Study area
The study site is an active mud volcano in Murono area, Niigata Prefecture in north-central Japan (Fig. 1). The mud volcano locates at 316 m a.s.l. near an anticline limb of the Naradate syncline (e.g., Noda 1962), with a small coverage area of approximately 130 m × 180 m. The surrounding area is characterized by many fold structures in the NE–SW direction and is known as one of the most important petroleum-producing areas in Japan. The mud volcano has therefore been attracted many interests for geophysical and geochemical researches (e.g., Onishi et al. 2009; Shinya and Tanaka 2009; Suzuki et al. 2009; Etiope et al. 2011). The domain of the mud volcano is used as a practice course for automobile driving, and many areas are paved but others are covered with vegetation including grass and trees (Fig. 1b).
Significant deformations of the mud volcano have been observed by several large earthquakes in the area. Onishi et al. (2009) reported a vertical movement of up to 400 mm in the area before and after the Niigata-ken Chuetsu-oki Earthquake in 2007 (Mj = 6.8) whose epicenter is 44 km away, using a GNSS survey. Such remarkable vertical deformation was also observed by the North Nagano Prefecture Earthquake (Mj = 6.7, epicenter 16 km away) in March 2011 (Matta et al. 2012), as well as by the Nagano-ken Kamishiro Fault Earthquake (Mj = 6.7, 76 km away) in November 2014 (Kusumoto et al. 2015). Furthermore, although the amount is less than those caused by large earthquakes, inter-seismic uplift of the mud volcano has also been observed by precise leveling survey at approximately 50 mm (Kusumoto et al. 2014).
However, such vertical movements have only been observed at sparse points by GNSS or leveling survey, and the detailed spatial characteristics of the deformation have been unknown. The dense measurement by TLS is therefore applied to reveal the spatial distribution of vertical displacements in the study area. The target of measurement is approximately 50 m × 60 m zone (Fig. 1b,c), where vertical displacement has been frequently occurring (Kusumoto et al. 2014, 2015).
Data acquisition and processing
On-site measurements using TLS were taken three times: June 2011 (110623), December 2013 (131205) and December 2014 (141204), where the date of survey is expressed as YYMMDD. The first measurement (110623) was just after the North Nagano Prefecture Earthquake occurred on March 12, 2011, whereas the third (141204) was after the Nagano-ken Kamishiro Fault Earthquake on November 22, 2014. There was no significant earthquake between the first (110623) and second (131205) measurements.
For the first measurement (110623), a medium-range, time-of-flight-type terrestrial laser scanner GLS-1500 by Topcon Co. was used (Fig. 2a). This device has an ability of measuring up to 500 m distance with a 1-mm minimum pitch of laser pulse, with a scan frequency of 30,000 points per second (Topcon 2010). The range accuracy of the laser measurement is 4 mm within 150 m of distance. The weight of the device is 16 kg (without batteries). Initial data management and basic processing of the point cloud obtained with this device are performed using Topcon ScanMaster v.2.0, which is bundled with the scanner. For the following measurements (131205 and 141204), we use a lightweight, phase-based short-range scanner Trimble TX5. This device is more suitable for such short-range measurements with denser point acquisition with an ability of measuring up to 120 m in distance at a scan frequency of up to 900,000 points per second (Trimble Navigation Limited 2012). The range accuracy of the laser measurement is 0.3–1.1 mm at 10–25 m of distance from scanner. The weight is only 5 kg. Trimble RealWorks v.8.1 is used for the initial data processing of point clouds by TX5. The RealWorks software enables not only the initial processing of the raw point cloud data but also the post-processing including cloud-based registration as noted later.
For each measurement, scanners were placed at multiple positions (4–7 for the target zone of 50 m by 50 m) to obtain the data from different sight of views. The point clouds from different scan positions are then registered to each other by several reference targets, as well as by key morphological features in the point cloud without positional changes such as tree trunks and poles in the surrounding areas. For the matching of the key features, cloud-based registration is applied based on the iterative closest point (ICP) algorithm built in the RealWorks software (Besl and McKay 1992; Bergevin et al. 1996). In this algorithm, one point cloud is fixed as a reference, and another cloud is transformed (shift and rotation) to best match the reference. Overlap areas with the same morphological features for both the reference and moving clouds are necessary to perform the matching. The transformation of the moving cloud is iteratively refined to minimize the distance between closest points in the moving and reference point clouds. The aligned point clouds are finally merged after achieving the least error of the cloud-based registration (at millimeter scales). Such a cloud-based registration method has an advantage in that many tie points (registration targets) are not necessarily set. This process is hereafter referred to as “internal registration.”
The merged point cloud is then georeferenced using target references whose geographical coordinates are obtained by GNSS measurement (in UTM Zone 54N projection). Checkerboard-type flat targets made by black and white tapes were placed on the paved ground, whose central point was identified in the TLS point cloud using the RealWorks software. The GNSS positioning data are post-processed by carrier-phase correction using a nearby base station data of GEONET (GNSS network in Japan) provided by Geospatial Authority of Japan. Accuracy of the GNSS measurement is on the order of centimeters with the fix solutions after post-processing.
The point clouds obtained at different times can be roughly aligned to each other by the georeferencing process above. However, since some centimeter-scale errors may remain in the GNSS positioning data, the registration is further refined based on the cloud-based registration with the same, unchanged key features in the point cloud using the ICP algorithm, which can reduce the errors down to millimeter scales (e.g., Pesci et al. 2007; Teza et al. 2007). The third measurement dataset (141204) was set as the reference because it showed the best fix-solution accuracy in GNSS positioning for the georeference: 12.1 mm in XY and 23.1 mm in Z directions with 11 GCP targets. Each of the other datasets is successively aligned to one following dataset. To perform the cloud-based ICP registration, it is important to find overlaps of the same objects in two clouds. The main target area of measurement is therefore cropped out, and stable, unchanged features in surrounding areas are used for the registration. For the best registration accuracy, the ICP registration procedure was repeatedly applied by limiting the overlap areas: Once the moving cloud is registered to the reference cloud, the points that are significantly away from the overlap areas of the moving and reference clouds are excluded, and the ICP registration is again applied. Because the exclusion of points out of the overlap areas forces the remaining points to represent the unchanged features more clearly, repeated application of this process enables to significantly refine the registration. This process is hereafter referred to as “external registration.”
The raw point cloud contains some unnecessary points, which can be derived from temporally located materials on ground (e.g., tripods, antenna rods and bags) and/or materials passing through the laser pulse (e.g., walking person, raindrops, birds and bugs). These anomalous, noisy features are readily removed either by automatic filtering or visual inspection because the presence of those materials is not so frequent during the measurements.
Digital elevation models (DEMs) with a fixed grid size are then generated from the point clouds. The resolution of DEMs is determined based on the spatial density of point cloud data. Although the linear interpolation by triangular irregular network is applied to fill voids, the uncertainty of the interpolation for the areas with insufficient point density is minimal because the target zone is selected with enough point density. Because areas apart from scanner position often have insufficient density of point cloud, those areas are excluded from the following analyses by setting a mask. The mask is also applied to vegetation areas covered by low-height (<40 cm) plants where the ground surface is rarely detected (Fig. 1c). Within the domain of the analysis, three section lines are set along which topographic profiles are extracted from the DEMs (Fig. 1c). The target area after the masking (2038 m2) is therefore smaller than the original point cloud extents.
The difference of DEMs is then computed. The periods of comparison are defined as below: Period I, 110623 to 131205; and Period II, 131205 to 141204. We use ESRI ArcGIS 10.3 software for the DEM data processing.
Since the bulging ground surface has exhibited apparent cracks at the time of measurement 110623, the cracks are traced using the generated DEM of this time. To support the tracing, crack features are highlighted by hillshade image and local variation of elevation (3 × 3 cell statistics) calculated from the DEM.