Using Uncertain DM-Chameleon Clustering Algorithm Based on Machine Learning to Predict Landslide Hazards

Abstract

Landslide hazard prediction is a difficult, time-consuming process when traditional methods are used. This paper presents a method that uses machine learning to predict landslide hazard levels automatically. Due to difficulties in obtaining and effectively processing rainfall in landslide hazard prediction, and to the existing limitation in dealing with large-scale data sets in the M-chameleon algorithm, a new method based on an uncertain DM-chameleon algorithm (developed M-chameleon) is proposed to assess the landslide susceptibility model. First, this method designs a new two-phase clustering algorithm based on M-chameleon, which effectively processes large-scale data sets. Second, the new E-H distance formula is designed by combining the Euclidean and Hausdorff distances, and this enables the new method to manage uncertain data effectively. The uncertain data model is presented at the same time to effectively quantify triggering factors. Finally, the model for predicting landslide hazards is constructed and verified using the data from the Baota district of the city of Yan’an, China. The experimental results show that the uncertain DM-chameleon algorithm of machine learning can effectively improve the accuracy of landslide prediction and has high feasibility. Furthermore, the relationships between hazard factors and landslide hazard levels can be extracted based on clustering results.

Landslide susceptibility mapping using O-CURE and PAM clustering algorithms

“…Rainfall factor was also selected because frequent rain waters can easily penetrate the soils due to sinkholes, loess structural joints, and weathering fractures, forming saturated areas on impermeable layers, which leads to a reduction in soil strength and gravity, creating pore water pressure and increase in the weight of the rocks and soil mass. The selection of this factor is also supported by the historic reports that recorded the frequency of landslide events during the rainy season (Zhang and Liu, 2010;Hu et al, 2019). Figure 4A-G represents the thematic maps for the factors whereby: maps for elevation, slope angle, slope aspect, and profile curvature were generated from DEM at a 25 m resolution and a scale of 1:10,000, NDVI and lithology maps were developed from ETM + remote sensing images, and geology map at a scale of 1: 50,000 respectively, and the rainfall map was created based on meteorological data at a scale of 1:50,000.…”

“…The average annual temperature and rainfalls are 10 °C and 550 mm respectively, and the heavy rainfall varies between 58 and 117 mm extending between June and October (Zhang and Liu, 2010). It has also been observed that rainfall triggers most landslides in the area (Mao et al, 2017;Hu et al, 2019;Mao et al, 2021a).…”

“…There is no standard for the selection of these factors, and different studies select their factors based on various criteria. Based on previous research studies in Baota District (Hu et al, 2019;Mao et al, 2021a;Mao et al, 2021b), we selected 7 landslideconditioning factors for modeling: elevation, slope angle, slope aspect, profile curvature, lithology, vegetation coverage index (NDVI), and rainfall. The elevation factor is associated with landslide occurrences, especially in plateau areas (Lee et al, 2018).…”

“…Landslide density [LD (Hu et al, 2019)] is computed using the number of landslides per square kilometer (km 2 ) of a mapping unit in a subclass and is applied to specify the susceptibility level of that subclass. When the number of landslides in a subclass is zero, which means the LD is also equal to zero, then, the attribute values (which describe more the characteristics of the area) based on geology expertise were applied to specify the susceptibility level.…”

“…Over decades, these methods have been widely used in other fields such as marketing research, pattern recognition, and image processing, but very rarely explored in LSM studies (Huang et al, 2020;Su et al, 2022). In recent years, making use of the advantages of these methods, some landslide researchers have also shown interest and conducted LSM studies using these methods (Wan et al, 2015;Wang et al, 2017;Hu et al, 2019;Mao et al, 2021a;Mao et al, 2021b;Hu et al, 2021;Pokharel et al, 2021;Yimin et al, 2021;Mao et al, 2022). From the analysis of these studies and other traditional clustering algorithms, some limitations were observed: the inability to detect subclasses with arbitrary shapes, sensitivity to noise, inability to perform well in large study areas with large datasets, and principally a standard method to process the uncertain data (rainfall) has not being obtained yet.…”