1. Introduction
The landfill, also known as garbage dump, is a common way to contain the waste materials generated from human activities. Mismanagement of the landfill may bring the risk of landfill instability and therefore cause a landfill landslide. Different from most landslides caused by natural factors, the landfill landslide is a man-made event that is mostly induced by human activities. The residue spoil caused by construction is a common problem faced by high-speed developing cities. A landfill distributed on the edge of the city is the most common solution. With the gradual increase in the number of years of landfilling, the residue soil volume of the landfill is approaching the peak storage capacity, and the production and life around the landfill are facing great safety risks. The amount of solid waste is growing quickly, especially in the developing countries at the rapid urbanization stage. For example, China produces approximately 30% of the world’s solid waste, and 40% is generated by construction, which reaches more than 200 million tons every year [
1]. Normally, there is a lack of physical barriers between the landfill area and the residential area. Excessive and overloaded landfills cause a very fast-moving debris flow of MSW. The failure of landfill can be disastrous if it occurs near communities of people. In 2000, a landslide at the Bayada landfill in Quezon City, Philippines, caused nearly 300 deaths. On 20 December 2015, a major landfill landslide occurred in Shenzhen, China [
2]. The accident directly caused 73 deaths, 4 missing, 17 injured, 33 buildings destroyed, and economic losses of 134 million US dollars. Due to the significant injuries and economic loss caused by landslides, more and more attention is being paid to landfill landslide disaster management for the purpose of mitigating the disaster [
3]. Landfill landslide monitoring, risk mapping, and process modeling are the main research aspects in current landslide disaster study.
As some characteristics of landslide can be witnessed before the failure, the frequent and large-scale monitoring and risk mapping would be helpful to avoid or mitigate the landslide disaster. Remote Sensing (RS) and Geographic Information System (GIS) are commonly used for monitoring, early warning, and risk assessment of landslide disasters. The landform features interpolated from remote sensing images, such as form, tone, and texture structure, can truly reflect the geomorphic environment. Therefore, RS are often used to identify the boundary, scale, morphological features, and disaster-generating environment of landslides. As early as 1995, SPOT satellite images, digital geologic maps, and digital numerical terrain models (DTM) were used to automatically draw landslide hazardous maps of Tahiti [
4]. Three aerial photographs were used to calculate the slip rate of the La Clapière landslide in the Mercantour Massif in eastern France from 1983 to 1999. By associating height information with high-resolution images, they visualized the evolution process of the landslide. GIS tools are then introduced to simulate the landslide movement process and assess the risk of slope damage [
5]. Recently, more and more advanced remote sensing sensors, platforms, and methods have increased the value of remote sensing in landslide monitoring [
6]. After the acquisition of landslide information, some typical spatial analysis methods in GIS, i.e., the analytic hierarchy process [
7,
8,
9] and self-similarity model [
10], are widely used to determine evaluation indicators for landslide risk management. Principal component analysis [
11], which uses cluster analysis, is also adopted to reduce the number of indicators. A logistic regression method was used to build a landslide sensitivity model with rainfall as the triggering factor for landslide risk assessment in Izmir, western Turkey [
12]. Several schemes, including probabilistic and deterministic methods, have been adopted to determine the rainfall thresholds for early warning of landslides induced by precipitation [
13]. The combination of multiple technologies helps to improve the accuracy of landslide risk calculations [
14]. Up to now, RS and GIS are widely used in many aspects of landslide research, including geomorphological environment identification, formation mechanism discovery, influencing factor analysis, and risk assessment.
RS and GIS can monitor and analyze the landslides in a relatively large scale, while the numerical mechanical modeling of landslides offer opportunities to observe the process with more details. The mechanical modeling is of great significance for disaster risk management as it can reveal the mechanism for landslide failure. According to the medium setting, they can be grouped into two types, i.e., the grid model and the non-grid model. The grid model is based on discrete grids, mainly including finite difference method (FDM), finite element method (FEM), finite volume method (FVM), etc. Early in 1970, a one-dimensional mud flow dynamic flood model was proposed to simulate the dam break of viscous debris flow [
15]. After that, a two-dimensional FLO-2D finite difference model was established in 1993. It integrates the Bingham fluid mechanics body model with the Bagnold model [
16]. The FLO-2D model was used to simulate the impact of varying degrees of precipitation on the landslide range and guide the construction for the retaining wall [
17]. Later, a two-dimensional MacCormack-TVD space–time finite difference method was proposed, which uses Coulomb friction and Volmi friction law to solve the coupling equations [
18]. The finite volume method is also adopted in many studies, to simulate the influence of various parameters (e.g., friction coefficients, flow rate, and critical slope) on landslide failure [
19,
20]. Some studies combined multiple grid models to simulate the landslide movement and the cascading damage [
21,
22]. Another type of numerical model uses a non-grid method based on a discontinuous medium model. They mainly include discrete element method (DEM), discontinuous deformation analysis (DDA), and smooth particle hydrodynamics (SPH). The discrete element method (DEM) has been used to calculate the jumping distance of the particle flow, the impact force on the obstacle, and the impact spectrum characteristic curve. It can also be used to simulate the debris flow and rock collapse [
23,
24,
25]. To simulate and analyze the disaster evolution, a multi-physics model considering the influence of different types of particle expansion rates on the movement of debris flow is constructed [
26,
27,
28]. With the in-depth research on fluid dynamics of debris rheology, the advanced numerical models have integrated more and more constitutive models, and can therefore approach the actual landslide process more precisely.
Like many disasters, the landfill landslide is a complex process composed of many subsequential sub-events, bringing continuous damages to the nearby communities. As the coupling of a hazardous event and damage happened in a sequential chain mode, many studies define this phenomenon as a disaster chain. Cascading effects are common in the chain of disaster. When the disaster moves to the next states, the initial impact can trigger other phenomena and amplify the effect [
29]. The casualties and property damage caused by a disaster chain are much more significant than the primary disaster event. Because of these subsequential events, many disasters bring about far more damages than people expect. Increasing interests move to studying the disaster in the framework of the disaster chain, aiming to get a full image of the disaster and its cascading effects. From the perspective of disaster risk management, landfill landslides can be divided into four key stages, from formation to occurrence: pre-failure stage, failure stage, post-failure destruction stage, and chain-cutting management stage [
30]. As reviewed above, many studies have been carried out on landslide monitoring, risk mapping, and modeling, although most of them treat the landslide as a single and independent event, and only focus on one stage, e.g., landslide monitoring at the pre-failure stage or physical modeling at the failure stage. However, to better meet the objectives of disaster risk reduction, it is required to analyze the full-period disaster chain of landslides.
Therefore, we propose a disaster chain analysis framework for a landfill landslide in terms of scenario simulation and chain-cutting modeling. Each stage of the landfill landslide is simulated or modeled using RS, GIS, or mechanical modeling techniques. Scenario simulation is one of the most cutting-edge scientific issues in the current disaster management field. Chain-cutting modeling based on observational data and mechanical models can predict the effect of landslide treatments and is very helpful for disaster prevention and emergency preparedness. In this work, we firstly reviewed a real case to summarize the general requirements for a disaster chain model for a landfill landslide. Guided by this model, we then proposed the specific steps for landfill landslide disaster chain analysis. The proposed framework is finally implemented in an undergoing landfill site to provide support for disaster prevention and mitigation.
2. Disaster Chain of Landfill Landslide
We firstly studied a typical landfill landslide case, the 2015 Shenzhen “1220” landslide, to generalize the disaster chain mode. Based on previous studies and historical data, we reconstructed the whole process of the Shenzhen 1220 landslide. The tempo-spatial characteristics of the disaster chain were discovered. We then summarized several possible slope treatments for this disaster. In our analysis framework, we divided the landfill landslide chain caused by solid waste into four stages following previous studies [
30], i.e., pre-failure, failure, post-failure, and chain-cutting stages. The corresponding urban landfill landslide disaster chain model is shown in
Figure 1.
The first stage is the pre-failure stage. At the beginning of the design of a landfill, the first consideration is its site selection. The landfill volume needs to be as large as possible to accommodate enough construction waste. When using natural pits for landfill, it is better to avoid the accumulation of large amounts of rainwater at the bottom. The foundation of the landfill should not be higher than the built-up area, and the exit should avoid directing to the communities. In the case of the 2015 Shenzhen landslide, the landfill materials were composed of residue spoils from the foundation excavation of engineering facilities and construction waste. However, the amount of residue spoils was far more than the construction wastes. The residue soil is soft and has a low compactness, which is likely to induce landfill settlement. The precipitation and surrounding surface runoff further infiltrate into landfill body, causing high moisture content in residue soil.
At the failure stage of this case, the high moisture content in the lower part of the residue soil landfill developed to an ooze floating layer. When the landfilling process keeps going, the surrounding precipitation and surface runoff continue to flow in, and the accumulation of water in the lower part becomes more and more serious. The groundwater level continues to rise and reach the retaining dam. As the ooze floating layer is not strong enough, the solid waste then cracks and slips out.
At the post-failure stage, great casualties and economic losses occur. In the 2015 Shenzhen landslide case, this accident caused 73 deaths, 4 missing, and 17 injured (3 serious injuries and 14 minor injuries). There were also 33 buildings (24 factory buildings and dormitories) destroyed and buried. The accident affected 90 companies and 4630 employees, causing a direct economic loss of 881.2223 million yuan. Some secondary disasters were also triggered, such as communication interruption and gas exploration, which brought great obstacles for the rescue activities.
For the chain-cutting purpose, we analyzed the disaster behavior of the 2015 Shenzhen landslide accident and deduced some disaster management measures. Specifically, at the pre-failure stage, we can reduce the hazard factors and prevent the occurrence of landslides by changing the environment of the disaster chain. Such management measures include, reducing the amount of landfill, monitoring the landfill volume frequently, carrying out waterproof treatment, constructing drainage ditches according to the mountain topography, reinforcing construction at key locations, and strengthening landfill management. At the failure stage, several measures can be taken to reduce the magnitude of landslides, such as constructing retaining dams to block the flow of soil, building drainage canals to delay the flow speed, and changing the direction of the landslide. At the post-failure stage, the main purpose for chain-cutting is to reduce casualties and property losses. As disasters have occurred, the disaster reduction methods at this stage are mainly to control the scope of the disaster’s impact, prevent the damage results from reacting to cause disasters to increase, and prevent various secondary and derivative accidents. Disaster reduction methods for chain-breaking include relocating dangerous factories or building retaining walls to suppress the spread of disasters, implementing an emergency plan system for rapid crowd evacuation, conducting drills, quickly organizing emergency rescue operations to evacuate residents, and avoiding other secondary and derivative disasters caused by building collapse or lifeline system damage.
3. Method
According to the disaster chain model generalized from this case, we propose a landslide disaster chain analysis framework to analyze the full-life period of a landfill landslide, for the purpose of supporting disaster prevention or mitigation. The processing procedure in this framework is composed of two parts: scenario simulation and chain-cutting modeling.
3.1. Scenario Simulation
Here, we propose a scenario simulation method covering the failure and post-failure stages of the landslide disaster chain. It includes the simulation of the landslide movement process during the failure and the simulation of the crowd evacuation after the disaster.
3.1.1. Failure Stage
Simulation of the failure process is implemented at this stage. The physical model of the landslide simulation is first established based on GIS and RS technology. As the landfill body is composed of residue soil and its failure depends on the topography and saturated water content, the landfill failure can be modeled as debris flow. Debris flow is a multiphase body, including rocks, gravel, silt, water, and other substances. These substances are mixed during the failure process and share the mechanical properties of fluid. The complex flow process follows the continuum assumption and the conservation of momentum principle in fluid movement. Our landfill failure simulations were carried out by the FLO-2D software package [
31], which uses the non-Newtonian fluid model and the central finite difference scheme to calculate the velocity, depth, and impact range of the debris flow.
Four types of parameters are required in the numerical simulation: terrain condition parameters, flow condition parameters, material parameters, and stability parameters of numerical difference calculations. Based on GIS and oblique photography technology, the digital terrain model (DTM) data can be created. The remaining three parameters can be calculated through the governing equations of the debris flow simulation mathematical model:
Continuity equation:
where
is rainfall intensity and
is mud depth.
Motion equation:
where
is the mud depth,
is the rainfall intensity,
is the time,
is the gravitational acceleration,
,
is the bed slope in the
,
direction, and
is the energy slope in the
,
direction.
Rheological equation:
where
is the viscous slope,
is the yield slope,
is the turbulence-diffusion slope,
is the yield stress,
is the viscosity coefficient,
is the relative density of the debris fluid,
is the laminar flow retardation coefficient, and
is the Manning coefficient.
The yield stress and viscosity coefficient of the fluid in the numerical model are given by the following two equations:
The BF (bulking factor) of the debris flow process line is given by:
where
is the volume concentration of the debris flow landslide. Referring to the definition of equilibrium concentration,
, the local conditions, and the experience of debris flow numerical simulation, the volume concentration of the debris flow numerical simulation and its expected range can be determined. Generally,
= 0.3~0.7 and
are acceptable. Equilibrium concentration,
, is given by:
where
is the density of clear water,
is the density of debris flow particles,
is the internal friction angle of the soil material,
is the slope, and
is the bulk density of the accumulated muck.
The appropriate parameter values can be selected according to the requirements for analysis accuracy in the FLO-2D mode. The following conditional expressions must be met during the calculation:
where
is the calculation time,
is the dynamic wave stability coefficient,
is the grid size,
is the bed slope, and
is the unit flow.
3.1.2. Post-Failure Stage
Crowd evacuation simulation is carried out at the post-failure stage. The people living around the landfill have been exposed to huge safety hazards when the landfill becomes unstable. The previous failure simulation can reveal the process and affected areas of the dynamic debris flow. The buildings, roads, and population located in the affected areas are all disaster-bearing bodies. We can collect the information about buildings and roads from satellite remote sensing images and retrieve the population data from cellular signaling data. The population data contain the population structure and spatial population density. Then, we used the PathFinder software [
32], which adopted the continuous model (Agent-base), to carry out the evacuation simulation. The input parameters include 3D geo-space, population density, population structure, locations of entrance and exit, and evacuation routes. The Steering mode of this software uses a combination of path planning, guidance mechanism, and collision handling to control the movement of people. It is capable to consider the real psychological state of people. The software supports three-dimensional modeling and can create evacuation obstacles using irregular triangular networks.
In the process of evacuation simulation, the A* algorithm and two-dimensional grid are used to arrange paths for evacuees. String pulling technology is used to smooth the path. The A* algorithm is a method of calculating the shortest path in a static grid. The calculation formula of the algorithm is given by:
where
is the evaluation function of node n from the starting point to the target node,
is the actual cost of starting to take you to node n, and
is the estimated cost of the optimal path from node n to the target node.
3.2. Chain-Cutting Modeling
Chain-cutting modeling is implemented by analyzing the causes of disasters at each stage of the disaster chain, from which the engineering treatment measures can be deduced correspondingly. Specifically, we use FLO-2D to simulate the flow velocity, flow depth, impact force, and impact range of this debris flow-like landslide under construction engineering measures. The results are compared with the situation without construction measures to evaluate the effect of management measures.
Through the analysis of the landslide disaster chain model derived from the 2015 Shenzhen Landslide, some treatment measures are suggested, and we can select appropriate management measures to model according to local conditions. At the pre-failure stage, the treatment methods include: (1) change the direction of the artificial waste landfill, (2) reinforce the pile, (3) reduce water volume, (4) strengthen management, and (5) increase the width of the waste landfill platform to reduce the slope rate. At the failure stage, treatment methods include: (1) building a retaining dam, (2) establishing a diversion channel, and (3) burying culvert pipes to reduce the moisture content of the slag soil. At the post-failure stage, treatment methods mainly include improving evacuation efficiency. The simulations of all these processes are performed separately. By comparing the simulation results with and without the measurements, we can verify their effectiveness and feasibility.
4. Study Area and Experiment Settings
We then applied the proposed method to study the Xinwuwei landfill in Nanshan District, Shenzhen, China. The site of the landfill is located on a mountain in a country park in the urban built-up area. The main body of the landfill is in the cove of the mountain. The average elevation of the landfill is 120 m, which is nearly 50 m higher than the built-up area below the mountain. The site is close to major urban traffic routes, such as expressways, urban arterial roads, secondary arterial roads, and rail transit (
Figure 1). The total land area of this landfill is 427,000 m
2, and the designed total storage capacity is 4.1 million m
3. According to the current topography and traffic conditions, the plan layout of the landfill project is divided into 1# plot and 2# plot (
Figure 2). Among them: 1# plot is a residue soil landfill area with a storage capacity of 3.91 million m
3, and 2# plot is a circulation area with a storage capacity of 190,000 m
3.
Referring to the disaster chain model of the 2015 Shenzhen “1220” accident, we hypothesized a disaster scenario of this landfill for the landslide simulation and chain-cutting modeling. The disaster setting is as follows: a continuous heavy rain is assumed to occur in this area, bringing rapid water accumulation in the landfill. The water is not drained in time, and results in saturation of the soil moisture in the landfill. At the same time, due to ultra-high landfill loading of the landfill and incomplete treatment of the slope, the landfill becomes instable and the landslide then happens, which causes great damage to the surrounding expressways, railways, and industrial parks. Many casualties and huge property losses are resulted correspondingly.
4.1. Failure Simulation Setting
The parameters required for the control equation of the mathematical model of the debris flow simulation are shown in
Table 1. The volume concentration and relative density of the landslide debris flow are given based on the field survey results and the empirical relationship formula of the debris flow. The K value of the retardation coefficient is related to the appearance of the debris flow. According to the field survey, the K value is set as 2285. The Manning coefficient reflects the impact of different land use types on the movement of debris flow. The value of n is 0.04, as recommended by the FLO-2D program, which is suitable for vegetation types such as shrubs and forest litter, pasture. The yield stress and viscosity coefficient are set by experience.
According to field investigation and UAV surveying, the volume concentration of this landfill is 0.5 and the volume is 4.1 million m
3. Historical rainfall data of the Shenzhen city suggest an average rainfall of 459.3 mm. By considering these basic data, we designed a debris flow process line for the landfill, as shown in
Figure 3.
4.2. Post-Failure Simulation Setting
Crowd evacuation is simulated at the post-failure stage to evaluate the response to the landslide. The settings for the crowd evacuation simulation are as follows:
(1) Number of evacuees. The on-site survey data shows that there are mainly 5 small industrial zones, about 20 office buildings, and 3 residential buildings below the landfill. The mobile phone signaling data suggest the evacuated population to be approximately 10,330. To include the visitors, we set the evacuation number to be 1.5 times larger than the actual number of people, which is equal to 15,495.
(2) Evacuation exit. Investigation in the landfill shows that the industrial area below the landfill is surrounded by mountains on three sides. It can only be accessed through its gate open to the municipal road. We therefore only put exits on the municipal road in the modeling, i.e., the east and west ends of the road. The evacuation destinations are set to be some emergency shelters, including the university city and elementary schools on the north side of the municipal road. As the landslide break is on the northwest side of the evacuation area, this evacuation simulation does not set an evacuation passage on the east and south sides.
(3) Evacuation time. The debris flow simulation will give a landslide alarm time value T1 in the area. We set the response time of the person to be evacuated to 10 s after receiving the alarm signal, and therefore the total delay time is (10 + T1) s.
4.3. Chain-Cutting Modeling Setting
At the failure stage, the main methods of landfill management include constructing retaining dams and diversion canals or burying of seepage culverts in the pile to reduce the moisture content of the muck. Here, we modeled the first three measures in this landfill, i.e., adding diversion channels, building a retaining dam, and combination of diversion channel and retaining dam. For better comparison of different treatments, the numerical simulation of debris flow in three conditions will be overlaid to the satellite image.
At the post-failure stage, some effective chain-cutting measures can be taken, which include reducing the number of people in the industrial zone and increasing the evacuation speed by improving early warning capabilities and conducting daily emergency drills. In the post-failure chain-cutting modeling, we kept the evacuated population the same as the current population but increased the evacuation speed of people. To evaluate the effect of chain-cutting, we will separately investigate the total evacuation time, the dynamic change of evacuated and un-evacuated people, and the flow of people in different evacuation routes.
6. Conclusions
The disaster chain represents hazards that have a causal relationship, similar to dominoes [
33], where a series of events related to the primary disaster behave in a sequential chain mode. It includes many secondary or derivative disasters triggered by the primary disaster, resulting from the interaction between the hazard-causing factors and the disaster-bearing body in the disaster environment. The purpose of studying disaster chains is to explore disaster mitigation measures for chain-cutting in key stages.
This paper applied the disaster chain model to the disaster scenario simulation and chain-cutting modeling. We carried out the simulation at different stages of landfill landslide disasters, aiming at understanding the basic occurrence and development characters of such disaster. Then, several chain-cutting treatments were proposed. We further modeled these treatments to find out the most effective ones for landfill disaster prevention and mitigation. The main contributions of this paper are as follows:
(1) We used disaster chain theory to guide scenario simulation and chain-cutting modeling. By dividing the disaster into several key stages, i.e., pre-failure, failure, post-failure, and chain-cutting, we clearly revealed the hierarchical structure of the landslide disaster chain. It facilitates the analysis and understanding of various hazardous factors in the disaster.
(2) We took advantage of various advanced techniques to implement the disaster chain analysis. We used remote sensing, 3DGIS, non-Newtonian fluid model, central finite difference scheme, and Agent-base steering model to establish a landslide disaster chain analysis framework. The proposed framework provides a multi-dimensional, dynamic, and interactive way for interpolating the spatio-temporal characteristics of the landfill landslide disaster.
(3) We deduced optimized disaster treatments by integrating scenario simulation and chain-cutting modeling. The scenario simulation suggested potential hazardous risks and treatments. Through chain-cutting modeling, the effectiveness and feasibility of these treatments were further validated, and the most optimized solution could be deduced.
There are diverse factors (e.g., meteorological, topographical, and social factors) that can trigger a landfill landslide. As this work is a preliminary study of the disaster chain analysis of landfill landslide, our study only treated a limited number of them, and the quality of data input for simulation still needs to be improved and more chain-cutting treatments can be modelled. In the future, we can extend our disaster chain analysis by including more factors not covered in this study. Also, we can further improve the time and space resolution of landfill topography by taking more frequent UAV field surveying and using more advanced cameras. The population in affected areas can also be counted more accurately by using surveillance video or household survey data. The treatment measures for chain-cutting can be further optimized, e.g., modeling multi-layer retaining dams to limit the debris flow to a smaller area.