Application of Markov-chain model for vegetation restoration assessment at landslide areas caused by a catastrophic earthquake in Central Taiwan
Introduction
The catastrophic Chi–Chi earthquake occurred on September 21, 1999 in Central Taiwan. It caused massive landslides – 2415 people died and 11,306 people were injured. Due to massive debris piled on the slope land after the Chi–Chi earthquake, severe disasters were caused during the typhoon season, such as debris flow, floods, sediment disasters, etc. Effectively monitoring landslides and deriving vegetation restoration information is a vital issue. Ten years have passed since the earthquake occurred. Vegetation restoration at landslide areas has been studied by many researchers (Singh et al., 2000, Lin et al., 2004, Lin et al., 2006, Lin et al., 2008). Integrating remote sensing (RS), geographic information system (GIS), and image classification technique for landslide hazard monitoring and vegetation restoration assessment is very important in the future.
There are many computer models for simulating the dynamics of ecological succession. The Markov chain is mathematically and conceptually the most straightforward succession model presently in use (Usher, 1979, Lepš, 1988). Anderson and Goodman (1957) developed methods for estimating the transitive matrix from observing the states of a system through time. Their methods have been widely developed and applied (Cox, 1972, Kalbfleisch and Lawless, 1985). This has led to its application to vegetation succession (Balzer, 2000, Logofet and Lesnaya, 2000, Benabdellah et al., 2003, White, 2005). Aaviksoo (1993) integrated aerial photograph, GIS, and the Markov chain model to simulate the changes in plant cover and land use types in Estonia (corresponding to 1950s, 1960s, and 1980s). Muller and Middleton (1994) used a first order Markov chain as a stochastic model to make quantitative comparisons of land cover changes between discrete time periods from 1935 to 1981 in the Niagara Region, Ontario, Canada. The result pointed out that the urbanization of agricultural land was the predominant land use change and a continuous exchange of land area occurs between wooded and agricultural land use categories. Boerner (1996) applied the Markov Inertia and Dynamism models on two contiguous Ohio landscapes. Jenerette and Wu (2001) developed the Markov cellular automata model to simulate land use changes in Central Arizona.
From the reasons mentioned above, this paper integrates the RS, GIS, image classification and vegetation succession model (Markov chain) to assess vegetation restoration at landslide areas. It is necessary to understand the reason for non-vegetation restoration at a site because such large mass movements of rocks and soil have already preconditioned the area for failure and may ultimately trigger the next potential hazard.
Section snippets
Chiufenershan
The Chiufenershan landslide is located about 15 km NNE of the epicenter (23.85°N and 120.81°E). The peak ground acceleration (PGA) recorded by a seismic monitoring station from Taiwan's Central Weather Bureau about 6 km north of this landslide was 465.3 gal in the east–west component, 370.5 gal in the north–south component, and 274.7 gal in the vertical component during the Chi–Chi earthquake. The landslide area lies between 23°58′08″N and 23°56′52″N latitudes and between 120°49′36″E and
Materials
Before vegetation restoration assessment, landslide elimination is the most important and first work. NDVI (Normalized difference vegetation index) can be used to quantify vegetation on the surface and is suitable for understanding land cover changes (Hsien, 1996). However, actual landslides may include the collapse of bare lands. If landslide elimination was considered only using image differencing, the bare land collapse areas would not be extracted. Chuang and Lin (2010) used pre- and
Training sample analysis
Training samples, whether good or bad, influence the image classification accuracy. Selecting good training samples for image classification is very important. The catastrophic earthquake caused massive landslides on the mountain slopes, making the growth conditions for plants much worse.
Generally, the growth conditions for grass are easier with a much faster growth rate than trees. Grass is considered as a pioneer species (or an important reference) for early vegetation succession. That is why
Conclusion
Grass can be considered a pioneer species at landslide areas and suitable for a reference index for vegetation restoration. Grass must be extracted individually. This study presented original wavebands plus vegetation index coupled with the back-propagation neural network to produce image classification at landslide areas. Kappa coefficients of image classification were more than 0.7 and thus could be applied for vegetation restoration assessment. Land cover from each stage was prepared using a
References (30)
- et al.
Markov chain models for forest successions in the Erzgebirge, Germany
Ecological Modelling
(2003) A review of assessing the accuracy of classifications of remotely sensed data
Remote Sensing Environment
(1991)- et al.
Vegetation recovery assessment at the Jou-Jou Mountain landslide area caused by the 921 Earthquake in Central Taiwan
Ecological Modelling
(2004) - et al.
Assessment of vegetation recovery and soil erosion at landslides caused by a catastrophic earthquake: a case study in Central Taiwan
Ecological Engineering
(2006) - et al.
Eco-environmental changes assessment at the Chiufenershan landslide area caused by catastrophic earthquake in Central Taiwan
Ecological Engineering
(2008) - et al.
The mathematics of Markov models: what Markov chains can really predict in forest succession
Ecological Modelling
(2000) - et al.
The effect of internal parameters and geometry on the performance of back-propagation neural networks: an empirical study
Environmental Modelling and Software
(1998) - et al.
Neural network classification of remote-sensing data
Computer Geoscience
(1995) - et al.
Analysis of the Chiufengershan landslide triggered by the 1999 Chi–Chi earthquake in Taiwan
Engineering Geology
(2003) An economic analysis of ecological monitoring
Ecological Modelling
(2005)
A new sub-pixel mapping algorithm based on a BP neural network with an observation model
Neurocomputing
Changes of plant cover and land use types (1950s to1980's) in three mire reserves and their neighbourhood in Estonia
Landscape Ecology
Statistical inference about Markov chains
The Annals of Mathematical Statistics
Markov chain models for vegetation dynamics
Ecological Modelling
Markov models of inertia and dynamism on two contiguous Ohio landscapes
Geographical Analysis
Cited by (27)
Vegetation patterns on a landslide after five years of natural restoration in the Loess Plateau mining area in China
2019, Ecological EngineeringCitation Excerpt :These results all indicated that the potential of natural restoration to promote the vegetation succession on landslide with native species in the Loess Plateau. In previous studies, some researchers have monitored the restoration of vegetation after a landslide that had been caused by an earthquake (Chuang et al., 2011; Huang et al., 2017). Most of the vegetation had been restored naturally, although some cases also demonstrated that revegetation could be affected by subsequent geological disasters.
Bilberry vs. cowberry in a scots pine boreal forest. II. Alternate modes of prediction
2019, Ecological ModellingCitation Excerpt :Markov-chain models have long been known for their immanent phenomenological nature (Jeffers, 1978, Ch. 5). Normally lacking any cause–effect mechanisms, the Markov-chain models of plant succession are also known for their extreme commitment to observation data – at the level of either community “relay” (Knyazkov et al., 1992; Rego et al., 1993; Vinogradov and Shitov, 1994; Logofet et al., 1997; Logofet and Lesnaya, 2000; reviews in Balzter, 2000, and Logofet, 2010; Chuang et al., 2011; Logofet et al., 2016), or the “tree-by-tree replacement” (Horn, 1975; Acevedo, 1981; a review in Acevedo et al., 1995), or at the fine scale (Hobbs, 1984; Samoylov and Tarkhova, 1985; Wootton, 2001; Hill et al., 2004; Baasch et al., 2010). Therefore, Markov-chain models provide for the most objective picture of what is going on in reality and of what follows from this picture for the future unless drastic changes in the environment exclude the chance of reliable predictions.
Bilberry vs. cowberry in a Scots pine boreal forest: Exclusion or coexistence in a post-fire succession?
2019, Ecological ModellingCitation Excerpt :However, the same kind of data are also available from 6 observations during 25 years, which makes it possible to do the same for five pairs of examination/re-examination, bearing in mind that the results may differ. In formal terms, it means that the transition matrix, P(t), of the Markov chain depends on the year, t, of examination, or the chain is inhomogeneous (or time-dependent), as it was in other projects with Markov models of succession (Usher, 1981; Knyazkov et al., 1992; Logofet et al., 1997; Benabdellah et al., 2003; Logofet, 2010; Logofet et al., 2016) or landscape dynamics (Baker, 1989; Vinogradov et al., 1989; Li, 1996; Wang and Zhang, 2001; Yemshanov and Perera, 2002; Chuang et al., 2011). Inevitable differences in the time-dependent transition matrices occur because of fluctuations in species abundances caused either by weather factors or by disturbances of various kinds including phytophagan impacts.
Dynamic simulation of vegetation abundance in a reservoir riparianzone using a sub-pixel markov model
2015, International Journal of Applied Earth Observation and GeoinformationCitation Excerpt :In addition, the Markov is also applied to vegetation dynamic simulation. For instance, using five land cover maps extracted from five SPOT satellite images, Chuang et al. (2011) applied the Markov model to the vegetation restoration assessment, and their research found that vegetation restoration is on-going in Central Taiwan after a catastrophic earthquake. To evaluate patterns and causes in landscape dynamics in the Niger Delta, Alex and Blackburn (2011) derived the transitive probability matrix from two forest landscape maps and performed a forward Markov simulation to generate the future status of forests.
Spatial modelling of natural disaster risk reduction policies with Markov decision processes
2014, Applied GeographyCitation Excerpt :Musa, Gajere, and Akinyede (2012) emphasised that the valleys of the Rivers Kamandagu Gana and Kamandugu Yobe are among the most vulnerable areas from desertification. Applied in the vegetation restoration assessment at landslide areas caused by catastrophic earthquake in Central Taiwan, the Markov-chain model showed that vegetation restoration at the Chiufenershan and Ninety-nine peaks landslide areas is ongoing, but has been disturbed by natural disasters (Chuang, Lin, Chien, & Chou, 2011). In modelling emergency evacuation for major hazard industrial sites, Georgiadou, Papazoglou, Kiranoudis, and Markatos (2007) used the Markov–Monte Carlo model to support decisions for emergency response concerning for example areas that must be evacuated or not in certain circumstances and for land use planning issues such as providing information about the need to increase transportation network capacity and safe shelters.
Understanding forest changes to support planning: A fine-scale Markov chain approach
2012, Developments in Environmental ModellingCitation Excerpt :Land use and cover have profound impacts on the environment, the economy, and society. Generally, there are major conflicts among the urban, agricultural, and natural uses of land (Araya and Cabral, 2010; Chen et al., 2010; Chuanga et al., 2011; Falahatkar et al., 2011; Muller and Middleton, 1994). In different countries, researchers and planners are required to recognize several different processes that influence land use and cover.