Susceptibility Zoning and Prioritization of the Factors Affecting Landslide Using MaxEnt, Geographic Information System and Remote Sensing Models (Case study: Lorestan Province)

The main objective of this research is to prioritize the factors affecting the occurrence of landslide and its susceptibility zoning in Lorestan province using the maximum entropy and MaxEnt models. To do this research, 11 factors affecting the occurrence of landslide including height, slope, aspect, surface curvature, distance from the stream, fault and road, lithology, land use, rainfall, and topographic humidity index have been used. In this research, 30, 40, 50, 60 and 70 percent of landslides were evaluated for validation to determine the sensitivity and accuracy of the model. For evaluation of the model, the relative recognition function curve (ROC) was used. From the total of 176 landslides, 70% of the data was used as the test data and 30% as the validation data using Mahalanobis distance method and the accuracy of the model in the testing and validation stages based on the curve level was reduced. The results showed that 35.5% of the province of Lorestan has a landslide sensitivity. Based on jackknife diagram, rainfall, distance from road, lithology and land use layers were the most important factors influencing the sensitivity of landslide. The AUC level based on the relative function recognition curve indicated a 90% accuracy (excellent) of the maximum entropy method at the training stage and 83% (very good) at the validation stage to determine the landslide susceptibility. The results of this study will be suitable for provincial administrators and managers in order to land planning and reduce the damage caused by landslide occurrence. Mass movements, including landslide, is one of the most important issues in natural hazards, because its occurrence can cause many human and economic losses, especially in mountainous areas (Symeonakis et al., 2016). Regarding the destructive effects of landslides on natural resources, as well as human habitats and erosion of significant volumes of valuable soils, the identification of susceptible areas and zoning of potential occurrence or landslide susceptibility is vital and very important (Zhang et al., 2019). In recent years, the use of GIS and remote sensing along with machine learning methods has created a new step in the zoning of landslide occurrences. Lorestan province is a vulnerable area to landslide hazard due to the mountainous and wetness conditions. Therefore, the main objective of this research was to prioritize the factors affecting the occurrence of landslide and its susceptibility zoning in Lorestan province using the maximum entropy and MaxEnt model.

Lorestan province with an area of 2829612 hectares is one of the major provinces in the west of the country. To do this research, 11 factors affecting the occurrence of landslide including altitude, slope, aspect, surface curvature, distance from the stream, fault, and road, lithology, land use, rainfall, and topographic humidity index have been used. The required maps were prepared using GIS and RS techniques. In this research, 30, 40, 50, 60 and 70 percent of landslides` division were evaluated for validation to determine the sensitivity and accuracy of the model. For evaluation of the model, the relative recognition function curve (ROC) was used. Using Mahalanobis distance method, from the total of 176 landslides, 70% of the data was used as the test data and 30% were utilized as the validation data for having the best classification. The difference of the current research with other similar studies was that in this study, use was made of Mahalanobis distance method for classification of validation data and training instead of random classification. The Mahalanobis distance helps to classify data richness and prevents random selection of points for validation. Maximum entropy method (MaxEnt model) is one of the methods of machine learning and one of the main advantages of MaxEnt model is the ability of this model to identify the most important variables and sensitivity analysis of variables using Jackknife method, which has been investigated in the current study.

The results showed that 35.5% of the province of Lorestan had landslide susceptibility. Based on Jackknife diagram, rainfall, distance from road, lithology and land use were, respectively, the most important factors influencing the susceptibility of landslide. The AUC level, based on the relative function recognition curve, indicated 90% accuracy (excellent) of the maximum entropy method at the training stage and 83% (very good) at the validation stage to determine the susceptibility of landslide occurrence.

Discussion and conclusion

Landslide is considered as one of the most dangerous natural disasters in the world. In this study, taking into account the affective environmental and human factors, and using the maximum entropy method, the map of landslide susceptibility of Lorestan province was prepared. The results showed that factors such as rainfall, distance from the road, lithology, land use, distance from the fault and slope were the most important factors influencing landslide susceptibility with the participation of over 60%, regarding which, land use management and road construction principles need human activity interventions. The drawn ROC curve showed that the accuracy of the model in the estimation of landslide susceptibility regions both in the stage of the test and in the validation stage was excellent and very good, which meant the excellent performance of the model. According to the obtained results, it can be said that MaxEnt model had a high ability to determine areas with landslide susceptibility and due to the speed and accuracy of the model,it is suggested that in similar researches, especially in developing countries, due to the lack of facilities and financial resources, as well as the time consuming of identifying areas with landslide susceptibility, it can be used. In addition to natural factors, some human factors such as road construction, play an important role in the occurrence of landslide, which requires avoiding ecosystem change as a disaster risk factor to reduce relative risks. The results of this research can be applicable to the decision making and management of provincial lands as well as urban planning, and they can have a significant role in preventing and reducing the damage caused by landslide.