Actes du colloque - Volume 3 - page 405

2209
Estimation and Prediction of Debris Flow Potential Using Discrimination Analysis
Estimation et prédiction du potentiel d’écoulement de boue utilisant une analyse discriminante
Lin M.L., Lin Y.S.
Department of Civil Engineering, National Taiwan University, Taiwan
ABSTRACT: Taiwan is situated at the juncture of tectonic plates, which caused complex and fragile geological condition with steep
mountain terrain. Being frequently struck by typhoons and earthquakes, the landslide and debris flow hazard occurs frequently. In this
research, the estimation model of regional debris flow potential was constructed based on the geo-morphological and hydrological
conditions of the research area. For constructing the estimation model of debris flow potential, the Fisher’s discrimination analysis
was used. A study area of Nantou County in Central Taiwan was selected. Influence factors were identified and a database for both
debris flow torrents and non-debris flow torrents were constructed. Estimation model was constructed using the Fisher’s analysis by
random sampling of the debris flow and non-debris flow torrents. The estimation model is validated and then used for prediction of
debris flow potential. The final model can be determined by evaluating the estimation stability and prediction rate with each
additional influence factor. The resulting potential estimation of the study area appears to be satisfactory. The influence factor
stability of the Fisher’s discriminant model and the prediction rate associated with the differences in influence factors were discussed.
RÉSUMÉ : Taiwan est situé à la jonction de plaques tectoniques, ce qui engendre un terrain escarpé avec des situations complexes et
des conditions géologiques fragiles. L’île étant régulièrement frappée par des typhons et des tremblements de terre, des glissements de
terrain et des écoulements de boue se produisent fréquemment. Pour cette étude, le modèle d'estimation des coulées de boue régionale
potentiel a été construit sur la base des conditions géomorphologiques et hydrologiques de la zone de recherche. Pour la construction
de celui-ci, l’analyse discriminante de Fisher a été utilisée. Une zone d'étude du comté de Nantou qui se situe au centre de Taïwan a
été choisie. Les facteurs d'influence ont été identifiés et des bases de données pour les torrents d'écoulement de boue et torrents
d'écoulement de non-boue ont été construites. Le modèle d'estimation a été construit en utilisant une analyse de Fisher par
échantillonnage aléatoire des torrents d'écoulement de boue et des torrents d'écoulement de non-boue. Le modèle d'estimation est
validé, puis utilisé pour la prédiction de potentiels écoulements de boue. Le modèle final peut être déterminé en évaluant l’estimation
de la stabilité et la fréquence prédite avec chaque facteur d'influence additionnel. L'estimation résultante potentielle de la zone d'étude
semble être satisfaisante. Le facteur d'influence de stabilité du modèle discriminant de Fisher et la fréquence prédite associée aux
différences des facteurs d'influence ont été discutés.
KEYWORDS: debris flow, potential estimation, Fisher’s discriminaton analysis, influence factor, prediction model.
1 INTRODUCTION
More than 70 percent of areas in Taiwan are in mountain region
and with steep and fragile slopes. The earthquakes and heavy
rainfall introduced by typhoons often induced significant
landslide and debris flow hazards in Taiwan, which lead to
significant loss of properties and lives. For effective mitigation
of the debris flow hazards, it is important to evaluate the
potential of debris flow torrents, which supports decision on
mitigation measures and priority.
This research is based on the data of the 1,420 debris flow
torrents published by Soil and Water Conservation Bureau in
2003. The fundamental data of the debris flow torrents for the
study area of Nantou County in Central Taiwan were collected,
and the basic database along with the related influence factors
were established utilizing the geographic information system
software, Arcview. The influence factors database included
watershed area, stream length, hypsometric integral, stream
mean slope, form factor, slope distribution, slope aspect and
geology category, were extracted from the fundamental data.
The statistic analysis was performed on all influence factors to
discuss their significances. The multi-variant discrimination
analysis was used to discriminate debris flows and non-debris
flows. The analysis model was verified, and accordingly the
potential of debris flow torrents in Nantou County was
evaluated.
2 STUDY AREA AND FUNDAMENTAL DATABASE
The Nantou County located in central Taiwan is selected as the
study area in this research, which has a large area with high
mountains and rugged topography. The debris flow hazard in
Nantou County came into great concern since 1996, when
Typhoon Herb caused severe losses of properties and lives. In
addition, severe debris flow hazard struck this area frequently
after the Chi-Chi earthquake, 1999. According to the data
published by Soil and Water Conservation Bureau (2003), the
number of debris flow torrents reaches 199 in Nantou County,
which are for the potential analysis. The distribution of the
debris flow torrents and the study area are as shown in Figure 1.
The fundamental database used for construction of related
database for analysis include: the digital elevation model
published in 1989 with a resolution of 40m x 40m, 1/500,000
Taiwan geology map produced by the Central Geological
Survey in 1986, aerial photographs, topographical map.
The primary geologic formations of the study area include slate,
phyllite, sandstone, and shale. According to the engineering
geology zonation proposed by Hung in 1997, the research area
contents C zone (metamorphic rocks, metamorphic sandstone,
shale, slate, phyllite), D zone (sedimentary rocks), E zone
(lateritic tableland), and G zone (basin and plain). As shown in
Figure 1. Study area and distribution of Debris Flow Torrents and Non-
Debris Flow Torrents
1...,395,396,397,398,399,400,401,402,403,404 406,407,408,409,410,411,412,413,414,415,...840