HYDELiDAR DTM - A digital terrain (elevation) model estimated from the 2004 leaff-on sparse LiDAR

Fig. 1. The coverage of the DTM is approximately 1.5 x 6 km. The Lapinkangs glacial delta is in the south. Most values in the DTM are "N/A". The HSV-coloring ranges from 140 m to 198 m.

Back to index,   Next

The DTM was made using Terra-software modules. The algorithm that filters terrain hits from the rest of the LiDAR data is described in Axelsson (1999, ISPRS JPRS). It has three major parameters that affect the performance. These were tried in 100 different combinations, and the best DTM was selected using some 8300 test points in Hyytiälä, which were tree butts in 27 permanent sample plots, where trees had been measured using a tacheometer.

The ground points formed a TIN in TerraModeler, which was further interpolated into a grid of 1-m resolution. This was made into a binary file that has the N60-elevations in single precision (4 byte) decimal numbers. Because of the high resolution (4054 x 9855 cells), the file size is 156 Mbytes (Fig. 1).

Residuals of the DTM under canopy
Fig.2. The residuals (x N60) of 8329 test points in 27 permanent sample plots representing stands with tree heights varying from 12 to 30 meters. The LiDAR-DTM was 0.14 m below the test points and the overall RMSE was 0.27 m. Note that part of this inaccuracy is due to field measurement errors in the reference values. The actual accuracy of the DTM is therefore better, app. 25 cm, but there seems to be underestimation of N60-elevations under canopy, whereas in seedling stands the LiDAR DTM is slightly above the true surface. This is likely due to the forest bottom flora that has its succession and phenology. The 2004 LiDAR that was used for DTM estimation was done in August, under leaff-on conditions.

Errors in one seedling stand
Fig. 3. Example of DTM errors in a 2-yr old seedling stand (250 x 100 m). Two small hills are missing from the DTM. The reference elevations here were obtained using VRS-GPS positioning.