1 INTRODUCTION
Tree height is considered one of the most useful variables, along
with stocking and diameter at breast height, in estimating forest
stand wood volumes and productivity. It also determines the light
penetration in the forest canopy and is of importance for certain
habitat studies. It is however one of the most difficult variable to
measure when forest covers are dense. Ground measurements can
take a few minutes per tree and are error prone. Alternatives
include automated photogrammetric approaches that consist in
finding the difference between tree altitude and nearby ground
altitude using stereo comparison. Because seeing the ground is of
critical importance, good results can only be obtained in open
forest covers, a situation seldom encountered in mature
commercial stands of the boreal forest. One-dimensional radar or
lidar resolve this problem by being able to penetrate the forest
covers to a good extent and getting a clear echo from the ground.
Lidar-based techniques were successfully used to estimate basal
area, volume and biomass (Nelson et al., 1997, Lefsky et al.,
1999) as well as accurate mean stand height estimations (Naesset,
1997) and percent canopy cover (Ritchie et al., 1993; Weltz et
al., 1994). Estimating tree height from small footprint one-
dimensional data can however only be accomplished by building
statistical relationships between spot height data and average tree
height in a stand since it is very difficult to establish if a
particular impulse was echoed by the top of a tree or by its side.
Moreover, mapping the forest cover by creating continuous
coverage is impossible. Recent technological developments in
laser remote sensing will most likely improve remote sensing
measurements of trees. Indeed, scanning laser altimetry can now
provide sub-meter resolution DTMs, i.e. continuous coverages, of
both canopy top and underlying terrain, with high horizontal and
vertical accuracy (Flood and Gutelius, 1997). High point densities
even enable the recognition of the shape of individual tree
crowns. While scanning laser altimetry provides a solution,
estimating the height on an individual tree using such an
approach requires that a laser spot fall near the point of maximum
height, an event that cannot normally be verified unless ancillary
data is available. The advent of high resolution satellites, such as
IKONOS, a 1-m resolution launched on September 24th 1999,
will help in obtaining imagery putting laser data in context.
Meanwhile, aerial imagery can be used for that purpose.
Our general objective is to evaluate the usefulness of combining
high density scanning laser altimetry to high resolution
multispectral data to measure the accuracy of the estimation of
individual tree height (this paper), wood volume per hectare,
productivity, for dense and mature stands of the boreal tree
forest. The combination of tree height data and variables such
as drainage derived from the laser DTM are also used to study
the relationships between ecological factors and productivity.
The general approach we follow consists in overlaying
multispectral imagery over interpolated laser data to help locate
trees and verify if a given tree was hit near the center of its
crown, i.e., at the most probable location of the highest point in
the tree. This paper is specifically concerned with the accuracy
assessment of laser prediction of individual tree heights. We
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
show how the laser data was obtained and geometrically
matched with 50 cm resolution multispectral imagery. We then
show how ground measurements were correlated with laser
measurements for trees that have a good hit coverage. Factors
determining height prediction errors are then discussed.
2 STUDY REGION
Data and methods have been developed and tested for the
Training and Research Forest of Lake Duparquet (TRFLD), a
80 square km of the boreal forest in the Abitibi region, Quebec
(79.3 W, 48.5 N), which is part of the Forest Ecosystem
Research Network of Sites. The boreal forest is the largest forest
biome, covering 17% of the terrestrial land. Its floristic
composition is rather simple, indeed, only nine species are
commonly found. The forest landscape at the TRFLD is
essentially composed of hardwood, softwood and mixed stands
aged from 50 to more than 230 years growing on a part of the
Canadian shield culminating at 382 m. Common species
include: Trembling aspen (populus tremuloides), White spruce
(picea glauca), White birch (betula paperifera), Balsam fir
(abies balsamea), severely attacked by spruce budworm, Jack
pine (pinus banksiana), Eastern cedar (thuya occidentalis), and
Black spruce (picea mariana). The study site was commercially
exploited until 1992 and bears regeneration areas. This test area
was chosen for its landscape and habitat diversity,
representative of the mixed boreal forest, the wide availability
of data, and the important collaborative effort between
universities, forest companies and the socio-economic
environment.
3 DATA ACQUISITION
Airborne scanning laser altimetry data
Scanning laser altimetry data was acquired on June 28th 1998
using Optech’s ALTM 1020 instrument on a Piper Navajo plane
flying at 700 m. Separate passes were needed for vegetation
(two passes) and terrain description (single pass) respectively.
Vegetation/terrain separation was carried out by the data
provider using Optech’s ALTM software. Flight, laser and GPS
characteristics are presented in Table 1 and a sample of the
derived imagery is presented in Figure 1. The impulse
frequency combined with the lowest sustainable flight speed
and altitude did not allow to achieve the desired laser hit
density of one hit every 50 cm. To alleviate this problem, each
flight line was flown twice in an effort to double point density
for the first return (vegetation). However, since some first pass
hits fall very near the second pass hits, the effective hit density
was increased but not usefully doubled. This does not constitute
a severe operational limitation because of the existence of
higher frequency instruments and the possibility of using a
helicopter. However, it does have consequences on this study S
results. Still it provides a very good database to carry out à
crossover study to assess laser accuracy (not presented in this
paper).
International /
Multispectral imagery
The area covered by I:
before (September 27tl
camera, functioning in
airplane. The videotape
multispectral digital im:
characteristics of the n
presented in table 2. A s:
in figure 2. Tree growtk
(multispectral data) and 1
(laser data) is minimal a
allowing for direct comy
shape) between the two ¢
Ground measures
The height of individua
using a standard clinorr
measures were taken fror
at least 90 degrees to i
measures. Trees for whic
more than 3 meters or by
errors in comparing lase;
be attributed to the laser 1
well measured trees wer
ground measurements. '
Trembling Aspen (Popi
(Picea glauca) but some
were measured. Of the aj
40 had been positioned c
at the time of paper subr
trees remained. Other g
breast height, crown ra
directions, species, and G
was used as a base
differential data on a c
served as the mobile ui
epochs were obtained f
maximum error after diffe
three meters for most tree
4 MI
Creation of the Canopy |
The Canopy Height Mode
the interpolated terrain al
altitudes. Triangulated irr
the X,Y,Z points was cor
using Surfer 6.04.
A TIN interpolation as
between points, a fact that
However, the choice of a
have to rely on appropriate
