orientation), and also for the establishment of ground control points for the processes of aerial digital photogrammetry. GPS measurements are performed using the real-time kinetic (RTK) method. A single base station receiver was located in a known (stable) location, connected to a radio modem and through mobile units were made displacement measurements at each vertex of the network. The distance of a vertex with its adjacent vertices is approximately 100 m.

    To monitor the test area (approximately 1 km2 sampling) a GPS measurement network was performed through 100 control points with a distance of 100 m between points. The processed network completely covered the areas and considered it as unstable, as well as the areas identified as stable becoming reference points for relative movement. The GPS data collection procedure consists of four steps: (ⅰ) identification and selection of reference points (zones without horizontal or vertical movement); (ⅱ) the linking of the selected points with reference points of the Mexican Geodetic Network and a topographic level bank belonging to the National Institute of Statistic and Geography (INEGI); (ⅲ) static measurement for 6 h at the new reference points to increase the accuracy of the measurements; and finally (iv) measurement at each vertex of the network. The spatial distribution of GPS points is shown in Fig. 6. Three measurement campaigns were carried out. The first to establish an initial state by coupling the model information, and the other two campaigns in the validation stage of the simulations of phase 2 of this methodology.

    Figure 6.  Implementation of the network of GPS control points. The height is specified in meters.

  • Aerial platforms are typically used to obtain spatial and temporal mapping with high resolution that cannot be achieved with ground or satellite systems. The spatial resolution of images acquired from an aerial platform is precisely controlled by varying the altitude of the aircraft concerning the land surface. In recent years, the use of civil unmanned aerial vehicles as remote sensing platforms has been increasing, mainly due to improved availability of accurate small GPS and Inertial Measurement Unit (IMU) systems (Turner et al., 2014), from which inertial stabilization systems are developed. Additionally, the UAV/drones are controlled by an on-board computer that allows interaction with the ground base station (UAV pilot) and attends precise instructions such as the execution of pre-programmed flights, trajectory tracking and automatic mapping of the land surface.

    For the generation of the 3D virtual environment, we used the principle of stereoscopy, which is based on the recreation of the effect of depth (3D) from images (2D) in binocular vision. Images acquired by a light DJI drone-multirotor were used. The sensor coupled to the aerial platform consists of a multispectral optical sensor with an operating range in the visible spectrum, and a resolution of 14 megapixels, the GPS unit on-board of the Phantom has an accuracy of 2.5 m in vertical direction and 0.8 m in the horizontal direction. Fifty controlled flights were performed to cover the total area. The planning and flight scheme are shown in Fig. 7. The drone used to perform the study was the Phantom 2 model.

    Figure 7.  Strategy for acquiring spatial data by drone.

    The procedure to obtain the information was through four stages, the drone used was the Phantom 2 Vision+ manufactured by DJI as shown in Fig. 7a; the image acquisition process is shown in Fig. 7b, the drone makes flights at an altitude of 50 m on average and a superposition between images of 70% to obtain a spatial resolution of 1 inch per pixel; Fig. 7c shows the segmentation/planning of the terrain by flight; and finally it displays in Fig. 7d. the ground control points used to georeference the virtual models generated (Grayson et al., 2018).

3.   RESULTS
  • The field measurements and information generated from the samples were integrated and represented in a 3D geo/spatial scheme. The pre-processing of the images as well as the elaboration of the point cloud data was carried out in the PIX4D software due to its robustness, performance, and quality of its results. The generated 3D model was saved as a standard.xyz file containing geospatial positions and texture values. The aggregation of the attribute layers (depth, the thickness of materials, density) was performed in the MATLAB software (R2015a). The workflow is as follows (Fig. 8).

  • Upload images: 2 500 images acquired by drone platform uploaded to PIX4D.

    Pre-processing data: Sensor calibration (camera positions and orientations), distortion corrections (fisheye effect), image alignment, and brightness adjustment and geolocation images.

    Dense point cloud development: Considering the principle of stereoscopy (overlapping of images) a 3D virtual model was constructed. The virtual model was represented by a point cloud with a density of ~1 478 points/m2.

    Re-optimization: Ground control points were identified and added to increase model certainty and scaling correction.

    Point cloud data export: Dense point cloud was exported in.xyz format for the incorporation of a priori information, its manipulation, and post-processing.

  • Georeferencing: Thematic maps obtained by geological observation and electric surveys were georeferenced using the virtual model acquired by the drone.

    Oversampling: An oversampling process was applied to the georeferenced maps to reach the spatial resolution of the point cloud (0.027 m)

    Layer stacking: A 3D information matrix was generated by spatially coupling the physical features/objects of the scene with its attributes (density, thickness, depth, etc.).

    Networking: Points belonging to the cloud of points under the neighbourhood criterion (neighbors nearby) generating the surface of the land were interconnected.

    Texturing: The virtual model was textured using the photographs acquired by the UAV.

    Figure 8 shows the information flow diagram and the stages implemented. In phase 1, the images acquired by UAV (1.1) were loaded. The preprocessing (1.2), where the data was corrected, was carried out. Later the clouds of dense points (1.3) were generated. These were adjusted with the control and validation points in the field through a re-optimization process (1.4) as well as the adaptation of the data in a convenient format for export. The second stage involves the assembly of the data with the initial georeferencing process (2.1) supported by the GPS monitoring network. In (2.2), scaling and oversampling procedures were applied to the data for correct spatial representativeness of the parameters considered (geology, resistivity, presence of fractures, slopes, etc.). In (2.3) the data was concentrated by stacking layers. In (2.4) an enriched surface model was generated, and finally, in (2.5) texture characteristics are added to the virtual model.

    Figure 8.  Scheme for the generation and strengthening of spatial data acquired through drone flights and geophysical/geological information.

    At the end of the processing, (ⅰ) a dense point cloud model(~1 478 points/m2), (ⅱ) a high-resolution orthomosaic (~1-inch spatial resolution), and (ⅲ) a very high-resolution digital surface model were obtained. The products derived from the process were reinforced with the physical-geological parameters of the area. Figures 9a and 9b show the orthomosaic generated with a coverage area of approximately 1 km2 and the digital surface model. These, as well as the point cloud model, represent the necessary inputs for the simulation stage in the virtual landslide laboratory. The development of the numerical model of landslide movements adapted to point cloud models (particle behavior analysis), the computational implementation, the experimental phases as well as the validation processes will be presented in another paper. The objective of this research is to establish the methodology, techniques and enough information to characterize areas prone to landslides by generating 3D models that faithfully represent the characteristics of the environment at surface and subsoil level.

    Figure 9.  High-resolution orthomosaic (a), with 1 inch of spatial resolution and (b) the high-resolution digital elevation model.

4.   DISCUSSION AND CONCLUSIONS
  • At this stage of the research, it is essential to mention that the strategy presented in this paper promotes the use and unification of multidisciplinary techniques in a set of geospatial information. The use of conventional geophysical methods, geological observations, GPS technology and cartography with UAV form a robust integrative methodology. Previously a work-oriented to the integration of different landslide monitoring technique was carried out (Zhong et al., 2012). The proposed framework was still in development at the time the results were published. The methods applied were GPS laser scanner and InSAR. In our approach, the acquired data were processed and adjusted so that they were compatible and coexisted as a whole, obtaining a new scheme or framework with higher capabilities. The formulated framework aims to create realistic models that incorporate information related to the physical, geological and mechanical properties of the different materials/layers present in the study area. The main features/ information incorporated into the virtual environment (digital terrain model) were: topographic values of centimeter precision, stratigraphic information (subsoil range ~50 m), identification of lithographic strata and determination of their depth and thickness. All these characteristics and aggregate information allow the generation of suitable products, prepared for advanced processes of simulation of sliding movements of real scenarios. These are complemented with direct data of the terrain in the superficial layer and at the level of the subsoil (up to approximately 50 m of depth), in addition to the physical and mechanical characteristics of the rock. Attributes such as types of lithology, depth, and thickness of layers. The apparent density of materials, among others, integrated into a very high- resolution virtual model provides a considerable advantage over conventional digital surface/terrain models and standard landslide monitoring techniques. In principle, the proposed methodology can be used for landslide monitoring as standard methods (GPS monitoring, InSAR technique, geological- topographic exploration). However, the real potential lies in the ability to provide, organize, adapt and standardize the relevant information of the area to be studied, presenting it in four different products that include all the data collected, packaged and ready to be exported and managed in some Geographical Information System (GIS) or programming environment. The available products are very high-resolution orthomosaic (1-inch approximately), digital surface/terrain model (1-inch spatial resolution) and dense point cloud model (~1 478 points/m2). Being the last product of particular interest, since the dense point cloud model is the elementary input information for the processes of simulation and recreation of sliding events and find the necessary conditions and causes that could trigger their occurrence, allowing to think that the fusion of paradigms of measurement of geophysical parameters. The use of drone technology as well as advanced techniques of photogrammetry and digital signal processing represents a new alternative for the study and knowledge of the phenomenon of landslides.

ACKNOWLEDGMENTS
  • We are grateful to the editors for editorial handling and the anonymous reviewers for their critical and constructive comments that have greatly improved the quality of this paper. This study was supported by the CONACYT Academic Fellowship (No. 308896). The final publication is available at Springer via https://doi.org/10.1007/s12583-019-1243-8.

Reference (25)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return