The incorporation of
PS
technol-
ogies in operational forest inven-
tories requires information on the
accuracy of methods implemented
in end-user tools. The accuracy of
individual tree diameter at 1.3 m
height (
DBH
) estimation using For-
eStereo has been demonstrated in
previous studies: Rodríguez-García
et al.
(2014) reported a root mean
square error of 0.015 m for
DBH
in
8 m radius plots, whereas Sánchez-
González
et al.
(2016) obtained a
Pearson coefficient between
DBH
field measurement and ForeSte-
reo estimates of 0.86. Sánchez-
González
et al.
(2016) also analyzed
stand level estimation of
N
,
BA
, and
mean
DBH
from ForeStereo data,
reporting high agreement between
field measurements and ForeSte-
reo estimates in the case of
BA
but
a notable underestimation of
N
,
especially for smaller
DBH
classes.
We present a comparative work in
which a ForeStereo dataset is used
to estimate plot level variables (
N
,
BA
, and
DBH
distribution) with the
aforementioned inference methods,
and compared with the noncor-
rected estimates. We adapt the
method proposed by Seidel and Ammer (2014) for estimating
the proportion of the sampling area shadowed in the case of
stereoscopic hemispherical images acquired with ForeStereo.
Material and Methods
Data Set
Data from mixed stands of
Pinus sylvestris
vatica
L. were used in this work. Stereo-pai
photos were captured using ForeStereo fro
plots randomly selected from a management inventory in a
public forest at Ansó, located in the Spanish Pyrenees (Figure
1), and covering a broad range of stand densities and light
conditions. The stand originates from natural regeneration
and timber is the main production, although it also contains
protected areas. The ForeStereo device was located in the plot
centre, displacing the center if necessary just to ensure that the
nearest tree was at least 0.5 m distant. In addition to the pair of
hemispherical photos, the field crew measured the
DBH
with a
calliper and the distance from the plot centre of all trees with
DBH
> 0.075 m was also measured using a Vertex IV (©Haglöf
Sweden 2016). These values were used to calculate basal area
and number of trees in fixed radius circular plots of 8 m and
9.8 m, to serve as ground truth reference. The mean
N
was 1167
trees/ha (
SD
= 681) and the mean
BA
was 41 m
2
/ha (
SD
= 18).
The ForeStereo device used for measurement is a pro-
totype of the
MU2005-01738
. It is composed of two Fujinon
FE185C057HA-1
fish-eye lenses and two Hitachi Kokusai Electric
Inc.
KP-FD500GV
cameras with 5 Mp, a 2/3' charge-coupled
device sensor wired to a laptop through Gigabit Ethernet
interface (Figure 2). The two cameras are 0.5 m apart and
mounted on a light framework. The optical axes of the lenses
are parallel and should be vertically aligned when shooting.
The image acquisition software was developed using
GENICAM
TM
with a shooting sequence that automatically acquires four
pairs of stereoscopic hemispherical photos at 1/9 s, 1/60 s,
1/200 s and 1/1000 s exposures, as proposed by Pueschel
(2012) for gap fraction retrieval from hemispherical photos.
Figure 1. Study area map.
Figure 2. ForeStereo prototype used in this study.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2019
495
