In: Wagner W„ Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
Figure 6: Segmentation result. The best finding was acquired
using all Ikonos’ bands.
Figure 6 shows the 50m buffered Ikonos scene. The riparian
forest consists of all trees within this buffer. The polygons in
green are validation data and those in red were obtained with
MAGIC.
The MAGIC segmentation result points to a high visual
correspondence with the validation data like in zoom window (a).
The MAGIC was able to segment some features like individual
trees (zoom window (d)). However, some individual features or
narrow areas were not segmented properly (zoom windows (c)
and (b)).
3.2 Biophysical Riparian Forest Modelling Results
From the initial 70 plots, only 62 were used to obtain the average
spectral and texture values. The remaining eight plots were
partially located outside the riparian mask and had to be
withdrawn.
The statistical correlations results (adjusted R 2 ) between spectral
data, textural data and the allometric and structural measurements
of the plots are presented in Table 7: band 3 indicates that the
texture features were computed from the red spectral band and
band 4 the infrared band. The red band is much more related with
allometric parameters than the infrared band for which only the
LAI had some success. Basal area and volume obtained the best
overall results with R 2 =0.61 and R 2 =0.66, respectively. The
results show better correlations when using an 11x11 pixel
window for the parameters DBH, Basal Area and Volume. The
most successful distance between pixels is d=4, which showed
better results with Basal Area, Volume, DBH and LAI. The best
mathematical model for each allometric parameter is presented in
Equation 2 to 8.
Band 3
u
QJ
QJ
E
j=
JM
X
CO
3
qj
u
<
QJ
E
3
£
a
</>
a g
o S
e 55
<
3
u
3
On
x
Q
3
00
3
CQ
o
>
QJ
Q
« Q.
U §
wll/d3
0.34
0.52
0.50
0.64
0.07
0.32
0.45
wll/d4
0.44
0.24
0.61
0.66
0.13
0.33
0.41
wl5/d3
0.39
0.30
0.36
0.42
0.20
0.44
0.46
wl5/d7
0.52
0.29
0.49
0.63
0.04*
0.34
0.40
w20/d4
0.49
0.34
0.52
0.48
0.16
0.20
0.40
w30/d3
0.25
0.19
0.21
0.27
0.45
0.39
0.45
Band 4
w30/d4
0.16
0.20
0.19
0.09
0.05
0.41
0.54
Table 7: The correlations results for band 3 and 4 (adjusted R 2
with p test value < 0.05). The left column shows the window
size (w) and the lag distance between pixels (d). Boxed values
are significant at p>0.05.
Height = 64.6 - 0.001 con 90 - 0.00574 enti 35 - 0.0055 asm«
- 0.0065 ent 90 + 0.00128 idm I35 + 0.00064 coni 35 (2)
DBH= 184 - 0.0397 ent 90 + 0.0584 B - 0.0662 R
+ 0.126 cor 0 - 0.0786 cor 135 - 0.0103 asm 90
+ 0.0037 idm ]3s - 0.005 idm 0 - 0.00246 con 45 (3)
Basal Area = 0.0569 + 0.000002 asm 135 + 0.000001 IR
- 0.000004 asm4 5 - 0.000013 ent 45 (4)
Volume = 80.7 + 0.00304 asm 13s - 0.00555 asm^
- 0.0187 ent 45 (5)
Density = 0.018 - 0.000287 con 90 + 0.000137 con 45
-0.000190 coi*i 35 + 0.000081 idm« - 0.000531 ent 135
+ 0.000094 con 135 + 0.000547 ent 45 (6)
Canopy Openness = - 1129 + 0.214 ent 90 + 0.136 R
- 0.0865 cor 135 + 0.0137 con 135 + 0.143 asm 90 - 0.101 G (7)
LAI= 0.556 - 0.00203 R + 0.000573 con 0 - 0.00337 cor 0
+ 0.000695 idmo - 0.000119 asm 90 + 0.00354 cor 45 (8)
The direction is not a determining factor in the models and
none appear to occur predominantly. It is also difficult to
pinpoint a single co-occurrence measurement that stands out.
In models with few parameters, the ASM seem to be
reoccurring (Eq. 4 and 5). Entropy seems to come in second
place. It is likely that the diversity of measurements is the best
asset of these models and accounts for their strength. When
all texture features are analyzed together, it can be verified
that Second Angular Moment and Entropy are predominant
for the best results (basal area and volume). These models are
but indicative of the condition of the riparian forest and are
probably not directly applicable in another region. However,
but they can be used regionally to orientate the riparian
restoration efforts that are currently being undertaken in
various watersheds of Northern Minas Gerais by the Forest
Institute of MG.