2211
Technical Committee 208 /
Comité technique 208
analyzed, the principal component analysis was performed on
slope distribution and aspect distribution. In Table 2, the PS2
factor indicated a resulting principal component of slope for the
whole area. As shown in Table 2, the significant factors are
quite different for the whole area, Zone C, and Zone D except
with the Hypsometric Integral, HI. This justified that the debris
flow potential would be better understood if separate analysis
were conducted for regions of different geological properties.
Table 2 The significant influence factors for different geological zones
Order of significance
Whole area
Zone C
Zone D
1
HI
HI
HI
2
PS2
E1
Mj
3
Q0
O1
FF
4
NE
WN
Q0
5
ES
Ml
ES
6
My
SD10
O3
7
Ml
-
SD10
4 ESTIMATION MODEL OF DEBRIS FLOW
POTENTIAL
In this research the multi-variant variables discrimination
analysis is used to establish the differential function for debris
flow torrents and non-debris flow torrents. The discrimination
analysis is to form a linear combination of variables for each
associated group to provide estimation values, where the
coefficient of each individual variable represents its
contribution to the associated group. The differential function of
discrimination analysis defines the line which differentiates two
groups, and its coefficients help to discriminate properties of
each group. This research uses the commercial statistic
software, SPSS, with Fisher’s discrimination analysis, and
analyses are performed for the whole area, Zone C and Zone D.
Random sampling of the debris flow and non-debris flow
torrents were used assuming normal distribution of each factor.
For each analysis, the contributing influence factor was added
following the order of significance, and the improvement of the
rate of accuracy was checked with each additional factor. The
definition of accuracy rate is expressed as the sum of accurately
estimated debris flow torrents and non-debris flow torrents
divided by the total number of torrents.
1. Whole area with combined geological zones: The
analysis was performed over the whole area using 87 sets
randomly sampled out of 199 and 175 debris and non-debris
flow torrents. It was found that the HI appeared to be the most
significant factor; the additional factors were added following
the significant sequence of PS2, Q0, NE, ES, My, and Ml, with
accuracy rate of 78.9%, 81%, 82.2%, 82.2%, 83.3%, 84.5%, and
85.1%. The resulting discrimination function,
y
, is:
y=5.108(HI)+0.090(PS2)
-
0.020(Q0)
-
0.027(NE)
-
0.065(ES)
-
0.018(
My
)+0.003 (
Ml
)
-
1.911
(1)
The accuracy rate increases more or less steadily with the
additional parameters, but the trend is not significant with
addition of NE, and the amount of increase in accuracy was not
steady, suggesting different contribution of the parameters
compared to their significance level.
2. Zone C: The analysis was performed for the Zone C using 40
sets randomly sampled out of 80 sets debris and non-debris flow
torrents. It was found that the HI appeared to be the most
significant factor; the additional factors were added following
the significant sequence of E1, O1, WN, Ml, SD10, with
accuracy rate of 82.5%, 83.8%, 85.0%, 85.0%, 83.8%, and
83.8%. The resulting discrimination function,
y
, is:
y=19.050(HI)
+
0.018(E1)
-
0.016(O1) +0.009(WN)
-
0.025(M1) + 0.082(SD10)
-
11.388
(2)
The accuracy rate increases with the additional parameters
up till O1, and then remains the same and decreases. Thus, the
amount of increase in accuracy does not increase beyond
parameter O1. Although the rest of the parameters appear to be
significant, they do not contribute to the estimation model
3. Zone D: The analysis was performed for the Zone D using
40 sets randomly sampled out of 54 sets debris and non-debris
flow torrents. It was found that the HI appeared to be the most
significant factor; the additional factors were added following
the significant sequence of Mj, FF, Q0, ES, O3, and SD10, with
accuracy rate of 61.3%, 70%, 68.8%, 75%, 80%, 80%, and 80%.
The resulting discrimination function,
y
, is:
y=
-
5.070(HI) +0.036(Mj)
-
1.516(FF)
-
0.015(Q0)
-
0.083(ES)
+
0.024(O3)
+
0.009(SD10)
+
4.731
(3)
The accuracy rate increases with the additional parameters
till ES and then remains the same. It suggests that the addition
of O3 and SD10 parameters does not improve the accuracy rate,
although both parameters are significant.
Observing the estimation models for the three regional
analyses, the accuracy rate has a tendency to increase with the
additional factors, and the HI factor appears to be the most
effective factor in all three models. For all three models, the
coefficient of each parameter indicates the contribution of the
parameter, and is consistent with the variation in accuracy rate.
However, the effectiveness of the influence factors is not fully
in accord with the order of factor significance shown in Table 2.
Therefore, the level of significance of the parameter could not
be correlated to the contribution of the parameter to the
estimation model.
5 VALIDATION AND PREDICTION
In order to verify the feasibilities of the potential estimation
model discussed previously, the data sets of debris flow torrents
and non-debris flow torrents not used in developing the
estimation models were used for validation and prediction. A
total of 112 debris flow torrents and 87 non-debris flow torrents
were used for the prediction of the whole area using Eq.1. A
total of 40 sets of debris flow and non-debris flow torrents were
used for the Zone C, and a total of 14 sets of debris flow and
non-debris flow torrents were used for prediction using Eqs. 2,
and 3, respectively. The prediction accuracy rates were
compared to the estimation accuracy rates for whole area with
combined geological zones, Zone C, and Zone D, as shown in
Figure 2, Figure 3, and Figure 4, respectively.
From Figure 2, it was found that the accuracy rate for
prediction increased steadily up to ES but then decreased with
additional factor for whole area with combined geological zones
compared to the estimation model. Therefore, the factors used
for the model are only up to ES, and the model is rectified as:
y=4.955(HI)+0.090(PS2)
-
0.0205(Qo)
-
0.027(NE)
-
0.065(ES)
-
1.741
(4)
Figure 2. Accuracy rates of estimation and prediction for whole area