In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
the synergy between the two sensors probably contributed 
complementary information in the model. 
Model 
R 2 
RMS 
error 
1. Texture parameters of AVNIR-2 all 
bands 
ME AB4 5, Ku AB2 9, CO AB4 9, 
TEN AB3 9, Sk AB2 5, Ske AB1 9 
0.79 
46.5 
2. Texture parameters of SPOT-5 all bands 
Sk SB3 9, ASM SB1 9, HO SB4 9, 
ID SB3 5, ID SB2 3, GASM SB4 5 
0.85 
38.5 
3. Texture parameters of both sensors 
combined 
ASM SB1 9, ASM AB4 9, 
HO AB4 7, Sk SB3 7, Var SB3 9, 
GEN SB4 7, MDM AB3 5 
0.90 
32.4 
4. Texture parameters from PCA both 
sensors 
ASM BPC1 9, CO BPC3 9, 
Sk BPC1 7, Var BPC2 9, 
Var BPC1 9, Std BPC1 5, 
MED BPC3 3/4 3 
0.85 
38.8 
5. Texture parameters from Average of both 
sensors 
Ku A4+S4 7, ASM A2+S1 9, 
Ku A2+S1 5, Sk A4+S3 7, 
Var A4+S3 9, ASM A4+S3 9, 
HO A3+S2 3 
0.91 
30.1 
6. Texture parameter ratio of AVNIR-2 
GEN AT 1/4 9, ASM AT2/3 7, 
GEN AT2/3 7, DI AT2/3 9, 
Std AT2/4 5, TME AT2/4 9, 
ME AT3/4 9, Ku ST2/3 5 
0.90 
32.0 
7. Texture parameter ratio of SPOT-5 
Sk ST3/4 9, DI ST2/4 7, 
Var ST3/4 9, ASM ST1/2 5, 
MDM ST3/4 7, CO ST2/4 9, 
GEN ST3/4 9 
0.92 
29.1 
8. Texture parameter ratio of both sensors 
DI ST2/4 7, Sk ST3/4 9, 
Var ST3/4 9, ASM ST 1/2 5, 
MDM ST3/4 7, 
CO ST2/4 9,GEN ST3/4 9, 
MDM aT2/3 5 CO AT2/3 7 
0.94 
24.8 
Table 2. Results of biomass estimation. For models (ME, Ku, 
CO etc, see Table 1. AB4_5 means AVNIR Band 4 with kernel 
5*5, and SB3_7 means SPOT Band 3 with 7*7 kernel. 
Finally, the ratio of texture parameters was found to be more 
effective for biomass estimation compared to the highest 
accuracies obtained from all previous steps. The accuracies 
obtained using all ratios of texture parameters of AVNIR-2 
(r 2 =0.899) (model 6 in Table 2), SPOT-5 (r^O.916) (model 7 in 
Table 2) and the texture ratios of both sensors together 
(r^O.939) (model 8 in Table 2) were considerably higher than 
for the simple texture models. Similar to the texture models, no 
multicollinearity effects were evident. 
This great improvement in biomass estimation observed in this 
study can be explained by the fact that we used three image 
processing techniques together as follows; 
(i) texture processing which had already shown potential for 
biomass estimation in many previous studies using 
optical (Fuchs et al, 2009; Lu, 2005) and SAR data 
(Santos et al, 2003; Kuplich et al, 2005). 
(ii) datasets from two different sensors were used in this 
processing. Although both datasets used are from optical 
sensors (AVNIR-2 and SPOT-5), there are differences in 
the wavebands, therefore it was anticipated that at least 
some complementary information could be obtained. 
(iii) finally we tested the ratio of texture parameters. We know 
from previous research that ratios, whether simple or 
complex, and whether between different bands, different 
polarizations, or different frequencies, can improve 
biomass estimation by minimizing features which are 
similar in both bands such as topographic and forest 
structural effects. 
AVNIR-2 Texture Ratio 
SPOT-5 Texture Ratio 
soo 
. R* = 0.899 * / 
500 
. R 2 = 0.916 •/* 
RMSE: 32.04 / 
RMSE: 29.09 / 
to 
fO 
400 
/ 
jC 400 
/ 
7 
• / 
M 300 
/* 
m 300 
A 
3 
,/* 
1 
/* 
Sr 200 
y 
?200 
7 
*U 
*D 
T3 
0) 
n 
9) 
S loo 
i 100 
0> 
*/» 
T# 
-Q 
J3 
O o 
O o 
100 200 300 400 500 6i 
)0 
100 200 300 400 500 600 
Predicted dry weight (t/ha) 
Predicted dry weight (t/ha) 
Figure 3. Relationship between field and model biomass 
5. CONCLUSION 
Data from two high resolution optical sensors were used in this 
research to establish a relationship between field biomass and 
remotely sensed observation parameters. The processing of data 
was conducted for each sensor individually and both sensors 
together. Spectral reflectance, texture parameters and ratio of 
texture parameters were evaluated for the improvement of 
biomass estimation. The results are promising, and except for 
the simple spectral reflectance, the accuracy (r^) of biomass 
varied estimation was higher than 80%, though this varied 
between the two sensors due to different band availability. The 
accuracy of SPOT-5 sensor was somewhat higher in all 
processing steps compared to AVNIR-2 except for the simple 
spectral reflectance because of the availability of SPOT’s SWIR 
band. However, better results were obtained using data from 
both sensors because of the complementary information. 
In this research we obtained accuracy (r^) ranging from 0.79 to 
0.94 using different processing steps, and the highest accuracy 
(r^=0.94) was obtained using the texture parameter ratio of both 
sensors. This accuracy is very promising, and this achievement