International Journal of

Geo-Information

Article
A Virtual Geographic Environment for Debris Flow
Risk Analysis in Residential Areas
Lingzhi Yin 1 , Jun Zhu 1, *, Yi Li 2 , Chao Zeng 3 , Qing Zhu 1 , Hua Qi 1 , Mingwei Liu 1 , Weilian Li 1 ,
Zhenyu Cao 3 , Weijun Yang 4 and Pengcheng Zhang 4
1 Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University,
Chengdu 611756, China; Linzyhn@[Link] (L.Y.); zhuq66@[Link] (Q.Z.); qi-3dgis@[Link] (H.Q.);
liumingwei@[Link] (M.L.); vgewilliam@[Link] (W.L.)
2 State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing and Digital Earth,
Chinese Academy of Sciences, Beijing 100101, China; liyi@[Link]
3 Sichuan Geomatics Center, Chengdu 610041, China; zeng3chao@[Link] (C.Z.); scgisczy@[Link] (Z.C.)
4 Guangzhou Urban Planning & Design Survey Research Institute, Guangzhou 510060, China;
fazeyang@[Link] (W.Y.); guangzhou2000@[Link] (P.Z.)
* Correspondence: zhujun@[Link]; Tel.: +86-130-8442-6186

Received: 9 October 2017; Accepted: 20 November 2017; Published: 22 November 2017

Abstract: Emergency risk assessment of debris flows in residential areas is of great significance
for disaster prevention and reduction, but the assessment has disadvantages, such as a low
numerical simulation efficiency and poor capabilities of risk assessment and geographic knowledge
sharing. Thus, this paper focuses on the construction of a VGE (virtual geographic environment)
system that provides an efficient tool to support the rapid risk analysis of debris flow disasters.
The numerical simulation, risk analysis, and 3D (three-dimensional) dynamic visualization of debris
flow disasters were tightly integrated into the VGE system. Key technologies, including quantitative
risk assessment, multiscale parallel optimization, and visual representation of disaster information,
were discussed in detail. The Qipan gully in Wenchuan County, Sichuan Province, China, was selected
as the case area, and a prototype system was developed. According to the multiscale parallel
optimization experiments, a suitable scale was chosen for the numerical simulation of debris flow
disasters. The computational efficiency of one simulation step was 5 ms (milliseconds), and the
rendering efficiency was approximately 40 fps (frames per second). Information about the risk area,
risk population, and risk roads under different conditions can be quickly obtained. The experimental
results show that our approach can support real-time interactive analyses and can be used to share
and publish geographic knowledge.

Keywords: virtual geographic environment; debris flow disaster; multiscale parallel optimization;
real-time interaction; emergency analysis

1. Introduction
A debris flow is a typical geomorphic spatiotemporal process with hazard potential in
mountainous areas and is trigged by intense rainfalls with landslides [1–3]. Debris flows create
hazard risks, including human injury and death, property losses, social instability, and environmental
destruction, and hinder the sustainable development of the economy in mountainous areas [4].
For example, the debris flow disasters that occurred in Zhouqu County on 7 August 2010, in Wenchuan
County on 11 July 2013, and in Jiuzhaigou County on 25 July 2016, all resulted in damaged houses,
traffic disruptions, a large number of casualties, and huge economic losses. Therefore, it is important
to implement a rapid risk analysis of debris flow disasters in residential areas that can provide
accurate disaster information. The disaster information consists of two parts: the disaster severity

ISPRS Int. J. Geo-Inf. 2017, 6, 377; doi:10.3390/ijgi6110377 [Link]/journal/ijgi

ISPRS Int. J. Geo-Inf. 2017, 6, 377 2 of 15

information and the disaster damage information. The disaster severity information includes the
arrival time, inundated area, flow depth, and flow velocity. The disaster damage information includes
the population and roads at risk. Such disaster information can provide scientific support for
emergency rescue response and is of considerable significance for debris flow disaster prevention and
reduction [5–7].
Risk assessment of debris flow disasters plays an important role in the analysis of the disaster
area, risk degree, and disaster loss [8,9]. Most of the existing risk assessment methods are based on
the relevant factor theory [10]. Assessment factors, such as the catchment area, relative elevation,
rainfall intensity, and debris flow density, are used to construct the risk assessment function, and these
assessment factors can be obtained from either a statistical analysis of the historical data and monitoring
data that describe the occurrence of debris flow disasters or an interpretation of RS (remote sensing)
images [11,12]. Although these methods can be used to quickly evaluate the risk of the debris flow
disaster, the selection of the risk assessment factors and weight assignments, as well as the accuracy
of the evaluation results are dependent on the experience of experts, and these methods can only
be used to quantify the debris flow risk level for a whole gully [10,13,14]. Numerical simulation
methods can quantitatively analyze the development process and the results of debris flow disasters
in various conditions. Thus, many researchers have applied numerical simulation methods to the
risk assessment of debris flow disasters [7,9,15,16]. However, the above numerical simulations are
separated from the risk assessment analyses. Because the simulation results have to be imported into
some professional software (e.g., ArcGIS) for a spatial analysis, this process requires a long time to
obtain risk assessment results.
Currently, researchers mainly use GIS (Geographic Information Science)-supported numerical
equations to construct a combination of one- and two-dimensional models for the simulation and
analysis of a debris flow routing process [17,18]. However, in this field of research, most GIS software
packages mainly use the CPU (central processing unit) serial computing mode to simulate the debris
flow routing process, which results in a very long calculation time [19,20]. The calculation efficiency will
be sharply reduced if high-resolution DEM (digital elevation model) data is used for high-resolution
simulations. Therefore, to eliminate the limitation of computing power, it is necessary to introduce the
parallel computing mode to the numerical simulation of debris flow disasters [21]. There have been
some studies on numerical simulations of debris flow disasters that are based on parallel computing,
including distributed computing based on sockets [22], parallel computing based on the CUDA
(compute unified device architecture) platform [23,24], and multicore parallel computing based on
OpenMP [25]. The method of multicore parallel computing based on OpenMP has the advantages of
simplicity, portability, flexibility, and cross-platform capability [26]. However, to ensure the accuracy
of the debris flow simulation, the above researches usually use high-resolution grid cell data for
parallel computing, often leading to a low computational efficiency. Therefore, it is critical to determine
how to quickly and accurately obtain disaster information, such as the flow depth, flow velocity,
and inundated area.
Abundant complex disaster information can be obtained through a numerical simulation and
risk assessment of debris flow disasters. However, it is difficult for such disaster information to
effectively support multisensory spatial cognition, communication of multilevel geographic knowledge,
and abstract representation and analysis beyond reality [27]. A VGE (virtual geographic environment)
can be used for multidimensional abstract representations of the real world, multimodal visualizations
with multiple viewpoints and multiple details, and analytical understandings [28]. Users can observe a
geographic phenomenon and its evolution process through a 3D (three-dimensional) virtual geographic
space [28,29]. At present, a VGE has been widely used in multiple fields, such as air pollution diffusion,
dam-break flooding, and digital city modelling, to express and share geographic knowledge [30–32].
However, it has not been applied in the field of emergency analysis of debris flow disasters.
Furthermore, most of the existing 3D visualization platforms are compatible only when installed
with plug-ins or additional software, and thus, it is difficult to achieve cross-platform compatibility,
ISPRS Int. J. Geo-Inf. 2017, 6, 377 3 of 15

real-time updating, and multiuser terminal versatility [33]. However, the 3D, dynamic visualization
of debris flow disasters in an emergency situation has higher cross-platform, real-time, and user
experience requirements for the visualization platform. With the development of HTML5 and WebGL
technologies, users can browse and observe 3D, dynamic scenes in a browser without plug-ins.
Currently, these technologies are supported by most browsers, such as Firefox, Chrome, and Apple
Safari. Cesium is an open-source JavaScript library for world-class 3D globes and maps that are based
on HTML5 and WebGL technologies. This paper will use Cesium to implement the 3D visualization of
debris flow disasters.
Based on high-performance computing and WebGL technologies, our research goal is to
integrate the simulation, analysis and visualization into a VGE framework, providing a tool for
the rapid analysis of debris flow disasters in residential areas. Key technologies, including multiscale
parallel optimization, quantitative risk assessment, and visual representation of disaster information,
are discussed in detail. All of these efforts aim to solve the following problems: low numerical
simulation efficiency, poor risk assessment capability, and difficulty in geographic knowledge sharing.
The remainder of this paper is organized as follows: Section 2 shows the methods of the rapid risk
analysis of the debris flow disaster, Section 3 describes the development of a prototype system and the
implementation of the rudimentary experimentation, and Section 4 provides a conclusion and a brief
discussion of further researches.

2. Methods

2.1. Overall Framework

The overall framework of the VGE is shown in Figure 1. It is divided into four layers: a basic data
layer, a service layer, a presentation layer, and a user layer. The basic data layer provides abundant
data resources to support the VGE system. The service layer is the core of the whole system. Based on
the methods of simulation, risk assessment, and spatial analysis, the service layer can provide a
professional service for the rapid simulation and interactive analysis to obtain disaster information.
The presentation layer supports the multidimensional perception of the virtual geographic scene.
It can improve the spatial cognitive efficiency for a debris flow disaster. The user layer is directly
oriented toward the end user. Users can set the simulation parameters of a debris flow disaster routing
process through an interactive interface and can perform a series of operations, such as querying and
interactive analysis, on the simulation results.

2.2. Multiscale Parallel Optimization

Numerical simulation is an effective method to reproduce the recurrence and redevelopment
of debris flows and can reflect the dynamic routing process and the spatial distribution of the flow
velocity and flow depth of the debris flow in the affected area [34]. The particle model is used to
simulate the debris flow disaster routing process in this paper. The particle model was proposed by
Wang et al., and was based on the Lagrangian-Euler numerical method [35]. The particle model regards
debris flow masses as multiple small particles, and each particle has the same size and shape. On this
basis, Hu et al. used the forward difference method to calculate the flow velocity and the displacement
of each particle at each time interval [10]. Then, the debris flow movement on alluvial fans can be
approximately described by the movement of a large number of particles. Presently, this method has
been used to simulate and analyze the debris flow disaster routing process [7,36].
Due to the real-time interactive analysis in the VGE, the computational efficiency of numerical
simulations needs to improve as soon as possible. According to the 3D visualization rendering
efficiency requirements (no less than 25 frames per second), one simulation step should be completed
within 40 ms [37,38]. The data with different grid cell sizes has a considerable effect on the efficiency
and accuracy of the simulation and, consequently, on the spatial analysis and visualization of debris
flow disasters. Thus, it is better to select data with different grid cell sizes when a parallel simulation
ISPRS Int. J. Geo-Inf. 2017, 6, 377 4 of 15

study of a debris flow disaster is to be performed. To meet the rapid simulation and visual analysis
requirements of a debris flow disaster in an emergency situation, an appropriate grid cell size is selected
on the premise of ensuring the accuracy of a debris flow disaster numerical simulation. We do not
change the model itself. In this paper, we use a parallel computing technology and select appropriate
grid
ISPRScell
Int. sizes to improve
J. Geo-Inf. 2017, 6, 377the computational efficiency. 4 of 14

Figure 1.
Figure 1. Overall
Overall framework.
framework.

2.2. Multiscale
The OpenMP Parallel Optimization
framework was used to implement the parallel optimization of simulations with
multiple grid cellsimulation
Numerical sizes, as shown in Figuremethod
is an effective 2. First, to
thereproduce
data, including the DEM,and
the recurrence roughness coefficient,
redevelopment of
breach
debris coordinates,
flows and can particle
reflectratio,
the peak
dynamicflow,routing
and initial flow and
process depththearespatial
initialized. Second,ofthe
distribution model
the flow
based
velocityonandCPU serial
flow depthcomputing is divided
of the debris flow inintothe serial computing
affected area [34].andTheparallel
particle computing
model is used parts.
to
The serial computing part is the main thread, including the calculations
simulate the debris flow disaster routing process in this paper. The particle model was proposed byof the discharge at the
breach,
Wang et flow
al.,velocity,
and was initial
basedposition
on theand flow depth. The parallel
Lagrangian-Euler numerical computing
method mainly
[35]. The involves mapping
particle model
the computing-intense update of particles for multithreading, including the
regards debris flow masses as multiple small particles, and each particle has the same size and shape. flow velocity in the
xOndirection andHu
this basis, y direction,
et al. usedthethedisplacement in the x direction
forward difference method to and y direction,
calculate and the
the flow current
velocity andtime.
the
Finally, after the
displacement of completion
each particleofatthread synchronization,
each time the number
interval [10]. Then, of particles
the debris in each grid
flow movement cell is
on alluvial
calculated
fans can beaccording
approximately to thedescribed
new flowbyvelocities,
the movementpositions, and coordinates
of a large of the particles
number of particles. Presently,in this
the
main
method thread. Then,
has been usedthetoflow depth,
simulate theanalyze
and flow velocity, andflow
the debris the disaster
inundated area in
routing each grid
process cell can
[7,36].
be calculated
Due to the through
real-timedispatching
interactivemultiple
analysisthreads at each
in the VGE, thetime interval and
computational then used
efficiency of to update
numerical
the particle state values for the next time interval. Therefore, these calculation
simulations needs to improve as soon as possible. According to the 3D visualization rendering results can be used to
dynamically analyze and(no
efficiency requirements visualize
less than the
25debris
framesflow
per disaster
second),information.
one simulation step should be completed
within 40 ms [37,38]. The data with different grid cell sizes has a considerable effect on the efficiency
and accuracy of the simulation and, consequently, on the spatial analysis and visualization of debris
flow disasters. Thus, it is better to select data with different grid cell sizes when a parallel simulation
study of a debris flow disaster is to be performed. To meet the rapid simulation and visual analysis
requirements of a debris flow disaster in an emergency situation, an appropriate grid cell size is
time. Finally, after the completion of thread synchronization, the number of particles in each grid cell
is calculated according to the new flow velocities, positions, and coordinates of the particles in the
main thread. Then, the flow depth, the flow velocity, and the inundated area in each grid cell can be
calculated through dispatching multiple threads at each time interval and then used to update the
particle
ISPRS Int. state values
J. Geo-Inf. for
2017, 6, 377the next time interval. Therefore, these calculation results can be used
5 of to
15
dynamically analyze and visualize the debris flow disaster information.

Figure 2.
Figure 2. Multiscale
Multiscale parallel
parallel optimization based on
optimization based on the
the OpenMP
OpenMP framework.
framework.

2.3. Quantitative Risk Assessment

2.3. Quantitative Risk Assessment
The risk assessment of debris flow disasters includes a comprehensive analysis of hazard and
The risk assessment of debris flow disasters includes a comprehensive analysis of hazard and
vulnerability [7]. Hazard assessment and vulnerability assessment are designed to estimate the
vulnerability [7]. Hazard assessment and vulnerability assessment are designed to estimate the damage
damage capability of the debris flow and the losses that are suffered by risk objects, which are of
capability of the debris flow and the losses that are suffered by risk objects, which are of considerable
considerable significance for disaster prevention and emergency response [39]. Damages from the
significance for disaster prevention and emergency response [39]. Damages from the debris flow
debris flow are caused by the impact and silt effects. The flow velocity is the key parameter to
are caused by the impact and silt effects. The flow velocity is the key parameter to determine the
determine the impact effect, and the flow depth reflects the deposition of the debris flow. Losses that
impact effect, and the flow depth reflects the deposition of the debris flow. Losses that are caused by a
are caused by a debris flow disaster mainly occur in the disaster inundated area and involve the
debris flow disaster mainly occur in the disaster inundated area and involve the damage degree and
damage degree and comprehensive value of the risk objects [36]. A quantitative risk assessment
comprehensive value of the risk objects [36]. A quantitative risk assessment method for debris flow
method for debris flow disasters is designed in this paper, as shown in Figure 3.
disasters is designed in this paper, as shown in Figure 3.
First, the quantitative risk assessment method is tightly integrated into the VGE framework. The
First, the quantitative risk assessment method is tightly integrated into the VGE framework.
numerical simulation of the debris flow disaster can be implemented after the parameters are set.
The numerical simulation of the debris flow disaster can be implemented after the parameters are set.
Then, the disaster information, such as the flow depth, flow velocity, inundated area, and maximum
Then, the disaster information, such as the flow depth, flow velocity, inundated area, and maximum
flow depth of each grid cell, can be obtained. Second, after this disaster information is combined with
flow depth of each grid cell, can be obtained. Second, after this disaster information is combined
the risk assessment model proposed by Zou et al. [36], the hazard value and vulnerability value of
with the risk assessment model proposed by Zou et al. [36], the hazard value and vulnerability
value of each grid cell can be obtained. In addition, these values can be divided into three levels,
namely, low, medium, and high grades, and are obtained using the standard deviation method.
Finally, integrated with the thematic data, including roads, residential areas and public facilities,
the disaster information, such as risk area, risk roads, risk population, and risk infrastructure at
different risk levels, is calculated individually by an overlay analysis.
each grid cell can be obtained. In addition, these values can be divided into three levels, namely, low,
medium, and high grades, and are obtained using the standard deviation method. Finally, integrated
with the thematic data, including roads, residential areas and public facilities, the disaster
information, such2017,
ISPRS Int. J. Geo-Inf. as 6,
risk
377 area, risk roads, risk population, and risk infrastructure at different6 risk
of 15
levels, is calculated individually by an overlay analysis.

Figure 3. Quantitative risk assessment of debris flow disasters.

Figure 3. Quantitative risk assessment of debris flow disasters.

2.4. Real-Time Visualization of Disaster Information

2.4. Real-Time Visualization of Disaster Information
A virtual geographic scene, which includes the virtual terrain, simulation results data, and
A virtual geographic scene, which includes the virtual terrain, simulation results data,
thematic information, can be used to better understand and analyze debris flow disasters.
and thematic information, can be used to better understand and analyze debris flow disasters.
Determining how to render these data in real time is very important for the interactive analysis of
Determining how to render these data in real time is very important for the interactive analysis
debris flow disasters.
of debris flow disasters.
First, local data and online data are the main data resources with which the virtual terrain scenes
First, local data and online data are the main data resources with which the virtual terrain scenes
of debris flow disasters are constructed. The local data includes high-resolution RS data and DEM
of debris flow disasters are constructed. The local data includes high-resolution RS data and DEM
data, and these data are organized into a local cache based on a pyramid model using leveling and
data, and these data are organized into a local cache based on a pyramid model using leveling and
block processing. Then, these data are loaded and rendered according to the LOD (level of detail)
block processing. Then, these data are loaded and rendered according to the LOD (level of detail)
requirements of the users. The online data includes a large number of high-resolution global satellite
requirements of the users. The online data includes a large number of high-resolution global satellite
images and a large amount of map data and rough elevation data. The system can support multiple
images and a large amount of map data and rough elevation data. The system can support multiple
map data services such as WMS (web map service), WCS (web coverage service), and TMS (tiled map
map data services such as WMS (web map service), WCS (web coverage service), and TMS (tiled map
service) to obtain and parse these online data and then load and render them in real time. It can also
service) to obtain and parse these online data and then load and render them in real time. It can also
support 3D visualization and browsing of a virtual earth with different resolutions.
support 3D visualization and browsing of a virtual earth with different resolutions.
Second, the simulation results of the debris flow disasters are stored in a two-dimensional array
Second, the simulation results of the debris flow disasters are stored in a two-dimensional array of
of the grid cell units. Each grid cell at different times contains data on the flow depth, elevation, and
the grid cell units. Each grid cell at different times contains data on the flow depth, elevation, and plane
plane coordinates, which can support the rendering of the debris flow scenes. Because some grid cell
coordinates, which can support the rendering of the debris flow scenes. Because some grid cell states
states may be not included, there are many “invalid” data points in the array. Thus, to reduce the
may be not included, there are many “invalid” data points in the array. Thus, to reduce the amount of
amount of data, only grid cells with flow depth data are extracted from the two-dimensional array,
data, only grid cells with flow depth data are extracted from the two-dimensional array, organized into
a triangular network structure, and later parsed and rendered. The visualization color of the debris
flow adopts a gray color scheme, which conforms to the public’s perception. Different flow depths are
visualized by different colors through hierarchical mapping, as shown in Figure 4a.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 7 of 14
ISPRS Int. J. Geo-Inf. 2017, 6, 377 7 of 14

organized
organized into
into aa triangular
triangular network structure, and later parsed and rendered. The visualization color
ISPRS Int. J. Geo-Inf. 2017, 6, 377 network structure, and later parsed and rendered. The visualization 7color of 15
of
of the debris flow adopts a gray
the debris flow adopts a gray color
color scheme,
scheme, which
which conforms
conforms to to the
the public’s
public’s perception.
perception. Different
Different
flow
flow depths
depths are are visualized
visualized by by different
different colors
colors through
through hierarchical
hierarchical mapping,
mapping, as as shown
shown in in Figure
Figure 4a.4a.
Finally, thematic
Finally, thematic data,data, including
including roads,
roads, residential
residential areas,
areas, and
and public
public facilities,
facilities, is
isloaded
loadedas as vector
Finally, thematic data, including roads, residential areas, and public facilities, is loaded asvector vector
layers, whichare
layers, are usedfor for riskassessment
assessment andspatialspatial analysisof of debrisflowflow disasters. The The riskareaarea
layers, which
which are used used for risk
risk assessment and and spatial analysis
analysis of debris
debris flow disasters.
disasters. The risk risk area
and risk population
and population can be be obtained through
through overlay analysis
analysis of the the inundated area. area. The disaster
disaster
and risk
risk population can can be obtained
obtained through overlayoverlay analysis of of the inundated
inundated area. The The disaster
degreecan
degree canbe bedetermined
determinedthrough throughthe therisk
riskassessment
assessment model. The color scheme ofrisk
a risk map can
degree can be determined through the risk assessment model. The color scheme of a risk map can
model. The color scheme of a map can be
be
be partly
partly basedbasedon on
the the combination
combination of of
an an early
early warning
warning color
color system
system
partly based on the combination of an early warning color system and the extension of an early
and
and the
the extension
extension of
of anan early
early
warning
warning color
warning color system
color system [40],
system [40], as
[40], as shown
as shown in
shown in Figure
in Figure 4b.
Figure 4b. Users
4b. Userscan
Users can obtain
can obtaindisaster
obtain disaster information,
disaster information,such
information, such as
such asthethe
as the
disaster
disasterdegree,
degree,risk riskroads,
roads,andand risk
risk population
populationin in the
the debris
debris flow
flow disaster
disaster routing
disaster degree, risk roads, and risk population in the debris flow disaster routing process through
routing process
process through
through
an
an interactive
an interactive query.
interactive query. All
query. Allof
All of the
of the efforts
the efforts aim
efforts aim at
aim at real-time
at real-time visualization
real-time visualization and
visualization and interactive
and interactive operation
interactive operation of
operation of
of
disaster
disaster information.
information.
disaster information.
0.0 2.0 4.0 6.0 8.0 10.0 (m)
0.0 2.0 4.0 6.0 8.0 10.0 (m)

(a)
(a)

(b)
(b)
Figure 4. Visualization color design: (a)
(a)real-time
real-timeflow depth and color mapping and (b) color grade
Figure 4. Visualization color design: (a) real-time flow depth and color mapping and (b) color grade
designed
designedininthe
therisk
riskmap.
map.

3.
3. Results
Results and Analysis
Results and
and Analysis
3.1.
3.1. Case
CaseArea
Case Area
The
The Qipan
Qipan gully
Qipan gully (30°45′–31°43′
gully (30◦ 450 –31◦ 43N,
(30°45′–31°43′
0 N,102°51′–103°44′
102◦ 510 –103◦ 44E)
N, 102°51′–103°44′
0 is approximately
E) E) is is approximately
approximately 55 km
5 km
km southwest
southwest
southwest of
of
Wenchuan
of Wenchuan County, which
County, is
which located
is in
located the
in southeast
the of
southeast Sichuan
of Province,
Sichuan
Wenchuan County, which is located in the southeast of Sichuan Province, China (Figure 5). The China
Province, (Figure
China 5).
(Figure The5).
mainstream
The mainstream
mainstream of the
of the Qipan
ofQipan gully
the Qipan is
gullygully 15.1 km
is 15.1
is 15.1 long and
km long
km long is
and and the
is the first
is first tributary
the first of
tributary
tributary the
of the Minjiang
of the River.
Minjiang
Minjiang Its
River.
River. Its
drainage
drainage basin
Its drainage
basin covers
basin
covers an
covers area
an anarea of
areaof 54.2
of km
54.2
54.2 km
2 and
km2 and has
2 and
has an
has
anan elevation
elevation
elevation ranging
ranging
ranging from
from1320
from 1320m
1320 mmtototo4360
4360m.
4360 m.m.
Because
Because
Because of
of a
a heavy
heavy rainfall
rainfall
rainfall from
from the
the night
night
night of
of 7–11
7–11 July
July 2013,
2013, the
the Qipan
Qipan gully
gully suffered
suffered a
a large
large debris
debris
flow
flow disaster
disaster [41].
[41].

Figure 5. Case area.

Figure 5. Case area.
Figure 5. Case area.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 8 of 15

3.2. Prototype System Development

The simulation of the debris flow routing process in this paper is carried out by using C++
language and OpenMP parallel computing framework. The 3D scene of the debris flow disaster is
constructed based on Cesium. The vector data in the disaster area is stored in the PostGIS spatial
database and is published by the GIS server GeoServer as a WFS (web feature service) under the
OGC (open geospatial consortium) standard. After being invoked by the system, the service can
be loaded and parsed. This system includes four functions. First, it can integrate the online map
data, which provides basic geographic information services. Second, it can provide a basic spatial
analysis function. Third, it can support rapid simulation and risk assessment of debris flow disasters.
Fourth, it can query the debris flow disaster information in real time. At present, the numerical
simulation of debris flow disasters is oriented to expert users, and other functions of visualization
and spatial analysis are for all of the users. The simulation analysis process is shown below. A visual
interface is first provided for expert users to set the simulation parameters and start the simulation
calculations. Then, a series of simulation results are generated automatically, and the disaster
information, such as the maximum flow depths, maximum flow velocities, and inundated area
can be obtained. Moreover, the risk population, risk roads, and risk residential area can be obtained
through an overlay analysis of the thematic data. All of these data are automatically converted to the
JSON (JavaScript object notation) format and are deployed as visual services. Finally, public users can
interactively query and browse this disaster information on the client.

3.3. Simulation Optimization Analysis

To compare the computational efficiency under different situations, the DEM data is resampled to
generate different grid cell sizes for analysis experiments. When the grid cell size is 40 m, the debris
flow model cannot run normally. Thus, the original terrain with 5 m spatial resolution is resampled
to several typical sizes, such as 10 m, 20 m, and 30 m. The simulation experiments on the debris
flow disaster routing processes are carried out by using CPU serial computing and multicore parallel
computing based on the OpenMP framework; then, the accuracy of the simulation results from grids
with different cell sizes can be compared.
The actual inundated area of the debris flow disaster is extracted from the post-disaster UAV
(unmanned aerial vehicle) image, and the inundated area is approximately 657,316 m2 . When the grid
cell sizes are 5 m, 10 m, 20 m, and 30 m, the simulated inundated areas are 635,125 m2 , 712,600 m2 ,
699,600 m2 , and 656,100 m2 , respectively. The simulation results are quite similar to the measured
results of the actual debris flow.
According to the field survey data of Zeng [42], flow velocities have been obtained at 11 cross
sections in the Qipan gully. The maximum flow velocities of cross sections under different grid cell
sizes are shown in Table 1, and the spatial distribution of the maximum flow velocities under different
grid cell sizes is shown in Figure 6. When the grid cell sizes are 5 m, 10 m, and 20 m, the simulated
flow velocities are consistent with the variation trend in the actual flow velocities. When the grid cell
size is 30 m, then the simulated flow velocities have larger errors.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 9 of 15

Table 1. Maximum flow velocities of cross sections under different grid cell sizes.

Simulated Simulated Simulated Simulated

Cross Actual Flow Flow Average Flow Average Flow Average Flow Average
Sections Velocities Velocities Deviation Velocities Deviation Velocities Deviation Velocities Deviation
(5 m) (10 m) (20 m) (30 m)
K1–K4 7.8–11.6 m/s 7.4–11.5 m/s 3.0% 8.1–12.0 m/s 3.6% 8.8–12.5 m/s 10.3% 11.0–12.5 m/s 24.4%
K4–K6 6.2–7.8 m/s 6.2–7.3 m/s 3.2% 6.1–8.3 m/s 4.0% 5.2–8.2 m/s 10.6% 7.0–12.3 m/s 35.3%
K6–K8 4.6–7.5 m/s 4.7–7.6 m/s 1.8% 5.2–7.5 m/s 6.5% 5.0–8.8 m/s 13.0% 4.8–12.1 m/s 32.8%
K8–K10 4.1–6.6 m/s 4.0–5.5 m/s 9.6% 4.6–6.5 m/s 6.9% 4.4–7.2 m/s 8.2% 6.3–12.1 m/s 68.5%
K10–K11 0–4.28 m/s 0–4.6 m/s 3.7% 0–4.7 m/s 4.9% 0–4.8 m/s 6.1% 0–9.8 m/s 64.5%
ISPRS Int. J. Geo-Inf. 2017, 6, 377 10 of 15
ISPRS Int. J. Geo-Inf. 2017, 6, 377 9 of 14

ISPRS Int. J. Geo-Inf. 2017, 6, 377 9 of 14

Figure 6. Spatial distribution of the maximum flow velocities under different grid cell sizes.
Figure 6. Spatial distribution of the maximum flow velocities under different grid cell sizes.
The Figure
actual [Link]
Spatial distribution
depths of fourofreference
the maximum flow
points in velocities
the Qipan under different
gully grid cell
have been sizes. by way
obtained
ofThe
fieldactual flow depths
investigation. of four reference
The maximum points
flow depths in the
of the Qipanpoints
reference gullyunder
have different
been obtained bysizes
grid cell way of
The actual flow depths of four reference points in the Qipan gully have been obtained by way
fieldare
investigation.
shown in Table The2, maximum flowdistribution
and the spatial depths of ofthethereference
maximum points
flow under
depths different grid cell
under different gridsizes
of field investigation. The maximum flow depths of the reference points under different grid cell sizes
are shown
cell sizesinare
Table 2, and
shown the spatial
in Figure distribution
7. When of the
the grid cell maximum
sizes are 5 m, 10flow
m, depths
and 20 m,under different grid
the simulation
are shown in Table 2, and the spatial distribution of the maximum flow depths under different grid
cell results
sizes areareshown
approximately
in [Link]
When as the
the actual flow
grid cell depth.
sizes areHowever,
5 m, 10 m, when
andthe20 grid cellsimulation
m, the size is
cell sizes are shown in Figure 7. When the grid cell sizes are 5 m, 10 m, and 20 m, the simulation
30
results m, the simulated
are approximately flow depths
thethe
same have larger
as the errors.
actual flow depth. However, when thethe
grid cell size is is
30 m,
results are approximately same as the actual flow depth. However, when grid cell size
the simulated flow depths have larger errors.
30 m, the simulated flow depths have larger errors.

Figure 7. Spatial distribution of the maximum flow depths under different grid cell sizes.

Figure 7. Spatial distribution of the maximum flow depths under different grid cell sizes.
Figure 7. Spatial distribution of the maximum flow depths under different grid cell sizes.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 11 of 15

Table 2. Maximum flow depths of the reference points under different grid cell sizes.

Simulated Simulated Simulated Simulated

Actual
Reference Flow Average Flow Average Flow Average Flow Average
Flow
Points Depths Deviation Depths Deviation Depths Deviation Depths Deviation
Depths
(5 m) (10 m) (20 m) (30 m)
A 5–8 m 5.3–7.7 m 4.9% 5.2–7.2 m 7.0% 4.2–7.1 m 13.6% 1.4–4.2 m 59.8%
B 2–4 m 2.2–4.5 m 11.3% 1.8–4.6 m 12.5% 1.7–3.7 m 15.0% 1.5–2.5 m 31.3%
C 4–6 m 3.3–5.9 m 9.6% 3.7–6.5 m 7.9% 3.6–6.5 m 9.2% 1.2–2.3 m 65.8%
ISPRS Int. J. Geo-Inf. 2017, 6, 377 10 of 14
D 3–6 m 3.1–6.5 m 5.8% 3.3–6.4 m 8.3% 2.8–7.2 m 13.3% 2.2–6.8 m 20.0%

Table 2. Maximum flow depths of the reference points under different grid cell sizes.
To analyze the parallel
Actual
computing performance,
Simulated Simulated the multicore
Simulated parallel computing
Simulated time is recorded
Reference Flow Average Flow Average Flow Average Flow Average
separately from Points the CPU
Flow computing
Depths time,
Deviation as
Depths shown in
Deviation Table
Depths 3. When
Deviation the
Depthsgrid Deviationsize is 5 m,
cell
Depths
(5 m) (10 m) (20 m) (30 m)
one simulation A
step is
5–8 m
completed
5.3–7.7 m
in approximately
4.9% 5.2–7.2 m
50
7.0%
ms, which
4.2–7.1 m
cannot
13.6%
meet
1.4–4.2 m
the need
59.8%
of real-time
interactive simulations. When the grid cell size is 10 m, one simulation step is completed in
B 2–4 m 2.2–4.5 m 11.3% 1.8–4.6 m 12.5% 1.7–3.7 m 15.0% 1.5–2.5 m 31.3%
C 4–6 m 3.3–5.9 m 9.6% 3.7–6.5 m 7.9% 3.6–6.5 m 9.2% 1.2–2.3 m 65.8%
approximatelyD 15 ms, 3–6 mwhich can
3.1–6.5 m meet 5.8% the need
3.3–6.4 mof real-time
8.3% interactive
2.8–7.2 m 13.3%simulations.
2.2–6.8 m In
20.0%fact, since the

time of the data reading and writing occupies a large proportion as compared to the calculation time,
To analyze the parallel computing performance, the multicore parallel computing time is
the speedup ratio will
recorded decrease
separately fromwith thecomputing
the CPU increase in theasgrid
time, showncellin sizes.
Table [Link],
When the grid cell wesize
canisuse5 different
grid cell sizes
m, one and different
simulation step computing
is completed inmodes together
approximately towhich
50 ms, analyze the
cannot meetcomputational
the need of real-time performance.
According interactive
to the above simulations.
accuracyWhen the grid
analysis cell size
results, is 10select
we can m, one simulation
different gridstep
cellissizes
completed in
to simulate debris
approximately 15 ms, which can meet the need of real-time interactive simulations. In fact, since the
flow disasters. When the grid cell size is 5 m, the CPU computing time is 43.07 min. When the grid cell
time of the data reading and writing occupies a large proportion as compared to the calculation time,
size is 20 m,
thethe parallel
speedup computing
ratio will decrease timewithisthe
0.29 min. The
increase in themaximum speedup
grid cell sizes. ratio isweapproximately
Furthermore, can use 148.
Thus, the multiscale
different grid parallel
cell sizesoptimization
and different can greatlymodes
computing improve
togetherthe to
computational efficiency.
analyze the computational
performance. According to the above accuracy analysis results, we can select different grid cell sizes
to simulate
Table debris flow disasters.
3. Comparison When
of parallel the grid cellperformance
computing size is 5 m, theunder
CPU computing
differenttime
gridiscell
43.07 min.
sizes.
When the grid cell size is 20 m, the parallel computing time is 0.29 min. The maximum speedup ratio
is approximately 148. Thus, the multiscale parallel optimization can greatly improve the
Grid Cell Size CPU (min) OpenMP (min) Time of a Cycle Calculation (ms) Speedup Ratio
computational efficiency.
5m 43.07 12.16 51 3.54
10 m Table5.53
3. Comparison of parallel
1.71 computing performance under15different grid cell sizes. 3.23
20 m 0.65CPU (min)
Grid Cell Size 0.29 (min)
OpenMP 5
Time of a Cycle Calculation (ms) Speedup Ratio 2.24
30 m 5m 0.23 43.07 0.10
12.16 51 3 3.54 2.30
10 m 5.53 1.71 15 3.23
20 m 0.65 0.29 5 2.24
3.4. Real-Time Interactive
30 m Analysis
0.23 0.10 3 2.30

According to the Interactive

3.4. Real-Time multiscale parallel optimization experiments, a suitable scale (20 m grid cell size)
Analysis
has been chosen for the numerical simulation of debris flow disasters. The computational efficiency of
According to the multiscale parallel optimization experiments, a suitable scale (20 m grid cell
one simulation stepbeen
size) has canchosen
be completed within 5simulation
for the numerical ms, and of
thedebris
rendering efficiency
flow disasters. The is approximately 40 fps.
computational
The VGE system
efficiencycan support
of one a real-time
simulation step can beinteractive analysis
completed within ofand
5 ms, debris flow disasters.
the rendering efficiency The
is disaster
informationapproximately
including the 40 fps.
riskThe VGErisk
area, system can support
residential a real-time
area, and risk interactive analysis
roads under of debris risk
different flow grades can
disasters. The disaster information including the risk area, risk residential area, and risk roads under
be quickly different
calculated, as shown in Figure 8 and Table 4.
risk grades can be quickly calculated, as shown in Figure 8 and Table 4.

Figure 8. Risk map of the debris flow disaster in Qipan gully.

Figure 8. Risk map of the debris flow disaster in Qipan gully.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 12 of 15

ISPRS Int. J. Geo-Inf. 2017, 6, 377 11 of 14

Table 4. Statistical results of the risk assessment of the debris flow disaster.
Table 4. Statistical results of the risk assessment of the debris flow disaster.
Risk Degree Risk Area Risk Residential Area Risk Roads
Risk Degree Risk Area Risk Residential Area Risk Roads
Low risk 420,000 m22 30,500 m22 4900 m
Low risk 420,000 m2 30,500 m 4900 m
Medium risk 120,000 m2 39,400 m22 1800 m
Medium risk
High risk 120,000 m
106,000 m2
39,400 m
36,700 m2
1800
900 mm
High risk 106,000 m2 36,700 m2 900 m

Meanwhile,
Meanwhile, the the debris
debris flow
flow routing
routing process,
process, inundated
inundated area,
area, and
and flow
flow depth
depth can
can be
be displayed
displayed
in
in real time in the virtual geographic scene. Users can visually obtain the risk levels of
real time in the virtual geographic scene. Users can visually obtain the risk levels of different
different
disaster
disaster areas
areas and
and thethe spatial
spatial distribution
distribution of of buildings, roads and
buildings, roads and other
other objects with different
objects with different risk
risk
degrees. Users can also query the disaster information such as the inundated area,
degrees. Users can also query the disaster information such as the inundated area, risk population, risk population,
and
and risk
riskroads,
roads,asasshown
shownin in
Figure 9. [Link]
Figure for emergency
is convenient rescuers
for emergency to obtain
rescuers to decision-making
obtain decision-
information quickly.
making information quickly.

Figure 9.
Figure 9. 3D visualization of
3D visualization of disaster
disaster information.
information.

4. Conclusions and Future Work

4. Conclusions and Future Work
Analytical GIS is one of the current focuses of geographic information science
Analytical GIS is one of the current focuses of geographic information science researches [28,43–45].
researches [28,43–45]. A VGE system has been developed to support the rapid risk analysis of debris
A VGE system has been developed to support the rapid risk analysis of debris flow disasters in
flow disasters in residential areas. A suitable scale has been chosen to implement the multiscale parallel
residential areas. A suitable scale has been chosen to implement the multiscale parallel computing of
computing of debris flow disasters. The computational efficiency and visual rendering efficiency have
debris flow disasters. The computational efficiency and visual rendering efficiency have been greatly
been greatly improved. The risk area, risk residential area, and risk roads under different conditions
improved. The risk area, risk residential area, and risk roads under different conditions can be quickly
can be quickly obtained. Thus, this system can support real-time simulation, visualization, and
obtained. Thus, this system can support real-time simulation, visualization, and interactive analysis.
interactive analysis. The main contributions of this paper are summarized as follows:
The main contributions of this paper are summarized as follows:
First, a multiscale parallel optimization method was designed to improve the simulation
First, a multiscale parallel optimization method was designed to improve the simulation efficiency
efficiency of debris flow disasters. The OpenMP framework was used to implement parallel
of debris flow disasters. The OpenMP framework was used to implement parallel optimization of
optimization of the debris flow simulation. Several typical grid cell sizes were tested in a sensitivity
the debris flow simulation. Several typical grid cell sizes were tested in a sensitivity analysis of the
analysis of the inundated area, maximum flow depths, and maximum flow velocities. The
inundated area, maximum flow depths, and maximum flow velocities. The experimental results show
experimental results show that the appropriate grid cell sizes within a range from 10 m to 20 m can
that the appropriate grid cell sizes within a range from 10 m to 20 m can be used for simulation analysis.
be used for simulation analysis. The computational efficiency of one simulation step can be completed
within 15 ms, and the maximum speedup ratio is approximately 148. Thus, the multiscale parallel
optimization can support real-time interactive analysis.
ISPRS Int. J. Geo-Inf. 2017, 6, 377 13 of 15

The computational efficiency of one simulation step can be completed within 15 ms, and the maximum
speedup ratio is approximately 148. Thus, the multiscale parallel optimization can support real-time
interactive analysis.
Second, the numerical simulation, risk assessment and 3D visualization of the debris flow disasters
were tightly integrated into the VGE system for real-time interactive operation and risk analysis.
A quantitative risk assessment method was deigned. 3D visualization based on the WebGL technology
was implemented to dynamically display the simulation results on a virtual earth. Users can obtain the
inundated area, flow depth, risk area, and spatial distribution of risk objects at different risk levels and
query the disaster information in the debris flow disaster routing process, such as the risk population,
risk residential area, and risk roads. The average efficiency of visual rendering is approximately 40 fps,
which can support real-time interactive analysis of debris flow disasters in a virtual geographic scene.
Third, a VGE prototype system was provided to support the emergency risk analysis of debris
flow disasters in residential areas. The Qipan gully in Wenchuan County, Sichuan Province, China,
was selected as the case area. This system can support the rapid simulation, interactive analysis,
and dynamic visualization of the debris flow disaster, which can provide scientific evidence for crowd
evacuation, temporary shelter arrangement, and rescue planning in emergency situations. In short,
our study provides a new paradigm that will help to promote the application of spatial information
technologies to emergency disaster simulations and analyses.
Despite the achievements described above, this paper has shortcomings. First, based on multiscale
optimization and parallel computing, the VGE system can support the real-time simulation of debris
flow disasters. However, due to the competition between reading and writing in the protocol
computing of the debris flow disaster simulation algorithm, complete parallel processing cannot be
carried out. According to Amdahl’s law, if there is a serial computing in parallel programs, the speedup
ratio cannot increase with the increase of computational kernel [46]. Thus, how to optimize the parallel
computing algorithm of debris flow disasters should be discussed in the future. Second, several typical
grid cell sizes are selected in this paper to determine a proper grid cell size. Other grid subdividing
methods will be used to obtain an optimal result. Third, our experimental work in this paper was aimed
at the Qipan gully in Wenchuan County, Sichuan Province, China. Thus, if necessary, the framework
and method that is presented in this paper will be used for other regions.

Acknowledgments: This paper was supported by the National Key Research and Development Program of
China (Grant No. 2016YFC0803105), the National High Technology Research and Development Program
of China (Grant No. 2015AA123901), the National Basic Surveying and Mapping Technology Project
(Grant No. 2017KJ0303),the Smart Guangzhou Spatio-temporal Information Cloud Platform Construction
(Grant No. GZIT2016-A5-147), and the 2015 Cultivation Program for the Excellent Doctoral Dissertation of
Southwest Jiaotong University.
Author Contributions: Lingzhi Yin and Jun Zhu provided the initial idea for this study, conceived the experiments,
and wrote the paper; Yi Li, Qing Zhu and Hua Qi contributed the experimental data and supplied the infrastructure
for the experiments; Lingzhi Yin, Mingwei Liu, Weilian Li and Chao Zeng performed the experiments and analyzed
the results; and Zhenyu Cao, Weijun Yang and Pengcheng Zhang provided important suggestions and reviews.
Conflicts of Interest: The authors declare no conflicts of interest.

References
1. Iverson, R.M.; Reid, M.E.; LaHusen, R.G. Debris-flow mobilization from landslides. Annu. Rev. Earth Planet. Sci.
1997, 25, 85–138. [CrossRef]
2. Borga, M.; Stoffel, M.; Marchi, L.; Marra, F.; Jakob, M. Hydrogeomorphic response to extreme rainfall in
headwater systems: Flash floods and debris flows. J. Hydrol. 2014, 518, 194–205. [CrossRef]
3. Hong, H.Y.; Pradhan, B.; Xu, C.; Bui, D.T. Spatial prediction of landslide hazard at the Yihuang area (China)
using two-class kernel logistic regression, alternating decision tree and support vector machines. Catena
2015, 133, 266–281. [CrossRef]
4. Hürlimann, M.; Copons, R.; Altimir, J. Detailed debris flow hazard assessment in Andorra: A multidisciplinary
approach. Geomorphology 2006, 78, 359–372. [CrossRef]
ISPRS Int. J. Geo-Inf. 2017, 6, 377 14 of 15

5. Aronica, G.T.; Biondi, G.; Brigandì, G.; Cascone, E.; Lanza, S.; Randazzo, G. Assessment and mapping of
debris-flow risk in a small catchment in eastern Sicily through integrated numerical simulations and GIS.
Phys. Chem. Earth Parts A/B/C 2012, 49, 52–63. [CrossRef]
6. Lin, J.W.; Chen, C.W.; Peng, C.Y. Potential hazard analysis and risk assessment of debris flow by fuzzy
modeling. Nat. Hazards 2012, 64, 273–282. [CrossRef]
7. Cui, P.; Xiang, L.Z.; Zou, Q. Risk assessment of highways affected by debris flows in Wenchuan earthquake
area. J. Mt. Sci. 2013, 10, 173–189. [CrossRef]
8. Fuchs, S.; Heiss, K.; Hübl, J. Towards an empirical vulnerability function for use in debris flow risk assessment.
Nat. Hazards Earth Syst. Sci. 2007, 7, 495–506. [CrossRef]
9. Liu, K.F.; Li, H.C.; Hsu, Y.C. Debris flow hazard assessment with numerical simulation. Nat. Hazards 2009,
49, 137–161. [CrossRef]
10. Hu, K.H.; Wei, F.Q.; He, Y. Application of particle model in risk zoning of debris flows. J. Mt. Sci. 2003, 21,
726–730.
11. Liu, X.; Lei, J. A method for assessing regional debris flow risk: An application in Zhaotong of Yunnan
province (SW China). Geomorphology 2003, 52, 181–191. [CrossRef]
12. Pradhan, B. Remote sensing and GIS-based landslide hazard analysis and cross-validation using multivariate
logistic regression model on three test areas in Malaysia. Adv. Space Res. 2010, 45, 1244–1256. [CrossRef]
13. Wei, F.Q.; Gao, K.C.; Hu, K.H.; Li, Y.; Gardner, J.S. Relationships between debris flows and earth surface
factors in Southwest China. Environ. Geol. 2008, 55, 619–627. [CrossRef]
14. Hu, K.; Wei, F.; Li, Y. Real-time measurement and preliminary analysis of debris-flow impact force at Jiangjia
Ravine, China. Earth Surf. Process. Landf. 2011, 36, 1268–1278. [CrossRef]
15. Gentile, F.; Bisantino, T.; Liuzzi, G.T. Debris-flow risk analysis in south Gargano watersheds (Southern-Italy).
Nat. Hazards 2008, 44, 1–17. [CrossRef]
16. Calvo, B.; Savi, F. A real-world application of Monte Carlo procedure for debris flow risk assessment.
Comput. Geosci. 2009, 35, 967–977. [CrossRef]
17. Wang, C.; Li, S.; Esaki, T. GIS-based two-dimensional numerical simulation of rainfall-induced debris flow.
Nat. Hazards Earth Syst. Sci. 2008, 8, 47–58. [CrossRef]
18. Ouyang, C.; He, S.; Tang, C. Numerical analysis of dynamics of debris flow over erodible beds in Wenchuan
earthquake-induced area. Eng. Geol. 2015, 194, 62–72. [CrossRef]
19. D’Ambrosio, D.; Spataro, W.; Iovine, G. Parallel genetic algorithms for optimising cellular automata models
of natural complex phenomena: An application to debris flows. Comput. Geosci. 2006, 32, 861–875. [CrossRef]
20. Lacasta, A.; Juez, C.; Murillo, J.; Garcia-Navarro, P. An efficient solution for hazardous geophysical flows
simulation using GPUs. Comput. Geosci. 2015, 78, 63–72. [CrossRef]
21. Sanders, J.; Kandrot, E. CUDA by Example: An Introduction to General-Purpose GPU Programming;
Addison-Wesley Professional: Upper Saddle River, NJ, USA, 2010.
22. Yang, F.K.; Guan, Q.; Zhang, Z.G.; Li, X.T. Debris-flow risky district decision system based on socket
distributed-computing. Comput. Eng. Des. 2010, 31, 4909–4912.
23. Yang, S.; Guan, Q. CUDA-based simulation of debris flow. Comput. Eng. Des. 2011, 32, 4231–4236.
24. Spataro, D.; D’Ambrosio, D.; Filippone, G.; Rongo, R.; Spataro, W.; Marocco, D. The new SCIARA-fv3
numerical model and acceleration by GPGPU strategies. Int. J. High Perform. Comput. Appl. 2017, 31, 163–176.
[CrossRef]
25. Oliverio, M.; Spataro, W.; D’Ambrosio, D.; Rongo, R.; Spingola, G.; Trunfio, G.A. OpenMP parallelization
of the SCIARA Cellular Automata lava flow model: Performance analysis on shared-memory computers.
Procedia Comput. Sci. 2011, 4, 271–280. [CrossRef]
26. Amritkar, A.; Tafti, D.; Liu, R.; Kufrin, R.; Chapman, B. OpenMP parallelism for fluid and fluid-particulate
systems. Parallel Comput. 2012, 38, 501–517. [CrossRef]
27. Yu, M.; Huang, Y.; Xu, Q.; Guo, P.; Dai, Z.L. Application of virtual earth in 3D terrain modeling to visual
analysis of large-scale geological disasters in mountainous areas. Environ. Earth Sci. 2016, 75. [CrossRef]
28. Lin, H.; Chen, M.; Lu, G.N.; Zhu, Q.; Gong, J.H.; You, X.; Wen, Y.N.; Xu, B.L.; Hu, M.Y. Virtual geographic
environments (VGEs): A new generation of geographic analysis tool. Earth Sci. Rev. 2013, 126, 74–84. [CrossRef]
29. Chen, M.; Lin, H.; Kolditz, O.; Chen, C. Developing dynamic virtual geographic environments (VGEs) for
geographic research. Environ. Earth Sci. 2015, 74, 6975–6980. [CrossRef]
ISPRS Int. J. Geo-Inf. 2017, 6, 377 15 of 15

30. Xu, B.L.; Lin, H.; Chiu, L.S.; Hu, Y.; Zhu, J.; Hu, M.Y.; Cui, W.N. Collaborative virtual geographic
environments: A case study of air pollution simulation. Inf. Sci. 2011, 181, 2231–2246. [CrossRef]
31. Zhu, J.; Yin, L.Z.; Wang, J.H.; Zhang, H.; Hu, Y.; Liu, Z.J. Dam-break flood routing simulation and scale effect
analysis based on virtual geographic environment. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2015, 8,
105–113. [CrossRef]
32. Liang, J.M.; Gong, J.H.; Sun, J.; Liu, J. A customizable framework for computing sky view factor from
large-scale 3D city models. Energy Build. 2017, 149, 38–44. [CrossRef]
33. Yin, L.Z.; Zhu, J.; Zhang, X.; Li, Y.; Wang, J.H.; Zhang, H.; Yang, X.F. Visual analysis and simulation of
dam-break flood spatiotemporal process in a network environment. Environ. Earth Sci. 2015, 74, 7133–7146.
[CrossRef]
34. Shao, S.D.; Wang, L.X. Debris flow simulation and hazard zone mapping in mountainous regions of Beijing.
J. Beijing For. Univ. 1999, 21, 11–16.
35. Wang, G.Q.; Shao, S.D.; Fei, X.J. Particle model for alluvial fan formation. In Debris-Flow Hazards Mitigation:
Mechanics, Prediction, and Assessment; ASCE: Reston, VA, USA, 1997; pp. 143–152.
36. Zou, Q.; Cui, P.; Zeng, C.; Tang, J.X.; Regmi, A.D. Dynamic process-based risk assessment of debris flow on a
local scale. Phys. Geogr. 2016, 37, 132–152. [CrossRef]
37. Qiao, C.J.; Li, J.S.; Tian, Z.S. A new approach for fluid dynamics simulation: The Short-lived Water Cuboid
Particle model. J. Hydrol. 2016, 540, 437–456. [CrossRef]
38. Jin, H.L.; Gao, J.X. An Algorithm for Real-Time Visualization of Large-Scale Terrain. J. Inst. Surv. Mapp. 2006,
23, 65–68.
39. Liu, G.X.; Dai, E.F.; Wu, S.H.; Wu, W.X. A study on theory and method in debris flow risk assessment.
Prog. Geogr. 2012, 31, 383–391.
40. Jiang, L.L.; Qi, Q.W.; Zhang, A. Color on emergency mapping. In Proceedings of the Geoinformatics 2007:
Cartographic Theory and Models, Nanjing, China, 25–27 May 2007; SPIE: Bellingham, WA, USA, 2007;
Volume 6751, p. 675104.
41. Zhu, J.; Tang, C.; Chang, M.; Le, M.H.; Huang, X. Field Observations of the Disastrous 11 July 2013
Debris Flows in Qipan Gully, Wenchuan Area, Southwestern China. In Engineering Geology for Society and
Territory—Volume 2; Springer: Cham, Switzerland, 2015; pp. 531–535.
42. Zeng, C. Vulnerability Assessment of Building to Debris Flow Hazard. Ph.D. Thesis, University of Chinese
Academy of Sciences, Chengdu, China, 2014.
43. Longley, P.A.; Batty, M. Spatial Analysis: Modelling in a GIS Environment; John Wiley & Sons: Hoboken, NJ,
USA, 1996.
44. Maguire, D.J.; Batty, M.; Goodchild, M.F. GIS, Spatial Analysis, and Modeling; Esri Press: Redlands, CA,
USA, 2005.
45. Lü, G.N. Geographic analysis-oriented Virtual Geographic Environment: Framework, structure and
functions. Sci. China Earth Sci. 2011, 54, 733–743. [CrossRef]
46. Wang, Q.J.; Gao, Y.; Li, C.H. Research on Multi-Core-Based Multitask parallel processing Technology.
Comput. Appl. Softw. 2012, 29, 141–143.

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license ([Link]