10:45 AM - 12:15 PM
[SVC31-P02] Quantifying surface activity on Mt. Tokachi by image analysis using UAV photogrammetry and principal component analysis
Keywords:UAV, Photogrammetry, Image analysis, Principle Component Analysis, Mt. Tokachi
Mt. Tokachi in central Hokkaido has had no significant eruptive activity for more than thirty years since its eruption in 1988. However, since around 2006, phenomena of volcanic unrest have been observed, including ground deformation indicative of inflation in the shallow part of the crater and magnetic field changes suggesting heat accumulation. In recent years, the change in fumarolic activity and the enlargement of vegetation death areas around crater 62-2 have been observed, suggesting that surface volcanic activity is increasing.
In this study, UAV photogrammetry and image analysis using Principal Component Analysis (PCA) were used to quantify the surface activity of Mt. Tokachi.
On 30 August 2022, we conducted UAV observations in the geothermal anomaly area of Mt. Tokachi, capturing 2,323 visible images. The structure from Motion (SfM) multi-view stereo photogrammetric technique (Agisoft Metashape Pro) created a three-dimensional terrain model, Digital Elevation Models (DEM), and an orthomosaic image (resolution: 2.96 cm/pix, coverage area: 1.08 km2). The area of surface activity, such as the area of vegetation death around the 62-2 crater, was included in the photographic coverage.
The accuracy of the DEM was verified by comparing the elevation values with the DEM produced by a laser survey carried out by Geospatial Information Authority of Japan in 2009. The elevation values in the northern part of the observation area, where ground control points (GCPs) were installed, were almost identical. However, in the southern area, where no GCPs were installed, the elevation values of the in-house DEM tended to be higher than those of the GSI DEM. This error in elevation values may have resulted from an uneven placement of the GCPs and may require uniform placement of GCPs throughout the observation area in future observations.
By applying PCA to the RGB values of the generated orthomosaic images, areas of optical anomalies (principal component values > mean + standard deviation) were extracted from each principal component (PC). PC1 showed a high contribution rate of 96.6%, and eigenvectors were obtained that were almost equal to the line connecting black (RGB: 0, 0, 0) and white (RGB: 255, 255, 255) in the RGB space. In PC2, white-to-grey pixels indicative of valley stripes were detected as areas of optical anomalies. The results are consistent with the findings of Muller et al. (2021), who performed a similar analysis on the orthomosaic of La Fossa crater on Vulcano Island, Italy. Thus, whitish pixels may be more likely to be recognized as areas of optical anomalies in PC2. The areas of optical anomalies in PC3 were found to correspond to zones of vegetation death. When PCA has applied again to the RGB values of the areas detected in PC3, the results showed that the areas of vegetation death were grouped into several small regions of optical anomalies. These clusters of optical anomalies may reflect volcanic activity in the shallow subsurface and warrant investigation for surface temperature and shallow subsurface resistivity.
Our results suggested that the use of photogrammetry and PCA in image analysis holds the potential to quantify surface activity in Mt. Tokachi accurately. In particular, the eigenvector of PC3 can serve as a marker and discriminator in identifying vegetation death areas. Through subsequent observations and analysis, there is the possibility of determining the changes in vegetation death areas both spatially and temporally. The application of PCA to orthomosaic imagery can be performed easily within ArcGIS, a well-established geographic information system. Therefore, the analysis is effective in conducting volcanic research using UAV photogrammetry.
In this study, UAV photogrammetry and image analysis using Principal Component Analysis (PCA) were used to quantify the surface activity of Mt. Tokachi.
On 30 August 2022, we conducted UAV observations in the geothermal anomaly area of Mt. Tokachi, capturing 2,323 visible images. The structure from Motion (SfM) multi-view stereo photogrammetric technique (Agisoft Metashape Pro) created a three-dimensional terrain model, Digital Elevation Models (DEM), and an orthomosaic image (resolution: 2.96 cm/pix, coverage area: 1.08 km2). The area of surface activity, such as the area of vegetation death around the 62-2 crater, was included in the photographic coverage.
The accuracy of the DEM was verified by comparing the elevation values with the DEM produced by a laser survey carried out by Geospatial Information Authority of Japan in 2009. The elevation values in the northern part of the observation area, where ground control points (GCPs) were installed, were almost identical. However, in the southern area, where no GCPs were installed, the elevation values of the in-house DEM tended to be higher than those of the GSI DEM. This error in elevation values may have resulted from an uneven placement of the GCPs and may require uniform placement of GCPs throughout the observation area in future observations.
By applying PCA to the RGB values of the generated orthomosaic images, areas of optical anomalies (principal component values > mean + standard deviation) were extracted from each principal component (PC). PC1 showed a high contribution rate of 96.6%, and eigenvectors were obtained that were almost equal to the line connecting black (RGB: 0, 0, 0) and white (RGB: 255, 255, 255) in the RGB space. In PC2, white-to-grey pixels indicative of valley stripes were detected as areas of optical anomalies. The results are consistent with the findings of Muller et al. (2021), who performed a similar analysis on the orthomosaic of La Fossa crater on Vulcano Island, Italy. Thus, whitish pixels may be more likely to be recognized as areas of optical anomalies in PC2. The areas of optical anomalies in PC3 were found to correspond to zones of vegetation death. When PCA has applied again to the RGB values of the areas detected in PC3, the results showed that the areas of vegetation death were grouped into several small regions of optical anomalies. These clusters of optical anomalies may reflect volcanic activity in the shallow subsurface and warrant investigation for surface temperature and shallow subsurface resistivity.
Our results suggested that the use of photogrammetry and PCA in image analysis holds the potential to quantify surface activity in Mt. Tokachi accurately. In particular, the eigenvector of PC3 can serve as a marker and discriminator in identifying vegetation death areas. Through subsequent observations and analysis, there is the possibility of determining the changes in vegetation death areas both spatially and temporally. The application of PCA to orthomosaic imagery can be performed easily within ArcGIS, a well-established geographic information system. Therefore, the analysis is effective in conducting volcanic research using UAV photogrammetry.