A heavy rain accompanied by a front hit Saga prefecture in Kyushu, Japan in August 2019, flooded the Rokkaku river and others, and brought an inundation disaster nearby. Geospatial Information Authority of Japan (GSI), which is one of the designated government oraganizations enacted by the Disaster Countermeasures Basic Act, took aerial photographs and produced inundation-depth estimation maps to support activities of relevant disaster management organizations. These products were also published on GSI's website for general use. This article shows major GSI's activities, especially for providing the geospatial products that are helpful for flood disaster management. In addition, outreach activities to government and public organizations that conduct disaster response activities are also reported.
We performed comparative verification of accuracy of dense cloud produced by UAV aerophotos to reveal the effect of the difference of absolute orientation methods. The methods of orientation are, (1) coordinates of GCP and platform GNSS are taken into aero-triangulation, (2) coordinates of GCP and platform GNSS are used for rigid transformation of dense cloud that was produced first. And in each methods, we produced dense clouds in two ways : whether to use GCP or not. Verification of accuracy was done by comparing of tree locations obtained from LiDAR data and each dense clouds. The result of horizontal accuracy of method (2) was significantly inferior to method (1). After projective transformation by using picked-up tree coordinates, horizontal accuracy of method (2) was greately improved. In terms of height accuracy, the error is under 10% of tree height in method (1) with GCP. But in method (1) with no GCP, overall deviation occurred. In method (2), affect of non-linear distortion of the dense cloud was suggested.
We compared estimated location precisions from different coordinate decision methods using dense cloud point data acquired by UAV derived images. We adopt four coordinate decision methods. The first method takes GNSS coordinates of reference marks and the platform into account when we estimate locations of images. In our second method, we generated dense point cloud first, and assigned coordinate values using GNSS coordinates of reference marks and the platform. For both methods, we added two variations by removing the reference mark coordinates. With these four methods, we generated dense point clouds and estimated tree locations and heights. We obtained tree locations and height data from airborne LiDAR data separately to investigate location precisions from the four decision methods.
We found that the coordinate decision methods using GNSS coordinates of reference marks and the platform superior to the decision method estimate coordinates afterward. However, we observed that we could remove systematic errors in the latter method by applying a projective transformation based on tree locations. The 95% of tree heights were within 10% error levels when we use both reference point coordinates and GNSS coordinates. When we remove reference point coordinates from our calculations, error levels became higher. We also found that we introduced additional errors caused by non-linear distortion when we estimate coordinates afterward.