L. GHAHRAMANY ET AL.

400

trees having diameter at breast height (DBH) (i.e. 1.3 m

above ground) greater than 5 cm were callipered in each

plot. Basal area in each plot was determined using field

data. Main bands, artificial bands such as vegetation in-

dices and principle component analysis (PCA) were stu-

died. Digital numbers related to each plot were extracted

from original and artificial bands. Correlation and re-

gression analysis were performed to study the statistical

relationships between standing basal area and digital

numbers of satellite data. All plots were ordinated by

major geographic aspects and the best fitted regression

models were determined for both study areas without

consideration of aspects and consideration of major geo-

graphic aspects by multiple regression analysis (step wise

regression). The use of square root of basal area as a de-

pendent variable in multivariate linear regression im-

proved the results. A number of 32 sample plots from the

319 measured plots were randomly selected and served

as control plots for verification of derived models. These

control plots were not used for correlation and regression

analysis.

3. Results

3.1. Suggested Model for Northern-Faced

Forests

In the northern-faced forests, the root of basal area

[SQRT(BA)] had highest coefficient of correlation with

band B1 (r = 0.67) among the other indices (Table 7). In

the model for the northern-faced forests, the band B1 is a

predictive variable. According to the ANOVA of regres-

sion, the hypothesis of “no linear relationship” was re-

jected with significant level of 99% (F = 69.88, P < 0.01).

The hypotheses of “The slope of the regression model =

0” and “The intercept of the regression model = 0” were

rejected at the significant level of 99% (Table 2). The

Kolmogrov-Smirnov test showed that the distribution of

residuals was normal (P = 0.66, K-S Z = 0.732). Nine out

of 10 control samples were accepted in validity test of

the model.

3.2. Suggested Model for Southern-Faced Forests

The root of basal area [SQRT(BA)] showed the highest

coefficient of correlation with RVI index (r = 0.68)

among the other indices (Table 7) in the southern-faced

forests. The RVI and B3 are predictive variables for the

model of these forests. Multiple correlation coefficient

between SQRT(BA) and predictive variables is 72%. The

ANOVA of regression model revealed that hypothesis of

“no linear relationship” is rejected with significant level

of 99% (F = 37.71, P < 0.01). The hypotheses of “The

slope of the regression model = 0” and “The intercept of

the regression model = 0” are laso rejected at the signifi-

cant level of 99% (Table 3). The distribution of residuals

Table 2. Regression coefficients and summery model for su-

ggested model for northern forests.

Beta P t SE CoefficientsModel

- <0.01 12.825 0.270 3.467 b0

–0.670 <0.01 –8.359 0.002 0.018 b1

SQRT(BA) = b0 + b1B1

N = 88 R2Adj.= 0.44 MSE = 0.073 F = 69.88 P < 0.01

Bias = –0.017 m2/ha Bias = –4.62%

Table 3. Regression coefficients and summery model for

suggested model for southern Forest.

Beta P t SE CoefficientsModel

- <0.01 7.705 0.513 3.955 b0

–0.381 <0.01 –2.930 0.002 –0.007 b3

–0.973 <0.01 –7.481 0.265 –1.980 RVI

SQRT(BA) = b0 + b1B3 + b2RVI

n= 72 R2Adj =0.51 MSE = 0.037 F = 37.71 P < 0.01

Bias = –0.035 m2/ha Bias = –2.94%

was normal (P = 0.67, K-S Z = 0.76) based on the Kol-

mogrov-Smirnov test. The whole control samples were

accepted in validity test of the model.

3.3. Suggested Model for Eastern-Faced Forests

Band B1 showed the highest coefficient of correlation (r

= –0.66) with the root of basal area [SQRT(BA)] in the

eastern-faced forests (Table 7) as the northern-faced for-

ests; however, the PCA1 and B1 are predictive vari-

ables in their model. Multiple correlation coefficient be-

tween SQRT(BA) and predictive variables are 66%.

Based on the ANOVA of regression, all hypotheses are

rejected at the significant level of 99% (Table 4). The Col-

mogrov-Smirnov test showed that the residuals are nor-

mally distributed (P = 0.73, K-S Z = 0.76). The whole

control samples were accepted in validity test of the model.

3.4. Suggested Model for Western-Faced Forests

The western-faced forests behaved similar to southern-

faced forests as they showed the highest coefficient of

correlation between the root of basal area [SQRT(BA)]

and the RVI index (r = –0.68), shown in Table 7; while

three predictive variables including RVI, PCA2 and

PCA3 are selected for their model. Multiple correlation

coefficient between SQRT(BA) and predictive variables

are 80%. All hypotheses were rejected as the other for-

ests at the significant level of 99% (Ta ble 5). The test of

normality by the Kolmogrov-Smirnov showed the nor-

mal distribution of the residuals (P = 0.72, K-S Z = 0.69).

The whole control samples were accepted in validity test

of the model.

Copyright © 2012 SciRes. IJG