International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
phenomenon that a large number of mountain materials 
suddenly downward slides along a sliding surface of the 
interior in the gravity by a variety of external factors, for 
instance, earthquake, volcano, river erosion, snow melt, rainfall 
and human activities (Kirschbaum et al, 2010). Especially, the 
destructive force of other secondary disasters induced by large 
landslide far exceeds its direct damage. In recent years, 
mudslides which were affected by global climate change 
occurred frequently. More than 90,000 landslide hidden 
dangers are located in 70 cities and counties of several southern 
and north-western provinces in China, and tens of millions of 
people live under the threat of landslide throughout the year. 
Landslide caused thousands of deaths and tens of billions of 
property loss (Zhou et al, 2005). Therefore, landslide 
monitoring, assessment and prediction are advanced subjects of 
international landslide disaster research and environmental 
engineering geology field nowadays. 
Many causes of landslide occurrence mainly include terrain, 
geology, geomorphy, weather conditions, land cover, and so on 
(Parry, 2011). Obtaining the relationships between these above 
factors and landslide occurrence is very important for the 
quantitative evaluation of landslide susceptibility (LS) and 
hazards. LS assessment is the quantitative or qualitative 
evaluation for the existing or potential type, volume, 
distribution of some area’s landslide, and LS mapping would 
conduce to us the space distribution of one regional slope 
instability probability (Mathew et al, 2008). It is the first and 
most step of landslide risk assessment, in order to make 
effective measures of landslide mitigation. Reliable 
susceptibility assessment depends on the quality and range of 
the available data and the selection of method for modelling to 
identify landslide, analyze landslide formation conditions and 
characteristics, show landslide detailed geometric description. 
LS assessment typically excludes the prediction of occurrence 
time, is an important feature of susceptibility evaluation 
(Ercanoglu, 2008). 
Based on historical data and practical experience, empirical 
landslide susceptibility assessments adopt the statistical 
approach and pattern recognition methods to construct 
empirical models. Meanwhile, empirical weights of landslide 
factors can be obtained by the initial analysis; and spatial 
associations between spatial factors and landslides will be 
showed in a GIS (Geographic Information System). Artificial 
neural network (ANN) can analyze complex data at different 
scales such as continuous, indexical and binary data (Chauhan 
46 
et al, 2010). Based on learning from data with known 
characteristics to obtain the weights of factors, ANN is used to 
recognize the unseen data (Pradhan, 2011). In this paper, we 
used the ANN black box by capturing the connection weights 
among various inputs, with multi-temporal ground and remote 
sensing satellite data for susceptibility evaluation and mapping 
of China’s landslide disaster. 
2. STUDY AREA AND DATA 
China lies in the east of the Asia-Europe Continent, on the 
western shore of the Pacific Ocean, and covers about 9.6 
million km2 land area (between latitudes 3.85°N and 53.56°N, 
between longitudes 73.55°E and 135.08°E). The terrain of 
China is high in east but low in west. Mountain, plateau and 
hills cover about 67% of the land area; basin and plain cover 
about 33%. China is one of the countries which are most 
vulnerable to landslide disaster. Some studies indicated that 
landslide disaster mainly happened in a steep slope such as the 
river and stream coast of bank slope zone, and the gorges with 
high level difference; in geological tectonic belt (fractures or 
structural zones); in the soft rock-soil (loose covering layer, 
loess, mudstone, shale, coal beds); in some areas with heavy 
rainfall (He et al, 2008). 
The reports of the landslide events for nearly 60 years were 
obtained mainly from online news reports, yearbooks, and 
hazard database, including: China Geological Environmental 
Information Network (CGEIN, http://www.cigem.gov.cn/); 
China Risk Network (CRN, http://www.irisknet.cn/); Geostress 
and Geological Disaster Querying Database (GGDQD, 
http://www.geomech.ac.cn/geo0503/); China Statistical 
Yearbook (1950-2011); China major landslide reports (from 
newspaper and media). Since the 1980s, numbers of reports 
started to increase, concerned with the government’s attention. 
To be emphasized, numbers of landslide reports of the south 
and southwest are more than that of other regions. 
Many causes of landslide occurrence previous mentioned can 
be divided into two main categories (Wu and Sidle, 1995): 
(1) Internal factors: those that have decision effects on 
landslide, including geology, geomorphology, slope gradient, 
slope aspect, elevation, soil property, vegetation cover, flow 
distribution, fracture, and so on. 
(2) External factors: they will trigger landslide suddenly, such 
as earthquake, rainfall. 
Int 
Obvious 
and usu 
act on ] 
Geomat 
Table 1. 
Geo 
Rive 
Seismi 
  
The qui 
region | 
the rele 
2010). 
the nur 
differer 
RPmn 
range n 
1 
Synthe: 
were c 
Accord 
conditi.