Determination of sensitive areas to Landslide Occurrence Using Shannon Entropy Model (CaseStudy: Chahardangeh Basin, Mazandaran Province)

IntroductionLandslide is a common geomorphic hazard with considerable economic and ecological consequences. Globally, landslides cause billions of dollars in damage and thousands of deaths and injuries each year. Developing countries mostly suffer from landslide up to 0.5% of their gross national product (GNP) per year. It is hence necessary to study landslide by susceptibility mapping, hazard mapping, and systematic risk assessment tools. These measures enable us to mitigate or to control the damages caused by landslides. Identification and mapping of landslides prone areas are two major steps in the elevation of environmental risks and play a positive role in watershed management. The main purpose of this paper is to determine landslide-prone areas in the Chahardangeh watershed using the Shannon entropy index.
Study AreaThe study area is located in the Southern part of Sari city, between 36°07 ′31″ to 36° 26′31″N, and 3° 11′01″ to 53° 58′57″E, enclosing an area of 1210.80 km2. The mean annual precipitation is approximately 464.62 mm. The Chahardangeh watershed is mainly covered with Paleozoic and the Holocene formations.
Material and MethodsLandslide susceptibility mapping in this research consists of four phases of data preparation, data correlation analysis, landslide susceptibility modeling, and model validation. The first step in landslide susceptibility analysis is to collect information about antecedent landslide incidents. For this purpose, the locations of landslides were identified on Google Earth images and during extensive field surveys. The model was trained using the randomly selected incidents on the landslide inventory map, from which 70% was used for modeling, and the remaining 30% was used for the model validation. The identification and selection of the relevant landslide conditioning factors are two necessary steps, to develop a reliable landslide susceptibility zonation map. The factors controlling instabilities considered in the present study are slope gradient, slope aspect, plan curvature, profile curvature, topographic wetness index (TWI), altitude (obtained from the DEM produced from the topographic map at the scale of 1:50,000), land use (provided by the Forest, Rangeland, and Watershed Management Organization), and NDVI (calculated from NDVI= (IR-R)/(IR) where IR is the infrared band and R is the red band). Lithology map was prepared from two geological maps of Kyasar and Polsefid at the scale of 1:100,000, in paper sheet format. These maps were digitized in ArcGIS and divided into 11 lithology groups. This information was even used to prepare the distance to faults and fault density maps. Maps of distance to streams, distance to roads, drainage density and road density were produced from the DEM.
An important consideration in landslide susceptibility maps is the correlation between independent variables. Tolerance and the variance inflation factor (VIF) are two important indicators of multi-collinearity. The entropy index serves as a measure of the extent of the instability, disorder, imbalance, and uncertainty of a system.
The entropy of a landslide refers to the extent that various factors influence its development. This index determines the weight of each parameter.
The final landslide susceptibility map was developed based on the following equation in ArcGIS.
Ls= ∑_(i=1)^n▒W_f *C_f
Where C_f is the conditioning factors on landslide occurrence
Verification is a fundamental step in the development of susceptibility maps and its verification. Among the 485 landslides identified, 340 (70%) locations were used for landslide susceptibility mapping, and the remaining 145 (30%) locations were used for model validation. In this study, the accuracy of the susceptibility maps was verified by the ROC curve. The range of values of the AUC by 0.5–1 was indicative of a good fit of the model.
Results and DiscussionCorrelation among independent variables is a significant concern in landslide susceptibility mapping. According to the results, the smallest tolerance and the highest variance inflation factor were 0.387 and 2.58, respectively. Therefore, there was not any multicollinearity between the independent factors.
We used the index of entropy (IoE) for landslide susceptibility mapping, This method allows the calculation of the weight for each input variable. The resultant weights obtained for each thematic map from IoE change in the following order: elevation (0.6)> slope gradient (0.52)> plan curvature (0.29)> landuse(0.215)> NDVI(0.199)> profile curvature (0.156)> TWI(0.155) > lithology(0.107)> distance to streams (0.106)> topographic position index (0.1). These factors exert the greatest influence over landslides occurrence in the study area. The susceptibility map, produced by IoE, was divided into four classes including low, moderate, high and very high (24.66, 26.39, 25.63 and 23.31). Approximately 51.05% of the area had high sensitivity to a landslide, which could be explained by the susceptibility of the geological formations to landslide occurrence in the central part of the Elburz Mountains. This part, which covers 65.32% of the total study area, contains conglomerates, sandstones with interbedded marl, sandstone, limestone, and shale intersected with marl. The indiscriminate exploitation of natural resources and the existence of numerous pathways are important factors in rendering this area sensitive to mass movements. Landslide susceptibility map validation using AUC obtained an index of entropy (IoE) of 0.766.
ConclusionLandslide is one of the most dangerous natural disasters worldwide. In this research, 16 factors were evaluated by employing the Shannon Entropy Index. These factors included slope gradient, slope direction, elevation classes, plan and profile curvature, topographic position index, topography wetness index, lithology, distance to faults, fault density, land use, NDVI, distance to streams, drainage density, distance to roads and road density. These factors were then used to plot the landslide sensitivity map using the index of entropy (IoE). For this purpose, 70% of landslide locations were used for modeling and the remaining 30% for model validation. In total, 51.5% of the area was highly susceptible to landslide occurrence. Among the factors affecting landslide, elevation, slope gradient, plan curvature, land use, NDVI, profile curvature, topographic wetness index and lithology had the highest weights and impact. Additionally, model accuracy assessment using the ROC curve with a 76.6% area under the curve, indicated a suitable model performance in the study area. Therefore, the landslide sensitivity map can play a major role in planning and the management of the area to prevent and reduce the damages caused by this hazard.