Application of terrestrial laser scanning for detection of ground surface deformation in small mud volcano (Murono, Japan)

Geodetic study often requires highly accurate and precise measurements of ground surface positions, and several approaches have been developed for the detection of ground surface deformation including leveling, total station and global satellite navigation system (GNSS) (e.g., Hubbard and Glasser 2005; Mottershead et al. 2008; McCormac et al. 2012). However, such previous approaches have often been limited to some specific points of measurement, and if one would like to know the characteristics of spatially variable deformation of ground surface, increasing the density of measurement points could be one solution which requires significant time of measurement in the field, or one needs to introduce InSAR (e.g., Antonielli et al. 2014).

Among else, terrestrial laser scanning (TLS) has been an emerging approach for the measurement of spatially varying topographic changes in environmental sciences (Heritage and Hetherington 2007; Heritage and Large 2009). TLS enables accurate measurements of highly dense point cloud on the ground surface, and detection of changes in surface morphology is feasible when multiple time series of measurement data are obtained. Although the accuracy assessment is crucial for determining the level of detection (Lane et al. 2003; Milan et al. 2011), the validity of application of TLS for geodetic measurements would be promising (DeLong et al. 2015).

Here, we examine the applicability of TLS measurement for an active mud volcano where both gradual and abrupt surficial changes have been observed with precise measurements using the leveling method (Kusumoto et al. 2014, 2015). Topographic features are explored from the high-definition data by TLS, and changes in ground surface elevation are examined by multi-temporal TLS measurements. Because the TLS approach generally provides morphological information of the surface of a target object (Heritage and Large, 2009) and not for its internal structures, here we focus on the surficial deformation of the mud volcano, whose amount is on the order of centimeters in a year (Kusumoto et al. 2014).

As noted, this third measurement dataset (141204) was set as the reference, and each dataset was aligned to one following dataset. The computed error values of the ICP registrations, expressed as an average distance between nearest points in point clouds on the overlap areas (hundreds to thousands for each pair), for the internal registration were 2.9–7.9 mm, while those for the external registration procedure were 6.2–15.7 mm (Table 1).

The cracks on the paved ground surface found in June 2011 were mapped (Fig. 3). The average direction of the crack lines was found to be 37.5°W.

Differences in elevation values of the DEMs were computed for each period (Fig. 4). Centimeter-scale vertical changes of the ground surface appear to be spatially variable. Positive changes in elevation for Period I are apparent around the western side of the cracks in 2011 (up to ca. 80 mm), whereas some negative changes are also found in the eastern side. Because the paved area with cracks was partially repaired after the 110623 measurement, these patchy decreases in elevation may have derived from such artificial modification. Nevertheless, the apparent increase in elevation is more likely natural changes because it is not realistic to fill up the cracked surface for the repair purpose. Also, anomalously large changes around the corners of the target area are supposed to be the interpolation errors due to insufficient point density, and they are ignored. For Period II, increase in ground surface elevation is clear around the central areas with previous cracks with up to ca. 60 mm, while the easternmost side shows negative changes in the ground surface elevation.

Figure 5 illustrates the topographic profiles of the target area at the different times. In section A–A’, the ground surface elevation seems to have continuously increased since 2011. However, the section B–B’ shows decrease in elevation in 2013, which again shows increase in 2014. Such contrasts of the western and eastern sides are also shown in the section C–C’.

To find the feasible level of detection of the vertical displacements in the ground surface of the mud volcano, the accuracy of the TLS measurement should be carefully assessed. In general, the measurement accuracy by TLS may be affected by (1) specification of scanner, (2) internal registration, (3) external registration and (4) filtering (Heritage et al. 2009; Hayakawa and Oguchi 2016). The assessment of spatial variability in such errors is also essential for areas with local topographic variations (Heritage et al. 2009; Wheaton et al. 2010; Milan et al. 2011). However, the area of this study has less variation in topographic condition with a smooth, paved ground surface. Because of this condition without significant spatial variability in errors due to the local topographic anomalies, we only consider uniform level of detection for centimeter-scale changes within the target zone.

Among the error sources listed above, the scanner-derived error may increase with the elongation of laser footprint due to lower hit angle of laser beams on target objects, as well as the bad reflectance of target material. As noted, the amount of this is given to be 1–4 mm for the scanners used. Regarding the internal registration errors, the cloud-based ICP refinement gave 2.9–7.9 mm of fit after repeated refinements (Table 1). Compared to the average point spacing of the point clouds obtained (7.8–50.9 mm), these values are small enough, whereas the external registration errors after the cloud-based refinement were 6.2–15.7 mm, which are twice as large as the internal registration errors. It is common that external registration gives significantly larger errors than scanner-derived and internal registration errors (Pfeifer et al. 2007), because the point clouds at different times may contain large amount of different features that are natively different. Nevertheless, the external registration errors observed in this study fall in a reasonable range, which is smaller than the average point spacing (7.8–50.9 mm). Summarizing the error values above as the square root of the sum of squares of maximum error values for each, the overall accuracy of the TLS measurement is supposed to be 18.0 mm. This would be small enough to detect centimeter-scale changes over 20 mm in the land surface.

In addition, conversion from the point cloud to DEM can also produce some more errors due to inappropriate interpolation between largely spacing points, i.e., in areas with low point density (Heritage et al. 2009). Although such areas are mostly excluded from the analysis of this study, some errors could have remained particularly around the edge of the data extent and at the bottom of scanner where point data cannot be obtained. Because such errors are small and visually identifiable (marked by “e” in Fig. 4), they are simply ignored as noted before.

These accuracies or errors indicate the level of detection plausible by these TLS-derived datasets. Because the error values listed above are maximal in three dimensions including XYZ components, the error values in one component can be similar to or smaller than those values. In Fig. 4, we showed changes over ±20 mm of Z component, i.e., vertical displacements, and this was enough to represent the actual vertical displacements which often exceeded this level. Although changes less than a centimeter may be difficult to detect by this method, the TLS approach effectively enables centimeter- to decimeter-scale topographic changes.

Compared to the leveling survey, one of the advantages of the TLS measurement is that it can obtain significantly dense points over an area. Because the leveling survey is based on measurements of fixed points whose density is generally sparse, it cannot provide information for a certain distance between points (at least several meters in the case of Kusumoto et al. 2014, 2015), and interpolation is necessary to obtain the intermediate information, whereas the dense point cloud data by TLS (7.8–50.9 mm in this study) can provide more aerial information, which enables to show spatial distribution of surficial displacements with direct measurement (Fig. 4). Topographic profiles can also be utilized to help the interpretation of the changes (Fig. 5).

Artificial modification (repair of pavement) after the earthquake-derived deformation has partially applied in the study area. The area of such artificial modification is clearly shown in the data, i.e., the surface in the eastern side of the target area has significantly been smoothed in Period I (Figs. 4a, 5a, c). In this side, the rough surface with cracks found in 2011 has been smoothened and the elevation decreased. Such artificial modification has also been applied to the western side of the target area where the cracks were smoothed (Fig. 5a), although the uplift is more active in this side as noted later.

The deformation of Murono mud volcano has been recognized to be particularly active when large earthquakes occur in this region (Onishi et al. 2009; Kusumoto et al. 2015), while it has been deformed even when without such strong earthquakes (Kusumoto et al. 2014). Kusumoto et al. (2014) reported +26 to −14 mm uplift and subsidence by means of leveling survey in June to December 2012, while no particular earthquake was observed in this period. They also reported seasonal variance in the uplift and subsidence throughout the measurement period. Such seasonal fluctuations, as well as the differences in the covering time span [Period I: June 2011 to December 2013, which is much longer than that by Kusumoto et al. (2014)] hinder direct comparison of the amount of displacement. Nevertheless, the spatial pattern of uplift and subsidence within the target seems to correspond well with those reported by Kusumoto et al. (2014) (Fig. 4a): The spatial distribution of the large (≥40 mm) uplift seems to follow the displacement in 2011 as shown in the crack distribution (Fig. 3). This suggests that the activity of the mud volcano similarly continues even after the 2011 earthquake. The amount of maximum total uplift of ~80 mm in 2.5 years of Period I corresponds to some extent with that in the half a year in 2012 (~26 mm; Kusumoto et al. 2014) if they are normalized over time. Such gradual changes are therefore supposed to be accumulated during the quiescent phase.

When the large earthquake in 2014 (Nagano-ken Kamishiro Fault Earthquake) hit the region, the vertical displacement of the Murono mud volcano area was more than 50 mm (Kusumoto et al. 2015). Period II of this study covers the earthquake event, and the locus of marked uplift observed in this period (Figs. 4b, 5) corresponds well with the point having the highest uplift by the earthquake (Kusumoto et al. 2015). Because the maximum amount of uplift in this Period II (ca. 60 mm) is almost the same as or slightly larger than that by the earthquake, it is plausible that the changes in Period II comprise both the activities by the earthquake and in the quiescent phase. Furthermore, the spatial distribution of vertical displacements in this period exhibits oval pattern (Fig. 4b), and the direction of its major axis also corresponds well with those of the cracks (37.5°W) found in 2011. Therefore, the activity pattern of the mud volcano by the 2014 earthquake is supposed to be similar to that by the 2011 earthquake. It is also noted that the largest uplift is observed in the western side in Period I (P in Fig. 4a), while that is found in the slightly eastern side in Period II (Q in Fig. 4). This may indicate that the central position of the activity of the mud volcano is being shifted from the west to east side within the target area.

The high-definition data were also quite useful for accurate mapping of small topographic features on the ground including the cracks in the pavement (Fig. 3). Such distinct cracks are well shown in the profiles (Fig. 5), whose depths are identifiable to be 100–200 mm. Although we only focused on the vertical displacements of the ground surface based on DEM analyses, it would also be possible to detect changes of horizontal components if such comparable feature extraction is effectively performed.

In this study, we demonstrated the effectiveness of TLS measurements for the detection of ground surface changes due to underground activities of a mud volcano. The results of error estimates, as well as good accordance with the leveling survey, support the validity of TLS measurements to detect centimeter-scale changes. Repeated measurements will also be taken to further examine long-term activities of the mud volcano. As noted, the TLS approach only provides information on surficial morphology and not internal structures, and mechanical details of the mud volcano remain to be examined. Combinations with other geophysical data will therefore allow us to unravel the subsurface mechanisms of the mud volcano.