Forest stand variables (mean diameter, mean height, stand volume, stand density, and stand carbon stock) were estimated in manmade coniferous forest stands comprising Hinoki (Chamaecyparis obtuse) and Sugi (Cryptomeria japonica), using low-density light detection and ranging (LiDAR). LiDAR data were obtained on a transect in western Shikoku Island using 2 pulses per square meter. We chose young to old plots along the transect and measured forest stand variables. LiDAR indices were derived for the first and last pulse of a digital canopy height model : average ; maximum ; coefficient of variation ; 10, 20, ..., 90 percentiles ; and 10, 20, ..., 90% canopy density. A logarithmic multiple regression analysis, a linear single regression analysis with variable selection, and a vegetation profile method were employed to compare LiDAR indices and forest stand variables with Root Mean Square Error (RMSE). Among those methods, the linear single regression was the most precise in terms of average height (RMSE : 1.5 meter), and the logarithmic multiple regression was most precise for mean diameter (RMSE : 3.1 centimeter), stand density (RMSE : 837 trees per hectare), stand volume (RMSE : 68.2 cubic meter per hectare), and stand carbon stock (RMSE : 18.0 ton per hectare). Although the vegetation profile method was less precise for stand volume (RMSE : 105.7 cubic meter per hectare), the slope of the vegetation profile equation was stable in the results.
Condition of vegetation is important factor for considering possibility of slope failure, as well as condition of topological structure such as slope degree and convexity. Previous studies have revealed that roots of trees might prevent erosion of surface soil which leads to slope failures. There are three factors that might be related to slope failure prevention, thought to be able to be acquired from LIDAR (Laser Imaging Detection and Ranging) data. That is height of trees, number of trees and breast high diameter. These factors were previously very hard to acquire widely as field survey was the only way to get accurate data, and data collecting area was limited. However, accumulation of LIDAR data has made it possible for wider area simultaneously. This study showed that height of trees can be obtained more correctly from LIDAR data in a state of relatively harmonized with possibility of slope failure occurrence. Also assuming number of trees is possible when considering the number might be estimated smaller in steep slopes compared to gentle slope. In contrast with those, possibility of acquiring information relating breast high diameter might be different among type of trees. Thus, while assuming accumulation of cross-sectional area at breast height were seemed to be acquirable at evergreen needle leaf forests such as plantation of cedars, those assumption were almost impossible at deciduous broad leaf forests such as secondary forests of Quercus serrata in the case of Izumozaki district where many slope failures brought by heavy rainfall in 2004.
Currently, a DEM (Digital Elevation Model) of high density range has been able to be created by spread of airborne laser scanner. However, when the DEM is created, the points of the DEM can not be interpolated accurately in some cases. Therefore, we propose a method of accurately interpolating the points of the DEM. The proposed method is performed by using joint singular value decomposition. Through the experiments using the DEM of 5m mesh, we demonstrate the effectiveness of our interpolation method.