A comparison of Support Vector Machines and Bayesian algorithms for landslide susceptibility modelling

Abstract

In this study, the main goal is to compare the predictive capability of Support Vector Machines (SVM) with four Bayesian algorithms namely Naïve Bayes Tree (NBT), Bayes network (BN), Naïve Bayes (NB), Decision Table Naïve Bayes (DTNB) for identifying landslide susceptibility zones in Pauri Garhwal district (India). First, landslide inventory map was built using 1295 historical landslide data, then in total sixteen influencing factors were selected and tested for landslide susceptibility modelling. Performance of the model was evaluated and compared using Statistical based index methods, Area under the Receiver Operating Characteristic (ROC) curve named AUC, and Chi-square method. Analysis results show that that the SVM has the highest prediction capability, followed by the NBT, DTNBT, BN and NB, respectively. Thus, this study confirms that the SVM is one of the benchmark models for the assessment of susceptibility of landslides.     

Spatial prediction of rainfall-induced shallow landslides using hybrid integration approach of Least-Squares Support Vector Machines and differential evolution optimization: a case study in Central Vietnam

This study represents a hybrid intelligence approach based on the differential evolution optimization and Least-Squares Support Vector Machines for shallow landslide prediction, named as DE–LSSVM_SLP. The LSSVM is used to establish a landslide prediction model whereas the DE is adopted to search the optimal tuning parameters of the LSSVM model. In this research, a GIS database with 129 historical landslide records in the Quy Hop area (Central Vietnam) has been collected to establish the hybrid model. The receiver operating characteristic (ROC) curve and area under the curve (AUC) were used to assess the performance of the newly constructed model. Experimental results show that the proposed model has high performances with approximately 82% of AUCs on both training and validating datasets. The model’s results were compared with those obtained from other methods, Support Vector Machines, Multilayer Perceptron Neural Networks, and J48 Decision Trees. The result comparison demonstrates that the DE–LSSVM_SLP deems best suited for the dataset at hand; therefore, the proposed model can be a promising tool for spatial prediction of rainfall-induced shallow landslides for the study area.     

Urban flood vulnerability assessment in a densely urbanized city using multi-factor analysis and machine learning algorithms

Theoretical and Applied Climatology - Flood is considered as the most devastating natural hazards that cause the death of many lives worldwide. The present study aimed to predict flood...     

Paleozoic subduction of the southern Dunhuang Orogenic Belt,northwest China: metamorphism and geochronology of the Shuixiakou area

Amphibolites in the Shuixiakou area of the southern Dunhuang Orogenic Belt, southernmost Central Asian Orogenic Belt (CAOB), occur as lenses within hornblende-biotite-plagioclase gneiss or pelitic schist, exhibiting block-in-matrix feature of tectonic mélange. Three generations of metamorphic mineral assemblages (M1, M2, and M3) have been recognized in the garnet-bearing amphibolite lenses. The metamorphic prograde assemblage (M1) is documented with inclusion trails (hornblende + plagioclase + quartz) within garnet porphyroblasts, and are estimated to be formed under 610–690 °C and 6.5–10.2 kbar. The metamorphic peak assemblage (M2) consists of garnet + hornblende + clinopyroxene + plagioclase + quartz in the matrix and records metamorphic peak P-T conditions of 720–750 °C and 13.4–14.7 kbar. The retrograde assemblage (M3) is represented by the symplectic assemblage (hornblende + plagioclase + quartz ± biotite ± magnetite) rimming the garnet porphyroblast, formed in the decompression stage under P-T conditions of 630–730 °C and 3.8–7.2 kbar. The derived metamorphic P-T paths show similar tight clockwise loops including nearly isothermal decompression processes, typical of orogenic metamorphism. SIMS dating of metamorphic zircons from the amphibolites confirm that the high-pressure metamorphism (M2) occurred at ca. 438–398 Ma.     

A Method to Represent a Well in a Three-Dimensional Discrete Fracture Network Model

In discrete fracture network (DFN) modeling, fractures are randomly generated and placed in the model domain. The rock matrix is considered impermeable. Small fractures and isolated fractures are often ignored to reduce computational expense. As a result, the rock matrix between fractures could be large and intersections may not be found between a well introduced in the model and the hydraulically connected fracture networks (fracture backbones). To overcome this issue, this study developed a method to conceptualize a well in a three-dimensional (3D) DFN using two orthogonal rectangular fractures oriented along the well's axis. Six parameters were introduced to parameterize the well screen and skin zone, and to control the connectivity between the well and the fracture backbones. The two orthogonal fractures were discretized using a high-resolution mesh to improve the quality of flow and transport simulations around and along the well. The method was successfully implemented within dfnWorks 2.0 (Hyman et al. 2015) to incorporate a well in a 3D DFN and to track particles leaving an injection well and migrating to a pumping well. Verification of the method against MODFLOW/MODPATH found a perfect match in simulated hydraulic head and particle tracking. Using three examples, the study showed that the method ensured the connectivity between wells and fracture backbones, and honored the physical processes of flow and transport along and around wells in DFNs. Recommendations are given for estimating the values of the six introduced well parameters in a real-world case study.     

A Semi-Evolutive Partially Local Filter for Data Assimilation

The singular evolutive extended Kalman (SEEK) filter has been proposed recently by Pham et al. (1997) for data assimilation into numerical oceanic models. This filter has been applied in different realistic ocean frameworks and has provided satisfactory results ( Pham et al., 1997; Verron et al., 1998). However, the SEEK filter remains expensive in real operational assimilation. To reduce cost and obtain a better representativity, we introduce the idea ‘local correction basis'. Such basis however cannot be made to evolve according to the model without destroying its locality property. Therefore we shall keep this basis fixed and we augment it by a few global basis vectors which evolve. The resulting semi-evolutive partially local filter is much less costly to implement than the SEEK filter and yet can yield better results. In the first application, validation twin experiments are conducted in a realistic setting of the OPA model over the tropical Pacific Ocean.     

Sources of anomalous transient electric signals (ATESs) in the ULF band in the Lamia region (central Greece): electrochemical mechanisms for their generation

Anomalous transient electric signals (ATESs) in the ultra low frequency (ULF) band have been often observed during magnetotelluric (MT) investigations [Nature 319 (1986) 310; Phys. Earth Planet. Int. 114 (1999) 141; Geophys. J. Int. 142 (2000) 948], but their origin was unknown until now. They have the same characteristics as the so-called seismic electric signals (SES) claimed to be earthquake precursors by the VAN group (e.g. [Tectonophysics 110 (1984) 73] and later works by this group). Our analysis suggests that the so-called SES could be of anthropic origin. Following the devastating 7 September 1999 Athens earthquake, the VAN group claimed that a SES had been recorded at LAM station (Lamia, central Greece) some days prior to the main shock and that a second SES, which might correspond to an impending even larger earthquake, had been observed after the main shock. In the 2 years after the Athens main shock, no subsequent large earthquakes have occurred near Athens. We conducted a campaign of measurement in the Lamia region in May and June 2001. The results show that ATESs, which look like SES, have several different sources: pump-stations for ground-water, high power electric lines, and factories located to the SE of Lamia city. The ATESs can be generated by two electrochemical mechanisms of metallic electrode polarization: the “galvanic cell” and the “ac electrolytic cell” which are studied by simulated field experiments and discussed in detail in Appendix A. These two mechanisms can occur over a wide range of length scales in the field. Any isolation failure in buried metallic conductors, such as electrical and telecommunication networks, oil, water and gas pipes, railways, high power electric lines, factories and so on, can produce a galvanic cell or an ac electrolytic cell, or both, which could generate, under some circumstances, an “overvoltage”, the ATES. Finally, the absence of a magnetic signal has been observed during ATES and does not constitute a firm criterion for SES [Acta Geophys. Pol. 44 (1996b) 301]. Thus, great care must be taken when claiming the existence of electric precursors of seismic or volcanic events.     

Mapping and surveillance of active fissure zones on a volcano by the self-potential method, Etna, Sicily

The self-potential (S.P.) method, often employed in base-metal exploration, has successfully been applied to volcanological problems.Test surveys on the site of Mount Etna (Sicily) have revealed some positive anomalies as high as a few hundred millivolts over recent and old eruptive fissures. The S.P. technique allows one to determine the precise location of the fissures.The interpretation of the geophysical results (including S.P.), taking into account the structure of Mount Etna, has shown two important zones upon which several S.P. investigations have been carried out.Studies of the causes of these S.P. anomalies have shown the importance of the streaming potential as the principal source mechanism.