Juha Hyyppä
3.4.2 Standwise attributes
Standwise estimates were defined by calculating single tree attributes within the specified area. Since standwise
estimates are typically expressed in units per hectare, a coefficient S relating the stand values to hectare-wise values had
to be formed
oe 10000 : (9)
R
where R is the area (in square meters) specified by the stand boundaries. Standwise volume V [m’/ha], basal area BA
[m?/ha] and mean height H [m] were expressed as
V 3S (10)
BA- Y gS (11)
US (12)
Bi
The mean height was calculated as Lorey's mean height; weighted by basal area of each tree. The number of stems N
[pc/ha] were correspondingly
N = max(i)S, (13)
where the function max(i) gives the number of stems within the stand area.
3.5 Evaluation Procedure
The above mentioned methods (Sections 3.1 to 3.4) were implemented in Matlab environment and the segmentation
program was obtained from Arboreal Oy. As an input to the segmentation procedure, a 0.5-m resolution tree height
model was created using the TopoSys-1 laser scanner data obtained during the Finnish campaign. The parameters of the
segmentation algorithm were fixed before the processing of the test, and same parameters were applied for all stands
selected. Therefore, the method was applied in an automatic manner.
The formula (3) relating stem diameter and tree height was calibrated by using 25 crown, stem diameter and height
measurements. The correlation coefficient of the formula was 0.65 and standard error 4.4 was cm. The use of several
tree species within the same model deteriorated the performance. In near future, automatic tree species classification is
included in the system to improve the total accuracy.
In order to evaluate the accuracy of the segmentation-based single tree estimation methods applied to standwise forest
inventory, mean squared error (abbreviated to MSE), was calculated.
As a reference material for standwise estimation, conventional forest inventory depicted in Section 2 was applied. Since
accuracy of the conventional forest inventory affects on the evaluation, the accuracy of conventional inventory was
assessed and the errors due to inaccuracy of the field inventory were removed from the mean squared errors. Since these
two errors can be assumed as independent, the corrected root mean squared error were expressed as
l
| 1
RMSE = |MSE — 23 Var(6;) , (14)
where Var(ó;) refers to variance of conventional field inventory error ó; for stand i. The accuracy of field inventory
measurement was verified in earlier study (Hyyppá et al., 2000a).
426 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.