Assessing tree vigor condition in Prunus species
using airborne lidar and hyperspectral remote sensing data
Youngkeun SONG
Key Words: tree health, tree vitality, vegetation indices, radiometric correction, Spectral Angle Mapper, canopy surface model
1.
INTRODUCTION
Tree is a critical element to manage environment. To survey tree vigor condition, ground work has traditionally been performed, but costly, time-consuming and laborious. In this respect, remote sensing technology proposes a way to collect trees¡ information more efficiently in wider areas. Hyperspectral remote sensing detects target¡s attribute ranging from visible to infrared region with narrow spectral band widths. Lidar (light detection and ranging) sensor measures three-dimensional information. Airborne platform can observe individual trees with high spatial resolution. This data set provides useful information to manage Prunus species which is one of the most popular trees but concerned about the declination at many places in Japan.
In this paper, we focused on (1) extracting more precisely corrected information in individual tree spectra using Lidar data and (2) developing a method to assess tree vigor condition in Prunus species from hyperspectral data..
2.
MATERIALS AND METHODS
We collected airborne hyperspectral and lidar data at Yamasina, Kyoto, on September 7, 2003, and ground survey data of 44 Prunus yedoensis and 22 Prunus jamasakura. To correct the unfavorable illumination variation induced by tree geometry and illumination angle, topographic correction methods were applied to canopy surface model generated from Lidar data. After tree pixels were calibrated to contain only the information of tree vigor condition, they were analyzed by Spectral Angle Mapper (SAM) algorithm as well as existing vegetation indices. The results were evaluated by correlation analysis with ground survey, and accuracy of classification (rank A to E) in discriminant analysis .
3.
RESULTS AND DISCUSSION
(1) After the topographic correction, coefficient of determination (r©) for regressions between incidence angle and reflectance of canopy area, decreased to approximately zero; Errors caused by the irregular shape of tree were successfully corrected.
(2) SAM results showed higher correlation than the conventional vegetation indices with tree vigor condition assessed by the ground survey; Comparing with SAM, the vegetation indices appeared to be inferior because of losing information of the other bands except selected 2 or 3 bands.
(3) SAM predicted tree vigor condition had 77.3% of overall fuzzy accuracy, which includes "correct" and "acceptable" ratings; Extremely good or poor rank (A and E) showed higher accuracy than intermediate levels (B,C and D) which have larger variation.
4.